Econophysics and Financial

Economics
 
 

Econophysics and
Financial Economics
An Emerging Dialogue

Franck Jovanovic
and
Christophe Schinckus

1
 

1
Oxford University Press is a department of the University of Oxford. It furthers
the University’s objective of excellence in research, scholarship, and education
by publishing worldwide. Oxford is a registered trade mark of Oxford University
Press in the UK and certain other countries.

Published in the United States of America by Oxford University Press

198 Madison Avenue, New York, NY 10016, United States of America.

© Oxford University Press 2017

All rights reserved. No part of this publication may be reproduced, stored in

a retrieval system, or transmitted, in any form or by any means, without the
prior permission in writing of Oxford University Press, or as expressly permitted
by law, by license, or under terms agreed with the appropriate reproduction
rights organization. Inquiries concerning reproduction outside the scope of the
above should be sent to the Rights Department, Oxford University Press, at the
address above.

You must not circulate this work in any other form

and you must impose this same condition on any acquirer.

CIP data is on file at the Library of Congress

ISBN 978–​0–​19–​020503–​4

1 3 5 7 9 8 6 4  2
Printed by Edwards Brothers Malloy, United States of America
 

C O N T E N TS

Acknowledgments╇ vii
Introduction╇ ix

1.╇ Foundations of Financial Economics: The Key

Role of the Gaussian Distribution╇ 1

2.╇ Extreme Values in Financial Economics: From Their Observation

to Their Integration into the Gaussian Framework╇ 25

3.╇ New Tools for Extreme-╉Value Analysis:

Statistical Physics Goes beyond Its Borders╇ 49

4.╇ The Disciplinary Position of Econophysics:

New Opportunities for Financial Innovations╇ 78

5.╇ Major Contributions of Econophysics to Financial Economics╇ 106

6.╇ Toward a Common Framework╇ 139

Conclusion: What Kind of Future Lies in Store for Econophysics?╇ 164

Notes╇ 167
References╇ 185
Index╇ 217

v
 
 

AC K N O W L E D G M E N TS

This book owes a lot to discussions that we had with Anna Alexandrova, Marcel
Ausloos, Françoise Balibar, Jean-​Philippe Bouchaud, Gigel Busca, John Davis, Xavier
Gabaix, Serge Galam, Nicolas Gaussel, Yves Gingras, Emmanuel Haven, Philippe Le
Gall, Annick Lesne, Thomas Lux, Elton McGoun, Adrian Pagan, Cyrille Piatecki,
Geoffrey Poitras, Jeroen Romboust, Eugene Stanley, and Richard Topol. We want to
thank them. We also thank Scott Parris. We also want to acknowledge the support of
the CIRST (Montréal, Canada), CEREC (University St-​Louis, Belgium), GRANEM
(Université d’Angers, France), and LÉO (Université d’Orléans, France). We also thank
Annick Desmeules Paré, Élise Filotas, Kangrui Wang, and Steve Jones. Finally, we wish
to acknowledge the financial support of the Social Sciences and Humanities Research
Council of Canada, the Fonds québécois de recherche sur la société et la culture, and
TELUQ (Fonds Institutionnel de Recherche) for this research. We would like to
thank the anonymous referees for their helpful comments.

vii
 
 

INTRODUCTION

Stock market prices exert considerable fascination over the large numbers of people
who scrutinize them daily, hoping to understand the mystery of their fluctuations.
Science was first called in to address this challenging problem 150 years ago. In 1863,
in a pioneering way, Jules Regnault, a French broker’s assistant, tried for the first time
to “tame” the market by creating a mathematical model called the “random walk” based
on the principles of social physics (­chapter 1 in this book; Jovanovic 2016). Since then,
many authors have tried to use scientific models, methods, and tools for the same pur-
pose: to pin down this fluctuating reality. Their investigations have sustained a fruitful
dialogue between physics and finance. They have also fueled a common history. In
the mid-​1990s, in the wake of some of the most recent advances in physics, a new ap-
proach to dealing with financial prices emerged. This approach is called econophysics.
Although the name suggests interdisciplinary research, its approach is in fact multi-
disciplinary. This field was created outside financial economics by statistical physicists
who study economic phenomena, and more specifically financial markets. They use
models, methods, and concepts imported from physics. From a financial point of view,
econophysics can be seen as the application to financial markets of models from par-
ticle physics (a subfield of statistical physics) that mainly use stable Lévy processes and
power laws. This new discipline is original in many points and diverges from previous
works. Although econophysicists concretized the project initiated by Mandelbrot in
the 1960s, who sought to extend statistical physics to finance by modeling stock price
variations through Lévy stable processes, econophysicists took a different path to get
there. Therefore, they provide new perspectives that this book investigates.
Over the past two decades, econophysics has carved out a place in the scientific
analysis of financial markets, providing new theoretical models, methods, and results.
The framework that econophysicists have developed describes the evolution of finan-
cial markets in a way very different from that used by the current standard financial
models. Today, although less visible than financial economics, econophysics influences
financial markets and practices. Many “quants” (quantitativists) trained in statistical
physics have carried their tools and methodology into the financial world. According
to several trading-​room managers and directors, econophysicists’ phenomenological
approach has modified the practices and methods of analyzing financial data. Hitherto,
these practical changes have concerned certain domains of finance: hedging, portfolio
management, financial crash predictions, and software dedicated to finance. In the
coming decades, however, econophysics could contribute to profound changes in the
entire financial industry. Performance measures, risk management, and all financial

ix
 

x Introduction

decisions are likely to be affected by the framework econophysicists have developed.

In this context, an investigation of the interface between econophysics and financial
economics is required and timely.
Paradoxically, although econophysics has already contributed to change practices
on financial markets and has provided numerous models, dialogue between econo-
physicists and financial economists is almost nonexistent. On the one hand, econo-
physics faces strong resistance from financial economists (­chapter 4), while on the
other hand, econophysicists largely ignore financial economics (­chapters 4 and 5).
Moreover, the potential contributions of econophysics to finance (theory and prac-
tices) are far from clear. This book is intended to give readers interested in econophys-
ics an overview of the situation by supplying a comparative analysis of the two fields in
a clear, homogenous framework.
The lack of dialogue between the two scientific communities is manifested in sev-
eral ways. With some rare exceptions, econophysics publications criticize (sometimes
very forcefully) the theoretical framework of financial economics, while frequently
ignoring its contributions (­chapters 5 and 6). In addition, econophysicists are parsi-
monious with their explanations regarding their contribution in relation to existing
works in financial economics or to existing practices in trading rooms. In the same
vein, econophysicists criticize the hypothetico-​deductive method used by financial
economists, starting from postulates (i.e., a hypothesis accepted as true without being
demonstrated) rather than from empirical phenomena (­chapter 4). However, econo-
physicists seem to overlook the fact that they themselves implicitly apply a quite sim-
ilar approach: the great majority of them develop mathematical models based on the
postulate that the empirical phenomenon studied is ruled by a power-​law distribution
(­chapter  3). Many econophysicists suggest a simple importing of statistical physics
concepts into financial economics, ignoring the scientific constraints specific to each
of the two disciplines that make this impossible (­chapters 1–​4). Econophysicists are
driven by a more phenomenological method where visual tests are used to identify the
probability distribution that fits with observations. However, most econophysicists
are unaware that such visual tests are considered unscientific in financial economics
(­chapters 1, 4, and 5). In addition, econophysics literature largely remains silent on
the crucial issues of the validation of the power-​law distribution by existing tests.
Similarly, financial economists have developed models (autoregressive conditional
heteroskedasticity [ARCH] -​type models, jump models, etc.) by adopting a phe-
nomenological approach similar to that propounded by econophysicists (­chapters 2,
4, and 5). However, although these models are criticized in econophysics literature,
econophysicists have overlooked the fact that these models are rooted in scientific
constraints inherent in financial economics (­chapters 4 and 5).
This lack of dialogue and its consequences can be traced to three main causes.
The first is reciprocal ignorance, strengthened by some differences in disciplinary
language. For instance, while financial economists use the term “Lévy processes” to
define (nonstable) jump or pure-​jump models, econophysicists use the same term to
mean “stable Lévy processes” (­chapter 2). Consequently, econophysicists often claim
that they offer a new perspective on finance, whereas financial economists consider
 

xi  Introduction

that this approach is an old issue in finance. Many examples of this situation can be
observed in the literature, with each community failing to venture beyond its own per-
spective. A key point is that the vast majority of econophysics publications are written
by econophysicists for physicists, with the result that the field is not easily accessible
to other scholars or readers. This context highlights the necessity to clarify the differ-
ences and similarities between the two disciplines.
The second cause is rooted in the way each discipline deals with its own scien-
tific knowledge. Contrary to what one might think, how science is done depends on
disciplinary processes. Consequently, the ways of producing knowledge are different
in econophysics and financial economics (­chapter 4): econophysicists and financial
economists do not build their models in the same way; they do not test their models
and hypotheses with the same procedures; they do not face the same scientific con-
straints even though they use the same vocabulary (in a different manner), and so
on. The situation is simply due to the fact that econophysics remains in the shadow
of physics and, consequently, outside of financial economics. Of course there are
advantages and disadvantages in such an institutional situation (i.e., being outside
of financial economics) in terms of scientific innovations. A methodological study
is proposed in this book to clarify the dissimilarities between econophysics and fi-
nancial economics in terms of modeling. Our analysis also highlights some common
features regarding modeling (­chapter 5) by stressing that the scientific criteria any
work must respect in order to be accepted as scientific are very different in these two
disciplines. The gaps in the way of doing science make reading literature from the
other discipline difficult, even for a trained scholar. These gaps underline the needs
for clear explanations of the main concepts and tools used in econophysics and how
they could be used on financial markets.
The third cause is the lack of a framework that could allow comparisons between
results provided by models developed in the two disciplines. For a long time, there
have been no formal statistical tests for validating (or invalidating) the occurrence of
a power law. In finance, satisfactory statistical tools and methods for testing power
laws do not yet exist (­chapter 5). Although econophysics can potentially be useful
in trading rooms and although some recent developments propose interesting solu-
tions to existing issues in financial economics (­chapter 5), importing econophysics
into finance is still difficult. The major reason goes back to the fact that econophysi-
cists mainly use visual techniques for testing the existence of a power law, while finan-
cial economists use classical statistical tests associated with the Gaussian framework.
This relative absence of statistical (analytical) tests dedicated to power laws in finance
makes any comparison between the models of econophysics and those of financial
economics complex. Moreover, the lack of a homogeneous framework creates difficul-
ties related to the criteria for choosing one model rather than another. These issues
highlight the need for the development of a common framework between these two
fields. Because econophysics literature proposes a large variety of models, the first step
is to identify a generic model unifying key econophysics models. In this perspective,
this book proposes a generalized model characterizing the way econophysicists statis-
tically describe the evolution of financial data. Thereafter, the minimal condition for
 

xii Introduction

a theoretical integration in the financial mainstream is defined (­chapter 6). The iden-
tification of such a unifying model will pave the way for its potential implementation
in financial economics.
Despite this difficult dialogue, a number of collaborations between financial econ-
omists and econophysicists have occurred, aimed at increasing exchanges between
the two communities.1 These collaborations have provided useful contributions.
However, they also underline the necessity for a better understanding of the discipli-
nary constraints specific to both fields in order to ease a fruitful association. For in-
stance, as the physicist Dietrich Stauffer explained, “Once we [the economist Thomas
Lux and Stauffer] discussed whether to do a Grassberger-​Procaccia analysis of some
financial data … I realized that in this case he, the economist, would have to explain
to me, the physicist, how to apply this physics method” (Stauffer 2004, 3). In the same
vein, some practitioners are aware of the constraints and perspectives specific to each
discipline. The academic and quantitative analyst Emanuel Derman (2001, 2009) is
a notable example of this trend. He has pointed out differences in the role of models
within each discipline: while physicists implement causal (drawing causal inference)
or phenomenological (pragmatic analogies) models in their description of the phys-
ical world, financial economists use interpretative models to “transform intuitive
linear quantities into non-​linear stable values” (Derman 2009, 30). These consider-
ations imply going beyond the comfort zone defined by the usual scientific frontiers
within which many authors stay.
This book seeks to make a contribution toward increasing dialogue between the
two disciplines. It will explore what econophysics is and who econophysicists are by
clarifying the position of econophysics in the development of financial economics.
This is a challenging issue. First, there is an extremely wide variety of work aiming to
apply physics to finance. However, some of this work remains outside the scope of
econophysics. In addition, as the econophysicist Marcel Ausloos (2013, 109) claims,
investigations are heading in too many directions, which does not serve the intended
research goal. In this fragmented context, some authors have reviewed existing econo-
physics works by distinguishing between those devoted to “empirical facts” and those
dealing with agent-​based modeling (Chakraborti et al. 2011a, 2011b). Other authors
have proposed a categorization based on methodological aspects by differentiating be-
tween statistical tools and algorithmic tools (Schinckus 2012), while still others have
kept to a classical micro/​macro opposition (Ausloos 2013). To clarify the approach
followed in this book, it is worth mentioning the historical importance of the Santa
Fe Institute in the creation of econophysics. This institution introduced two compu-
tational ways of describing complex systems that are relevant for econophysics: (1)
the emergence of macro statistical regularity characterizing the evolution of systems;
(2) the observation of a spontaneous order emerging from microinteractions be-
tween components of systems (Schinckus 2017). Methodologically speaking, stud-
ies focusing on the emergence of macro regularities consider the description of the
system as a whole as the target of the analysis, while works dealing with an emerging
spontaneous order seek to reproduce (algorithmically) microinteractions leading the
system to a specific configuration. These two approaches have led to a methodological
 

xiii  Introduction

scission in the literature between statistical econophysics and agent-​based econophys-

ics (Schinckus 2012). While econophysics was originally defined as the extension of
statistical physics to financial economics, agent-​based modeling has recently been as-
sociated with econophysics. This book mainly focuses on the original way of defining
econophysics by considering the applications of statistical physics to financial markets.
Dealing with econophysics raises another challenging issue. The vast majority of
existing books on econophysics are written by physicists who discuss the field from
their own perspective. Financial economists, for their part, do not usually clarify their
implicit assumptions, which does not facilitate collaboration with outsider scientists.
This is the first book on econophysics to be written solely by financial economists. It
does not aspire to summarize the state of the art on econophysics, nor to provide an
exhaustive presentation of econophysics models or topics investigated; many books
already exist.2 Rather, its aim is to analyze the crucial issues at the interface of financial
economics and econophysics that are generally ignored or not investigated by scholars
involved in either field. It clarifies the scientific foundations and criteria used in each
discipline, and makes the first extensive analytic comparison between models and re-
sults from both fields. It also provides keys for understanding the resistance each dis-
cipline has to face by analyzing what has to be done to overcome these resistances. In
this perspective, this book sets out to pave the way for better and useful collaborations
between the two fields. In contrast with existing literature dedicated to econophysics,
the approach developed in this book enables us to initiate a framework and models
common to financial economics and econophysics.
This book has two singular characteristics.
The first is that it deals with the scientific foundations of econophysics and financial
economics by analyzing their development. We are interested not only in the presenta-
tion of these foundational principles but also in the study of the implicit scientific and
methodological criteria, which are generally not studied by authors. After explaining
the contextual factors that contributed to the advent of econophysics, we discuss the
key concepts used by econophysicists and how they have contributed to a new way of
using power-​law distributions, both in physics and in other sciences. As we demon-
strate, comprehension of these foundations is crucial to an understanding of the current
gap between the two areas of knowledge and, consequently, to breaking down the
barriers that separate them conceptually.
The second particular feature of this book is that it takes a very specific perspec-
tive. Unlike other publications dedicated to econophysics, it is written by financial
economists and situates econophysics in the evolution of modern financial theory.
Consequently, it provides an analysis in which econophysics makes sense for financial
economists by using the vocabulary and the viewpoint of financial economics. Such a
perspective is very helpful for identifying and understanding the major advantages and
drawbacks of econophysics from the perspective of financial economics. In this way,
the reasons why financial economists have been unable to use econophysics models
in their field until now can also be identified. Adopting the perspective of financial ec-
onomics also makes it possible to develop a common framework enabling synergies
and potential collaborations between financial economists and econophysicists to be
 

xiv Introduction

created. This book thus offers conceptual tools to surmount the disciplinary barriers
that currently limit the dialogue between these two disciplines. In accordance with
this purpose, the book gives econophysicists an opportunity to have a specific discipli-
nary (financial) perspective on their emerging field.
The book is divided into three parts.
The first part (­chapters 1 and 2) focuses on financial economics. It highlights the
scientific constraints this discipline has to face in its study of financial markets. This
part investigates a series of key issues often addressed by econophysicists (but also
by scholars working outside financial economics): why financial economists cannot
easily drop the efficient-​market hypothesis; why they could not follow Mandelbrot’s
program; why they consider visual tests unscientific; how they deal with extreme
values; and, finally, why the mathematics used in econophysics creates difficulties in
financial economics.
The second part (­chapters 3 and 4) focuses on econophysics. It clarifies econo-
physics’ position in the development of financial economics. This part investigates
econophysicists’ scientific criteria, which are different from those of financial econo-
mists, implying that the scientific benchmark for acceptance differs in the two com-
munities. We explain why econophysicists have to deal with power laws and not with
other distributions; how they describe the problem of infinite variance; how they
model financial markets in comparison with the way financial economists do; why
and how they can introduce innovations in finance; and, finally, why econophysics and
financial economics can be looked on as similar.
The third part (­chapters  5 and 6)  investigates the potential development of a
common framework between econophysics and financial economics. This part aims at
clarifying some current issues about such a program: what the current uses of econo-
physics in trading rooms are; what recent developments in econophysics allow pos-
sible contributions to financial economics; how the lack of statistical tests for power
laws can be solved; what generative models can explain the appearance of power laws
in financial data; and, finally, how a common framework transcending the two fields by
integrating the best of the two disciplines could be created.
 

1
F O U N DAT I O N S O F   F I N A N C I A L EC O N O M I C S
T H E K E Y R O L E O F T H E G AU S S I A N D I ST R I B U T I O N

This chapter scrutinizes the theoretical foundations of financial economics. Financial

economists consider that stock market variations1 are ruled by stochastic processes
(i.e., a mathematical formalism constituted by a sequence of random variables). The
random-​walk model is the simplest one. While the random nature of stock market vari-
ations is not called into question in the work of econophysicists, the use of the Gaussian
distribution to characterize such variations is firmly rejected. The strict Gaussian distri-
bution does not allow financial models to reproduce the substantial variations in prices
or returns that are observed on the financial markets. A telling illustration is the occur-
rence of financial crashes, which are more and more frequent. One can mention, for
instance, August 2015 with the Greek stock market, June 2015 with the Chinese stock
market, August 2011 with world stock markets, May 2010 with the Dow Jones index,
and so on. Financial economists’ insistence on maintaining the Gaussian-​distribution
hypothesis meets with incomprehension among econophysicists. This insistence might
appear all the more surprising because financial economists themselves have long been
complaining about the limitations of the Gaussian distribution in the face of empirical
data. Why, in spite of this drawback, do financial economists continue to make such
broad use of the normal distribution? What are the reasons for this hypothesis’s po-
sition at the core of financial economics? Is it fundamental for financial economists?
What benefits does it give them? What would dropping it entail?
The aim of this chapter is to answer these questions and understand the place of
the normal distribution in financial economics. First of all, the chapter will investigate
the historical roots of this distribution, which played a key role in the construction of
financial economics. Indeed, the Gaussian distribution enabled this field to become a
recognized scientific discipline. Moreover, this distribution is intrinsically embedded
in the statistical framework used by financial economists. The chapter will also clarify
the links between the Gaussian distribution and the efficient-​market hypothesis.
Although the latter is nowadays well established in finance, its links with stochastic
processes have generated many confusions and misunderstandings among financial
economists and consequently among econophysicists. Our analysis will also show that
the choice of a statistical distribution, including the Gaussian one, cannot be reduced
to empirical considerations. As in any scientific discipline, axioms and postulates2 play
an important role in combination with scientific and methodological constraints with
which successive researchers have been faced.

1
 

2╇ Econophysics and Financial Economics

1.1.╇ FIRST INVESTIGATIONS AND EARLY

ROOTS OF FINANCIAL ECONOMICS: THE KEY
ROLE OF THE GAUSSIAN DISTRIBUTION
Financial economics’ construction as a scientific discipline has been a long process
spread over a number of stages. This first part of our survey looks back at the origins
of financial tools and concepts that were combined in the 1960s to create financial
economics. These first works of modern finance will show the close association be-
tween the development of financial ideas, probabilities theory, physics, statistics, and
economics. This perspective will also provide reading keys in order to understand the
scientific criteria on which financial economics was created. Two elements will get our
attention: the Gaussian distribution and the use of stochastic processes for studying
stock market variations. This analysis will clarify the major theoretical and methodo-
logical foundations of financial economics and identify justifications for the use of the
normal law and the random character of stock market variations produced by early
theoretical works.

1.1.1.╇ The First Works of Modern Finance

1863: Jules Regnault and the First Stochastic Modeling
of Stock Market Variations
Use of a random-╉walk model to represent stock market variations was first proposed
in 1863 by a French broker’s assistant (employé d’agent de change), Jules Regnault.3
His only published work, Calculation of Chances and Philosophy of the Stock Exchange
(Calcul des chances et philosophie de la bourse), represents the first known theoretical
work whose methodology and theoretical content relates to financial economics.
Regnault’s objective was to determine the laws of nature that govern stock market
fluctuations and that statistical calculations could bring within reach.
Regnault produced his work at a time when the Paris stock market was a leading
place for derivative trading (Weber 2009); it also played a growing role in the whole
economy (Arbulu 1998; Hautcœur and Riva 2012; Gallais-╉Hamonno 2007). This
period was also a time when new ideas were introduced into the social sciences. As
we will detail in Â�chapter 4, such a context also contributed to the emergence of finan-
cial economics and of econophysics. The changes on the Paris stock market gave rise
to lively debates on the usefulness of financial markets and whether they should be
restricted (Preda 2001, 2004; Jovanovic 2002, 2006b). Regnault published his work
partly in response to these debates, using a symmetric random-╉walk model to dem-
onstrate that the stock market was both fair and equitable, and that consequently
its development was acceptable ( Jovanovic 2006a; Jovanovic and Le Gall 2001). In
conducting his demonstration, Regnault took inspiration from Quételet’s work on
the normal distribution ( Jovanovic 2001). Adolphe Quételet was a Belgian math-
ematician and statistician well known as the “father of social physics.”4 He shared
with the scientists of his time the idea that the average was synonymous with per-
fection and morality (Porter 1986) and that the normal distribution,5 also known
 

3  Foundations of Financial Economics

as “the law of errors,” made it possible to determine errors of observation (i.e., dis-
crepancies) in relation to the true value of the observed object, represented by the
average. Quételet, like Regnault, applied the Gaussian distribution, which was con-
sidered as one of the most important scientific results founded on the central-​limit
theorem (which explains the occurrence of the normal distribution in nature),6 to
social phenomena.
Precisely, the normal law allowed Regnault to determine the true value of a se-
curity that, according to the “law of errors,” is the security’s long-​term mean value.
He contrasted this long-​term determination with a short-​term random walk that was
mainly due to the shortsightedness of agents. In Regnault’s view, short-​term valua-
tions of a security are subjective and subject to error and are therefore distributed
in accordance with the normal law. As a result, short-​term valuations fall into two
groups spread equally about a security’s value:  the “upward” and the “downward.”
In the absence of new information, transactions cause the price to gravitate around
this value, leading Regnault to view short-​term speculation as a “toss of a coin” game
(1863, 34).
In a particularly innovative manner, Regnault likened stock price variations to a
random walk, although that term was never employed.7 On account of the normal
distribution of short-​term valuations, the price had an equal probability of lying
above or below the mean value. If these two probabilities were different, Regnault
pointed out, actors could resort to arbitrage8 by choosing to systematically follow
the movement having the highest probability (Regnault 1863, 41). Similarly, as in
the toss of a coin, rises and falls of stock market prices are independent of each other.
Consequently, since neither a rise nor a fall can anticipate the direction of future
variations (Regnault 1863, 38), Regnault explained, there could be no hope of short-​
term gain. Lastly, he added, a security’s current price reflects all available public infor-
mation on which actors base their valuation of it (Regnault 1863, 29–​30). Therefore,
with Regnault, we have a perfect representation of stock market variations using a
random-​walk model.9
Another important contribution from Regnault is that he tested his hypothesis of
the random nature of short-​term stock market variations by examining a mathemat-
ical property of this model, namely that deviations increase proportionately with the
square root of time. Regnault validated this property empirically using the monthly
prices from the French 3 percent bond, which was the main bond issued by the gov-
ernment and also the main security listed on the Paris Stock Exchange. It is worth
mentioning that at this time quoted prices and transactions on the official market of
Paris Stock Exchange were systematically recorded,10 allowing statistical tests. Such
an obligation did not exist in other countries. In all probability the inspiration for this
test was once again the work of Quételet, who had established the law on the increase
of deviations (1848, 43 and 48). Although the way Regnault tested his model was
different from the econometric tests used today ( Jovanovic 2016; Jovanovic and Le
Gall 2001; Le Gall 2006), the empirical determination of this law of the square root
of time thus constituted the first result to support the random nature of stock market
variations.
 

4  Econophysics and Financial Economics

It is worth mentioning that Regnault’s choice of the Gaussian distribution was

based on three factors: (1) empirical data; (2) moral considerations, because this law
allowed him to demonstrate that speculation necessarily led to ruin, whereas invest-
ments that fostered a country’s development led to the earning of money; and (3) the
importance at the time of the “law of errors” in the development of social sciences,
which was due to the work of Quételet based on the central-​limit theorem.
In conclusion, contemporary intuitions and mainstream ideas about the random
character of stock market prices and returns informed Regnault’s book.11 Its pioneering
aspect is also borne out with respect to portfolio analysis, since the diversification
strategy and the concept of correlation were already in use in the United Kingdom and
in France at the end of the nineteenth century (Edlinger and Parent 2014; Rutterford
and Sotiropoulos 2015). Although Regnault introduced foundational intuitions about
the description of financial data, his idea of a random walk had to wait until Louis
Bachelier’s thesis in 1900 to be formalized.

1900: Louis Bachelier and the First Mathematical Formulation

of Brownian Motion
The second crucial actor in the history of modern financial ideas is the French mathe-
matician Louis Bachelier. Although the whole of Bachelier’s doctoral thesis is based on
stock markets and options pricing, we must remember that this author defended his
thesis in a field called at this time mathematical physics—​that is, the field that applies
mathematics to problems in physics. Although his research program dealt with math-
ematics alone—​his aim was to construct a general, unified theory of the calculation of
probabilities exclusively on the basis of continuous time12—​the genesis of Bachelier’s
program of mathematical research most certainly lay in his interest in financial markets
(Taqqu 2001, 4–​5; Bachelier 1912, 293). It seems clear that stock markets fascinated
him, and his endeavor to understand them was what stimulated him to develop an ex-
tension of probability theory, an extension that ultimately turned out to have other
applications.
His first publication, Théorie de la spéculation, which was also his doctoral thesis,
introduced continuous-​time probabilities by demonstrating the equivalence be-
tween the results obtained in discrete time and in continuous time (an application
of the central-​limit theorem). Bachelier achieved this equivalence by developing two
proofs: one using continuous-​time probabilities, the other with discrete-​time prob-
abilities completed by a limit approximation using Stirling’s formula. In the second
part of his thesis he proved the usefulness of this equivalence through empirical inves-
tigations of stock market prices, which provided a large amount of data.
Bachelier applied this principle of a double demonstration to the law of stock
market price variation, formulating for the first time the so-​ called Chapman-​
Kolmogorov-​Smoluchowski equation: 13

+∞
p( z ,t ) = ∫ p( x ,t1 ) p( z − x ,t 2 )dx , with t = t1 + t 2 ,  (1.1)
−∞
 

5╇ Foundations of Financial Economics

where Pz ,t +t designates the probability that price z will be quoted at time t1 + t2, know-
1 2

ing that price x was quoted at time t1. Bachelier then established the probability of
transition as σWt—╉where Wt is a Brownian movement:14
x2
1 −
p( x ,t ) = (1.2)
2
e 4 πk t ,
2 πk t 

where t represents time, x a price of the security, and k a constant. Bachelier next ap-
plied his double-╉demonstration principle to the “two problems of the theory of spec-
ulation” that he proposed to resolve:  the first establishes the probability of a given
price being reached or exceeded at a given time—╉that is, the probability of a “prime,”
which was an asset similar to a European option,15 being exercised, while the second
seeks the probability of a given price being reached or exceeded before a given time
(Bachelier 1900, 81)—╉which amounts to determining the probability of an American
option being exercised.16
His 1901 article, “Théorie mathématique du jeu,” enabled him to generalize the
first results contained in his thesis by moving systematically from discrete time to
continuous time and by adopting what he called a “hyperasymptotic” point of view.
The “hyperasymptotic” was one of Bachelier’s central concerns and one of his major
contributions. “Whereas the asymptotic approach of Laplace deals with the Gaussian
limit, Bachelier’s hyperasymptotic approach deals with trajectories,” as Davis and
Etheridge point out (2006, 84). Bachelier was the first to apply the trajectories of
Brownian motion, making a break from the past and anticipating the mathematical
finance developed since the 1960s (Taqqu 2001). Bachelier was thus able to prove the
results in continuous time of a number of problems in the theory of gambling that the
calculation of probabilities had dealt with since its origins.
For Bachelier, as for Regnault, the choice of the normal distribution was not only dic-
tated by empirical data but mainly by mathematical considerations. Bachelier’s interest
was in the mathematical properties of the normal law (particularly the central-╉limit the-
orem) for the purpose of demonstrating the equivalence of results obtained using math-
ematics in continuous time and those obtained using mathematics in discrete time.

Other Endeavors: A Similar Use of the Gaussian Distribution

Bachelier was not the only person working successfully on premium/╉option pricing
at the beginning of the twentieth century. The Italian mathematician Vinzenz Bronzin
published a book on the theory of premium contracts in 1908. Bronzin was a professor
of commercial and political arithmetic at the Imperial Regia Accademia di Commercio
e Nautica in Trieste and published several books (Hafner and Zimmermann 2009,
chap.  1). In his 1908 book, Bronzin analyzed premiums/╉options and developed a
theory for pricing them. Like Regnault and Bachelier, Bronzin assumed the random
character of market fluctuations and zero expected profit. Bronzin did no stochastic
modeling and was uninterested in stochastic processes (Hafner and Zimmermann
2009, 244), but he showed that “applying Bernoulli’s theorem to market fluctuations
 

6  Econophysics and Financial Economics

leads to the same result that we had arrived at when supposing the application of the
law of error [i.e., the normal law]” (Bronzin 1908, 195). In other words, Bronzin used
the normal law in the same way as Regnault, since it allowed him to determine the
probability of price fluctuations (Bronzin 1908 in Hafner and Zimmermann 2009,
188). In all these pioneering works, it appears that the Gaussian distribution and the
hypothesis of random character of stock market variations were closely linked with
the scientific tools available at the time (and particularly the central-​limit theorem).
The works of Bachelier, Regnault, and Bronzin have continued to be used and
taught since their publication (Hafner and Zimmermann 2009; Jovanovic 2004,
2012, 2016). However, despite these writers’ desire to create a “science of the stock ex-
change,” no research movement emerged to explore the random nature of variations.
One of the reasons for this was the opposition of economists to the mathematization
of their discipline (Breton 1991; Ménard 1987). Another reason lay in the insufficient
development of what is called modern probability theory, which played a key role in
the creation of financial economics in the 1960s (we will detail this point later in this
chapter).
Development of continuous-​time probability theory did not truly begin until
1931, before which the discipline was not fully recognized by the scientific community
(Von Plato 1994). However, a number of publications aimed at renewing this theory
emerged between 1900 and 1930.17 During this period, several authors were working
on random variables and on the generalization of the central-​limit theorem, including
Sergei Natanovich Bernstein, Alexandre Liapounov, Georges Polya, Andrei Markov,18
and Paul Lévy. Louis Bachelier (Bachelier 1900, 1901, 1912), Albert Einstein (1905),
Marian von Smoluchowski (1906),19 and Norbert Wiener (1923)20 were the first to
propose continuous-​time results, on Brownian motion in particular. However, up until
the 1920s, during which decade “a new and powerful international progression of the
mathematical theory of probabilities” emerged (due above all to the work of Russian
mathematicians such as Kolmogorov, Khintchine, Markov, and Bernstein), this work
remained known and accessible only to a few specialists (Cramer 1983, 8). For ex-
ample, the work of Wiener (1923) was difficult to read before the work of Kolmogorov
published during the 1930s, while Bachelier’s publications (1901, 1900, 1912) were
hardly readable, as witnessed by the error that Paul Lévy (one of the rare mathemati-
cians working in this field) believed he had detected.21 The 1920s were a period of very
intensive research into probability theory—​and into continuous-​time probabilities in
particular—​that paved the way for the construction of modern probability theory.
Modern probability theory was properly created in the 1930s, in particular
through the work of Kolmogorov, who proposed its main founding concepts: he in-
troduced the concept of probability space, defined the concept of the random vari-
able as we know it today, and also dealt with conditional expectation in a totally new
manner (Cramer 1983, 9; Shafer and Vovk 2001, 39). Since his axiom system is the
basis of the current paradigm of the discipline, Kolmogorov can be seen as the father
of this branch of mathematics. Kolmogorov built on Bachelier’s work, which he con-
sidered the first study of stochastic processes in continuous time, and he generalized
on it in his 1931 article.22 From these beginnings in the 1930s, modern probability
 

7╇ Foundations of Financial Economics

theory became increasingly influential, although it was only after World War II that
Kolmogorov’s axioms became the dominant paradigm in the discipline (Shafer and
Vovk 2005, 54–╉55).
It was also after World War II that the American probability school was born.23 It
was led by Joseph Doob and William Feller, who had a major influence on the con-
struction of modern probability theory, particularly through their two main books,
published in the early 1950s (Doob 1953; Feller 1957), which proved, on the basis of
the framework laid down by Kolmogorov, all results obtained prior to the 1950s, ena-
bling their acceptance and integration into the discipline’s theoretical corpus (Meyer
2009; Shafer and Vovk 2005, 60).
In other words, modern probability theory was not accessible for analyzing stock
markets and finance until the 1950s. Consequently, it would have been exceedingly
difficult to create a research movement before that time, and this limitation made the
possibility of a new discipline such as financial economics prior to the 1960s unlikely.
However, with the emergence of econometrics in the United States in the 1930s, an
active research movement into the random nature of stock market variations and their
distribution did emerge, paving the way for financial econometrics.

1.1.2.╇ The Emergence of Financial Econometrics in the 1930s

The stimulus to conduct research on the hypothesis of the random nature of stock
market variations arose in the United States in the 1930s. Alfred Cowles, a victim
of the 1929 stock market crash, questioned the predictive abilities of the port-
folio management firms who gave advice to investors. This led him into contact
with the newly founded Econometric Society—╉an “International Society for the
Advancement of Economic Theory in its Relation with Statistics and Mathematics.”
In 1932, he offered the society financial support in exchange for statistical treatment
of his problems in predicting stock market variations and the business cycle. On
September 9 of the same year, he set up an innovative research group: the Cowles
Commission.24
Research into application of the random-╉walk model to stock market variations
was begun by two authors connected with this institution, Cowles himself (1933,
1944) and Holbrook Working (1934, 1949).25 The failure to predict the 1929 crisis
led them to entertain the possibility that stock market variations were unpredictable.
Defending this perspective led these researchers to oppose the chartist theories, very
influential at the time, that claimed to be able to anticipate stock market variations
based on the history of stock market prices. Cowles and Working undertook to show
that these theories, which had not foreseen the 1929 crisis, had no predictive power. It
was through this postulate of unpredictability that the random nature of stock market
variations was reintroduced into financial theory, since it allowed this unpredictability
to be modeled. Unpredictability became a key element of the first theoretical works in
finance because they were associated with econometrics.
The first empirical tests were based on the normal distribution, which was still con-
sidered the natural attractor for the sum of a set of random variables. For example,
 

8  Econophysics and Financial Economics

Working (1934) started from the notion that the movements of price series “are
largely random and unpredictable” (1934, 12). He constructed a series of random re-
turns with random drawings generated by a Tippett table26 based on the normal distri-
bution. He assumed a Gaussian distribution because of “the superior generality of the
‘normal’ frequency distribution” (1934, 16). This position was common at this time
for authors who studied price fluctuations (Cover 1937; Bowley 1933): the normal
distribution was viewed as the starting point of any work in econometrics. This pre-
sumption was reinforced by the fact that all existing statistical tests were based on
the Gaussian framework. Working compared his random series graphically with the
real series, and noted that the artificially created price series took the same graphic
shapes as the real series. His methodology was similar to that used by Slutsky ([1927]
1937)  in his econometric work, which aimed to demonstrate that business cycles
could be caused by an accumulation of random events (Armatte 1991; Hacking 1990;
Le Gall 1994; Morgan 1990).27 Slutsky proposed a graphical comparison between
a random series and an observed price series. Slutsky and Working considered that,
if price variations were random, they must be distributed according to the Gaussian
distribution.
The second researcher affiliated with the Cowles Commission, Cowles himself,
followed the same path: he tested the random character of returns (price variations),
and he postulated that these price variations were ruled by the normal distribution.
Cowles (1933), for his part, attempted to determine whether stock market profes-
sionals (financial services and chartists) were able to predict stock market variations,
and thus whether they could realize better performance than the market itself or than
random management. He compared the evolution of the market with the perfor-
mances of fictional portfolios based on the recommendations of 16 professionals.
He found that the average annual return of these portfolios was appreciably inferior
to the average market performance; and that the best performance could have been
attained by buying and selling stocks randomly. It is worth mentioning that the desire
to prove the unpredictability of stock market variations led authors occasionally to
make contestable interpretations in support of their thesis ( Jovanovic 2009b).28 In
addition, Cowles and Jones (1937), whose article sought to demonstrate that stock
price variations are random, compared the distribution of price variations with a
normal distribution because, for these authors, the normal distribution was the
means of characterizing chance in finance.29 Like Working, Cowles and Jones sought
to demonstrate the independence of stock price variations and made no assumption
about distribution.
The work of Cowles and Working was followed in 1953 by a statistical study by
the English statistician Maurice Kendall. Although his work used more technical sta-
tistical tools, reflecting the evolution of econometrics, the Gaussian distribution was
still viewed as the statistical framework describing the random character of time series,
and no other distribution was considered when using econometrics or statistical
tests. Kendall in turn considered the possibility of predicting financial-​market prices.
Although he found weak autocorrelations in series and weak delayed correlations
between series, Kendall concluded that “a kind of economic Brownian motion” was
 

9  Foundations of Financial Economics

operating and commented on the central-​limit tendency in his data. In addition, he

considered that “unless individual stocks behave differently from the average of similar
stocks, there is no hope of being able to predict movements on the exchange for a week
ahead without extraneous information” (1953, 11). Kendall’s conclusions remained
cautious, however. He pointed out at least one notable exception to the random nature
of stock market variations and warned that “it is … difficult to distinguish by statis-
tical methods between a genuine wandering series and one wherein the systematic
element is weak” (1953, 11).
These new research studies had a strong applied, empirical, and practical dimen-
sion: they favored an econometric approach without theoretical explanation, aimed
at validating the postulate that stock market variations were unpredictable. From the
late 1950s on, the absence of theoretical explanation and the weakness of the results
were strongly criticized by two of the main proponents of the random nature of stock
market prices and returns: Working (1956, 1961, 1958), and Harry V. Roberts (1959),
who was professor of statistics at the Graduate School of Business at the University
of Chicago.30 Each pointed out the limitations of the lack of theoretical explanation
and the way to move ahead. Roberts (1959, 15)  noted that the independence of
stock market variations had not yet been established (1959, 13). Working also high-
lighted the absence of any verification of the randomness of stock market variations.
In his view, it was not possible to reject with certainty the chartist (or technical) anal-
ysis, which relied on figures or graphics to predict variations in stock market prices.
“Although I may seem to have implied that these ‘technical formations’ in actual prices
are illusory,” Working said, “they have not been proved so” (1956, 1436).
These early American authors’ choice of the randomness of stock market varia-
tions derives, then, from their desire to support their postulate that variations were
unpredictable. However, although they reintroduced this hypothesis independently
of the work of Bachelier, Regnault, and Bronzin and without any “a priori assump-
tions” about the distribution of stock market prices,31 their works were embedded in
the Gaussian framework. The latter was, at the time, viewed as the necessary scientific
tool for describing random time series (­chapter 2 will also detail this point). At the end
of the 1950s, Working and Roberts called for research to continue, initiating the break
in the 1960s that led to the creation of financial economics.

1.2.  THE ROLE OF THE GAUSSIAN FRAMEWORK

IN THE CREATION OF FINANCIAL ECONOMICS
AS A SCIENTIFIC DISCIPLINE
Financial economics owes its institutional birth to three elements: access to the tools
of modern probability theory; a new scientific community that extended the analysis
framework of economics to finance; and the creation of new empirical data.32 This birth
is inseparable from work on the modeling of stock market variations using stochastic
processes and on the efficient-​market hypothesis. It took place during the 1960s at a
time when American university circles were taking a growing interest in American fi-
nancial markets (Poitras 2009) and when new tools became available. An analysis of
 

10  Econophysics and Financial Economics

this context provides an understanding of some of the main theoretical and method-
ological foundations of contemporary financial economics. We will detail this point in
the next section when we study how the hard core of this discipline was constituted.

1.2.1.  On the Accessibility of the Tools

of Modern Probability Theory
As mentioned earlier, in the early 1950s Doob and Feller published two books that
had a major influence on modern probability theory (Doob 1953; Feller 1957). These
works led to the creation of a stable corpus that became accessible to nonspecialists.
Since then, the models and results of modern probability theory have been used in
the study of financial markets in a more systematic manner, in particular by scholars
trained in economics. The most notable contributions were to transform old results,
expressed in a literary language, into terms used in modern probability theory.
The first step in this development was the dissemination of mathematical tools
enabling the properties of random variables to be used and uncertainty reasoning to
be developed. The first two writers to use tools that came out of modern probability
theory to study financial markets were Harry Markowitz and A. D. Roy. In 1952 each
published an article on the theory of portfolio choice theory.33 Both used mathemat-
ical properties of random variables to build their model, and more specifically, the
fact that the expected value of a weighted sum is the weighted sum of the expected
values, while the variance of a weighted sum is not the weighted sum of the variances
(because we have to take covariance into account). Their works provided new proof
of a result that had long been known (and which was considered as an old adage,
“Don’t put all your eggs in one basket”)34 using a new mathematical language, based
on modern probability theory. Their real contribution lay not in the result of portfolio
diversification, but in the use of this new mathematical language.
In 1958, Modigliani and Miller proceeded in the same manner: they used random
variables in the analysis of an old question, the capital structure of companies, to dem-
onstrate that the value of a company is independent of its capital structure.35 Their
contribution, like that of Markowitz and Roy, was to reformulate an old problem using
the terms of modern probability theory.
These studies launched a movement that would not gain ground until the
1960s:  until then, economists refused to accept this new research path. Milton
Friedman’s reaction to Harry Markowitz’s defense of his PhD thesis gives a good il-
lustration since he declared: “It’s not economics, it’s not mathematics, it’s not business
administration.” Markowitz suffered from this scientific conservatism since his first
article was not cited before 1959 (Web of Science). It was also in the 1960s that the
development of probability theory enabled economists to discover Bachelier’s work,
even though it had been known and discussed by mathematicians and statisticians
in the United States since the 1920s ( Jovanovic 2012). The spread of stochastic pro-
cesses and greater ease of access to them for nonmathematicians led several authors to
extend the first studies of financial econometrics.
The American astrophysicist Maury Osborne suggested an “analogy between ‘fi-
nancial chaos’ in a market, and ‘molecular chaos’ in statistical mechanics” (Osborne
 

11  Foundations of Financial Economics

1959b, 808). In 1959, his observation that the distribution of prices did not follow
the normal distribution led him to perform a log-​linear transformation to obtain the
normal distribution. According to Osborne, this distribution facilitated empirical tests
and linked with results obtained in other scientific disciplines. He also proposed con-
P 
sidering the price-​ratio logarithm, log  t +1  , which constitutes a fair approximation
 Pt 
of returns for small deviations (Osborne 1959a, 149). He then showed that deviations
in the price-​ratio logarithm are proportional to the square root of time, and validated
this result empirically. This change, which leads to consideration of the logarithmic
of returns of stocks rather than of prices, was retained in later work, because it pro-
vides an assurance of the stationarity of the stochastic process. It is worth mention-
ing that such a transformation was already suggested by Bowley (1933) for the same
reasons: bringing back the series to the normal distribution, the only one allowing the
use of statistical tests at this time. This transformation shows the importance of math-
ematical properties that authors used in order to keep the normal distribution as the
major describing framework.
The random processes used at that time have also been updated in the light of
more recent mathematics. Samuelson (1965a) and Mandelbrot (1966) criticized
the overly restrictive character of the random-​walk (or Brownian-​motion) model,
which was contradicted by the existence of empirical correlations in price move-
ments. This observation led them to replace it with a less restrictive model:  the
martingale model. Let us remember that a series of random variables Pt adapted to
( Φ;0 ≤ n ≤ N ) is a martingale if E(Pt+1 Φ t ) = Pt , where E(. / Φt ) designates the condi-
tional expectation in relation to (Φt) which is a filtration.36 In financial terms, if one
considers a set of information Φt increasing over time with t representing time and
Pt ∈Φ t , then the best estimation—​in line with the method of least squares—​of the
price (Pt+1) at the time t + 1 is the price (Pt) in t. In accordance with this definition, a
random walk is therefore a martingale. However, the martingale is defined solely by
a conditional expectation, and it imposes no restriction of statistical independence
or stationarity on higher conditional moments—​in particular the second moment
(i.e., the variance). In contrast, a random-​walk model requires that all moments in
the series are independent37 and defined. In other terms, from a mathematical point
of view, the concept of a martingale offers a more generalized framework than the
original version of random walk for the use of stochastic processes as a description
of time series.

1.2.2.  A New Community and the Challenge

to the Dominant School of the Time
The second element that contributed to the institutional birth of financial eco-
nomics was the formation in the early 1960s of a community of economists dedi-
cated to the analysis of financial markets. The scientific background of these econo-
mists determined their way of doing science by defining specific scientific criteria
for this new discipline.
 

12  Econophysics and Financial Economics

Prior to the 1960s, finance in the United States was taught mainly in business
schools. The textbooks used were very practical, and few of them touched on what
became modern financial theory. The research work that formed the basis of modern
financial theory was carried out by isolated writers who were trained in economics
or were surrounded by economists, such as Working, Cowles, Kendal, Roy, and
Markowitz.38 No university community devoted to the new subjects and methods
existed prior to the 1960s. During the 1960s and 1970s, training in American busi-
ness schools changed radically, becoming more “rigorous.”39 They began to “acade-
micize” themselves, recruiting increasing numbers of economics professors who
taught in university economics departments, such as Merton H. Miller (Fama 2008).
Similarly, prior to offering their own doctoral programs, business schools recruited
PhD students who had been trained in university economics departments ( Jovanovic
2008; Fourcade and Khurana 2009). The members of this new scientific community
shared common tools, references, and problems thanks to new textbooks, seminars,
and to scientific journals. The two journals that had published articles in finance, the
Journal of Finance and the Journal of Business, changed their editorial policy during the
1960s: both started publishing articles based on modern probability theory and on
modeling (Bernstein 1992, 41–​44, 129).
The recruitment of economists interested in questions of finance unsettled teach-
ing and research as hitherto practiced in business schools and inside the American
Finance Association. The new recruits brought with them their analysis frameworks,
methods, hypotheses, and concepts, and they were also familiar with the new math-
ematics that arose out of modern probability theory. These changes and their conse-
quences were substantial enough for the American Finance Association to devote part
of its annual meeting to them in two consecutive years, 1965 and 1966.
At the 1965 annual meeting of the American Finance Association an entire ses-
sion was devoted to the need to rethink courses in finance curricula. At the 1966
annual meeting, the new president of the American Finance Association, Paul Weston,
presented a paper titled “The State of the Finance Field,” in which he talked of the
changes being brought about by “the creators of the New Finance [who] become im-
patient with the slowness with which traditional materials and teaching techniques
move along” (Weston 1967, 539).40 Although these changes elicited many debates
( Jovanovic 2008; MacKenzie 2006; Whitley 1986a, 1986b; Poitras and Jovanovic
2007, 2010), none succeeded in challenging the global movement.
The antecedents of these new actors were a determining factor in the institution-
alization of modern financial theory. Their background in economics allowed them
to add theoretical content to the empirical results that had been accumulated since
the 1930s and to the mathematical formalisms that had arisen from modern prob-
ability theory. In other words, economics brought the theoretical content that was
missing and that had been underlined by Working and Roberts. Working (1961,
1958, 1956) and Roberts (1959) were the first authors to suggest a theoretical ex-
planation of the random character of stock market prices by using concepts and
theories from economics. Working (1956) established an explicit link between the
unpredictable arrival of information and the random character of stock market price
 

13╇ Foundations of Financial Economics

changes. However, this paper made no link with economic equilibrium and, prob-
ably for this reason, was not widely circulated. Instead it was Roberts (1959, 7) who
first suggested a link between economic concepts and the random-╉walk model by
using the “arbitrage proof ” argument that had been popularized by Modigliani and
Miller (1958). This argument is crucial in financial economics: it made it possible
to demonstrate the existence of equilibrium in uncertainty when there is no oppor-
tunity for arbitrage. Cowles (1960, 914–╉15) then made an important step forward
by identifying a link between financial econometric results and economic equilib-
rium. Finally, two years later, Cootner (1962, 25)  linked the random-╉walk model,
information, and economic equilibrium, and set out the idea of the efficient-╉market
hypothesis, although he did not use that expression. It was a University of Chicago
scholar, Eugene Fama, who formulated the efficient-╉market hypothesis, giving it its
first theoretical account in his PhD thesis, defended in 1964 and published the next
year in the Journal of Business. Then, in his 1970 article, Fama set out the hypothesis of
efficient markets as we know it today (we return to this in detail in the next section).
Thus, at the start of the 1960s, the random nature of stock market variations began to
be associated both with the economic equilibrium of a free competitive market and
with the building of information into prices.
The second illustration of how economics brought theoretical content to mathe-
matical formalisms is the capital-╉asset pricing model (CAPM). In finance, the CAPM
is used to determine a theoretically appropriate required rate of return for an asset, if
the asset is to be added to an already well-╉diversified portfolio, given the asset’s nondi-
versifiable risk. The model takes into account the asset’s sensitivity to nondiversifiable
risk (also known as systematic risk or market risk or beta), as well as the expected
market return and the expected return of a theoretical risk-╉free asset. This model is
used for pricing an individual security or a portfolio. It has become the cornerstone of
modern finance (Fama and French 2004). The CAPM is also built using an approach
familiar to economists for three reasons. First, some sort of maximizing behavior on
the part of participants in a market is assumed;41 second, the equilibrium conditions
under which such markets will clear are investigated; third, markets are perfectly com-
petitive. Consequently, the CAPM provided a standard financial theory for market
equilibrium under uncertainty.
In conclusion, this combination of economic developments with the probability
theory led to the creation of a truly homogeneous academic community whose actors
shared common problems, common tools, and a common language that contributed
to the emergence of a research movement.

1.2.3.╇ The Creation of New Empirical Data

Another crucial advance occurred in the 1960s: the creation of databases containing
long-╉term statistical data on the evolution of stock market prices. These databases al-
lowed spectacular development of empirical studies used to test models and theories
in finance. The development of these studies was the result of the creation of new sta-
tistical data and the emergence of computers.
 

14  Econophysics and Financial Economics

Beginning in the 1950s, computers gradually found their way into financial institu-
tions and universities (Sprowls 1963, 91). However, owing to the costs of using them
and their limited calculation capacity, “It was during the next two decades, starting
in the early 1960s, as computers began to proliferate and programming languages
and facilities became generally available, that economists more widely became users”
(Renfro 2009, 60). The first econometric modeling languages began to be developed
during the 1960s and the 1970s (Renfro 2004, 147). From the 1960s on, computer
programs began to appear in increasing numbers of undergraduate, master’s, and doc-
toral theses. As computers came into more widespread use, easily accessible databases
were constituted, and stock market data could be processed in an entirely new way
thanks to, among other things, financial econometrics (Louçã 2007). Financial econ-
ometrics marked the start of a renewal of investigative studies on empirical data and
the development of econometric tests. With computers, calculations no longer had
to be performed by hand, and empirical study could become more systematic and
conducted on a larger scale. Attempts were made to test the random nature of stock
market variations in different ways. Markowitz’s hypotheses were used to develop spe-
cific computer programs to assist in making investment decisions.42
In addition, computers allowed the creation of databases on the evolution of stock
market prices. They were used as “bookkeeping machines” recording data on phe-
nomena. Chapter 2 will discuss the implications of these new data on the analysis of
the probability distribution. Of the databases created during the 1960s, one of the
most important was set up by the Graduate School of Business at the University of
Chicago, one of the key institutions in the development of financial economics. In
1960, two University of Chicago professors, James Lorie and Lawrence Fisher, started
an ambitious four-​year program of research into security prices (Lorie 1965, 3). They
created the Center for Research in Security Prices (CRSP). Roberts worked with
them too. One of their goals was to build a huge computer database of stock prices
to determine the returns of different investments. The first version of this database,
which collected monthly prices from the New  York Stock Exchange (NYSE) from
January 1926 through December 1960, greatly facilitated the emergence of empirical
studies. Apart from its exhaustiveness, it provided a history of stock market prices and
systematic updates.
The creation of empirical databases triggered a spectacular development of finan-
cial econometrics. This development also owed much to the scientific criteria pro-
pounded by the new community of researchers, who placed particular importance
on statistical tests. At the time, econometric studies revealed very divergent results
regarding the representation of stock market variations by a random-​walk model with
the normal distribution. Economists linked to the CRSP and the Graduate School of
Business at the University of Chicago—​such as Moore (1962) and King (1964)—​
validated the random-​walk hypothesis, as did Osborne (1959a, 1962), and Granger
and Morgenstern (1964, 1963). On the other hand, work conducted at MIT and
Harvard University established dependencies in stock market variations. For example,
Alexander (1961), Houthakker (1961), Cootner (1962), Weintraub (1963), Steiger
(1963), and Niederhoffer (1965) highlighted the presence of trends.43 Trends had
 

15╇ Foundations of Financial Economics

already been observed by some of the proponents of the random-╉walk hypothesis,

in particular by Cowles, who, in his 1937 and 1944 articles, had observed a bias that
opened up the possibility of predicting future stock market price variations. In re-
sponse to a remark by Working (1960)44 concerning the statistical explanation for
these “alleged trends,” Cowles redid his calculations and once more validated the ex-
istence of trends (1960, 914).
These databases changed the perception of stock markets, and it also paved the way
to their statistical analysis. However, several drawbacks must be pointed out. First, it
must be borne in mind that the incompleteness, the nonstandardization, errors, and
misregistrations of data before the 1960s limited empirical investigations as well as the
trustworthiness of their results. For instance, many trades took place outside the offi-
cial markets and therefore were not recorded; the records generally focused on high
market value, leading to underevaluated returns because the higher returns offered
by firms with low market value are missing (Banz 1981; Annaert, Buelens, and Riva
2015). Second, data recorded were averages of prices (higher and lower day price,
for instance) or closing prices. When the first databases were created, they did not
collect daily data that are more time consuming than collecting monthly data. For in-
stance, the original CRSP stock file contained month-╉end prices and returns from the
NYSE starting from December 1925, while daily data have only been provided since
July 1962. Consequently, the volatility of stock market prices/╉returns recorded at that
time was necessary lower than the volatility observed during a market day. Chapter 2
will detail the implications of these drawbacks on the probability distribution analysis,
particularly the choice for the Gaussian distribution by financial economists.

1.3.╇ ROLE AND PLACE OF THE GAUSSIAN

DISTRIBUTION IN FINANCIAL ECONOMICS
1.3.1.╇ Stochastic Processes and the Efficient-╉Market Hypothesis
The overlapping of the mathematical formalisms that emerged from modern prob-
ability theory, and economics theory in particular, was a crucial factor in the birth
of financial economics. In this movement, the efficient-╉market hypothesis had a very
specific place, which is unclear to most econophysicists and financial economists.
Fama developed his intuition that a random-╉walk model would verify two properties
of competitive economic equilibrium: the absence of marginal profit and the equali-
zation of a stock’s price and value, meaning that the price perfectly reflects the avail-
able information. This project was undeniably a tour de force: creating a hypothesis
that made it possible to incorporate econometric results and statistics on the random
nature of stock market variations into the theory of economic equilibrium. It is
through this link that one of the main foundations of current financial economics was
laid down and that the importance of the random-╉walk model, or Brownian motion,
and thus of the normal distribution, can be explained: validating the random nature of
stock market variations would in effect establish that prices on competitive financial
markets are in permanent equilibrium as a result of the effects of competition. This is
 

16  Econophysics and Financial Economics

what the efficient-​market hypothesis should be, but this hypothesis does not really
reach this goal.
To establish this link, Fama extended the early theoretical thinking of the 1960s
and transposed onto financial markets the concept of free competitive equilibrium on
which rational agents would act (1965b, 56). Such a market would be characterized
by the equalization of stock prices with their equilibrium value. This value is deter-
mined by a valuation model the choice of which is irrelevant for the efficient-​market
hypothesis.45 The latter considers that the equilibrium model valued stocks using all
available information in accordance with the idea of competitive markets. Thus, on
an efficient market, equalization of the price with the equilibrium value meant that
all available information was included in prices.46 Consequently, that information is
of no value in predicting future price changes, and current and future prices are inde-
pendent of past prices. For this reason, Fama considered that, in an efficient market,
price variations should be random, like the arrival of new information, and that it is
impossible to achieve performances superior to that of the market (Fama 1965a, 35, 98).
A random-​walk model thus made it possible to simulate dynamic evolution of prices
in a free competitive market that is in constant equilibrium.
For the purpose of demonstrating these properties, Fama assumed the existence
of two kinds of traders: the “sophisticated traders” and the normal ones. Fama’s key
assumption was the existence of “sophisticated traders” who, due to their skills, make
a better estimate of the intrinsic/​fundamental value than other agents do by using all
available information. Moreover, Fama assumes that “although there are sometimes
discrepancies between actual prices and intrinsic values, sophisticated traders in ge-
neral feel that actual prices usually tend to move toward intrinsic values” (1965a, 38).
According to Fama’s hypothesis, “sophisticated traders” are better than other agents
at determining the equilibrium value of stocks, and since they share the same valu-
ation model for asset prices and since their financial abilities are superior to those of
other agents (Fama 1965a, 40), their transactions will help prices trend toward the
fundamental value that these sophisticated traders share. Fama added, using arbitrage
reasoning, that any new information is immediately reflected in prices and that the ar-
rival of information and the effects of new information on the fundamental value are
independent (1965a, 39). The independence of stock market fluctuations, the inde-
pendence of the arrival of new information, and the absence of profit made the direct
connection with the random-​walk hypothesis possible. In other words, on the basis of
assumptions about the existence of these sophisticated traders’ having financial abili-
ties superior to those of other agents, Fama showed that the random nature of stock
market variations is synonymous with dynamic economic equilibrium in a free com-
petitive market.
But when the time came to demonstrate mathematically the intuition of the link
between information and the random (independent) nature of stock market varia-
tions, Fama became elusive. He explicitly attempted to link the efficient-​market hypo-
thesis with the random nature of stock market variations in his 1970 article. Seeking
to generalize, he dropped all direct references to fundamental value. The question of
the number of “sophisticated traders” required to obtain efficiency (which Fama was
 

17  Foundations of Financial Economics

unable to answer) was resolved by being dropped. Consequently, all agents were as-
sumed to be perfectly rational and to have the same model for evaluating the price of
financial assets (i.e., the representative-​agent hypothesis). Finally, he kept the general
hypothesis that “the conditions of market equilibrium can (somehow) be stated in
terms of expected returns” (1970, 384). He formalized this hypothesis by using the
definition of a martingale:

Pj ,t+1 − Pj ,t
( )  (  )
E Pj ,t+1 Φ t = 1 + E rj ,t+1 Φ t  Pj ,t , with rj ,t+1 =
Pj ,t
,

(1.3)

where the tilde indicates that the variable is random, Pj and rj represent the price and
return of a period of the asset j, E(./​.) the conditional expectation operator, and Φt all
information at the time t.
This equation implies that “the information Φt would be determined from the par-
ticular expected return theory at hand” (1970, 384). Fama added that “this is the sense
in which Φt is ‘fully reflected’ in the formation of the price Pj,t” (1970, 384). To test the
hypothesis of information on efficiency, he suggested that from this equation one can
obtain the mathematical expression of a fair game, which is one of the characteristics
of a martingale model and a random-​walk model. Demonstration of this link would
ensure that a martingale model or a random-​walk model could test the double charac-
teristic of efficiency: total incorporation of information into prices and the nullity of
expected return.
This is the most well-​known and used formulation of the efficient-​market hypo-
thesis. However, it is important to mention that the history of the efficient-​market
hypothesis went beyond the Fama (1970) article. Indeed, in 1976, LeRoy showed
that Fama’s demonstration is tautological and that his hypothesis is not testable. Fama
answered by changing his definition and admitted that any test of the efficient-​market
hypothesis is a test of both market efficiency and the model of equilibrium used by in-
vestors (Fama 1976). Moreover, he modified his mathematical formulation and made
his definition of efficiency more precise:

( ) ( )
Em R j ,t |Φ mt −1 = E R j ,t |Φ t −1 ,  (1.4)

( )
where Em R j ,t |Φ mt −1 is the equilibrium expected return on security j implied by the
( )
set of information used by the market at t –​ 1, Φ mt −1 , and E R j ,t |Φ t −1 is the true ex-
pected return implied by the set of information available at t –​ 1, Φ t −1 . From then on,
efficiency presupposes that, using Fama’s own terms, the market “correctly” evaluates
the “true” density function conditional on all available information. Thus, in an effi-
cient market, the true model for valuing the equilibrium price is available to agents.
To test efficiency, Fama reformulated the expected return by introducing a distinction
between price—​defined by the true valuation model—​and agents’ expectations. The
test consisted in verifying whether the return expected by the market based on the in-
formation used, Φ mt −1 , is equal to the expectation of true return obtained on the basis of
all information available, Φ t −1 . This true return is obtained by using the “true” model
 

18  Econophysics and Financial Economics

for determining the equilibrium price. Fama proposed testing the efficiency in two
ways, both of which relied on the same process. The first test consisted in verifying
whether “trading rules with abnormal expected returns do not exist” (1976, 144). In
other words, this was a matter of checking that one could obtain the same return as that
provided by the true model of assessment of the equilibrium value on the one hand
and the set of available information on the other hand. The second test would look
more closely at the set of information. It was to verify that “there is no way to use the
information Φ t −1 available at t − 1 as the basis of a correct assessment of the expected
return on security j which is other than its equilibrium expected value” (1976, 145).
At the close of his 1976 article, Fama answered LeRoy’s criticisms: the new defini-
tion of efficiency was a priori testable (we will make this point more precise hereafter).
It should be noted, however, that the definition of efficiency had changed: it now re-
ferred to the true model for assessing the equilibrium value. For this reason, testing
efficiency required also testing that agents were using the true assessment model for
the equilibrium value of assets.47 The test would, then, consist of using a model for set-
ting the equilibrium value of assets—​the simplest would be to take the model actually
used by operators—​and determining the returns that the available information would
generate; then to use the same model with the information that agents use. If the same
result were obtained—​that is, if equation (1.4) was verified—​then all the other infor-
mation would indeed have been incorporated into prices. It is striking to note that this
test is independent of the random nature of stock market variations. This is because, in
this 1976 article, there is no more talk of random walk or martingale; no connection
with a random process is necessary to test efficiency. Despite this important conclu-
sion, Fama’s article (1976) is rarely cited. Almost all authors refer to the 1970 article
and keep the idea that to validate the random nature of stock market variations means
validating market efficiency.
The precise linkage proposed by Fama was, however, only one of many possible
linkages, as subsequent literature would demonstrate. LeRoy (1973) and Lucas
(1978) provided theoretical proofs that efficient markets and the martingale hypo-
thesis are two distinct ideas:  a martingale is neither necessary nor sufficient for an
efficient market. In a similar way, Samuelson (1973), who gave a mathematical proof
that prices may be permanently equal to the intrinsic value and fluctuate randomly,
explained that the making of profits by some agents cannot be ruled out, contrary to
the original definition of the efficient-​market hypothesis. De Meyer and Saley (2003)
showed that stock market prices follow a martingale even if all available information is
not reflected in the prices.
A proliferation of theoretical developments combined with the accumulation of
empirical work led to a confusing situation. Indeed, the definition of efficient markets
has changed depending on the emphasis placed by each author on a particular feature.
For instance, Fama et al. (1969, 1) defined an efficient market as “a market that adjusts
rapidly to new information”; Jensen (1978, 96) stated that “a market is efficient with
respect to information set θt if it is impossible to make economic profit by trading on
the basis of information set θt”; and according to Malkiel (1992), “The market is said
to be efficient with respect to some information set … if security prices would be
 

19  Foundations of Financial Economics

unaffected by revealing that information to all participants. Moreover, efficiency with

respect to an information set … implies that it is impossible to make economic profits
by trading on the basis of [that information set].” The confusing situation is similar
regarding tests: the type of test used depends on the definition used by the authors.
However, it is worth mentioning that all of these tests shared the hypothesis of
normality (Gaussian distribution). Indeed, all statistical tests have been based on
the central-​limit theorem, which cannot be separate from the Gaussian framework.
Financial economists, particularly Fama and Mandelbrot, discussed this characteristic
and its consequences in the 1960s (as will be explained in ­chapter 2). Even today, the
vast majority of statistical tests are developed in this Gaussian framework (­chapters 4
and 5 will come back to this point). Moreover, some authors have used the weakness
of theoretical definitions to criticize the very relevance of efficient markets. For in-
stance, Grossman and Stiglitz (1980) argued that because information is costly, prices
cannot perfectly reflect all available information. Consequently, they considered per-
fectly informationally efficient markets to be impossible.
In retrospect it is clear that the theoretical content of the efficient-​market hypo-
thesis refers to its suggestion of a link between a mathematical model, some empir-
ical results, and the economic equilibrium. For analyzing the connection between the
economic concept of equilibrium and the random character of stock market prices/​
returns, Fama assumed that information arrives randomly. In this perspective, a sto-
chastic process describing the evolution of prices/​returns should be able to test if a
financial market is a free competitive market perpetually assumed to be at its equi-
librium (such a framework means that the market is efficient). The choice for the
Gaussian distribution/​framework reflects the will of financial economists to keep sta-
tistical tests that make sense with their economic hypotheses (for instance, the fact
that a security should have one price and not a set of prices). However, it is important
to emphasize that this demonstration of the link between a random-​walk model, or
Brownian motion, and a competitive market perpetually assumed to be at its equi-
librium (as predicted by the efficient-​market hypothesis) holds only if information
arrives randomly.
Three consequences can be deduced from the previous remarks. First, the random
character of stock market prices/​returns must be separated from the efficient-​market
hypothesis. In this context, the impossibility of obtaining returns higher than those
of the market (i.e., making a profit) is sufficient for validating the efficient-​market hy-
pothesis. Second, statistical tests cannot reject this hypothesis because it is an ideal
to strive for. Precisely, in economics a free competitive market is a market that would
optimize global welfare; it is a theoretical ideal picture that must respect several condi-
tions (a large number of buyers and sellers; no barriers of entry and exit; no transaction
costs, etc.). Financial economists are generally aware that empirical examples contra-
dict the ideal of freely competitive stock markets. However, despite these empirical
contradictions, most financial economists hold onto this theoretical ideal. Faced with
these contradictions, they try to adopt rules for going through a more free, compet-
itive market. Such apriorism is well documented in the economics literature, where
several authors have studied its potential consequences on the financial industry
 

20╇ Econophysics and Financial Economics

(Schinckus 2008, 2012; McGoun 1997; Macintosh 2003; Macintosh et  al. 2000).
Third, the choice of the Gaussian framework is directly related to the need to de-
velop statistical tests, which, due to the state of science, cannot be separated from the
Gaussian distribution and the central-╉limit theorem.
Finally, the efficient-╉market hypothesis represents an essential result for financial
economics, but one that is founded on a consensus that leads to acceptance of the
hypothesis independent of the question of its validity (Gillet 1999, 10). The reason
is easily understood: by linking financial facts with economic concepts, the efficient-╉
market hypothesis enabled financial economics to become a proper subfield of eco-
nomics and consequently to be recognized as a scientific field. Having provided this
link, the efficient-╉market hypothesis became the founding hypothesis of the hard core
of financial economics.

1.3.2.╇ The Gaussian Framework and the Key Models of Finance

The 1960s and 1970s were years of “high theory” for financial economics (Daemen
2010) in which the hard core of the discipline was laid down.48 The efficient-╉market
hypothesis was a crucial building block of modern financial economics. If markets
are efficient, then techniques for selecting individual securities will not generate
abnormal returns. In such a world, the best strategy for a rational person seeking
to maximize expected utility is to diversify optimally. Achieving the highest level
of expected return for a given level of risk involves eliminating firm-╉specific risk
by combining securities into optimal portfolios. Building on Markowitz (1952,
1959), Treynor (1961), Sharpe (a PhD student of Markowitz’s) (1964, 1963),
Lintner (1965a, 1965b), and Mossin (1966) made key theoretical contributions to
the development of the capital-╉asset pricing model (CAPM) and the single-╉factor
model, and few years after, Ross (1976a, 1977)  suggested the arbitrage pricing
theory (APT), which is an important extension of the CAPM. A new definition of
risk was provided. It is not the total variance of a security’s return that determines
the expected return. Rather, only the systematic risk—╉that portion of total vari-
ance which cannot be diversified away—╉will be rewarded with expected return.
An ex ante measure of systematic risk—╉the beta of a security—╉is proposed and the
single-╉factor model used to motivate ex post empirical estimation of this param-
eter. Leading figures of the modern financial economics network, such as Miller,
Scholes, and Black, examined the inherent difficulties in determining empirical es-
timates and developed important techniques designed to provide such estimates.
A collection that promoted these important contributions was the volume edited
by Jensen (1972).
The combination of these three essential elements—╉the efficient-╉market hypo-
thesis, the Markowitz mean-╉variance portfolio optimization model, and the CAPM—╉
constitutes the core of analytical progress on modern portfolio theory during the
1960s. Just as a decade of improvement and refinement of modern portfolio theory
was about to commence, another kernel of insight contained in Cootner (1964) came
to fruition with the appearance of Black and Scholes’s work (1973).49 Though the
 

21  Foundations of Financial Economics

influential paper by Samuelson (1965b) was missing from the edited volume, Cootner
(1964) did provide, along with other studies of option pricing, an English translation
of Bachelier’s 1900 thesis and a chapter by Case Sprenkle (1961) where the partial-​
differential-​equation-​based solution procedure employed by Black and Scholes was
initially presented (MacKenzie 2003, 2007). With the aim of setting a price for op-
tions, Black and Scholes took the CAPM as their starting point, using this model of
equilibrium to construct a null-​beta portfolio made up of one unit of the underlying
asset and a certain quantity of sold options.50
Black and Scholes (1973) marked the beginning of another scientific movement—​
concerned with contingent securities pricing51—​that was to be larger in practical
impact and substantially deeper in analytical complexity. The Black-​Scholes-​Merton
model is based on the creation of a replicating portfolio that, if the model is clearly
specified and its hypotheses tested, holds out the possibility of locally eliminating risk
in financial markets.52 From a theoretical point of view, this model allows for a partic-
ularly fruitful connection with the Arrow-​Debreu general-​equilibrium model, giving
it a degree of reality for the first time. Indeed, Arrow and Debreu (1954) and later
Debreu (1959) were able to model an uncertain economy and show the existence
of at least a competitive general equilibrium—​which, moreover, had the property of
being Pareto-​optimal if as many markets as contingent assets were opened. When a
market system is in equilibrium according to Arrow-​Debreu’s framework, it is said to
be complete. Otherwise, it is said to be incomplete. Black-​Scholes-​Merton’s model
gave reality to this system of complete markets by allowing that any contingent claim
asset is replicable by basic assets.53 This model takes on special theoretical importance,
then, because it ties results from financial economics more closely to the concept of
equilibrium from economic science.
The theories of the hard core of financial economics have had a very strong
impact on the practices of the financial industry (MacKenzie and Millo 2009; Millo
and Schinckus 2016). The daily functioning of financial markets today is conducted,
around the clock, on concepts, theories, and models that have been defined in finan-
cial economics (MacKenzie 2006). What operators on today’s financial markets do
is based on stochastic calculation, benchmarks, informational efficiency, and the ab-
sence of arbitrage opportunities. The theories and models of financial economics have
become tools indispensable for professional activities (portfolio management, risk
measurement, evaluation of derivatives, etc.). Hereafter we give some examples to il-
lustrate this influence.
According to efficient-​market hypothesis, it is impossible to outperform the
market. Together with the results of the CAPM, particularly regarding the possibility
of obtaining a portfolio lying close to the efficiency frontier, this theory served as the
basis for the development, from 197354 on, of a new way of managing funds—​passive,
as opposed to active, management. Funds managed this way create portfolios that
mirror the performance of an externally specified index. For example, the well-​known
Vanguard 500 Index fund is invested in the 500 stocks of Standard & Poor’s 500
Index on a market-​capitalization basis. “Two professional reports published in 1998
and 1999 [on portfolio management] stated that ‘the debate for and against indexing
 

22  Econophysics and Financial Economics

generally hinged on the notion of the informational efficiency of markets’ and that
‘managers’ various offerings and product ranges [note: indexed and nonindexed prod-
ucts] often depended on their understanding of the informational efficiency of mar-
kets’ ” (Walter 2005, 114).55
Further examples of the changes brought about by the hard core of financial eco-
nomics are the development of options and new ways of managing risks. The Chicago
Board Options Exchange (CBOE), the first public options exchange, began trading
in April 1973, and since 1975, thousands of traders and investors have been using
the Black and Scholes formula every day (MacKenzie 2006; MacKenzie and Millo
2009) on the CBOE to price and hedge their option positions. By enabling a dis-
tinction to be made between risk takers and hedgers, the Black and Scholes model
directly influenced the organization of the CBOE by defining how market makers
can be associated with the second category, the hedgers (Millo and Schinckus 2016.
Between 1974 and 2014, annual volumes of options exchanged on the CBOE rose
from 5.6 million to 1.275 billion (in dollars, from $449.6 million to $579.7 billion
billion) in Chicago alone. OTC derivatives notional amounts outstanding totaled
$630 trillion at the end of December 2014 (www.bis.org). In 1977 Texas Instruments
brought out a handheld calculator specially programmed to produce Black-​Scholes
options prices and hedge ratios. Merton (1998) pointed out that the influence of
the Black-​Scholes option theory on finance practice has not been limited to financial
options traded in markets or even to derivatives generally. It is also used to price and
evaluate risk in a wide array of applications, both financially and nonfinancially.
Moreover, the Black and Scholes model totally changed approaches to apprais-
ing risk since it allows risk to be individualized by giving a price to each insurance
guarantee rather than mutualizing it, as was done previously. This means that a
price can be put on any risk, such as the loss of the use of a famous singer’s voice,
which would clearly not be possible when risks are mutualized (Bouzoubaa and
Osseiran 2010).
Last, we would point out that financial-​market regulations increasingly make
reference to concepts taken from financial economics, such as the efficiency of
markets, that directly influence US legislative policies (Dunbar and Heller 2006)56
As Hammer and Groeber (2007, 1) explain, the “efficient-​market hypothesis is the
main theoretical basis for legal policies that impact both Fraud on the Market and
doctrines in security regulation litigation.” The efficient-​market hypothesis was in-
voked as an additional justification for the existing doctrine of fraud on the market,
thereby strengthening the case for class actions in securities-​fraud litigation. The
efficient-​market hypothesis demonstrated that every fraudulent misrepresentation
was necessarily reflected in stock prices, and that every investor could rely solely
on those prices for transactions ( Jovanovic et al. 2016). Chane-​Alune (2006)
emphasizes the incidence of the efficient-​market hypothesis on accounting stand-
ardization, while Miburn (2008, 293) notes that the theory directly influences the
international practice of certified accountants: “It appears that arguments typically
put forward by the International Accounting Standards Board and the FASB for the
 

23  Foundations of Financial Economics

relevance of fair value for financial reporting purposes do imply a presumption of

reasonably efficient markets.”

1.3.3.  The Mathematical Foundations

of the Gaussian Framework
This final section looks at the role of the mathematical properties of Brownian
motion (continuity and normal distribution) and their importance in the creation
of the hard core of financial economics. The importance of Brownian motion clearly
emerges from the work of Harrison, Kreps, and Pliska, who laid down the mathemat-
ical framework—​using a probability framework based on the measure theory—​for
much of current financial economics, and for the Black-​Scholes-​Merton model and
the efficient-​market hypothesis in particular.
Harrison and Kreps (1979), Kreps (1981), and Harrison and Pliska (1981) pro-
posed a general model for the valuation of contingent claim assets with no arbitrage
opportunity. They showed that on any date the price of an asset is the average of
its discounted terminal flows, weighted by a so-​called risk-​neutral probability.57 In
order to obtain a single risk-​neutral probability and thus to have a complete market,
these authors hypothesized that variations in the underlying asset followed Brownian
motion. If the price did not follow Brownian motion, then the market would not be
complete, which would imply that the option price was not unique and that exact rep-
lication by a self-​financing strategy (i.e., there is neither investment nor withdrawal
of money) would no longer be possible.58 Unique price and exact replication are two
central hypotheses of financial economics. The uniqueness of the price of a good
or asset originates from the “law of one price” in economics, which is a constituent
part of financial economics. Exact replication by means of a self-​financing strategy
is based on the one-​price hypothesis associated with arbitrage reasoning that, as we
have seen, enables an equilibrium situation on a market to be obtained—​thus making
the absence of arbitrage opportunity the financial-​economics counterpart of equilib-
rium in economics.
The efficient-​market hypothesis also has roots in Brownian motion (section 1.3.1).
The definition put forward by Jensen (1978) and found in Malkiel (1992) is without
doubt one of the easiest to apply: “A market is efficient with respect to information
set θt if it is impossible to make economic profit by trading on the basis of informa-
tion set θt.” This definition is equivalent to the no-​arbitrage principle as defined by
Harrison, Kreps, and Pliska, according to which it should not be possible to make a
profit with zero net investment and without bearing any risk. The absence of arbitrage
opportunities as defined by Harrison, Kreps, and Pliska indeed made it possible to
give a mathematical definition of the theory of informational efficiency proposed by
Fama. Now, as we have just pointed out, the demonstration by Harrison, Kreps, and
Pliska relies on the mathematical properties of Brownian motion.
The capital-​asset pricing model, the arbitrage pricing theory, and modern port-
folio theory, also components of the hard core of financial economics, were built on
 

24╇ Econophysics and Financial Economics

the mean-╉variance optimization developed by Markowitz. This optimization owes its

results to the hypothesis that the returns of financial assets are distributed according
to the normal distribution. Similarly, assuming that the returns of an efficient stock
market are influenced by a large number of firm-╉specific economic factors that should
add up to something resembling the normal distribution, the creators of CAPM took
their hypothesis from the central-╉limit theorem.
Without any question, the normal distribution, and its cousin, Brownian motion,
or the Wiener process, are fundamental hypotheses for reasoning within the math-
ematical framework of financial economics:59 “While many quantitative financiers
would gladly dispose of the Brownian motion, the absence of arbitrage, or a free lunch,
is a cornerstone principle few could do without. In the light of these discoveries, the
researcher wishing to reject Brownian diffusion as description of the evolution of re-
turns must first invent an alternative mechanism, which would include the concept of
arbitrage. This is not impossible but it requires a very radical conceptual revision of
our current understanding of financial economics.”60

1.4.╇CONCLUSION
This chapter analyzed the theoretical and methodological foundations of financial ec-
onomics, which are embedded in the Gaussian framework. The historical, mathemat-
ical, and practical reasons justifying these foundations were investigated.
Since the first works in modern finance in the 1960s, the Gaussian distribution
has been considered to be the law ruling any random phenomena. Indeed, the au-
thors based their stochastic models on results deduced from the central-╉limit theorem,
which led to the systematic use of the Gaussian distribution. In this perspective, the
major objective of these developments was to “reveal” the Gaussian distribution in
the data. When observations did not fit with the normal distribution or showed some
extreme values, authors commonly used a log-╉linear transformation to obtain the
normal distribution. However, it is worth reminding that, in the 1960s, prices were
recorded monthly or daily, implying a dilution of price volatility.
In this chapter, we explained that financial econometrics and statistical tests
became key scientific criteria in the development of financial economics. Given that
the vast majority of statistical tests have been developed in the Gaussian framework,
the latter was viewed as a necessary scientific tool for the treatment of financial data.
Finally, this chapter clarified the links between the Gaussian distribution and the
efficient-╉market hypothesis. More precisely, the random character of stock market
prices/╉returns must be separated from the efficient-╉market hypothesis. In other
words, any stochastic process, including the Gaussian one, will not provide an empir-
ical validation for this hypothesis.
For all these reasons, the Gaussian distribution becomes a key element of financial
economics. The following chapter will study how financial economists have dealt with
extreme values given the scientific constraints dictated by the Gaussian framework.
 

2
E X T R E M E VA LU E S I N   F I N A N C I A L EC O N O M I C S
F R O M T H E I R O B S E RVAT I O N TO T H E I R I N T EG R AT I O N
I N TO T H E G AU S S I A N F R A M E W O R K

The previous chapter explained how the Gaussian framework played a key role in the
development of financial economics. It also pointed out how the choice for the normal
distribution was directly related to the kind of data available at that time. Given the
inability of the Gaussian law to capture the occurrence of extreme values, ­chapter 2
studies the techniques financial economists use to deal with extreme variations on
financial markets. This point is important for the general argument of this book be-
cause the existence of large variations in the stock prices/​returns is often presented
by econophysicists as the major justification for the importation of their models in
finance. While financial economists have mainly used stochastic processes with
Gaussian distribution to model stock variations, one must not assume that they have
ignored extreme variations. On the contrary, this chapter will show that the possi-
bility of modeling extreme variations has been sought since the creation of financial
economics in the early 1960s. From an econophysics viewpoint, this statement may
surprise: there are countless publications on extreme values in finance. However, few
econophysicists seem to be aware of them, since they usually ignore or misunderstand
the solutions that financial economists have implemented. Indeed, statistical analysis
of extreme variations is at the heart of econophysics, and the integration of these vari-
ations into stochastic processes is the main purpose of this discipline, as will be shown
in ­chapter 3. From this perspective, a key question is, how does financial economics
combine Gaussian distribution with other statistical frameworks in order to charac-
terize the occurrence of extreme values?
This chapter aims to investigate this question and the reasons that financial econ-
omists decided to keep the Gaussian distribution. First of all, this chapter will review
the first observations of extreme values made by economists. Afterward, we will an-
alyze their first attempts to model these observations by using stable Lévy processes.
The difficulties in using these processes will be detailed by emphasizing the reasons
that financial economists did not follow this path. We will then study the alterna-
tive paths that were developed to consider extreme values. Two major alternatives
will be considered here: the ARCH-​type models and the jump-​diffusion processes.
To sum up, this chapter shows that although financial economists have integrated
extreme variations into their models, they use different stochastic processes than
econophysicists.

25
 

26╇ Econophysics and Financial Economics

2.1.╇ EMPIRICAL EVIDENCES

2.1.1.╇ Extreme Values in the Current Stock Price Variations
Although anyone who looks at stock current movements will observe some extreme
variations, they are relatively rare. Their occurrence can be observed in all kinds of
listed companies (public or state-╉owned) and over any time scale. For example,
monthly variations of returns for the private company IBM listed on the New York
Stock Exchange (NYSE) show several such extreme variations (figure 2.1).

0.3

0.2

0.1
Returns

0.0

–0.1

–0.2

–0.3

0 100 200 300 400 500

Index

Figure 2.1╇ Time plot of monthly log returns of IBM stock from March 1967 to December 2012
Source: Google Finance.

Extreme variations are also present in daily returns of the same stock (figure 2.2).

0.1

0.0
Log return

–0.1

–0.2

1970 1980 1990 2000

Year

Figure 2.2╇ Time plot of daily log returns of IBM stock from July 1962 until December 1998
Source: Google Finance.
 

27  Extreme Values in Financial Economics

Similar observations exist for most stocks, currencies, and other commodities listed on
any financial market around the world. The challenge for theoreticians is therefore to find
the most appropriate statistical framework for describing variations as observed empirically.
From a statistical point of view, the occurrence of these extreme values is gener-
ally associated with what statisticians call the leptokurticity of empirical distribution.
Schematically, leptokurtic distributions (such as the dotted line in figure 2.3) have higher
peaks (characterized by long tails on both sides of the mean) around the mean than does
the normal distribution (the solid line in figure 2.3), which has short statistical tails.

Leptokurtic distribution
Gaussian distribution
Probability

–1.980 1.980
95 % of values
Values

–2.580 2.580
Probability of Cases 99 % of values
in portions of the curve

Standard Deviations
From The Mean –40 –30 –20 –10 0 +10 +20 +30 +40
Cumulative % 0.1 % 2.3 % 15.9 % 50 % 84.1 % 97.7 % 99.9 %

Figure 2.3  Visual comparison between the Gaussian distribution (solid line) and a more leptokurtic
distribution (dotted line) for an infinity of observations

Long tails observed in leptokurtic distributions signify that portion of the distribu-
tion which has a large number of occurrences far from the mean. For instance, with the
Gaussian distribution the probability of a fluctuation 2.6 times the standard deviation is
less than 1 percent. This implies that extreme values are extremely rare. By contrast, with
the more leptokurtic distribution represented in figure 2.3, such a fluctuation is more
probable and therefore more important (depending on the value of parameters associ-
ated with the distribution), meaning that the possibility of large variations is greater in
the distribution. In other words, long tails in a distribution describe the behavior of a
variable for which extreme variations occur more often than in the case of the Gaussian
distribution. For illustrative purposes, consider that the occurrence of a financial crash is
equivalent to a fluctuation five times the standard deviation suggested by the Gaussian
framework, in which a financial crash has a probability of less than 0.000057 percent
(that would mean a crash every 10,000 years, according to Mandelbrot 2004).1
Although the Gaussian distribution has been widely used by financial economists,
mainly for its interesting statistical and mathematical properties (­chapter 1), it does
not provide a full description of empirical data (figure 2.4a). In contrast, a leptokurtic
distribution implies that small changes are less frequent than in the Gaussian case, but
that extreme moves are more likely to occur and are potentially much larger than in the
Gaussian distribution, as shown in figure 2.4b.
 

(a) 0.3

0.2

0.1

–0.1

–0.2

–0.3

(b) 0.3

0.2

0.1

–0.1

–0.2

–0.3

(c) 0.3

0.2

0.1

0
Log return

–0.1

–0.2

–0.3

1970 1980 1990 2000

Year
Figure 2.4  (a) Gaussian simulation describing the dynamics of financial data
(b) Paretian simulation (based on a leptokurtic distribution) describing the dynamics of financial data
(c) Empirical evolution of financial data (time plot of daily log returns of IBM stock from July 1962
until December 1998)
Source: Google Finance and authors.
 

29╇ Extreme Values in Financial Economics

In this figure we see that using a stochastic process with a Gaussian distribution
(figure 2.4a) does not allow extreme variations of stock prices or returns to be re-
produced (figure 2.4c). In contrast, a Paretian simulation opens a room for the sta-
tistical characterization of extreme variations (figure 2.4b). Obviously, this inability
to capture the occurrence of extreme variations is a major limitation of the Gaussian
framework for reproducing stock variations and consequently for analyzing risk or
for theoretical models. In this challenging context, financial economists had to find
a way to capture the occurrence of extreme values in the evolution of stock prices/╉
returns.

2.1.2.╇ Early Evidence of Extreme Values

Chapter 1 pointed out that the first authors to study stock price fluctuations did not per-
form systematic analysis of the distribution.2 The Gaussian distribution was for them
simply the most obvious distribution to use, first because it was commonly used in sci-
ence at the time, and second because of its mathematical properties, particularly the fact
that Gaussian law was seen as the asymptotic limit of statistical laws in common use. In
1863, Regnault introduced the Gaussian distribution because it was the “scientific law”
in social sciences, and also for ethical reasons. Regnault is one of the authors studied in
Â�chapter 1 who analyzed the distribution of stock price variations as such. The empirical
data he had were imprecise (the highest and lowest price for each month), but they were
sufficient to confirm the normal distribution and the fact that stock price variations fol-
lowed the square root of time. In 1900, Louis Bachelier in turn used the normal distri-
bution for mathematical reasons, this distribution having the right mathematical proper-
ties for the demonstration he wished to make. Bachelier did not study the distribution of
stock prices as such; he used the same data as Regnault (the French 3 percent bond) to
validate his mathematical demonstration. In 1908, Vinzenz Bronzin justified the Gaussian
distribution for modeling stock price variations for the same reason as Bachelier: “Total
mathematical expectations of gains and losses are equal to one another at the moment
when the respective deals are struck” (Hafner and Zimmermann 2009, 156). Like
Bachelier, he did not study the distribution of stock prices as such. Finally, the first au-
thors (Working, Cowles) to make empirical studies on stock prices in the United States in
the 1930s sought to demonstrate the independence of stock price variations. Here again,
except for the work of Cowles and Jones (1937),3 research was not focused on the distri-
bution but aimed at confirming the independence of stock price variations. Paradoxically,
the first analyses of the distribution of stock market fluctuations were not developed by
these authors, but by macroeconomists and statisticians whose research was conducted
independently of early research dedicated to financial markets.
From the late nineteenth century until the outbreak of World War I, there was mas-
sive development of measurement techniques.4 As Morgan and Klein (2001) showed,
this “age of economic measurement” involved the use of various instruments to de-
scribe empirical phenomena. Inside this movement, economists who were interested
in macroeconomic forecasts via barometers studied stock price variations and their dis-
tribution too.5 The first American economic barometers, designed for the study and
forecasting of business cycles, appeared in the early twentieth century (Armatte 1992,
 

30  Econophysics and Financial Economics

2003; Chancelier 2006b, 2006a; Morgan 1990). The leader was Harvard’s forecast-
ing barometer, created in 1916 by the Economic Office of Harvard and supervised by
Professor Warren Persons.6 Harvard’s barometer became famous for its three curves
(labeled A, B, and C), representing stock prices, the level of business, and the interest
rate on the money market. This approach spread to Europe in the 1920s, when French
statistician Lucien March constructed a French business barometer based on index
numbers that was seen as an equivalent to the Harvard barometer ( Jovanovic and Le
Gall 2001). With the development of barometers, economists—​for example Fisher7
and Mitchell8 (Armatte 1992; Banzhaf 2001; Diewert 1993)—​became increasingly in-
terested in the construction of indices, among them the stock price index. According to
barometer analysis, stock price movements were an important indicator of economic
activity, and this belief led economists to study stock price variations and their distribu-
tions. Economists interested in barometers were among the first to notice that stock
price variations were not necessarily distributed according to the normal distribution.
Among them, Mitchell (1915) seems to have been the first to empirically detect and
describe extreme variations in wholesale price time series. He found that distribution of
percentage changes in prices consistently deviated from normal distribution (Mitchell
1915, 19–​21). Mills (1927, chap. 3) also stressed the leptokurtic character of finan-
cial distributions. However, two points must be underlined here: first, neither Mitchell
nor Mills studied stock prices per se (they were more interested in the direction of the
prices than in the volatility of the series); second, the distribution was not the core of
their analysis, which, like business barometers, focused on macroeconomic predictions.
Three other authors deserve our attention. Irving Fisher (1922) did some analysis
of stock market prices in his book on index numbers; he noted in particular the ab-
sence of correlation in stock prices (Dimand and Geanakoplos 2005; Dimand 2007).
Moore must also be mentioned. In his book, published in October 1917, he analyzed
the price of cotton quoted on the stock market and compared the actual distribution
of spot cotton prices to the theoretical normal distribution (Le Gall 1999, 2006). Even
though the histogram created from real price changes has high peaks and fat tails com-
pared to the Gaussian distribution (Moore 1917, 26), Moore ignored these observa-
tions and considered that the variations in cotton prices were normally distributed.
Later, Arthur Lyon Bowley (1933) discussed the distribution of prices, particularly
stock prices, taking the length of time studied into account. He found that for short
periods (e.g., one year), a normal distribution provides a good approximation of stock
prices, but that “when we take the longer period 1924 to 1930, resemblance to the
normal curve is not present” (Bowley 1933, 369). To circumvent this difficulty while
preserving the Gaussian distribution (and consequently its mathematical properties),
he proposed considering the logarithm of the price in some cases (Bowley 1933, 370).9
These were not the only authors of the period to identify the leptokurtic nature of the
distribution of stock market fluctuations. This finding was shared by a number of statisti-
cians and French engineers (known at the time as ingénieurs économistes). As explained
by Jovanovic and Le Gall (2001), these ingénieurs économistes developed an indisputable
talent for measurement in economics. Their involvement in measurement and more gen-
erally in statistics became particularly apparent around 1900: Statistique Générale de la
France (SGF), the main government statistical agency at that time, became something
 

31╇ Extreme Values in Financial Economics

of an engineers’ fortress. Although the SGF remained small in comparison with other
European statistical bureaus, its early-╉twentieth-╉century achievements—╉in data col-
lection and in formulating explanations of what had been measured—╉remain impres-
sive. A large part of its work can be attributed to Lucien March (Huber 1937). March
(1921) was the first author to analyze whether the probability calculus and the law of
large numbers were relevant for studying, in particular, whether relative prices are dis-
tributed around their mean according to Gaussian law. He pointed out that that distri-
bution did not conform to Gaussian law: occurrences far from the mean occurred too
frequently and were too large. In addition, during the 1910s, SGF statisticians under
Marcel Lenoir became interested in the effects of stock market fluctuations on economic
activity (Chaigneau and Le Gall 1998). In 1919 Lenoir developed the first long-╉term
indices for the French stock market. These examples should not cause us to forget that
analysis of price distribution was still rare when another author, Maurice Olivier, pub-
lished his doctoral thesis in 1926, wherein he correctly pointed out that “a series of in-
dices should never be published unless the distribution of the data used to calculate the
index is published at the same time” (1926, 81). Olivier can be considered to have pro-
vided the first systematic analysis of price distribution. He obtained the same result as
March: the distribution of price variations did not fit with normal distribution. Several
decades later, Kendall (1953) studied a number of price series by analyzing their varia-
tions (see Â�chapter 1) and pointed out that some of them were leptokurtic.
Although these authors noted the fact that distributions of stock price variations
were not strictly Gaussian, there was no systematic research into this topic at the time.
Distributions were only a matter of observation and were not mathematically mod-
eled by a specific statistical framework. Econometrics was in its beginnings and not yet
organized as a discipline (Le Gall 1994, 2006; Morgan 1990), and economists were
interested in empirical investigations into dynamic business cycles (i.e., barometers).
Moreover, even Vilfredo Pareto (1848–╉1923), who was widely known for his research
on the leptokurtic nature of distribution of wealth in Italy at the beginning of the twen-
tieth century (Barbut 2003), did not analyze the dynamics of price variations (he made
a static analysis). Only one exception to this pattern can be mentioned, since Luigi
Amoroso used a Paretian distribution in a dynamic analysis of incomes (Tusset 2010).

2.1.3.╇ Confirmations in the 1960s and the 1970s

These early findings had very little impact because financial economics was not yet
established as a scientific discipline. Moreover, econometrics was a very young disci-
pline and the data collected had several drawbacks. With some exceptions, each author
worked with his own empirical data. Consequently, standard data did not exist and a
comparison between results was difficult. The low quality of the data was also a real
problem. For instance, the volume and price of all transactions were not compulsorily
recorded. Moreover, on the NYSE, when the volume of transactions was exceptionally
large, successive transactions made at the same price were recorded as a single transac-
tion (Godfrey, Granger, and Morgenstern 1964, 7). In addition, it is worth mentioning
that available data were computed as a monthly average, which dilutes the volatility.
Although weekly data were collected beginning in the 1920s (1921 in the United
 

32  Econophysics and Financial Economics

Kingdom and 1923 in the United States), they were available only for the wholesale
prices of a small number of foods.10 Finally, the quantity of data was limited, and con-
sequently the opportunity to observe extreme values or rare events was very limited.
The situation changed completely in the 1960s, as discussed in ­chapter  1. The
creation of financial economics as a scientific discipline, on the one hand, and the
building of the first comprehensive database (through the creation of the CRSP),
on the other, made the development of work in financial econometrics possible.
Moreover, with the creation of financial economics, empirical validation became
a prerequisite for publication of research. The introduction of computers into uni-
versity departments also facilitated changes. Thus, from the 1960s onward, financial
economists and especially MA and PhD students systematically tested the statistical
indicators of the random character of stock market fluctuations, such as the indepen-
dence, the distribution, and the stability of the process. Some of them also developed
new tests to validate the models used in “hard core” financial economics.11 However,
as Jovanovic (2008) has shown, empirical results indicating the random character
of stock market prices are inseparable from the underlying theoretical assumptions
guiding the gathering and interpretation of data, especially the assumption of per-
fect markets, defended at the Chicago Graduate School of Business, or of imperfect
markets, defended at MIT. Opinion on the randomness of stock market fluctuations
was far from unanimous: authors from the Chicago Graduate School of Business vali-
dated randomness empirically, while authors from the MIT did not. The same was
true of the identification and analysis of extreme values: while econometric studies
established several important results, there was no universally shared result regarding
leptokurtic distributions.
Some authors pointed out extreme values in stock market fluctuations. For in-
stance, Larson (1960, 380) identified “an excessive number of extreme values” that did
not accord with a Gaussian process. Houthakker (1961, 168) concluded that “the dis-
tribution of day-​to-​day changes in the logarithms of prices does not conform to the
normal curve. It is not significantly skew, but highly leptokurtic… . The variance of
price changes does not seem to be constant over time… . The leptokurticity men-
tioned above may be related to the changing variance.” Studying a sample of 85,000
observations, Brada, Ernst, and van Tassel observed that “the distributions, in all cases,
are excessively peaked. However … there are no fat tails. At this point we suggest that
stock prices differenced across single transactions are not normally distributed, not be-
cause they have fat tails, but because they are too peaked” (1966, 337). Let us also men-
tion Sprenkle (1961, 1964), who also noted that the stock price distribution was not
Gaussian.
It is clear that with the creation of financial economics and the development of
stock price databases, the distribution of stock price variations, and their leptokurtic
character in particular, began to be apparent and widely analyzed by financial econo-
mists. Although each author presented his empirical results with confidence, these
results should primarily be considered as statistical indices, because empirical data
and tests were still new at that time. However, this field of study underwent a radical
change when the leptokurtic nature of changes in stock prices became the subject of
mathematical investigations from 1962 onward.
 

33  Extreme Values in Financial Economics

2.2.  FINANCIAL ECONOMISTS AND

PARETO-​L ÉVY DISTRIBUTIONS
Mathematical treatment of extreme variations in financial economics gained impetus
from Benoit Mandelbrot’s work on stable Lévy processes.

2.2.1.  Interesting Properties of Stable Lévy

Processes for Extreme Variations
As we saw in ­chapter  1, various authors subscribed, explicitly or implicitly, to the
normal-​distribution hypothesis on account of its mathematical properties and more
particularly of its links with the central-​limit theorem, according to which distribution
of a set of identically distributed independent random variables will converge toward
a Gaussian distribution. But this was not the only mathematical property of interest
to these authors. The Gaussian framework implies that distribution of the process is
“stable” (by addition) over time.
The stability of a random process means that any linear combination of the inde-
pendent and identically distributed random variables of this process will generate a
random variable that has the same probability distribution. For instance, the sum of
any number of Gaussian random variables is itself a Gaussian random variable.12 This
statistical property is very useful for financial modeling because it allows data to be
analyzed independently from the time scale: if variables are independent and distrib-
uted in accordance with normal law, the estimate of statistical characteristics observed
at a given time scale (monthly) can be applied to another analysis time scale (annual
or daily). Thus, the stability of the process makes it possible to analyze the monthly
return of a stock as the sum of its daily returns.
In his efforts to model extreme values, the mathematician Benoit Mandelbrot noted
that these mathematical properties are not specific to Gaussian processes, as the French
mathematician Paul Lévy had shown in the 1920s. Mandelbrot was a student of Lévy
and worked on the mathematical infinity of empirical phenomena assuming infinite di-
visibility of observations. Lévy had defined a category of stochastic processes, X = X (t )
with t ≥ 0, that have right-​continuous paths whose increments are independent and
identically distributed (i.i.d.), and that satisfy the following conditions:

(1) X has independent and stationary increments.

(2) X(0) = 0 with a probability one.
(3) X is stochastically continuous, that is, ∀t > 0 and ∀s ≥ 0 ,

lim t →s P(( X (t ) − X (s)) > 0 )= 0.

The general characteristic function Φ (t ) of these processes is13

 α α
{ πα
}
 −σ t 1 − iβsign(t )tan 2 + iγ t , α ≠ 1
(2.1)
{ }
log Φ(t ) =  .
 − σ α t 1 − iβsign(t ) 2 log t + iγ t , α = 1
 π 
 

34  Econophysics and Financial Economics

This function requires four parameters (α, β, γ, δ) to be described; the parameter δ

does not need to be shown here.14 Lévy’s definition makes it possible to define a great
number of processes (Gauss, Poisson, Cauchy, Pareto, Gamma, generalized hyper-
bolic processes), depending on the distribution.
Through this definition, one can show that the stability of the distribution of the
stochastic process depends on the value of these four parameters: a stochastic process
is said to be stable if  0 < α ≤ 2, γ ≥ 0, −1 ≤ β ≤ 1, δ ∈ ℜ. As Nolan (2009) reminded
the only necessary condition concerns the value of the α:15 all Lévy processes with
0 < α ≤ 2 are said to be stable. Consequently, following Feller (1971), the definition
of a stable process can simply be formulated in terms of distribution (rather than in
terms of characteristic function): a random variable X has a stable distribution if, for
n independent copies Xi of X, there exists a positive number Cn and a real number
Dn such that the following relationship is valid for  0 < α ≤ 2:
d
X1 + X 2 +  + X n = Cn X + Dn .  (2.2)

Finally, Lévy’s work made it possible to introduce a set of stable stochastic pro-
cesses that can be used to model stock price variations from an empirical estimate
of the four parameters (α, β, γ, δ). Here is a brief reminder of the interpretation of
these parameters. The parameter α (called the “characteristic exponent”) is the index
of stability of the distribution. The value of this exponent determines the shape of
the distribution: the smaller this exponent is, the fatter the tails are (extreme events
have a higher probability of occurring). In other words, the lower α is, the more often
extreme events are observed. In financial terms, this parameter is an indicator of risk,
since it describes how often substantial variations can occur. The parameter β, termed
“skewness,” provides information about the symmetry of the distribution. If β = 0, the
distribution is symmetric. If β < 0 it is skewed toward the left (totally asymmetric
toward the left if β = −1) while β > 0 indicates a distribution skewed towards the right
(totally asymmetric towards the left if β = −1). The parameter γ is the scale factor,
which can be any positive number. It refers to the “random size,” that is, the size of var-
iance whose regularity is given by the exponent α. This parameter describes the size of
the variations (whose regularity is given by α). Finally, the parameter δ is a localization
factor: it shifts the distribution right if δ > 0 and left if δ < 0.
Thanks to the value of these parameters, stable Lévy distribution provides a ge-
neral framework that makes it possible to characterize both the Gaussian distribution
(one that does not include extreme values) and a Paretian distribution (one that in-
cludes extreme values).

2.2.2.  The First Attempts to Model Extreme

Variations with Pareto-​Lévy Distributions
The first attempts to model extreme stock market price variations using stable Lévy
processes were the work of Mandelbrot. After graduating in France, he left for the
 

35  Extreme Values in Financial Economics

United States permanently in 1958 and began working for IBM. From the beginning
of his research, Mandelbrot aimed to extend statistical physics to other fields, in-
cluding social sciences. In his 1957 report, he set out his “interdisciplinary” project: “It
appears that the author’s activity has by now been durably oriented along the lines of
a long-​range program, having two aspects: the study of physical and non-​physical ap-
plications of thermodynamics, and the study of the ‘universal’ foundations” (1957, 4).
His approach was “to distinguish the part of thermodynamics, which is so general
that it could also conceivably apply to systems more general than sets of molecules”
(1957, 2). Mandelbrot started his ambitious program of investigations (nonphysical
applications of thermodynamics) with the structure of language, texts, and commu-
nication, moving on to distributions of species, and then to income distributions. In
this perspective, he proposed to expand Pareto’s work on income distributions by rep-
resenting these distributions by Lévy distribution in the late 1950s.16 In 1961, he was
invited to present his work on income distribution at a Harvard University seminar.
On this occasion, Mandelbrot entered Hendrik Houthakker’s office and came face to
face with a graphical representation of the distribution of changes in cotton prices that
was similar to the results he had obtained with incomes: “Houthakker’s cotton chart
looked like my income chart. The math was the same” (Mandelbrot 1963, 394 n. *).
This parallelism led him to apply the stable Lévy process to cotton prices, which was
the longest, most complete time series with daily quotations available at the time, and
then to stock price movements. He published a first article on the topic in 1962 and
gave lectures at Harvard in 1962, 1963, and 1964.
Mandelbrot’s starting point was to postulate an infinite variance. It is clear that in-
finite variance cannot be observed in the real world:

In practice, it is of course impossible to ever attain any limit corresponding to an infinitely

large T, or to ever verify any condition relative to an infinitely large u. Hence, in practice, one
must consider that a month or a year are [sic] infinitely long, and that the largest observed
daily changes of log Z(t) are infinitely large.” (Mandelbrot 1962b, 28)

Mandelbrot justified his postulate by the need to distinguish physical infinity,

which cannot be observed, from mathematical infinity, which can be postulated
to model real-​world phenomena: “Let me begin by reassuring you that an infinite
variance is not the consequence of my mistakenly taking the logarithm of zero. The
concept of infinity, as used in engineering and physics, is not the same as the math-
ematician’s infinity. … there is a range of cases where a variance that is really finite
can be safely considered infinite. … I  was led to postulate an ‘infinite variance’
as the simplest way of mathematically describing the outliers, and of emphasizing
their very strong contribution and their very erratic behavior from one year to the
next” (Mandelbrot 1966b, 2).
Since we cannot observe infinite variance in practice, we can conceive it by as-
suming that time is infinitely divisible and that this infinite divisibility does not affect
the properties of a distribution. In other words, the distribution of a random process
must be stable (or “invariant”).17 For stock prices, this means that the distribution of
 

36  Econophysics and Financial Economics

the result of changes in prices of a given frequency (hour, day, week, etc.) remains the
same regardless of the length of time considered (one day, several weeks, one year) or,
in other words, whatever the number of observations. Thus, thanks to the stability of
probability distributions, a daily observation can be thought of as 24 hourly observa-
tions or 1,440 per-​minute observations or even 86,400 per-​second observations, and
so on. By conceptualizing the infinite divisibility of time, one can give real meaning to
the mathematical hypothesis of infinity.
In sum, drawing on Lévy’s work (1924) on the stability of probability distribu-
tions and that of Gnedenko and Kolmogorov (1954)18 on the generalization of the
central-​limit theorem, Mandelbrot used a Pareto-​Lévy distribution for two reasons
(Mandelbrot 1960, 80):

1. It is a possible limit distribution with infinite variance of the sums of random in-
dependent and identically distributed variables (i.e., generalized central-​limit
theorem).
2. It is stable (i.e., it is time invariant) in its Pareto form.

Mandelbrot pointed out several advantages of using stable Pareto-​Lévy processes

with infinite variance. First, these processes can characterize the leptokurticity of stock
price or return distributions. Second, they can explain some dependence in stock
price variations, particularly the “dependencies” detected by Alexander (1961). Third,
they can explain some breaks in the structure of stock price variations (the Noah
effect). Fourth, they can be used even though the Gaussian distribution seems to rep-
resent some empirical observations quite well. This last advantage would have been
of interest to most financial economists during the 1960s and the 1970s. Because the
Gaussian distribution is the only stable distribution with a finite variance, Mandelbrot
necessarily had to reject it. However, because most financial economists of the time
used the Gaussian distribution, Mandelbrot took time to explain why it is a specific
case of the Pareto-​Lévy distribution. In particular, he explained that while, for a small
sample, the observed distribution can be approximated by a Gaussian distribution,
extending the sample will necessitate dropping it in favor of a Pareto-​Lévy distribu-
tion. In other words, from Mandelbrot’s perspective, as the amount of stock market
data increased, empirical validation of the Pareto-​Lévy distribution becomes essential.
The lack of data at that time made the approach adopted by Mandelbrot prospective.
However, he had a clear theory: a growing amount of data would necessarily cause a
sample to converge toward a Pareto-​Lévy distribution.19 We will return to this predic-
tion in ­chapter 4.
Fama (1963) was the first author to attempt to connect Mandelbrot’s work with
the core of financial economics. In his first paper, Fama presented the statistical prop-
erties of Pareto-​Lévy processes from an economic point of view. Two years later, he de-
veloped a general model for portfolio analysis in a market where stock returns can be
described with a Pareto-​Lévy distribution. While he gave a mathematical reinterpreta-
tion of the modern portfolio theory developed by Markowitz (together with Sharpe’s
diagonal model), he was unable to provide a complete economic interpretation
 

37  Extreme Values in Financial Economics

of his work because a clear computational definition of all parameters did not exist
(Fama 1965b, 405). However, in his article, Fama aimed to give an economic mean-
ing to diversification in a Pareto-​Lévy framework, even if the parameter related to the
dispersion of the return on the portfolio could no longer be described by the vari-
ance. From this perspective, he observed that a general increase in diversification
has a direct impact on the evolution of the scale factor (γ) of the stable distribution.
More precisely, increasing diversification reduces the scale of the distribution of the
return on a portfolio (only when the characteristic exponent α > 1).20 Fama proposed,
therefore, to replace variance with the scale factor (γ) of a stable distribution in order
to approximate the dispersion of financial distributions. It is worth mentioning that
Fama’s work on the stable Lévy framework in the 1960s does not conflict with his
efficient-​market hypothesis (which we discussed in c­ hapter  1), because this hypo-
thesis assumes the perfect integration of information in financial prices with no partic-
ular hypothesis regarding the statistical process describing the evolution of financial
returns. Basically, the only necessary condition for a statistical process to be consistent
with the efficient-​market hypothesis is the use of independent variables (and by exten-
sion i.i.d. variables), ensuring that past information is not necessary for forecasting the
future evolution of variables. This is the case for stable Lévy processes.
In the same vein, Samuelson (1967) provided an “efficient portfolio selection for
Pareto-​Lévy investments” in which the scale factor (γ) was used as an approximation
of variance (because the scale parameter is proportional to the mean absolute devia-
tion). Samuelson presented the computation of the efficiency frontier as a problem of
nonlinear programming solvable by Kuhn-​Tucker techniques. However, even though
he demonstrated the theoretical possibility of finding an optimal solution for a stable
Lévy distribution, Samuelson gave no example or application of his technique. Other
economists followed the path opened by Fama and Mandelbrot toward stable Lévy
processes: Fama and Roll (1968, 1971), Blattberg and Sargent (1971), Teichmoeller
(1971), Clark (1973), and Brenner (1974) were among them. Moreover, following
the publications by Mandelbrot and Fama, the hypothesis of the Pareto-​Lévy pro-
cesses was frequently tested during the 1960s and the 1970s (Brada, Ernst, and van
Tassel 1966; Godfrey, Granger, and Morgenstern 1964; Officer 1972) as explained in
the following section.

2.3.  MATHEMATICAL TREATMENTS OF EXTREME

VALUES BY FINANCIAL ECONOMISTS
2.3.1.  Difficulties Encountered with Stable
Lévy Processes in the 1960s and 1970s
While research into Pareto-​Lévy processes in financial economics began to gather mo-
mentum in the 1960s, the subject rapidly lost its importance among financial econo-
mists, who identified four major limitations on the use of stable Lévy processes.
First, empirical tests did not validate with certainty either infinite variance or the
Pareto-​Lévy processes. Not everyone found infinite variance a convincing postulate.
 

38  Econophysics and Financial Economics

According to Godfrey, Granger, and Morgenstern, “No evidence was found in any of
these series [of the NYSE prices] that the process by which they were generated be-
haved as if it possessed an infinite variance” (1964, 6). Brada, Ernst, and van Tassel
observed that “the distributions [of the stock price differences], in all cases, are ex-
cessively peaked. However, contrary to The Stable Paretian hypothesis, there are no
fat tails” (Brada, Ernst, and van Tassel 1966, 337). Officer (1972) wrote that stock
markets have some characteristics of a non-​Gaussian process but also emphasized
that there was a tendency for sums of daily stock returns to become “thinner-​tailed”
for large sums, even though he acknowledged that the normal distribution did not
approximate the distribution of his sample. Blattberg and Gonedes (1974) showed
that while stable Pareto-​Lévy distributions had better descriptive properties than the
Gaussian one, the Student distribution21 fit the data better. Even Mandelbrot’s stu-
dent, Fama, reached a similar conclusion: “Distributions of monthly returns are closer
to normal than distributions of daily returns. This finding was first discussed in detail
in Officer 1971, and then in Blattberg and Gonedes 1974. This is inconsistent with the
hypothesis that return distributions are non-​normal symmetric stable, which implies
that distributions of daily and monthly returns should have about the same degree of
leptokurtosis” (Fama 1976, 38).
Nor did stability of the distribution, which is one of the major statistical proper-
ties of stable Lévy processes, appear certain.22 In the 1950s, Kendall (1953, 15) had
already noticed that variance was not always stationary. Investigations carried out
in the 1960s and 1970s seemed to confirm this point. Officer (1972) explained that
the sum of independently distributed random variables from a stable process did not
give a stable distribution with the same characteristic exponent, as required by the
stability property.23 Financial data “have some but not all properties of a stable pro-
cess,” and since several “inconsistencies with the stable hypothesis were observed,”
the evolution of financial markets cannot be described through a stable Lévy process
(Officer 1972, 811). Even Fama seemed to share this view: “Contrary to the implica-
tions of the hypothesis that daily and monthly returns conform roughly to the same
type of stable non-​normal distribution, monthly returns have distributions closer to
normal than daily returns” (Fama 1976, 33). Upton and Shannon (1979) also con-
firmed that stable processes were not appropriate for the analysis of empirical of time
series.24
The second limitation to the use of stable Lévy processes in finance was the fact
that, until the end of the 1970s, the theory of probability related to stable Lévy pro-
cesses was still unformed (Nolan 2009). Several developments were necessary in
order to apply these processes to the study of stock market fluctuations and to test
them adequately. While Mandelbrot opened up a stimulating avenue for research, the
formulation of the mathematical hypothesis (i.e., the Pareto-​Lévy process) was not
immediately followed by the development of statistical tools. A direct result of this un-
formed knowledge on stable Lévy processes was the absence, at the time, of statistical
tests to check the robustness of results provided by these processes. Therefore, authors
focused on visual tests to validate the hypothesis of non-​Gaussian distribution. This
situation is very similar to the first empirical work that suggested the random character
 

39  Extreme Values in Financial Economics

of stock prices before the creation of financial economics (­chapter  1). However,
without appropriate statistical tests, it is simply impossible to evaluate to what extent
these outcomes are statistically reliable (­chapters 4 and 5 will come back to this point).
Available statistical tests at the time either were constructed for the Gaussian distri-
bution or assumed normality: examples are the Pearson test (also known as the chi-​
square test), which was used to test the compliance of the observed distribution with
a theoretical distribution; and Student’s t-​test, which was used for comparing param-
eters such as the mean and for estimating the parameters of a population from data
on a sample. Similarly, as Mandelbrot (1962, 35) explained, methods based upon the
minimization of the sum of squares of sample deviations, or upon the minimization
of the expected value of the square of the extrapolation error, cannot be reasonably
used for non-​Gaussian Lévy processes. This was true for the test of the distribution
as well as for the other components of the hard core of financial economics, as Fama
pointed out:

“… There are admittedly difficult problems involved in applying [portfolio models with
stable Paretian distributions] to practical situations. Most of these difficulties are due to the
fact that economic models involving stable Paretian generating processes have developed
more rapidly than the statistical theory of stable Paretian distributions.” (Fama 1965b, 418)

For this reason, authors such as Mandelbrot and Fama insisted on the need “to de-
velop more adequate statistical tools for dealing with stable Paretian distributions”
(Fama 1965b, 429). Even in 1976, 14 years after Mandelbrot’s first publication, Fama
considered that “statistical tools for handling data from non-​normal stable distribu-
tions are [still] primitive relative to the tools that are available to handle data from
normal distribution” (Fama 1976, 26). It is worth keeping in mind the importance
that statistical tests hold for financial economists since they founded their approach
on the use of such tests, in opposition to the visual techniques used by chartists or the
accounting-​based analysis adopted by fundamentalists.
The third reason for explaining the meager use of stable Lévy processes in finance
in the 1970s refers to the difficulty of estimating their parameters. Indeed, the use of
these processes requires the identification of the four parameters that define the stable
distribution. This is done by using parameterization techniques. Unfortunately, pa-
rameterization techniques were nonexistent in the 1960s and still in their infancy in
the 1970s (Borak, Hardle, and Weron 2005). As Fama explained,

The acceptability of the stable Paretian hypothesis will be improved not only by further
empirical documentation of its applicability but also by making the distributions them-
selves more tractable from a statistical point of view. At the moment very little is known
about the sampling behavior of procedures for estimating the parameters of these distribu-
tions. (1963, 428–​29)

For instance, Fama (1965b, 414) emphasized that the application of stable Lévy pro-
cesses in practical situations is very complicated because “the difficulty at this stage
is that the scale parameter of a stable Paretian distribution is, for most values of α, a
 

40  Econophysics and Financial Economics

theoretical concept.” That is, the mathematical statistics of stable Paretian distribution
was not yet sufficiently developed to give operational or computational meaning to γ
in all cases. The first estimation for symmetric stable Lévy processes was proposed by
Fama and Roll (1968), who, for simplicity, assumed symmetry of stable distributions,
meaning that three parameters could be equal to zero (μ = 0, β = 0, γ = 0), while α was
given by the following approximate formula:

 n x 
α = 1 + n  ∑ ln i  , (2.3)
 i=1 x min  

where xi are the quantiles-​based measures and x min is the minimum value of x. This
formula is an approximation, because the parameterization of factors depends on the
size of the sample. It is worth mentioning that this parameterization technique25 was
the only one available at the end of the 1960s. Because results given by quantiles-​
based methods depend directly on the size of the sample, Press (1972) proposed
a second method based on their characteristic function. Given that the generalized
central-​limit theorem requires a huge (theoretically an infinite) amount of data, Press
combined observations from the empirical samples with extrapolations made from
these data. He was therefore able to generate a large quantity of data.26
All the methods that emerged in the 1970s for parameterizing these stable Lévy
processes had one serious drawback:  while quantiles-​based (called nonparametric)
methods directly depended on the size of the sample, the characteristic-​function-​
based (called spectral) methods depended on the extrapolation technique used to
minimize the difference between empirical results given by the sample and theoretical
results. The impossibility of parameterizing stable Lévy processes explained why fi-
nancial economists were not inclined to use them.
Finally, the fourth barrier to the use of stable Lévy processes in finance concerns the
infinity of variance, which has no theoretical interpretation congruent with the theo-
retical framework of financial economics. As discussed in ­chapter 1, variance and the
expected mean are two of the main variables for the theoretical interpretations of finan-
cial economists, which associated risk with variance and return with the mean.27 From
this perspective, if variance is infinite, it is impossible to understand the notion of risk
as financial economists define it. In other words, the statistical characteristics of stable
Lévy processes could not be integrated into the theoretical framework of financial ec-
onomics, which provided significant operational results. Financial economists focused
on statistical solutions that were theoretically compatible with the Gaussian framework.
Fama and Roll (1971, 337) emphasized this difficulty of working with a non-​Gaussian
framework in the social sciences, where “economists, psychologists and sociologists fre-
quently dismiss non-​normal distributions as data models because they cannot believe
that processes generating prices, breakdowns, or riots, can fail to have second moments.”
It is worth mentioning that, in the beginning of the 1970s, financial economics
was a young field trying to assert its ability to provide scientific reasoning about finan-
cial reality, and financial economists did not necessarily want to deal with a scientific
 

41╇ Extreme Values in Financial Economics

puzzle (the interpretation of infinite variance) that could discredit the scientific rep-
utation of their emerging field. Knowing these limitations, and because the Gaussian
framework had allowed the development of financial economics while achieving re-
markable theoretical and practical results (Â�chapter  1), several authors,28 including
Fama, proposed acceptance of the Gaussian framework as a good approximation:

Although the evidence also suggests that distributions of monthly returns are slightly lepto-
kurtic relative to normal distributions, let us tentatively accept the normal model as a work-
ing approximation for monthly returns…â•›. If the model does well on this score [how well
it describes observed relationships between average returns and risk], we can live with the
small observed departures from normality in monthly returns, at least until better models
come along. (Fama 1976, 38)

Fama added:

Although most of the models of the theory of finance can be developed from the assumption
of stable non-╉normal return distributions … the cost of rejecting normality for securities
returns in favor of stable non-╉normal distributions are substantial.” (Fama 1976, 26)

In other words, at the end of the 1970s, Mandelbrot’s hypothesis could not yet be
applied for modeling extreme values in financial markets. However, as Mandelbrot
(1962) had already pointed out, the four major limitations evoked above do not nec-
essarily contradict the Pareto-╉Lévy framework; they simply made it impossible to
use this approach in the 1970s. This conclusion can also be drawn from the paper by
Officer (1972), who mentioned that his sample was not large enough to justify the use
of a nonnormal distribution, as observed by Borak et al. (2005, 13). Parameterization
techniques also directly depends on the size of the sample used in the statistical a-
nalysis. The bigger the sample, the better the estimation of the parameters. Despite
such conclusions, research on integrating the leptokurtic character of distributions
was continued by financial economists, who developed alternative models in order to
describe the occurrence of extreme values.

2.3.2.╇ Alternative Paths Explored to Deal with Extreme Values

Financial economists have taken extreme variations into account without jettisoning
the Gaussian framework. Two methods have been explored giving rise to two litera-
tures, thereby allowing the Gaussian approach to persist in financial economics: jump-╉
diffusion and ARCH (autoregressive conditional heteroskedasticity)-╉type models.
These two types of models arose from two different methodologies: jump-╉diffusion
models characterize extreme values by combining the statistical properties of different
distributions, while ARCH models deal with extreme values by modeling the residues
observed in the Gaussian framework. Technically, the former propose a mathematical
solution for the analysis of extreme values, while the latter propose econometric pro-
cessing of these values.
 

42╇ Econophysics and Financial Economics

2.3.2.1.╇ Jump-╉Diffusion Processes

Jump-╉diffusion processes were the first class of models developed by financial econo-
mists to take extreme variations into account. These models start from the premise that
increments of process are independent but not identically distributed. They attempt
to reproduce empirical observations by breaking stock price variations into frequent
variations of small amplitude and rare variations of very great amplitude. The lepto-
kurtic nature of price distributions is therefore a reflection of this double movement.
The response of these financial economists was the hypothesis that the observed
distribution of prices can be divided into two: a Gaussian and a non-╉Gaussian dis-
tribution. The non-╉Gaussian distribution (referring to variations) can be described
through any distribution for which the mean is finite, including a Pareto-╉Lévy distri-
bution.29 Conceptually, we can use the graph in figure 2.5 in order to summarize this
reasoning.

Leptokurtic distribution
Gaussian distribution
Probability

Jump part

–1.980 1.980
95 % of values
Values

–2.580 2.580
Probability of Cases in 99 % of values
portions of the curve
Heavy
Standard Deviations
–30 tails
From The Mean –40 –20 –10 0 +10 +20 +30 +40
Cumulative % 0.1 % 2.3 % 15.9 % 50 % 84.1 % 97.7 % 99.9 %

Gaussian part

Figure 2.5╇ Conceptual schematization of jump-╉diffusion processes for an infinite number of

observations: while the empirical data seem to follow a leptokurtic distribution, financial economists
described the leptokurticity by means of an improvement of the Gaussian framework that they
combined with other processes

The statistical description of a long-╉tails distribution can be decomposed into two

blocks:  a Gaussian description (solid line) of the middle part of the main part of
the distribution (the trend or the diffusion) and a “jump part” (associated with
the dotted line distribution) describing the heavy tails that are not in the Gaussian
regime. So doing, financial economists opened a window for modeling the occur-
rence of extreme variations on the financial markets. The three graphs in figure 2.6
�
respectively compare a Gaussian simulation characterizing the dynamics of financial
data with a jump process-╉based simulation and the empirical data recorded on the
financial markets.
 

(a) 0.3

0.2

0.1

–0.1

–0.2

–0.3

(b) 0.3

0.2

0.1

–0.1

–0.2

–0.3

(c) 0.3

0.2

0.1

0
Log return

–0.1

–0.2

–0.3

1970 1980 1990 2000

Year

Figure 2.6  (a) Gaussian simulation describing the dynamics of financial data

(b) Jump-​process-​based simulation based on a combination of the Gaussian distribution with a
Poisson law for the description of the jump part of the process
(c) Empirical evolution of financial data (time plot of daily log returns of IBM stock from July 1962 until
December 1998)
Source: Google Finance and authors.
 

44  Econophysics and Financial Economics

The first models, introduced by Press (1967), combined the normal distribution
with a Poisson law (figure 2.6b). Although the combination of these two distributions
does not describe accurately the empirical data (figure 2.6c), it offers, at least, a possi-
bility to deal with a dynamics exhibiting a high volatility (by opposition to the Gaussian
distribution reproduced on figure 2.6a). Press was a PhD student in the GBS of Chicago
at the same time as Fama and opened debates about Mandelbrot’s hypothesis (evoked
in the previous section). He wanted to solve the issue of leptokurticity while keeping
finite variance, in line with the data he observed: “Sample evidence cogently reported
by Fama (1965a) supports the hypothesis of non-​zero mean, long-​tailed, peaked, non-​
Gaussian distributions for logged price changes. However, there is no need to conclude,
on the basis of this evidence, that the variance is infinite” (Press 1967, 319). Within
this framework, Press explained that the Gaussian distribution was not appropriate for
describing the empirical data.30 To address this shortfall, he built a model in which

the logged price changes are assumed to follow a distribution that is a Poisson mixture of
normal distributions. It is shown that the analytical characteristics of such a distribution
agree with what has been found empirically. That is, this distribution is in general skewed,
leptokurtic, [and] more peaked at its mean than the distribution of a comparable normal
variate. (Press 1967, 317)

From this perspective, the occurrence of extreme values is associated with jumps
whose probability follows a Poisson law:
N (t )
X t = C + ∑ Yk + Wt , (2.4)
k =1


where C is the initial price X(0), ( Y1 ,… Yk ) are independent random variables fol-
lowing a common law N (0, σt2 ), N(t) is a Poisson process counting the number of
random events, and Wt is a classical Weiner process.
Some years later, Merton, who introduced the term “jump,” popularized this ap-
proach by applying it to the analysis of option pricing, which is a major element in the
hard core of financial economics (­chapter 1). In his paper, Merton (1976) provided an
extension of Black, Scholes, and Merton’s 1973 option-​pricing model with a two-​block
model. The evolution of stock prices was therefore described by the following process:

St = S0 e Xt ,  (2.5)

where X t was characterized with a combination of Brownian motion and a Poisson

process to describe the occurrence of jumps, which took the following form:

Nt
X t = µt + σBt + ∑ Yi ,  (2.6)
i =1

where µ is the mean (drift) of the process, σ is the variance, and Bt is a Brownian
motion with Bt = Bt − B0 ~ N ( 0 ,t ) . {Yi } refers to the jumps block modeled with a
 

45╇ Extreme Values in Financial Economics

compound Poisson process with t ≥ 0. In his article, Merton explained the need to de-
velop a model “consistent with the general efficient market hypothesis of Fama (1970)
and Samuelson (1965b)” (Merton 1976, 128) by taking “the natural process for the
continuous component of the stock-╉price change: the Wiener process” (128). In other
words, Merton founded his reasoning on the Gaussian framework since the Wiener
process is also called Brownian motion (Doob 1953).
The methodology used by Press and Merton opened the path to a vast literature
on what are now called “jump-╉process models.” A large category of models were de-
veloped, using different kinds of statistical processes.31 From this category of models
emerged another kind of model, called “pure-╉jump processes,” introduced by Cox and
Ross in 1976. They do not describe the occurrence of jumps through a double-╉block
model but rather through a single block in which a high rate of arrival of jumps of dif-
ferent sizes obviates the need to use a diffusion component.32 Over two decades, these
processes have generated a prolific literature in which the number of models is impor-
tant: the generalized hyperbolic distribution (Eberlein and Keller 1995), the variance
gamma model (Madan, Carr, and Chang 1998), and the CGMY process (named after
the authors of Carr, Geman, Madan, and Yor 2002).33
This category of models allowed financial economics to characterize extreme values in
an improved Gaussian framework even though they are by definition nonstable. Indeed,
all these pure-╉jump processes are related to Brownian motion because of its properties
of normality and continuity (Geman 2002). In other terms, the randomness in opera-
tional time is supposed to generate randomness in the evolution of financial returns. By
associating the process describing the evolution of financial returns with the evolution
of volume characterized by Brownian motion, Clark (1973) showed that the variance
characterizing the extreme values observed in the dynamics of financial returns is always
finite. Consequently, pure-╉jump models offer very technical solutions for describing the
occurrence of extremes values while maintaining an improved Gaussian framework.

2.3.2.2.╇ ARCH-╉Type Models
The second category of financial economic models that take into account extreme var-
iations is ARCH-╉type models.
As we have seen, some of the first empirical investigations during the 1960s and 1970s
pointed out that the variance of stock price fluctuations appeared not to be stable over
time and that some dependence in price variations seemed to exist. Mandelbrot (1963,
1962)  had already discussed this empirical result and had suggested in 1966 (as did
Samuelson 1965) replacing Brownian motion with a martingale model (see Â�chapter 1).
This is because martingale models, unlike Gaussian processes, allow a process with a var-
iance that is not independent. The first exhaustive empirical study of this kind of depend-
ence was by McNees in 1979, who showed that variance “associated with different forecast
periods seems to vary widely over time” (1979, 52). To solve this problem, Engle (1982)
introduced ARCH-╉types models based on statistical processes whose variance directly de-
pends on past information. Within this framework, variance is considered a random var-
iable with its own distribution that can be estimated through a defined average of its past
values. Such models can therefore reproduce extreme variations, as Â�figure 2.7 suggests.
 

(a) 0.3

0.2

0.1

–0.1

–0.2

–0.3
(b) 0.3

0.2

0.1

–0.1

–0.2

–0.3

(c) 0.3

0.2

0.1

0
Log return

–0.1

–0.2

–0.3

1970 1980 1990 2000

Year

Figure 2.7  (a) Gaussian simulation describing the dynamics of financial data

(b) ARCH model-​based simulation characterizing the dynamics of financial data
(c) Empirical evolution of financial data (time plot of daily log returns of IBM stock from July 1962
until December 1998)
Source: Google Finance and authors.
 

47  Extreme Values in Financial Economics

These representations indicate that an ARCH-​based simulation provides a better

estimate of the dynamics ruling the evolution of financial data. However, figure 2.7b
also shows that this kind of model does not capture the occurrence of large varia-
tions that sporadically appear in reality (figure 2.7c). While ARCH-​type models are
based on a non-​Gaussian property (time dependence), financial economists aimed
to describe the occurrence of extreme values in a Gaussian perspective. Indeed, they
still postulate that financial returns follow a Gaussian framework. The distribution of
these returns is therefore considered as granted (for this reason it is called uncondi-
tional distribution). The variability of this unconditional distribution can be described
through a new distribution (called conditional distribution) derived from specific
time-​dependent dynamics.34 From a statistical point of view, it is important to empha-
size that only the variable referring to the variance is not independent and identically
distributed. In other words, variables (financial returns) associated with the uncon-
ditional distribution are still assumed to have a Gaussian behavior. More formally,
the evolution of financial returns is described through the following unconditional
normal law:

X t =N ( µ + σt2 ) + ε t ,  (2.7)

where μ is the mean, σ² is the variance, and E is the statistical error. This statistical
equation can also be expressed as

X t =µt + σt2 + ε t .  (2.8)

ARCH-​type models focus on the variance and statistical errors that they decompose
into an unconditional and conditional part. The first ARCH model was introduced by
Engle (1982), who characterized the variance as follows:

1 n 2
σt2 = ασ + (1 − α ) ∑ Ri . 
n i=1
(2.9)

The unconditional dimension of variance is defined by the statistical parameter

α > 0 and the long-​term variance σ, while the conditional part is characterized by the
Si − St −1
weighted sum of the last n returns, Ri2 = . This model was improved through
Si
GARCH (generalized autoregressive conditional heteroscedasticity) models by
Bollerslev (1986), who showed that all last n returns did not influence the current
variance in the same way, by using an exponentially weighted moving average esti-
mate in which greater weight is assigned to the more recent returns. Statistically, this
dependence of conditional variance on the past (which refers to the distribution of
statistical errors or innovation in statistical terms) can be characterized using various
potential statistical processes (Kim et al. 2008) that generated a huge literature pro-
viding a variety of time-​dependence dynamics. Depending on this specific dynamic,
one finds in the literature several types of ARCH models (IGARCH, EGARCH,
 

48╇ Econophysics and Financial Economics

GARCH, NGARCH, etc.).35 Moreover, notice that, in line with the literature on jump
processes, ARCH models can describe the occurrence of extreme variations within a
Gaussian framework.

2.4.╇CONCLUSION
This chapter showed that financial economists have considered extreme variations in
the analysis of stock prices/╉returns by using three statistical approaches: Pareto-╉Lévy,
jump diffusion, and ARCH types. In the 1960s, the Pareto-╉Lévy model was the first
alternative proposed for characterizing the occurrence of extreme values in finance.
However, as explained, this path of research was not investigated further for tech-
nical and theoretical reasons. Consequently, since the beginning of the 1980s, only
the jump-╉diffusion and the ARCH-╉types models have been developed in financial ec-
onomics. As discussed, these two approaches are totally embedded in the Gaussian
framework. This point is important because Â�chapter 1 showed that the latter defines
the foundations of financial economics (modern portfolio theory, the capital-╉asset
pricing model, the arbitrage pricing theory, the efficient-╉market hypothesis, and the
Black and Scholes model).
It is interesting to note that the Gaussian framework was also shared by alternative
approaches that emerged in financial economics. For instance, extreme-╉values analysis
was the starting point for one of the major theoretical alternatives developed in finan-
cial economics: behavioral finance. In 1981, Shiller showed that market volatility was
too high to be compatible with a Gaussian process and that, therefore, it could not de-
scribe the occurrence of extremes values. However, although behavioral finance offers
a psychological explanation for the occurrence of extreme values, it does not question
the Gaussian statistical framework (Schinckus 2009).
The first two chapters highlighted the foundations of financial economics and how
this field deals with the extreme variations. Similarly, the two following chapters will
introduce econophysics and investigate its foundations.
 

3
N E W TO O L S F O R   E X T R E M E -​V A LU E A N A LY S I S
STAT I ST I C A L P H Y S I CS G O E S B E YO N D I TS B O R D E R S

The previous chapter studied how financial economists deal with extreme values
given the constraints imposed by their theoretical and methodological framework, the
major constraint being variance as a measure of risk. In this context, we explained that
stable Lévy distributions, which generate an infinite variance, considerably compli-
cate their application in finance. Surprisingly, since the 1990s statistical physicists have
found a way to apply stochastic models based on stable Lévy distributions to deal with
extreme values in finance. Many financial economists have found the incursion from
statistical physics for the study of finance difficult to accept. How can this reluctance
be explained? It appears that the major reason comes from the divergence in terms of
scientific criteria, and a way of “doing science” that differs from financial economists’
practices. Moreover, other differences can be mentioned, including approach, meth-
odology, and concepts used. In the same vein, the core mathematics used in econo-
physics models are still not easily accessible for noninitiated scholars.
The aim of this chapter is to outline the theoretical and methodological foundations
of statistical physics, which led some physicists to apply their approach and models to
finance. This clarification is a crucial step in the understanding of econophysics from
an economic point of view. This chapter will trace the major developments in statis-
tical physics, explaining why physicists extended their new methods out of their field.
This extension is justified by four key physical notions: critical points, universality
class, renormalization group theory, and the Ising model. These elements will be clari-
fied through an intuitive presentation. We will then explain how the combination of
these key notions leads statistical physicists to consider their framework as the most
appropriate one to describe the occurrence of extreme values in some phenomena. In
finance, econophysicists describe the observation of extreme variations with a power
law. This chapter will show the connections between these power laws and the current
knowledge in financial economics. As c­ hapter 2 explained, financial economists aban-
doned stable Lévy processes due to a technical issue related to the infinity of variance.
This chapter will explain how physicists solve this problem by introducing truncation
techniques. Their original motivation was to make stable Lévy distributions physically
plausible, simply because all physical systems have finite parameters and therefore
finite measures. To sum up, this chapter studies the theoretical tools developed since
the 1970s in statistical physics that led to the creation of econophysics.

49
 

50╇ Econophysics and Financial Economics

3.1.╇ WHY DID STATISTICAL PHYSICS

GO BEYOND ITS BORDERS?
For most economists, including most financial economists, it is not perfectly clear why
statistical physicists find it obvious that their recent models can be applied to financial
markets or other areas of economics.
The influence of physics on economics—╉and on social science in general—╉is
nothing new. A number of writers studied the “physical attraction”1 exerted by eco-
nomics on hard sciences: Mirowski (1989) extensively highlighted contributions of
physics to the development of marginalist economics and mathematical economics.
Ingrao and Israel (1990) drew renewed attention to the influences of mechanics
in the conceptualization of equilibrium in economics. Ménard (1981), Schabas
(1990), and Maas (2005) also highlighted the role of physics in the economic works
of Cournot and of Jevons. Similarly, we saw in Â�chapters 1 and 2 that financial eco-
nomics, and more generally finance, is also subject to the influence of physics. Yet
despite these theoretical and historical links, econophysics is a fundamentally new
approach. Its practitioners are not economists taking their inspiration from the work
of physicists to develop their discipline, as has been seen repeatedly in the history of
economics. This time, it is physicists who are going beyond the boundaries of their
discipline, using their methods and models to study various problems thrown up by
economics.
The current movement involves statistical physics. It is rooted in changes that oc-
curred in physics but also in other scientific disciplines, including social sciences, in
the 1970s. We will analyze these changes in order to understand the origins and the
theoretical foundations of econophysics.

3.1.1.╇ Statistical Physics’ Golden Age

A turning point in the recent history of physics took place in the 1970s: it concerns
the new connection between the theories of statistical mechanics—╉also called sta-
tistical physics—╉and particle physics. Statistical physics’ main purpose is to explain
the macroscopic behavior of a system and its evolution, in terms of physical laws
governing the motion of the microscopic constituents (atoms, electrons, ions, spins,
etc.) that make it up. Statistical physics (which is the contemporary label used to
characterize thermodynamics) distinguishes itself from other fields of physics by its
methodology based on statistics. This is due to the enormous number of variables
with which statistical physicists have to work. As Fitzpatrick (2012)2 explains, in
areas of physics other than thermodynamics, physicists are able to formulate some
exact, or nearly exact, set of equations—╉resulting from physical laws and theories—╉
that govern the system under investigation. Therefore, they are able to analyze the
system by solving these equations, either exactly or approximately. In thermody-
namics, physicists have no problem in formulating the governing equations and
writing down the exact laws of motion, including all the interatomic forces. Their
 

51  New Tools for Extreme-Value Analysis

problem is the gigantic number of variables—​as many as Avogadro’s number, 6 ×

1023—​and therefore the gigantic number of equations of motion that have to be
resolved.3 This enormous number of relationships makes an analysis strictly based
on equations unworkable, even for a computer. “Quite plainly, this is impossible
[to solve the system of equations] … [the] subject is so difficult that [physicists]
are forced to adopt a radically different approach to that employed in other areas of
physics” (Fitzpatrick 2012, 4).
Fortunately, physicists are not interested in knowing the position and ve-
locity of every particle in a system at any time. Instead, they want to know the
properties of the system (volume, temperature, density distribution, etc.) at a
given moment. Therefore, the number of pieces of information they require is
minuscule compared to the number of pieces of information that would be
needed to completely specify the internal motion of all single elements. In
this perspective, the quantities that physicists are interested in do not depend
on the motions of individual particles. They rather depend on the average
motions of all the particles in the system combined with the correlations between
these movements in a spatiotemporal description. In other words, these quanti-
ties depend on the statistical properties of each particle’s motion. Moreover, the
gigantic quantity of data makes it possible to use most statistical laws and theo-
rems generally founded on asymptotic calculus (Batterman 2002).4 The methods
used in statistical physics are thus essentially dictated by the complexity of the
systems due to the enormous number of constituents. This situation leads sta-
tistical physicists to start with statistical information about the motions of the
microconstituent properties of the system in order to infer statistically some
macro properties for this system. The statistical approach is so common that
“in most situations physicists can forget that the results are statistical at all, and
treat them as exact laws of physics” (Fitzpatrick 2012, 6).5 The turning point
that occurred in the 1970s is a direct result of this problematic of extremely
voluminous data.
In 1982, the (high-​energy or elementary-​particle-​trained) physicist Kenneth
Wilson received the Nobel Prize in Physics for his contribution to the connection
between macroscopic and microscopic levels. More precisely, Wilson was awarded
the prize for having developed the renormalization group theory for critical phe-
nomena in connection with phase transitions.6 The systematic study of such crit-
ical phenomena emerged in the 1960s when physicists observed the emergence
of macroregularities in the evolution of complex systems. Before going further, it
is worth explaining what physicists mean by “critical phenomena.” This concept is
used to describe systems whose configuration evolves through a dynamics of crit-
ical states. A critical state is a particular configuration of the system in which two
phases (or two states) are about to become one. The most telling example is water.
Water is commonly known to be liquid in a room condition, but when the temper-
ature or the pressure of this environment changes, the state of water changes as well
(figure 3.1).
 

52  Econophysics and Financial Economics

Pressure
solid phase

critical pressure
critical point
liquid
phase

triple point gaseous phase

vapour
critical
temperature

Temperature

Figure 3.1  Temperature-​pressure phase diagram for a fluid

Source: Adapted from Batterman 2002, 38. Reproduced by permission of Oxford University
Press, USA.

The transition of a state into another one is due to the gradual change of an external
variable (temperature or pressure); it is simply called “phase transition” in physics.
This transformation can be likened to the passage from one equilibrium (phase)7 to
another. When this passage occurs in a continuous way (for instance, a continuous
variation of temperature), the system passes through a critical point defined by a crit-
ical pressure and a critical temperature and at which neither of the two states is real-
ized (figure 3.1). This is a kind of nonstate situation with no real difference between
the two configurations of the phenomenon—​both gas and liquid coexist in a homoge-
nous phase. Indeed, physicists have observed that at the critical point, the liquid water,
before becoming a gas, becomes opalescent and is made up of liquid water droplets,
made up of a myriad of bubbles of steam, themselves made up of a myriad of droplets
of liquid water, and so on. This is called critical opalescence. In other words, at the crit-
ical point, the system appears the same at all scales of analysis. This property is called
“scale invariance,” which means that no matter how closely one looks, one sees the
same properties. In contrast, when this passage occurs in a discontinuous way (i.e.,
the system “jumps” from one state to another), there is no critical point. Phenomena
for which this passage is continuous are called critical phenomena (in reference to the
critical points).
Since the 1970s, critical phenomena have captured the attention of physicists due to
several important conceptual advances in the characterization of scale invariance through
the theory of renormalization,8 on the one hand, and to the very interesting properties
that define them, on the other. Among these properties, the fact that the dynamics of crit-
ical states can be characterized by a power law deserved special attention, because this law
is a key element in econophysics’ literature (the next part of this chapter will come back to
this point). As Serge Galam (2004) explains in his personal testimony, the years 1975–​80
appeared to be a buoyant period for statistical physics, which was blossoming with the
 

53  New Tools for Extreme-Value Analysis

exact solution of the enigma of critical phenomena, one of the toughest problems in phys-
ics. The so-​called modern theory of phase transitions, along with renormalization group
techniques, brought condensed-​matter physics into its golden age, leading hundreds of
young physicists to enter the field with a great deal of excitement.
As previously mentioned, Wilson won the Nobel Prize for his method of renor-
malization, used to demonstrate mathematically how phase transitions occur in crit-
ical phenomena. His approach provides a conceptual framework explaining critical
phenomena in terms of phase transitions and enabling exact resolutions.

The development of [the renormalization group] technique undoubtedly represents

the single most significant advance in the theory of critical phenomena and one of the
most significant in theoretical physics generally since the 1970s. (Alastair and Wallace
1989, 237)

The renormalization group theory has been applied in order to describe critical phe-
nomena. As explained above, the latter are characterized by the existence of a crit-
ical state in which the phenomenon shows the same properties independently of the
scale of analysis. The major idea of the renormalization group theory is to describe
mathematically these common features through a power-​law dependence. As we will
show, this theory played a key role in the extension of statistical physics to other social
sciences. Therefore, we will introduce this method in order to understand some of
the connections econophysicists make with finance.9 As mentioned above, the renor-
malization method deals with scale invariance. While the concept of invariance refers
to the observation of recurrent characteristics independently of the context, the
notion of scale invariance describes a particular property of a system/​object or law
that does not change when scales of length, energy, or other variables are multiplied
by a common factor. In other words, this idea of scale invariance means that one re-
current feature (or more than one) can be found at every level of analysis. Concretely,
this means that a macroscopic configuration can be described without describing all
microscopic details. This aspect is a key point in the renormalization theory devel-
oped by Wilson, who extended Widom’s (1965a, 1965b) and Kadanoff ’s (1966)
discovery of “the importance of the notion of scale invariance which lies behind all
renormalisation methods” (Lesne 1998, 25). More precisely, his method considers
each scale separately and progressively connects contiguous scales to one another.
This makes it possible to establish a connection between the microscopic and the
macroscopic levels by decreasing the number of interacting parts at the microscopic
level until one obtains the macroscopic level (ideally a system with one part only). In
this perspective, the idea of scaling invariance allows physicists to capture the essence
of a complex phenomenon by identifying key features that are not dependent on the
scale of analysis.
Consider a phenomenon whose combination of microcomponents can be
described with the sequence X = X1 + X 2 +  + X kn , composed of kn random in-
dependent variables and identically distributed by a stable Lévy distribution
(i.e., power law) such as that used in finance. The renormalization group method
 

54  Econophysics and Financial Economics

consists in using a scaling transformation to group the kn random variables into n

blocks of k random variables. The transformation Sn takes the sequence X into a
new sequence of random variables—​still independent and identically distributed
by a Lévy distribution and therefore still stable. This transformation becomes truly
fruitful when it is iterated, when each renormalization leads to a reduction in the
number of variables, leading to a system that contains fewer variables while keep-
ing the characteristics of the original system—​thanks to the fact that the system
stays independent, identically distributed, and stable.10 For instance, considering
the previous sequence X with kn = 8, n = 4, and k = 2, we can renormalize the se-
quence three times in order to obtain a single random variable that characterizes
the initial sequence (figure 3.2).

Figure 3.2  Renormalization group method applied to a stochastic process

Source: Sornette 2006, 53. Reproduced by permission of the author.

When applied several times, this pairing method allows modelers to “climb” the
scales by reducing the number of variables (kn) without losing key features of the phe-
nomena, which are captured in the scaling invariance of the process. In other words,
this technique allows us to group random variables into (n) blocks of variables in order
to reduce the initial complexity of the phenomenon. Roughly speaking, the technique
can be summarized by the equation
n
Sn ([ X ] , α ) = n − α ∑ X j ,
j =1

where Sn is the sequence at a specific scale of analysis describing the phenomena, while
Xj refers to the number variables used at that level of analysis. The α quantity is called
the “critical exponent” and describes the scale invariance of the phenomena. In other
terms, this exponent describes a universal property observed without regard for scale.
Considering the renormalization group method, the system at one scale is said to con-
sist of self-​similar copies of itself when viewed at a smaller scale, but with different
 

55  New Tools for Extreme-Value Analysis

(“rescaled”) parameters describing the components of the system. In other words, at

the end of the 1970s, statistical physics had established precise calculation methods
for analyzing phenomena characterized by scale invariance.
The scale-​invariance assumption was not new in physics,11 but the properties
allowing the mathematical demonstration of invariance were only established at the
end of the 1970s. This demonstration makes it possible to study mathematically
macroscopic regularities that occur as a result of microscopic random interactions
without having to study these microscopic interactions.12 The focus is therefore on
the macroscopic level, which is directly observable for physical phenomena. In other
words, since the 1970s, thanks to scale invariance, physicists can infer from micro-
scopic constituents some key parameters that allow the capture and description of
the dynamics of macroscopic behaviors without studying, in detail, what happens
at the microscopic level. For these reasons, scale invariance is the foundation of any
modern approach to statistical physics aimed at understanding the collective beha-
vior of systems with a large number of variables that interact with each other.
Since the end of the 1960s, research into critical phenomena and scale invariance
benefited from another very fruitful connection: the Ising model. The latter is a math-
ematical model of ferromagnetism used to study phase transitions and critical points.
This model is considered to be the simplest description of a system with a critical point;
it played a central role in the development of research into critical phenomena, and it
occupies a place of importance in the mind of econophysicists. Briefly, the Ising model
consists of discrete variables that represent magnetic moments of atomic spins that can
take one of two states, +1 (“up”) or −1 (“down”), the two states referring to the direc-
tion taken by the spins. The concept of spin characterizes the circular movement of
particles (electrons, positrons, protons, etc.), implying that they have a specific rotation
as shown in figure 3.3.

Figure 3.3  Schematization of a particle’s spin

There is no way to speed up or slow down the spin of an electron, but its direc-
tion can be changed, as modeled by the Ising model. The interesting element is
that the direction of one spin directly influences the direction of its neighbor spins
(figure 3.4).
 

56  Econophysics and Financial Economics

Figure 3.4  Schematization of the interaction between particles’ spins

This influence can be by captured through a function of correlation that measures to

what extent the behaviors of spins are correlated. The major idea of the Ising model
is to describe this interaction between particles’ spins. In this perspective, the spins
are arranged in a graph, usually a lattice, in which each spin exerts an influence on
its neighbors. This influence is measured by the distance over which the direction of
one spin affects the direction of its neighbor spins. This distance is called the correla-
tion length; it has an important function in the identification of critical phenomena.
Indeed, the correlation length measures the distance over which the behavior of one
microscopic variable is influenced by the behavior of another. Away from the critical
point (at low temperatures), the spins point in the same direction. In such as situation,
the thermal energy is too low to play a role; the direction of each spin depends only on
its immediate neighbors, making the correlation length finite (figure 3.5).

T < Tc

Subcritical

Figure 3.5  Two-​dimensional Ising model at low temperature

Source: Binney et al. 1992, 19. Reproduced by permission of Oxford University Press.

But at the critical point, when the temperature has been increased to reach the critical
point, the situation is completely different. The spins no longer point in the same di-
rection because thermal energy influences the whole system, and the magnetization
 

57  New Tools for Extreme-Value Analysis

spin-​spin vanishes. In this critical situation, spins point in no specific direction and
follow a stochastic distribution.

T ~ Tc

Critical

Figure 3.6  Two-​dimensional Ising model at the critical temperature.

Source: Binney et al. 1992, 19. Reproduced by permission of Oxford University Press.

As we can see in figure 3.6, there are regions of spin up (black areas) and regions of
spin down (white areas), but all these regions are speckled with smaller regions of
the opposite type, and so on. In fact, at the critical point, each spin is influenced by
all other spins (not only a particle’s neighbors) whatever their distance. This situation
is a particular configuration in which the correlation length13 is very important (it is
considered to be infinite). At this critical state, the whole system appears to be in a
homogeneous configuration characterized by an infinite correlation length between
spins, making the system scale invariant. Consequently, the spin system has the same
physical properties whatever the scale length considered. The renormalization group
method can then be applied, and by performing successive transformations of scales
on the original system, one can reduce the number of interacting spins and therefore
determine a solution from a finite cluster of spins.
Beyond the ability to describe the spins’ movement, there is another point of in-
terest in the Ising model. Because of its very simple structure, it is not confined to the
study of ferromagnetism. In fact, “Proposed as a model of ferromagnetism, it ‘pos-
sesses no ferromagnetic properties’ ” (Hughes 1999, 104)! Its abstract and general
structure has enabled its use to be extended to the study of many other problems or
phenomena:

The Ising model is employed in a variety of ways in the study of critical point phenomena.
To recapitulate, Ising proposed it … as a model of ferromagnetism; subsequently it has been
used to model, for example, liquid-​vapour transitions and the behaviour of binary alloys.
Each of these interpretations of the model is in terms of a specific example of critical point
behaviour… . [T]‌he model also casts light on critical point behaviour in general. Likewise,
the pictures generated by computer simulation of the model’s behaviour illustrate … the
whole field of scale-​invariant properties. (Hughes 1999, 124–​25)
 

58  Econophysics and Financial Economics

For these reasons, statistical physicists consider the Ising model the perfect illus-
tration of the simplest unifying mathematical model. Their looking for such models
is rooted in the scientific view of physicists, for whom “the assault on a problem of
interest traditionally begins (and sometimes ends) with an attempt to identify and
understand the simplest model exhibiting the same essential features as the phys-
ical problem in question” (Alastair and Wallace 1989, 237). The Ising model meets
this requirement perfectly. Its use is not restricted to statistical physics because “the
specification of the model has no specific physical content” (Hughes 1999, 99); its
content is mathematical. Therefore, this model is independent of the underlying phe-
nomenon studied, and it can be used to analyze any empirical data that share the same
characteristics.
With these new theoretical developments, statistical physicists had powerful math-
ematical models and methods that could solve crucial problems in physics. They were
able to establish the behavior of systems at their macroscopic level from hypotheses
about their microscopic level, but without analyzing this microscopic level. The com-
bination of the renormalization theory and the Ising model offers statistical physicists
a unified mathematical framework that can analogically be used for the study of phe-
nomena characterized by a large number of interacting microcomponents.

3.1.2.  The Temptation to Apply the Methods

of Statistical Physics outside Physics
Encouraged by the results obtained in the 1970s, certain physicists began investigat-
ing correspondences with collective behaviors of any kind of phenomena that appear
critical, including social phenomena. For statistical physicists, as we will see now, this
temptation to extend their models and methods outside physics seemed relevant.
First, we have to be aware of some features of physicists’ scientific viewpoint. As
we have just seen, physicists are very concerned by the identification of a simple uni-
fying mathematical model. The Ising model, which is primarily a mathematical entity
even if it has been developed for solving various types of problems in physics, is a per-
fect example of such a model. It can be used to solve a universality class of problems.
This concept of universality class is used in physics to describe a category of phe-
nomena that have the same behaviors regarding the dynamics of their critical states
(even though they refer to very different realities). A  common example of such a
class is the association between the occurrence of earthquakes and financial crashes,
the laws describing these events having the same statistical characteristics (we will
detail this point in section 3.2.3). Research on critical phenomena offered statistical
physicists the possibility of identifying universality classes of critical phenomena that
share the same statistical behavior at their critical point. In this case, systems that are
different microscopically have identical macroscopic behavior. Here again, the bed-
rock of universality classes is looking for the simplest unifying mathematical model.
To understand this point, recall that the dynamics of critical points are character-
ized by a power law. Physicists associate each power law with a mathematical model
 

59  New Tools for Extreme-Value Analysis

describing the statistical features of a large variety of phenomena that have the same
critical exponent. In other words, the idea of “universality” refers here to the fact that
these phenomena can be studied with the same mathematical model independently
of the context.
The use of critical phenomena analysis and its extension to social sciences suggest
changes in scientific methodology that developed in the twentieth century. Giorgio
Israel (1996) identifies a major change in the way of doing science through “mathe-
matical analogies.” These are based on the existence of unifying mathematical simple
models that are not dedicated to the phenomena studied. Mathematical modeling then
uses mathematical analogies by means of which the same mathematical formalism is
able to account for heterogeneous phenomena. The latter are “only interconnected
by an analogy that is expressed in the form of a common mathematical description”
(Israel 1996, 41). The model then is an effective reproduction of reality without on-
tology, one that may provide an explanation of phenomena. The Ising model is a per-
fect illustration of these simple unifying mathematical models. Israel (1996) stressed
that such mathematical analogies strongly contribute to the increasing mathematiza-
tion of reality.
Mathematical analogies illustrate the origin of the temptation for statistical physi-
cists to extend their models to analyzing critical phenomena beyond physics. First,
they looked for phenomena with large numbers of interacting units whose micro-
scopic behaviors would not be observed directly but which can generate observable
macroscopic results. These results are consistent with the microscopic motions de-
fined by a set of mathematical assumptions (which characterize random motion).14
Therefore, modelers can look for statistical regularities often characterized by
power laws:

Since economic systems are in fact comprised of a large number of interacting units having
the potential of displaying power-​law behaviour, it is perhaps not unreasonable to examine
economic phenomena within the conceptual framework of scaling and universality. (Stanley
and Plerou 2001, 563)

This search led some statistical physicists to create new fields that were called “so-
ciophysics” or “econophysics” depending on the topics to which their methods and
models were applied.
A first movement, sociophysics,15 emerged in the 1980s. One of the reasons that
physicists attempt to explain social phenomena stems from the mathematical power of
the new models borrowed from statistical physics:

During my research, I started to advocate the use of modern theory phase transitions to de-
scribe social, psychological, political and economical phenomena. My claim was motivated
by an analysis of some epistemological contradictions within physics. On the one hand, the
power of concepts and tools of statistical physics [was] enormous, and on the other hand,
I was expecting that physics would soon reach the limits of investigating inert matter. (Galam
2004, 50)
 

60╇ Econophysics and Financial Economics

In the 1990s statistical physicists turned their attention to economic phenomena, and
particularly finance, giving rise to econophysics. As two of its leading authors, Stanley
and Mantegna, put it, econophysics is “a quantitative approach using ideas, models,
conceptual and computational methods of statistical physics” (2000, 2). As men-
tioned above, these models and methods allow the study of macroscopic behaviors
of any kind whose components behave randomly. This point is interesting because it
echoes the efficient-╉market hypothesis (the cornerstone of financial economics, as ex-
plained in Â�chapter 1), in particular Fama’s 1970 formulation based on the assumption
of a representative agent. This version of the efficient-╉market hypothesis is perfectly
compatible with the statistical-╉physics approach that makes no hypothesis about spe-
cific behaviors of investors. Moreover, the renormalization group method seems to be
an appropriate answer to finance, because it makes it possible to move from the micro-
scopic level (i.e., agents whose behaviors are reduced to mathematical hypotheses) to
the macroscopic level (i.e., the evolution of the financial prices empirically observed).
Thus, by analogy, statistical physicists could consider the evolution of financial mar-
kets as the statistical and macroscopic results of a very large number of interactions at
the microscopic level. The application of statistical physics to economics is not limited
to finance; it touches on a number of other economic subjects.16 However, an analysis
of the articles published by econophysicists indicates that research conducted in this
field mainly concerns the study of financial markets, marginalizing other fields. It is no
accident, then, that econophysics is often associated with themes dealing with finan-
cial data (Rickles 2007, 4).

3.1.3.╇ The Computerization of Science

The temptation to apply statistical physics to economic phenomena has had support
from an important ally: the increasing use of computers in science and the acceler-
ation of the computerization of scientific activities. In �chapter 1 we mentioned the
importance of computers in the development of financial economics. However, their
role in econophysics is even greater, for two reasons at least. First, econophysicists are
data driven (�chapter 4). Second, the main statistical tool used, power laws, need an
enormous quantity of data to be identified due to their asymptotic properties (i.e.,
the properties of power laws are verified for extremely large samples, ideally infinite).
Thus, although partially independent of the theoretical developments that occurred
in statistical physics, the increasing use of computers has enabled two major features:
(1) the collection and the exploitation of very large databases; (2) the development
of a new scientific approach supported by the extension of statistical physics outside
its original borders.
As one can observe nowadays, computers have become a physical and intellec-
tual extension in the process of providing data about the world. Computerization
has provided an increasing quantity of data in a large number of fields, particularly
fields related to social phenomena. Financial markets occupy a very specific place in
this movement, because the financial databases are the largest bases for social phe-
nomena. Indeed, since the end of the 1970s, all the major financial markets have been
 

61  New Tools for Extreme-Value Analysis

progressively automated thanks to computers,17 making Fischer Black’s wish (1971a,

1971b) for fully automated stock exchanges come true. In addition, some markets, like
the foreign exchange market, became active 24 hours a day with electronic trading.
Automation has allowed all transactions and all prices quoted to be recorded, leading
to storage of a huge amount of financial data. Moreover, since the 1990s, use of com-
puters has enabled the development of high-​frequency transactions, and therefore the
creation of high-​frequency data (also called “intraday” data).18 Previously, statistical
data on financial markets were generally made up of a single value per day obtained
by the average price or the last quotation of the day (­chapter 2). In other words, the
“daily” data were based on averages, which, by construction, dilute the intraday vola-
tility, as figure 3.7 shows.

20
10 (a) S & P 500 (10 min data)
0
–10
–20
Price returns

10 (b) S & P 500 (monthly data)

0
–10
–20
10 (c) Gaussian noise
0
–10
–20
Time

Figure 3.7  Intraday and monthly price return volatility

Source: Gopikrishnan et al. 1999, 5306. Reproduced by permission of the authors.

While the first time series shows the evolution of 10 minutes of data, the second
one describes the evolution of the same data but recorded on a monthly interval.
These graphs show that larger time intervals reduce the volatility of data, making
them appear to more closely match the Gaussian framework (simulated in the third
time series). Nowadays, with the recording of “intraday data,” all prices quoted
and tens of thousands of transactions are conserved every single day (Engle and
Russell 2004).
The increasing quantity of data and the computerization of financial markets led to
notable changes in techniques for detecting new phenomena. Intraday data brought
to light new phenomena that could not be detected or did not exist with monthly or
daily data. Among these are strategic behaviors that influence price variations.19 More
importantly, the new data have exhibited values more extreme than could be detected
before. Indeed, monthly or daily prices recorded were generally the last prices quoted
during a month or a day, or a mean of the prices quoted during a period, and therefore
extreme values in these data are generally smaller and less frequent than in intraday
data. As the ­figure 3.7 suggests, intraday data have increased interest in research on
 

62  Econophysics and Financial Economics

extreme values. They have brought new challenges in the analysis of stock price varia-
tions that have required the creation of new statistical tools to characterize them.
The increased quantity of statistical data on financial markets favors the grow-
ing interest in extending the methods and models of statistical physics into finance.
Indeed, as previously mentioned, most of the results obtained in statistical physics are
based on huge amounts of data, which makes this question crucial:

Economics systems differ from often-​studied physical systems in that the number of subunits is
considerably smaller in contrast to macroscopic samples in physical systems that contain a huge
number of interacting subunits, as many as Avogadro’s number, 6 × 1023. In contrast, in an eco-
nomic system, one initial work was limited to analysing time series comprising of order of mag-
nitude 103 terms, and nowadays with high-​frequency data the standard, one may have 108 terms.
(Stanley and Plerou 2001, 563–​64)

The advent of intraday data has made it possible to build samples that are suffi-
ciently broad to provide evidence to support the application of power-​law distribution
analysis to the evolution of prices and returns on financial markets. This explosion of
financial data—​which has no equivalent in other social sciences fields—​comes closer
to the standards to which statistical physicists generally work, making market finance
“a natural area for physicists” (Gallegati et al. 2006, 1).
Computers have also transformed scientific research on distributions of stock
market variations. Their ability to perform calculations more rapidly than humans
paved the way for an analysis of new phenomena and the study of old phenomena in
new ways. In this context, old issues debated in the 1970s by financial economics can
be updated and re-​evaluated. This is particularly true for stable Lévy processes.20 As
explained in c­ hapter 2, in general there are no closed-​form formulas for Lévy α-​stable
distributions—​except for Gaussian, Pareto, and Cauchy distributions—​which made
their use in finance difficult in the 1970s. This point is still true today: to work on such
distributions and the associated processes, one has to calculate the value of the param-
eters, and such complex calculations require numerous data. Of course, such calculus
cannot be done by hand. However, computer simulations have changed the situation
because they allow research when no analytic solution actually exists. Moreover, they
make it possible to chart step-​by-​step the evolution of a system whose dynamics are
governed by nonintegrable differential equations (i.e., there is no analytic solution).
They have also provided a more precise visual analysis of empirical data and of the re-
sults obtained by mathematical tools. By allowing simulations with the different param-
eters of the Lévy α-​stable distributions, they have facilitated work with these distribu-
tions, making possible visual research that could have appeared vague before (Mardia
and Jupp 2000). Statistical and mathematical programs have been developed to com-
pute stable densities, cumulative distribution functions, and quantiles, resolving most
computational difficulties in using stable Lévy distributions in practical problems. In
conclusion, then, the increasing use of computers created a context for reconsidering
the use of stable Lévy distributions (and their power laws) in finance. The following
section will detail this point.
 

63╇ New Tools for Extreme-Value Analysis

3.2.╇ A NEW TOOL FOR ANALYZING EXTREME

VALUES: POWER-╉L AW DISTRIBUTIONS
One of the legacies of the theoretical results obtained in statistical physics since the
end of the 1970s and the new computer techniques is extensive research into power
laws. Generally, any work in econophysics advances empirical results to demonstrate
that the phenomena studied are ruled by a distribution of variables or observables
following a power law (Newman 2005).21 This statistical approach can be considered
one of the major characteristics of these studies. Its origin is rooted in the theoretical
and methodological background of econophysicists clarified in the previous section.
Power laws have attracted particular attention for, on the one hand, their interesting
mathematical and theoretical properties, directly linked with critical phenomena, and,
on the other hand, for empirical observations suggesting that their distribution fits
with observations of a very large variety of natural and man-╉made phenomena. Given
the place of power laws in econophysics, they can be considered the key tool of the
econophysics approach.
This section will clarify the reasons that a vast majority of econophysics works are
based on power laws. Afterward, we will study the links that can be made with finance
in order to explain the use of such laws in financial economics.

3.2.1.╇ The Key Role of Power Laws in Statistical Physics

As explained in the previous sections, power laws, in statistical physics, are closely
linked with critical phenomena.22 This link means that, from the viewpoint of statis-
tical physicists, power laws are synonymous with complex systems, which makes their
study particularly stimulating:

Why do physicists care about power laws so much? … The reason … is that we’re condi-
tioned to think they’re a sign of something interesting and complicated happening. The first
step is to convince ourselves that in boring situations, we don’t see power laws.23

Bouchaud (2001, 105) expresses a similar idea:

Physicists are often fascinated by power laws. The reason for this is that complex, collective
phenomena give rise to power laws which are universal, that is, to a large degree independent
of the microscopic details of the phenomenon. These power laws emerge from collective
action and transcend individual specificities. As such, they are unforgeable signatures of a
collective mechanism.

These remarks underline the fascination of power laws for econophysicists and ex-
plain the numerous empirical studies of power laws (this point will be detailed in
section 3.2.3). As previously mentioned, the link between power laws and critical
phenomena occurs at three levels: correlation lengths, scale invariance, and univer-
sality classes.
 

64  Econophysics and Financial Economics

The first level is correlation lengths. Section 3.1.1 presented the Ising model by intro-
ducing the concept of correlation length between spins (or interacting components) of a
system. In this context, critical phenomena evolve into a fragile configuration of critical
states in which the large correlation lengths that exist in the system at the critical point are
distributed like a power law. Traditionally, physicists have characterized the correlations be-
tween the constituents of− rthe system they analyze (at a given temperature T) as decaying
like an exponential law e ξ(T ) (i.e., the correlation function), where r is the distance between
two components and ξ(T ) is the correlation length.24 With the purpose of characteriz-
ing the divergence observed at the critical point, physicists added to the exponential law
−r

a power law, taking the following for r − α e ξ(T ) . At the critical point,
−r
as the previous section
explained, correlations are infinite, making the exponential, e ξ(T ) , equal to zero. This situa-
tion means that the correlation function is only distributed according to a power law,25 r−α. In
other words, away from the critical point, the correlation length between two constituents,
− x− y

x and y, decays exponentially,26 e ξ(T ) . But as the critical point is approached, the correlation
length increases, and right at the critical point, the correlation length goes to infinity and
decays in accordance with the power of the distance, |x − y|−α. As Shalizi explains intuitively
in his notes, far from the critical point the microscopic dynamics mean that the central-​limit
theorem can be applied. In this case, the fluctuations are approximately Gaussian. As one
approaches the critical point, however, giant, correlated fluctuations begin to appear. In this
case, the fluctuations can deliver a non-​Gaussian stationary distribution.27
The second level of influence between power laws and critical phe-
nomena is scale invariance. At their critical point, the phenomena become
independent of the scale; there is, therefore, a scaling invariance. The lack
of a characteristic scale implies that the microscopic details do not have
to be considered in the analysis. Scaling invariance is the footprint of crit-
ical phenomena; and, statistically speaking, power-​ law distribution is the
sole distribution that has a scale-​invariance property.28 Indeed, consider the
definition of scale invariance—​no matter how closely you look, you see the same
thing. Thus, one must have the same probability distribution for price variations
on the interval [110, 121] when the price quoted is 110 as for price variations on
the interval [100, 110] when the price quoted is 100. Mathematically speaking,29
we have

 x
Pr [ Y ≥ x X ] = Pr  Y ≥  , with x > X . (3.1)
 X 

 x  F ( x1 ) (3.2)
⇔ F 1  = .
 x2  F ( x 2 ) 

If F(x) is a power law, F = X a , then

 

65  New Tools for Extreme-Value Analysis

α
 x1  x1α (3.3)
 x  = x α . 
2 2

As one can see, a power-​law distribution respects the scale-​invariance property: what

X
is true for the combination 1 is true for the elements ( X1, X 2 ) composing this com-
X2
bination. A power law is the sole statistical framework to that respects this property. In
other words, at the critical point, if there is scaling, “The observable quantities in the
system should adopt a power-​law distribution” (Newman 2005, 35). This means that
the shape of the phenomenon’s size distribution curve does not depend on the obser-
vation scale, implying that the results are exactly the same for the micro and the macro
levels. For this reason, power laws are sometimes called scaling laws.
The third association between power laws and critical phenomena is the existence
of “universality classes.” In statistical physics, a universality class refers to a collection
of events likely to be described with mathematical models sharing the same statistical
features through the same scaling function. The characteristic exponent of a power law
directly determines its statistical characteristics. In this perspective, all phenomena
whose dynamics can be described through the same power law belong to a specific
category of phenomena sharing common statistical features. In other words, the i-
dentification of the critical exponent for a power law allows modelers to regroup all
phenomena characterized by this particular power law in the same universality class.
Again, the “universal aspect” mainly refers to the recurrence of the statistical charac-
teristics whose measures appear to be the same whatever the context in which they
have been observed. These parameters can therefore provide information about the
behavior of these phenomena at their critical points.
Having explained the theoretical origins of the use of power laws by statistical
physicists, we can now express these relations from a mathematical viewpoint. This
mathematical formulation will be used in the next section in order to express the con-
nections with finance. A finite sequence y = (y1, y2, …, yn) of real numbers, assumed
without loss of generality always to be ordered such that y1 ≥ y2 ≥ … ≥ yn, is said to
follow a power law if

k = c yk − α ,  (3.4)

where k is (by definition) the rank of yk, c is a fixed constant, and α is called the critical
exponent or the scaling parameter.30 In the case of a power-​law distribution, the tails
decay asymptotically according to α—​the smaller the value of α, the slower the decay
and the heavier the tails. A more common use of power laws occurs in the context
of random variables and their distributions. That is, assuming an underlying proba-
bility model P for a nonnegative random variable X, let F(x)  =  P[X ≤ x] for x ≥ 0
denote the (cumulative) distribution function of X, and let F (x) =1− F(x) denote the
complementary cumulative distribution function. In this stochastic context, a random
 

66╇ Econophysics and Financial Economics

variable X or its corresponding distribution function F is said to follow a power law or

is scaling with index α > 0 if, as x → ∞,

P [ X > x ] = 1 − F ( x ) ≈ cx − α  (3.5)

for some constant 0 < c < ∞ and a tail index α > 0. For 1 < α < 2, F has infinite variance
but finite mean, and for 0 < α ≤ 1, F has infinite variance and infinite mean. In general,
all moments of F of order β ≥ α are infinite.

3.2.2.╇ Power Laws and Their Links with Financial Economics

From the mathematical definition of a power law proposed above, we can now express
three connections with finance.
The first connection is that power laws are easily deduced from the financial defi-
nition of returns. Considering the price of a given stock, pt, the (log-╉) stock return rt is
the change of the logarithm of the stock price in a given time interval Δt,

rt = ln pt − ln pt −∆t .  (3.6)

Therefore, the probability of having a return r higher than the return x, P[r > x], can be
written as ln P[r > x] = −α ln x + c, which can be rewritten as a power-╉law expression
by using the exponential of both sides of the equation, P[r > x] = c x−α. This way of in-
terpreting financial returns shows that power laws are an appropriate tool with which
to characterize the evolution of financial prices.
The second connection is that power laws are easily linked with stochastic pro-
cesses used in financial economics, which describe the evolution of a variable X (price,
return, volume, etc.) over time (t). Knowing that a power law is a specific relationship
between two variables that requires no particular statistical assumption, we may asso-
ciate the evolution between the variables X and t with a power law. In this case, this
evolution is said to be a stable Lévy process when

C
1+ µ (for x → ±∞), 
P(x)  (3.7)
x

where C is a positive constant called the tail or scale parameter and the exponent μ is
between 0 and 2 (0 < μ ≤ 2). It is worth mentioning that among Lévy processes, only
stable Lévy processes can be associated with power laws31 because the stability pro-
perty is a statistical interpretation of the scaling property. As explained in Â�chapter 2,
the stability of a random variable implies that there exists a linear combination of two
independent copies of that variable with the same statistical properties (distribution
form and location parameters—╉i.e., mean, variance). This property is very useful in fi-
nance because a monthly distribution can be seen as a linear combination of a weekly
 

67  New Tools for Extreme-Value Analysis

or daily distribution, for example, meaning that statistical characteristics can easily be
estimated for every time horizon. While stochastic processes used in financial eco-
nomics are based on the Gaussian framework, as Mandelbrot (1963), Pagan (1996),
and Sornette (2006, 97)  explained, taking a Brownian random walk (i.e., Gaussian
process) as a starting point, a no-​normal diffusion law can be obtained by keeping
the independence and the stationarity of increments but by characterizing their dis-
tribution through a distribution exhibiting fat tails. A telling example of such a law is a
power law whose exponent μ will be between 0 and 2.
The third and final connection between power laws and financial economics
refers to the work of Mandelbrot since the 1960s, which we discussed in ­chapter 2.
The link between power laws and Lévy processes allows econophysicists to anchor
their work in that of Mandelbrot, who tried to improve the first canonical models
of financial economics. Econophysicists have in a sense concretized Mandelbrot’s
project (and they refer systematically to his work). However, while Mandelbrot and
econophysicists arrive at the same result—​modeling stock price variations using
Lévy stable processes—​they do not take the same path to get there. Mandelbrot starts
his analysis from the stability of stochastic process. His suggestion was to generalize
Gaussian processes by using Lévy stable distributions. For doing this, Mandelbrot
singled out the Gaussian framework for its properties of stability. He considered sta-
bility the most important hypothesis for a process because without it we cannot use
certain mathematical results (such as the central-​limit theorem) to produce new in-
teresting results in finance:

It is widely believed that price series are not stationary, in the sense that the mechanism that
generates them does not remain the same during successive time periods. … Statisticians
tend to be unaware of the gravity of this conclusion. Indeed, so little is known of nonstation-
ary time series, that accepting nonstationarity amounts to giving up any hope of performing
a worthwhile statistical analysis. (Mandelbrot 1966b, 2)

Mandelbrot’s path was to use the generalized central-​limit theorem, which is com-
patible with Lévy stable distributions. In contrast, econophysicists’ starting point
is critical phenomena and the results obtained from renormalization group meth-
ods that can demonstrate stability for non-​Gaussian stable processes. More pre-
cisely, the property of being distributed according to a power law is conserved
under addition, multiplication, and polynomial transformation. When we combine
two power-​law variables, the one with the fatter-​tailed distribution (that is, the one
with the smaller exponent) dominates. The new distribution is the minimum of the
tail exponents of the two combined distributions. Renormalization group meth-
ods focus on the scaling property of the process. According to Lesne and Laguës,
such an approach led to the creation of a new mathematical approach. In this case,
the attractor32 is no longer the Lévy stable distribution—​as it was in Mandelbrot’s
approach—​but the critical point. The latter is the attractor in the sense of the scal-
ing invariance.
 

68╇ Econophysics and Financial Economics

Scaling laws are a new type of statistical law on the same level as the law of large numbers and
the central-╉limit theorem. As such they go far beyond the scope of physics and can be found
in other domains such as finance, traffic control and biology. They apply to global, macro-
scopic properties of systems containing a large number of elementary microscopic  units.
(Lesne and Laguës 2011, 63)

This difference between Mandelbrot and econophysicists explains why the latter
start systematically from power-╉law distributions, while Mandelbrot starts systemati-
cally from a stable Lévy distribution. The stable Lévy distribution is a specific case of
the power-╉law distribution, since stable Lévy distribution is associated with a power
law whose increments are independent. However, we must remember that when
Mandelbrot began his research on the extension of physical methods and models to
nonphysical phenomena, results from renormalization group theory had not yet been
established. While this connection between Mandelbrot and econophysics seems to
be an ad hoc creation, it does suggests that the roots of econophysics lie in an active re-
search movement that emerged with the creation of financial economics (Â�chapter 1).

3.2.3.╇ Power-╉Law Distribution as a Phenomenological Law

The key role that power laws have played in developments in statistical physics since
the end of the 1970s gave impetus to empirical research into power-╉law distributions.
Since the critical exponent is deduced from the power-╉law distribution that is ob-
tained empirically, econophysicists have provided a huge number of empirical results
on power-╉law distributions. As Stanley and Plerou (2001, 563) explained, research
was first focused on finding which phenomena display scaling invariance, and then on
identifying empirical values of the exponent α. Proceeding in this way, statistical physi-
cists have identified which phenomena have the same exponents and hence belong to
the same universality class—╉even if none of these empirical researches explained in
detail the reasons for this association between power laws and universality class. This
first step has led to a large number of empirical investigations that have stressed the
phenomenological universality of power-╉law distributions.
As mentioned, econophysics literature places strong emphasis on the statistical de-
scription of empirical observations. From their observations, econophysicists aim to
identify a distribution or a mathematical model that they consider universal and that
comes from physics.33 This identification is the signal that they can apply their models
and methods to the study of the underlying phenomenon. This is why most of the
studies in econophysics exhibit empirical results to demonstrate that the phenomena
studied are ruled by a power-╉law distribution.
First at all, we must mention that empirical investigations into phenomena dis-
tributed according to power laws (also often referred to as fat-╉tail distributions, Pareto
distributions, Zipfian distributions, etc.) are nothing new. They date back to Pareto
(1897) and have been regularly observed and studied for numerous phenomena since.
Statistical physicists thus joined a larger movement, which has reinforced their percep-
tion that power laws and critical phenomena constitute important tools for analyzing
 

69  New Tools for Extreme-Value Analysis

empirical phenomena. Recently, authors have provided historical perspectives on the

use of power laws and log-​normal distributions in sciences.34 We will not present here
all phenomena that may be analyzed by means of power laws—​they are extremely nu-
merous. Instead, with the aim of understanding the emergence and the foundations of
econophysics, we will point out some of the most important empirical results from a
historical perspective. This will provide a general overview of the diversity and the prec-
edence of these empirical observations.
In order to identify whether a phenomenon is distributed according to a power
law, authors have traditionally used a visual analysis based on a double logarithmic
axes histogram. Indeed, as mentioned previously, by taking the logarithm of both sides
of equation (3.6), we see that the power-​law distribution obeys ln P[r > x] = −α ln x + c,
implying a straight line on a double-​logarithmic plot whose slope is given by the scal-
ing parameter α. Therefore, for any critical phenomenon, we expect to find the plot
whose form directly depends on the value of the α (figure 3.8).

100

10–1
alpha = 1.5
10–2 alpha = 2
alpha = 2.5
10–3

10–4
Pr (X ≥ X)

10–5

10–6

10–7

10–8

10–9

10–10
10–1 100 101
X

Figure 3.8  The slope of the graph (log-​log axis) characterizing a power-​law directly depends on the
characteristic exponent

A potential infinity of power laws exists. Therefore, the common way of identifying
power-​law behavior consists in checking visually on a histogram whether the fre-
quency distribution approximately falls on a straight line, which therefore provides
an indication that the distribution may follow a power law (with a scaling parameter α
 

70  Econophysics and Financial Economics

given by the absolute slope of the straight line).35 To illustrate this visual test, one can
consider, for instance, Gabaix et al. (2006). These authors combined on the same his-
togram the graph of a power law with ones describing empirical data from the Nikkei,
the Hang Seng Index. and the S&P 500 index (figure 3.9).

100

10–1
Cumulative distribution

10–2

α
10–3

Nikkei ’84–’97
10–4 Hang–Seng ’80–’97
S&P 500 ’62–’96

10–5
1 10
Normalized daily returns

Figure 3.9  Empirical cumulative distribution function of the absolute value of the daily returns of
the Nikkei (1984–​97), the Hang-​Seng (1980–​97) and the S&P 500 (1962–​96)
Source: Gopikrishnan et al. 1999, 5311. Reproduced by permission of the authors.

In their identification of the power law, Gabaix and al. (2006) varied the scaling pa-
rameter α of the power laws until they obtained the best fit between the graph of the
power law and the ones describing the empirical data. In so doing, they approximately
estimate the critical exponent. After having observed that the distributions of returns
for three different country indices were similar, the authors concluded that the equa-
tion given by their power law, whose critical exponent has been estimated to 3, appears
to hold for the data studied independently from the context. Finally, it is worth men-
tioning that due to the asymptotic dimension of power laws, these laws make sense
only for a very large quantity of data.
This type of visual investigation has guided econophysicists’ empirical research on
power-​law distribution. As mentioned above, this kind of approximation is nothing
new, since Pareto (1897) was the first author to identify such empirical linearity for
income distribution in the population—​where many people seemed to have low in-
comes while very few seemed to have high incomes. Since then, this linearity has been
 

71  New Tools for Extreme-Value Analysis

observed in a wide variety of phenomena. The physicist Felix Auerbach (1913) pointed
out that city sizes are also distributed according to a power law—​there are many towns,
fewer large cities, and very few metropolises.36 The linguist Zipf (1949) refined this
observation, hence the term “Zipf ’s Law” frequently used to refer to the idea that city
sizes follow a Pareto distribution as a function of their rank. Interestingly, Auerbach
suggested that this law of city sizes was “only a special case of a much more general law
which can be applied to the most diverse problems in the natural sciences, geography,
statistics, economics, etc.” (Auerbach 1913, 76, in Rybski 2013, 1267).37 In the same
the vein, Estoup (1916), and some years later Condon (1928) and then Zipf (1935),
observed this linear relationship in the occurrence of words in the vast majority of texts,
regardless of the language.38 In 1922, Willis and Yule (1922) discovered that the fre-
quency of the sizes of biological genera collected by Willis (1922) satisfies a power-​law
distribution. Yule (1925) explained that the distribution of species among genera of
plants also follows a power law. Lotka (1926) observed this distribution for the fre-
quency of publications by scientists. Kleiber (1932) and Brody (1945) found that the
mass-​specific metabolic rate of various animals is a linear function of their body mass.
We can find many other authors who have made similar observations in a variety
of fields—​including earthquakes, DNA, and human memory retrieval. The number
of these observations has considerably increased with the spread of computerized
databases.39 Power laws have become more and more popular in science, as Clauset
explains:

In the mid-​1990s, when large data sets on social, biological and technological systems were
first being put together and analyzed, power-​law distributions seemed to be everywhere … .
There were dozens, possibly hundreds of quantities, that all seemed to follow the same pat-
tern: a power-​law distribution. (2011, 7)

The result of this evolution was such that some “scientists are calling them more
normal than normal [distribution; therefore,] the presence of power-​law distributions
in data … should be considered as the norm rather than the exception” (Willinger,
cited in Mitchell 2009, 269).
This kind of relationship was also observed in financial and economic phe-
nomena, in addition to Pareto’s observations on income distribution.40 As explained
in c­ hapter 2, Mandelbrot was the first to identify this linear relationship in stock
price variations. This observation led him to apply the stable Lévy process to stock
price movements in the early 1960s. Although financial economists did not follow
Mandelbrot’s research due to mathematical difficulties (­chapter 2), economists have
always used power laws as a descriptive framework to characterize certain economic
phenomena. The relationship between the size of firms, cities, and organizations and
one of their characteristics (number of employees, inhabitants, growth, profits, etc.)
is an example. Recently, financial economists have shown a new interest in this dis-
tribution. For instance, Gabaix (2009) showed that the returns of the largest compa-
nies on the New York Stock Exchange exhibit the same visual linearity (figure 3.10).
 

72  Econophysics and Financial Economics

10–1

10–2
Distribution function P(|return| > x)

10–3

10–4

10–5

10–6

10–7
100 101 102
x (units of standard deviation)

Figure 3.10  Empirical cumulative distribution of the absolute values of the normalized 15-​minute
returns of the 1,000 largest companies in the Trades and Quotes database for the two-​year period
1994–​95 (12 million observations)
Source: Gabaix et al. 2003, 268. Reproduced by permission of the authors.

These observations of statistical distributions have led to an emphasis on the phenom-

enological universality of power laws. These observations have in addition supported
the idea that methods and models from statistical physics can be applied outside phys-
ics. However, we must stress that the association of these empirical regularities with
a power law is not free of presuppositions directly inherited from physics. Indeed, as-
sociating empirical observations with a power law implies a specific theoretical read-
ing,41 since the search for a straight line on this kind of graph refers to the idea that the
scaling property is appropriate for describing the studied phenomenon. However, it is
worth mentioning that a perfect straight line on a log-​log plot graph is rarely observed.
In contrast, data rather imply convex or concave lines, suggesting the development
of analytical techniques to deal with the inflection points. This situation is indirectly
due to the asymptotic character of power laws, whose emergence requires a very large
(theoretically infinite) amount of data. For a finite sample, the linear aspect of power
laws often has to be corrected by specific techniques.42

3.3.  HOW PHYSICISTS MADE POWER

LAWS PHYSICALLY PLAUSIBLE
Because they see power laws as a key element, statistical physicists have prioritized
statistical processes that can integrate these laws into the statistical characterization
 

73╇ New Tools for Extreme-Value Analysis

of physical systems. However, while they are interested in power laws for the statis-
tical description of large fluctuations, some physicists, and particularly econophysi-
cists (Koponen 1995; Mantegna 1991; Mantegna and Stanley 1994) have emphasized
empirical difficulties in using these processes. They have pointed out oppositions
between the mathematical properties of these processes and their physical applica-
tions. With the objective of resolving these oppositions, these authors have developed
truncated techniques that have proved necessary for using power laws and associated
stochastic processes.

3.3.1.╇ The Necessary Truncation of Power Law

The necessity to truncate power laws comes from one of their mathematical proper-
ties, namely their asymptotic behavior. Non-╉Gaussian stable Lévy processes, whose
evolution of variables follows an asymptotic power law, have infinite variance. Because
of their asymptotic dimension, power laws make sense only for a very large quantities
of data (we could say an “infinite amount” of data). The greater the amount of data, the
greater the assurance that power-╉law behavior is correctly identified. Unfortunately,
empirical samples are always finite. Moreover, real phenomena have neither infinite
parameters nor asymptotic behavior. Thus, although these processes make it possible
to keep scale invariance, which is a key theme in physics, their infinite variance is em-
pirically never observed and therefore is not physically plausible. In this regard, some
physicists have asserted that these processes are inapplicable, claiming that they have
“mathematical properties that discourage a physical approach” (Gupta and Campanha
1999, 32). Specifically,

Stochastic processes with infinite variance, although well-╉defined mathematically, are

extremely difficult to use and, moreover, raise fundamental questions when applied to
real systems. For example, in physical systems, the second moment is often related to the
system temperature, so infinite variance implies an infinite temperature. (Mantegna and
Stanley 2000, 4)

Consequently, at the end of the 1980s physicists were faced with a contradiction: while
power laws (stable Lévy processes) appeared to be supported by visual analysis of em-
pirical data, their major statistical feature (infinite variance) did not fit with a phenom-
enological description of physical systems, which all exhibit strictly finite measures.
Van der Vaart (2000) explained that two conceptual problems can emerge when
an asymptotic regime is used to describe behaviors of finite systems: first, the number
of empirical data is always finite, and second, the theoretical properties associated
with an asymptotic regime are in opposition with data observed, which necessarily
have finite variance due to the finitude of the sample. To solve these problems, statisti-
cians have suggested truncation techniques. These techniques allow an adaptation of
the asymptotic regime to empirical finite samples. Therefore, they have made power
laws physically plausible.
 

74╇ Econophysics and Financial Economics

The first reason for truncation is the need to fill the gap between the finiteness
of every empirical sample of empirical data and the asymptotic framework de-
fining the power-╉law regime. The truncation of a distribution is a statistical tech-
nique allowing modelers to transform an asymptotic and theoretical distribution
into a finite and empirical one.43 In other words, truncation is a conceptual bridge
between empirical data and asymptotic results. This idea of truncation is nothing
new in physics,44 but research into this topic has accelerated during the past two
decades. During this period, econophysicists have developed sophisticated sta-
tistical tools (truncation techniques) to escape the asymptotic regime and retain
power laws to describe empirical data. Again, this need to hold onto power laws
derives from their theoretical link with scaling invariance, exposed in the previous
section.
The second reason for truncating power laws associated with non-╉Gaussian
stable Lévy processes involves the need to transform these laws into a “physically
plausible description” of reality. As we know, physical systems refer to real phe-
nomena that can have neither asymptotic behavior nor an infinite parameter. Thus,
using non-╉Gaussian stable Lévy processes to describe a finite sample with finite pa-
rameters, although they are based on an asymptotic regime generating an infinite
variance, seems complicated. Truncation techniques allow modelers to transform
an asymptotic regime into a finite sample by cutting off the asymptotic part of the
distribution, thus providing finite variance. We will illustrate this process in the fol-
lowing sections.

3.3.2.╇ Truncation Techniques

While truncation makes the use of power laws physically plausible, we have to point
out that all truncations resulting from the finite dimension of physical systems nec-
essarily imply a gap between asymptotical results and empirical data. In statistical
terms, the finiteness of samples means that the number of data suddenly reaches a
limit, implying an abrupt truncation of the distribution describing them.45 However,
very few physical phenomena come to an abrupt end. In this respect, the manner of
truncating power laws must also make sense for physicists, leading them to develop
progressively different truncations in order to make them physically plausible and
theoretically justified. The development of truncation techniques is closely bound
up with the emergence of econophysics46 in the early 1990s, when statistical physics
initiated research in order to solve the theoretical problem related to the infiniteness
of variance.
In order to fit their statistical tools to financial distributions, econophysicists
have to solve two theoretical problems: on the one hand, they look for finite var-
iance, and on the other hand, they want to work in a nonasymptotic framework.
Implicitly, the truncation of distribution has existed for a long time, because we have
always worked with the finite asymptotical regime. By developing theoretical so-
lutions for truncated distributions and by defining a function that allows a switch
 

75  New Tools for Extreme-Value Analysis

from one to the other, econophysicists provided a first theory of this changeover.
Therefore, physicists applied asymptotical properties without misrepresenting
them since they provided a specific formulation of the gap between these proper-
ties and empirical results. In this way, physicists clarified the bridge between the
asymptotic and nonasymptotic regimes by making the switch physically plausible
and theoretically justified.
Econophysicists resolved the challenge of obtaining stable Lévy processes with
finite variance by using a “cutoff.” The first truncation technique was established by
Mantegna (1991), who developed his solution using financial data. He justified his
choice to use financial data because they offer time series that take into account the oc-
currence of extreme variations. Later, Mantegna and Stanley (1994) generalized this
approach with a truncated stable Lévy distribution with a cutoff at which the distri-
bution begins to deviate from the asymptotic region. This deviation can take different
forms, as shown in the equations below. With small amounts of data, that is, for a small
value of x, P(x) takes a value very close to what one would expect for a truncated stable
Lévy process. But for large values of observations, n, P(x) tends to the value predicted
for a nontruncated stable Lévy process. In this framework, the probability of taking a
step of size (x) at any time is defined by

P (x ) = L(x ) g (x),  (3.8)

where L(x) is a stable distribution and g(x) a truncation function. In the simplest
case (abrupt truncation), the truncation function g(x) is equal to a constant k and the
abrupt truncation process can be characterized by

kL(x) if x ≤ I 
g (x)  , (3.9)
 0 if x > I  

where I is the value from which the distribution is truncated. If x is not large enough
(the physically plausible regime), a Lévy distribution behaves very much like a trun-
cated stable Lévy distribution since most values expected for x fall in the Lévy-​like
region. When x is beyond the crossover value, we are in the nontruncated regime
to which the generalized central-​limit theorem (i.e., asymptotic regime) can be
applied.
The goal is to reduce the fat tails of the non-​Gaussian stable Lévy distribution
without deforming the central part of the distribution in order to decompose it into
a truncated stable part dedicated to the short and medium term and a nontruncated
part dedicated to the long term.47 This temporal decomposition is required because
in the (very) long term, we necessarily have an asymptotic regime. Therefore, thanks
to the truncation, in the short and medium term, we can leave the asymptotic regime.
In other words, the truncation makes it possible to decompose the stable distribution
into a physically plausible regime (truncated part) and an asymptotic one (nontrun-
cated part) (figure 3.11).
 

76  Econophysics and Financial Economics

n
l
Truncated stable Non Truncated
Lévy regime stable Lévy
regime

Physically plausible regime Asymptotic regime

(1) (2)

Truncated distribution
(3)

Figure 3.11  The idea of truncation for stable Lévy processes, where l is the selected cutoff and n the
number of data

Thanks to the truncated non-​Gaussian stable Lévy process, physicists can have a finite
variance and hence a more physically plausible description of empirical systems. The
idea is to avoid the statistical properties that are in opposition to empirical observation
(example: an infinite variance) by keeping those that are perceived as useful (example:
stability). It is worth mentioning that a truncated Lévy distribution taken as a whole is
no longer stable since it is characterized by a statistical switch of regimes (the third arrow
in figure 3.11). However, the truncated part of the distribution (first arrow in figure
3.11) keeps the statistical property of stability by offering a finite variance (Nakao 2000).
This first truncation technique based on a “cutoff” parameter provides a solution to the
problem of infinite variance. However, this specific technique produces a distribution that
truncates abruptly. Some physicists have claimed that this kind of truncation is not physi-
cally plausible enough because the physical system rarely changes abruptly:48 “In general,
the probability of taking a step [a variation] should decrease gradually and not abruptly, in a
complex way with step size due to limited physical capacity” (Gupta and Campanha 1999,
232). This generalized empirical principle has led physicists to go beyond abruptly trun-
cated stable Lévy processes, which do not have a sufficient physical basis (Mantegna and
Stanley 2000). With this purpose, physicists have developed statistical techniques to solve
this constraint related to the physically plausible dimension of the truncation technique.
Among them were Gupta and Campanha (1999), who considered the truncation
with a cutoff that is a decreasing exponential function49 (also called an exponential
cutoff). The idea was to switch from a physically plausible regime to an asymptotic
one through a gradual or an exponential cutoff after a certain step size, which may be
due to the limited physical capacity of the systems under study. We can express the
exponential truncation function50 by using equation (3.10):

1 if x ≤ I
 β (3.10)
g (x) =    x − I  
exp  −  k   if x > I 
  
 

77╇ New Tools for Extreme-Value Analysis

where I is the cutoff parameter at which the distribution begins to deviate from the
  x − I  β 
Lévy distribution, exp  −    is a decreasing function, and k and β are constants
  k  
related to truncation.51 Using this truncation function, Gupta and Campanha (1999)
defined the probability of taking a step of size (x) at any time as being given by

 k L(x) if x ≤ I
 β
P( x ) =    x − I   . (3.11)
k L(x ) exp  −  k   if x ≤ I 
  

Like abruptly truncated Lévy distributions, exponentially (or gradually) truncated

Lévy distributions have a finite variance,52 but they offer a more physically plausible
framework for describing finite systems (Gupta and Campanha 2002).53

3.4.╇CONCLUSION
The major contribution of chapter is to present the theoretical and methodolog-
ical foundations of econophysics. The chapter explained the theoretical origins
of the power law used in econophysics models. This objective is valuable simply
because these foundations are usually not explicitly exposed in the literature. It is
worth reminding ourselves that our analysis is based on a financial economist’s
viewpoint.
As detailed, the roots of econophysics come from the research on critical phe-
nomena and from the scientific changes observed in the 1970s statistical physics.
The major key themes of this discipline were presented: the renormalization group
theory, the Ising model and scaling properties. We also discussed scale invariance,
which is at the heart of these three themes. With the purpose of understanding
econophysics, we explained the connections between power laws, scale invari-
ance, and financial economics. Power laws make sense in relation to the definition
of financial returns; they can also describe the evolution of financial fluctuations
through stable Lévy processes. This is where stable Lévy processes play a key role
for econophysicists. Readers were reminded that power laws are not new in finan-
cial economics, since Mandelbrot—╉for different reasons—╉introduced them into
finance in the 1960s.
As this chapter showed, power laws are not perfectly adapted to describe phys-
ical (finite) systems, because they theoretically generate an infinite variance. To solve
this problem, physicists have developed truncation techniques. Econophysicists are
thereby able to obtain a finite variance while keeping the notion of scale invariance.
As mentioned, these truncation techniques are closely associated with the emergence
of econophysics.
The following chapter will define econophysics and explain how this discipline is
related to financial economics.
 

4
T H E D I S C I P L I N A RY P O S I T I O N
O F   EC O N O P H Y S I C S
N E W O P P O RT U N I T I E S
F O R F I N A N CI A L I N N OVAT I O N S

The previous chapter traced the origins of econophysics in statistical-​physics. It ex-

plained econophysicists’ method for modeling extreme values. Since the 1990s, the
introduction of intraday data combined with a series of financial crises has shed light
on the issue of extreme values. One of the most telling examples is the collapse of
Long-​Term Capital Management (LTCM) in 1998. This fund was founded by two
key authors of finance (Merton and Scholes), implying risk management based on
the core models of financial economics. However, it appears that “fund managers at
LTCM lacked a method for assessing the likelihood of more extreme risks in the long
term” (Buchanan 2004, 6). Models and results provided by econophysicists could
offer an interesting opportunity to reconsider the models used in financial economics.
However, the dialogue between the two communities has almost been nonexistent
since the birth of econophysics. How have we ended up with such situation? How can
we explain this lack of dialogue?
The aim of this chapter is to identify the reasons for this lack of dialogue between
financial economists and econophysicists. In this perspective, the way both commu-
nities produce their specific scientific ideas will be investigated. This chapter will set
out the progressive institutionalization of econophysics in terms of the organization
and diffusion of knowledge. It will show that although econophysics still lies outside
financial economics, this discipline has progressively developed its autonomy from
physics. This specific position of econophysics in the disciplinary space explains why,
despite the fact that financial economists and econophysicists share the same research
topics, dialogue between them is fraught with difficulty. Moreover, this chapter will
investigate two kinds of consequences of this institutional position. A first direct con-
sequence is that econophysicists have difficulty getting their research published in
financial-​economics journals. Another interesting consequence is that being outside
financial economics, econophysicists have not been constrained by the former disci-
pline’s theoretical and methodological framework. As section 4.2 will detail, this situ-
ation has allowed econophysics to introduce some scientific innovations that have not
been developed within financial economics.

78
 

79╇ The Disciplinary Position of Econophysics

4.1.╇ THE EMERGENCE OF ECONOPHYSICS

The first part of this chapter deals with the birth and the institutionalization of econo-
physics. After having presented some contextual elements that favored the creation of
this field, we will investigate the way econophysics has progressively been crystallized.
With this purpose, a bibliometric analysis will be offered in order to clarify the disci-
plinary position of this new area of knowledge.

4.1.1.╇ The Official Birth

As mentioned in the general introduction, the term econophysics generally refers to the
extension of statistical physics to the study of problems commonly considered to fall
within the sphere of economics, and particularly problems in finance. From a financial
economist’s viewpoint, econophysics aims to provide models that reproduce the statis-
tical behaviors of stock price or return variations, including their extreme values, and
then to apply these models to the study of financial products and strategies, such as op-
tions pricing. The movement’s official birth announcement came in a 1996 article by
Stanley et al. (1996), which coined the term econophysics.1 However, following Kutner
and Grech (2008), we can trace the informal birth of the movement to a paper published
by Mantegna (1991) that studied the evolution of financial in terms of stable Lévy pro-
cesses. As detailed in Â�chapter 3, this birth finds some of its origins in the changes that
have occurred since the 1970s in statistical physics and on financial markets.
The growing interest of physicists in economics and finance that led to the appear-
ance of econophysics coincided with what Kaiser (2012) called the “second bubble of
Physics Ph.D.s” observed in the 1980s. This situation was enhanced by defense policy
under the Reagan administration combined with increasing fears of economic com-
petition with Japan, triggering greater spending in biotech, engineering, and physical
sciences. This second bubble clearly appears when one considers the number of phys-
ics PhDs defended during the twentieth century (figure 4.1). Moreover, this bubble is
directly linked with the “golden age” of physics mentioned in Â�chapter 3.

1,800

1,500
Physics Ph.D.s per year

1,200

900

600

300

0
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000

Figure 4.1╇ Number of physics PhDs granted by US institutions, 1900–╉2005.

Source: Kaiser 2012, 299. Reproduced by permission of the author.
 

80╇ Econophysics and Financial Economics

The rapid rise in the public funding in the 1980s for young physicists generated a form
of “déjà vu” since it resembled the first bubble observed during the “Sputnik era” of the
1960s (also visible in figure 4.1).2 The second bubble was mainly favorable to physicists
involved in condensed-╉matter physics because this area of knowledge benefited from
the significant theoretical results obtained in the 1970s (Â�chapter 3). Moreover, “[In this]
field of research the line between fundamental physics and its practical applications was
so close that it was often blurred” (Cassidy 2011, 131). This rhetoric was directly in line
with the political community’s expectations in the 1990s, leading to a higher number
of public-╉funding opportunities for projects developed in this area of knowledge. This
trend was strengthened in the 1990s with the second bubble, when this field became
the first choice for new PhD students in physics to choose condensed-╉matter physics,
generating the second bubble evoked above. In 2000, for instance, 41 percent of doctor-
ates in physics were in condensed-╉matter physics (Cassidy 2011).
A point must be clarified here. In the previous chapter, we associated the founda-
tions of econophysics with statistical physics. It is worth mentioning that statistical
physics refers to a set of theoretical tools that can be applied to different physical
systems, while condensed matter is the branch of physics dealing with the physical
properties of matter phases. In other terms, condensed-╉matter physicists investigate
the behavior of matter and how it can evolve toward different phases. In their study
of matter, these physicists often use statistical methods (i.e., statistical physics). This
context explains that several fathers of econophysics (e.g., McCauley and Stanley) and
the vast majority of research centers providing PhD opportunities in this field are as-
sociated with condensed-╉matter physics. Telling examples are the Center for Polymer
Studies in Boston and the Santa Fe Institute, which have been important actors in the
organization and the diffusion of econophysics.
The following section will analyze the disciplinary position of econophysics in this
particular context. Did econophysicists produce their knowledge in physics or finan-
cial economics (or both)? Did they publish their works in financial economics jour-
nals? These are the kind of questions we will investigate hereafter.

4.1.2.╇ The Institutionalization of the New Discipline

While the 1990s saw the emergence of econophysics, the next decade witnessed
a growing institutionalization of the field. To gain recognition for their field of
research, econophysicists have adopted various strategies for spreading their know-
ledge. Specialized journals have been created, symposia have been organized and spe-
cific courses have been set up by physics departments in order to promote scientific
recognition of the new approach. All these strategies have played a part not only in
disseminating econophysics but also in creating a shared scientific culture, which is a
key element in the creation of a new discipline (Nadeau 1995).
The first publications in econophysics date from the 1990s. The founding article
by Stanley et al. (1996) strongly influenced physicists and mathematicians who de-
veloped a non-╉Gaussian approach to the study of financial returns (Kutner and Grech
2008). The proportion of articles devoted to econophysics has since grown steadily.
 

81  The Disciplinary Position of Econophysics

This trend appears clearly in the journal Physica A (figure 4.2), one of the three jour-
nals that publish the great majority of articles on the subject (we will return to this
point in the next section).

25%

20%

15%

10%

5%

0%
96

97

98

99

00

01

02

03

04

05

06

07

08

09
19

19

19

19

20

20

20

20

20

20

20

20

20

20
Figure 4.2  Number of articles on econophysics published in Physica A between 1996 and 2010
Source: Web of Science.

It is worth mentioning that the nonlinearity of the trend shown in figure 4.2 is mainly
due to regularly published special issues devoted to econophysics. The trend observed
for Physica A is also found in the three other major journals that have published articles
in econophysics (International Journal of Modern Physics C, Physical Review E, and the
European Journal of Physics B). This sustained growth in the number of articles published
each year earned econophysics official recognition as a discipline by the Physics and
Astrophysics Classification Scheme (PACS) in 2003—​10 years after the first articles.
The emerging editorial activity in econophysics has followed a relatively clear
line:  at the beginning, econophysicists preferred to publish and gain acceptance in
journals devoted to a preexisting theoretical field in physics (statistical physics) rather
than create new journals outside a preexisting scientific space and structure. Moreover,
these journals are among the most prestigious in physics. This editorial orientation
results from the methodology used by econophysicists (derived from statistical phys-
ics, as was mentioned in the previous section) but also from the new community’s
hope, on the one hand, to quickly gain recognition from the existing scientific com-
munity and, on the other hand, to reach a larger audience. Next, econophysicists cre-
ated journals with a title directly associated with economics or financial economics.
One can consider, for instance, Quantitative Finance, created in 2001, a journal essen-
tially devoted to questions of econophysics. The Journal of Economic Interaction and
 

82  Econophysics and Financial Economics

Coordination (JEIC), which started in 2006, is open to econophysics although its main
focus, the concept of interaction, is not a key theme in econophysics. Econophysicists
have also increased their influence in economic journals that already existed. Among
these is the Journal of Economic Dynamics and Control, created in 1979, which has been
open to papers related to econophysics since the publication in 2008 of a special issue
devoted to this theme.
The 1990s, then, stand as the decade in which econophysics emerged thanks
to growth in the number of publications. Textbooks on econophysics were not far
behind. The first textbook, Introduction to Econophysics by Mantegna and Stanley,
was published in 2000. Several more have appeared since—​by Roehner (2002) and
McCauley (2004), for example. The publication of textbooks is a very important
step in the development of a new field. They do not have the same status as arti-
cles or collections of articles. The latter are frequently aimed at spreading know-
ledge in an approach to the subject matter that remains exploratory and not unified.
Textbooks, on the other hand, are more grounded in a unified analysis. Their con-
tents therefore require a period of homogenization of the discipline, which is why
they represent an additional step in the institutionalization process. They “contain
highly elaborate models of linguistic forms for students to follow” (Bazerman 1988,
155). Textbooks play a sociological and educational role for neophytes by defining
the patterns of learning and formulating statements appropriate to the community
they wish to join. Given that collections of articles are published before textbooks,
the interval between the publication of the former and that of the latter gives an indi-
cation of the discipline’s evolution ( Jovanovic 2008): econophysics appears, there-
fore, as a theoretical approach that is evolving relatively rapidly. Barely a decade was
required to see the appearance of the first textbooks presenting econophysics as a
unified and coherent field. The swiftness of the development of this discipline can
be gauged by noting that it took twice as long, that is, two decades, for the first text-
books devoted to another recent specialty in finance, behavioral finance, to appear
(Schinckus 2009).
Another way to spread knowledge related to a new field is to organize workshops
and colloquiums. The first conference devoted to econophysics took place at the
University of Budapest in 1997 and was organized by the Department of Physics. Two
years later, the European Association of Physicists officially endorsed the first con-
ference on applications, Applications of Physics in Financial Analysis (APFA), which
was held in Dublin. The APFA colloquium was entirely dedicated to econophysics and
was held annually until 2007. Today, several annual conferences dedicated to econo-
physics exist, such as the Nikkei Econophysics Research Workshop and Symposium
and the Econophysics Colloquium. Combined with publications of papers and text-
books, these events contribute to the stabilization and spread of a common scientific
culture among econophysicists.
Another important component of a truly institutionalized research field is the cre-
ation of academic courses and BA, MA, and PhD programs devoted solely to that field.
Here again, physics has served as the institutional basis. Courses in econophysics have
been offered exclusively in physics departments since 2002 (the universities of Ulm in
 

83  The Disciplinary Position of Econophysics

Sweden, Fribourg in Switzerland, Munster in Germany, Silesia and Wroclaw in Poland,

and Dublin in Ireland). Most of the time, these courses are framed for physicists and
focus on statistical physics applied to finance. An additional step in the institutional-
ization of econophysics has been the creation of full academic programs totally dedi-
cated to this field. The first universities to introduce complete degree programs were
Polish (Miskiewicz 2010). Indeed, since 2001, the University of Silesia has offered a
bachelor’s program in this field, and a master’s program was created in 2009. Six other
Polish universities also offer programs in econophysics, for example, Warsaw with a
bachelor’s program and Wroclaw with a master’s. In 2006, the University of Houston
in the United States was the first to coordinate a PhD in econophysics,3 while a great
number of physics departments offer a PhD with a major in econophysics.4 In order to
familiarize students with the reality of finance, these programs provide some courses
on the financial reality, but they do not familiarize students with the theoretical basis
of financial economics.
Finally, this process of institutionalization was reinforced because of econophysi-
cists’ ability to connect with other research themes. In 2006, the Society for Economic
Science with Heterogeneous Interacting Agents (ESHIA) was created to promote
interdisciplinary research combining economics, physics, and computer science
(essentially artificial intelligence). Of course, this project does not directly focus on
econophysics, since the analysis of the heterogeneity and interaction of agents is an
approach that covers a wider field, including experimental psychology and artificial
intelligence. However, ESHIA aims at promoting an interdisciplinary approach, which
is precisely one of the major characteristics of econophysics. Besides, the society’s new
journal (Journal of Economic Interaction and Coordination) has been inviting authors
to submit papers devoted to econophysics. A  further sign of the growing influence
of econophysics is the International Conference on Econophysics, a platform for the
presentation of interdisciplinary ideas coming from different communities, especially
economics, finance, and physics. Finally, let us mention the existence of an interna-
tional forum dedicated to econophysics where scholars can share their opinions about
the evolution of the field.5 This forum is a way of diffusing knowledge; it also lists all
available academic positions related to econophysics.

4.2.  THE POSITION OF ECONOPHYSICS

IN THE DISCIPLINARY SPACE
While econophysics was officially recognized as a new discipline in 2003,6 it is cru-
cial to situate it in the disciplinary space and to understand its connections with
financial economics. An analysis of econophysics’ position in the disciplinary
space gives relevant information for understanding how it has developed its own
models and methodology for studying finance. The current investigation is based
on a bibliometric analysis (Gingras and Schinckus 2012). To analyze the position
of econophysics in the disciplinary space, the most influential authors in econo-
physics were identified. Then their papers in the literature were tracked by using
the Web of Science database of Thomson-​Reuters.7 The period studied—​from 1980
 

84╇ Econophysics and Financial Economics

to 2008—╉allows an analysis of the evolution of the field since the changes that oc-
curred in its prehistory period and makes it possible to measure the impact of its
birth (in 1996). The objective of the following section is to trace the birth and the
beginning of econophysics in order to understand the context in which the field
emerged. From this perspective, we will investigate the early period of econophysics
between 1996 and 2008. Another reason for focusing our attention on this period
refers of the fact that econophysics in its infancy was clearly defined (statistical
physics applied to financial economics). As Chakraborti et al. (2011a, 2011b) and
Schinckus (2012, 2017) showed, the econophysics literature has become more and
more scattered since 2009–╉10.

4.2.1.╇ In the Shadow of Physics

Let us consider first the disciplines in which the source papers have been published.
More than 70 percent of the key papers that have been published since 1996 appear
in physics journals, while only 21.6 percent have found their place in economics
or finance journals (table 4.1). During the previous period (1980–╉95) there were
very few papers written in journals of physics. They were mainly written in finance
and economics journals and were not really based on an approach originating in
physics.8

Table 4.1╇ Disciplines in which the source papers have been published

Discipline 1980–╉95 % 1996–╉2008 % Total %

Physics 8 32.0% 153 70.5% 161 66.5%

Economics and finance 13 52.0% 47 21.6% 60 24.2%
Mathematics 0 0.0% 9 4.1% 9 3.7%
Other fields 1 16.0% 3 3.8% 4 5%
Total 25 100% 217 100% 242 100%

Source: Web of Science.

These observations suggest that, although finance and economics journals did publish
articles on econophysics, the field did not hold a large place in financial economics. In
contrast, econophysics became more and more important in the discipline of physics.
This observation corroborates the fact that all academic programs dedicated to econo-
physics have developed outside economics.
The centrality of physics for econophysics (between 1996 and 2008) is clearly vis-
ible in figure 4.3, which maps the network of co-╉citations between journals cited in
papers citing our 242 source papers in econophysics. The dense core of the network
is composed of physics journals, while economics and finance journals are peripheral
(northwest of the map) and Quantitative Finance is in between.
  
Review of Financial Studies
Journal of Economics and Finance
Journal of Econometrics
American Economic Review Journal of Empirical and Finance

Cours dEconomie poltique

J Economic Dynamics Control

Quarterly Journal of Economics

Econometrica Int J Theo Applied Finance

Journal of Finance

Journal of Political Economy Macroeconomic Dynamics

J Economic Behavior Organ

Physics Letters A

Quantitative Finance Physical A Economics Letters

Journal of Business
P Nat Acad Science USA

Nature
Int J Modern Physics B Physical Review Physical Review A

Introduction Econophysics Physical Review Letter

The European Physical Journal B Science

Journal de Physique I
Reviews of Modern Physics
Theo Fin Risk Derivative Pricing

Europhysics Letters Journal of Physics A

Int J Moren Physics C Journal of Statistical Physics

Physical D
Physics Reports
Fractals

Figure 4.3  Most co-​cited journals (and manuals) in papers citing our 242 source articles in econophysics (100 co-​citations +)
Source: Gingras and Schinckus (2012). Reproduced by permission of the authors.
 

86  Econophysics and Financial Economics

Another way to look at the centrality of physics journals is provided in table 4.2, which
shows that between 1996 and 2008 only 12  percent of the citations of the source
papers came from economics or finance journals. Interestingly, this trend was similar
in the previous period (1980–​95), even though more than half of the papers had been
published in economics and finance journals.

Table 4.2  Disciplines citing the source papers

Discipline 1980–​95 % 1996–​2008 % Total

Physics 16 76.2% 2489 76.1% 2505

Economics and finance 2 9.5% 399 12.2% 401
Mathematics 1 4.8% 112 3.4% 113
Other fields 1 9.5% 63 8.3% 64
Total 21 100% 3272 100% 3293

Source: Web of Science.

Econophysics is thus essentially discussed in physics journals, a result confirmed by

table 4.3, which shows that, for both periods, about three-​quarters of the citations come
from papers published in physics journals usually devoted to condensed matter and sta-
tistical mechanics. The growing presence of econophysics in the pages of physics jour-
nals explains the official recognition of the discipline by the Physics and Astrophysics
Classification Scheme in 2003. This concentration inside physics, together with the fact
that the audience targeted by econophysics is increasingly made up of physicists, sug-
gests that econophysics has been developed by physicists for physicists.

Table 4.3  Main journals citing the source papers

Journals 1980–95 % 1996–​2008 % Total %

Physica A 3 14.3% 1213 37.1% 1216 36.9%

European Physical Journal B 0 0.0% 326 10.0% 326 9.9%
Physical Review E 2 9.5% 279 8.5% 281 8.5%
International Journal of Modern Physics C 1 4.8% 143 4.4% 144 4.4%
Quantitative Finance 0 0.0% 110 3.4% 110 3.3%
Journal of Economic Dynamics 0 0.0% 68 2.1% 68 2.1%
and Control
Journal of Economic Behavior 1 4.8% 60 1.8% 61 1.9%
and Organization
Acta Physica Polonica B 0 0.0% 42 1.3% 42 1.3%
Physical Review Letters 1 4.8% 36 1.1% 37 1.1%
Chaos Solitons and Fractals 0 0.0% 35 1.1% 35 1.1%
 

87  The Disciplinary Position of Econophysics

Journals 1980–95 % 1996–​2008 % Total %

Journal of Physics A: Mathematical 1 4.8% 33 1.0% 34 1.0%
and General
Macroeconomic Dynamics 0 0.0% 33 1.0% 33 1.0%
Journal of the Korean Physical Society 0 0.0% 30 0.9% 30 0.9%
Europhysics Letters 0 0.0% 29 0.9% 29 0.9%
Proceedings of the National Academy of 0 0.0% 25 0.8% 25 0.8%
Sciences of the United States of America
Advances in Complex Systems 0 0.0% 24 0.7% 24 0.7%
Physics Reports—​review section 0 0.0% 24 0.7% 24 0.7%
of Physics Letters
Computer Physics Communications 0 0.0% 20 0.6% 20 0.6%
EPL 0 0.0% 20 0.6% 20 0.6%
International Journal of Bifurcation 0 0.0% 20 0.6% 20 0.6%
and Chaos
Reports on Progress in Physics 0 0.0% 19 0.6% 19 0.6%
International Journal of Modern Physics B 0 0.0% 18 0.6% 18 0.5%
Journal of Statistical Mechanics: Theory 0 0.0% 15 0.5% 15 0.5%
and Experiment

Source: Web of Science.

A more precise investigation of the bibliometric data shows that econophysics has
not merely grown in the shadow of physics: it has progressively developed as an au-
tonomous subfield in the discipline. Indeed, although most of the publications citing
the source papers have appeared in journals of physics since 1996, we note that they
are concentrated in only two journals: Physica A (devoted to “statistical mechanics
and its applications”) and European Physical Journal B (devoted to condensed matter
and complex systems). Together these two journals account for 47.1  percent of
the publications (table 4.3). In addition, table 4.4 shows that Physica A published
by far the largest number of econophysics papers, with 41.5 percent of the total in
the second period (1996–​2008). It has thus become the leading journal of this new
field. In second place is another physics journal, European Physical Journal B. This
observation must be compared to the relative unimportance of Physica A in the first
period: only 4 percent of the key papers were published in this journal between 1980
and 1996. With the European Physical Journal B, Physica A published 53.9  percent
of the key papers between 1996 and 2008. In addition to the two journals already
identified as the core publishing venues for econophysics, we find Physical Review E,
the major American physics journal devoted to research on “statistical, nonlinear and
soft-​matter physics.”
 

88╇ Econophysics and Financial Economics

Table 4.4╇ Journals where the source papers have been published

Journals 1980–╉95 % 1996–╉2008 % Total %

Physica A 1 4.0% 90 41.5% 91 37.6%

European Physical Journal B 0 0.0% 27 12.4% 27 11.2%
Journal of Economic Behavior 2 8.0% 9 4.1% 11 4.5%
and Organization
Quantitative Finance 0 0.0% 10 4.6% 10 4.1%
Physical Review E 0 0.0% 8 3.7% 8 3.3%

Source: Web of Science.

This concentration in physics journals, and the fact that econophysicists have a
strong tendency to refer essentially to each other (Pieters and Hans 2002), suggest
that econophysics has now established its autonomy within physics.

4.2.2.╇ Outside Financial Economics

While econophysics has grown in the shadow of physics, its situation vis-╉à-╉vis financial ec-
onomics is also singular. Until 1996, econophysicists published mainly in finance and eco-
nomics journals, but since then econophysics has developed outside financial economics
(table 4.1). The only economics-╉related journals citing econophysics are Quantitative
Finance, the Journal of Economic Dynamics and Control, the Journal of Economic Behavior
and Organization, and Macroeconomic Dynamics (table 4.3). Since the appointment of
J. B. Rosser9 as editor-╉in-╉chief in 2002, the Journal of Economic Behavior and Organization
has begun publishing regular articles on the issue of complexity in economics, allowing
econophysicists to publish their work in that journal. Quantitative Finance, a relatively new
journal created in 2001, appears to be the main economics journal to publish papers de-
voted to econophysics (table 4.4). It can be considered one of the first nonphysics journals
specifically devoted to the new field: its editorial board includes many econophysicists,
and the editors are two econophysicists (Jean-╉Philippe Bouchaud and Doyne Farmer)
and a mathematician (Michael Dempster). Another journal, the Journal of Economic
Interaction and Coordination, was created in 2006 to promote research combining eco-
nomics, physics, and computer science. This publication is mainly directed by physicists,
and its editorial team features a substantial number of physicists and artificial intelligence
specialists. Interestingly, in 2008 the most cited journal in this publication was Physica A,
followed by Quantitative Finance itself, the Journal of Economic Dynamics and Control, and
then by two physics journals (European Physical Journal B and Physical Review E).10
Lastly, a 2008 special issue of the Journal of Economic Dynamics and Control deserves
mention:  “Applications of Statistical Physics in Economics and Finance” explicitly pro-
posed to “overcome the lack of communication between economists and econophysicists”
(Farmer and Lux 2008, 3). Doyne Farmer and Thomas Lux11 were the guest editors for this
special issue, and its 12 articles devoted to econophysics were written by financial econo-
mists and physicists. We should emphasize that the economics-╉related journals citing
 

89  The Disciplinary Position of Econophysics

econophysics cannot really be considered mainstream economics journals, but rather as

what Backhouse (2004, 265) called “orthodox dissenter” journals, that is, journals that
while rooted in mainstream theory are open to other approaches.12 All this suggests that
econophysics has found its place outside the major journals in financial economics. The
complete absence of top-​tier economic journals from table 4.4 again confirms that, between
1996 and 2008, econophysics developed beyond the margins of financial economics.
Since 2009, the term econophysics has been used more and more in the literature
(Chakraborti et al. 2011a, 2011b; Schinckus 2012, 2017). However, the meaning of this
label has been extended by articles that implement ARCH-​type models (Anagnostidis
and Emmanouilides 2015; Queiros and Tsallis 2005) or classical agent-​based mod-
eling (Abergel et al. 2014) and claim to be related to econophysics. This increasing
fragmentation of the literature makes a bibliometric analysis difficult. Moreover, the
professional climate has also opened doors for potential (sometimes not well-​defined)
collaboration between financial economists and econophysicists. In this perspective,
it is worth mentioning that the online journal Economics has (since 2009) a specific
section dedicated to econophysics, while the International Review of Financial Analysis
appointed a topic editor devoted to econophysics. Let us also mention that this journal
has published two special issues dedicated to this field in the last seven years (Li and
Chen 2012; McCauley et al. 2016 [forthcoming]). This recent evolution shows that
potential collaborations can be found between financial economics and econophysics.
However, with the purpose of understanding the disciplinary position of the latter, this
section has aimed at studying the period during which econophysics was progressively
been crystallized—​this target led us to focus on the years between 1996 and 2008.
Despite this growth in the field, table 4.5 shows that developments in economics
and finance were still matters of concern for econophysicists, since nearly half the
citations (46.5 percent) were to journals from these disciplines. Physics remains an
important reference, with about a third of the citations going to papers published in
physics journals, followed by mathematics journals (about 7  percent) and a tail of
many different science journals (13 percent). During the first period (1980–​95) more
than 56 percent of the references cited were to economics or finance journals. We thus
observe a decreasing dependence of econophysics on the economics literature and a
growing presence of physics journals as a source of knowledge for econophysics, up
from 19.2 percent to 32.6 percent.

Table 4.5  Disciplines cited in the source papers (two citations or more)

Discipline 1980–​95 % 1996–​2008 % Total %

Economics and finance 168 56.8% 2721 46.5% 2889 47.3%

Physics 56 19.2% 1943 33.3% 1999 32.6%
Mathematics 21 7.2% 419 7.2% 440 7.2%
Other fields 47 15.9% 752 13% 799 12.9%
Total 292 100% 5835 100% 6127 100%

Source: Web of Science.
 

90  Econophysics and Financial Economics

This trend can also be observed in table 4.6, which lists the main journals cited in the
source papers. While economics journals (e.g., American Economic Review) were often
cited in the key papers written between 1980 and 1995, physics journals became the main
source of knowledge for the papers published after 1996. As the table shows, between
1996 and 2008, among the first 10 journals listed, physics journals represent 22.6 percent,
while economics and finance journals represent 10.7 percent of papers published.

Table 4.6  Main journals cited in the source papers (two citations or more)

Journals 1980–​95 % 1996–​2008 % Total %

Physica A 3 1.0% 551 9.4% 554 9.0%

European Physical Journal B 0 0.0% 260 4.5% 260 4.2%
Physical Review E 0 0.0% 196 3.4% 196 3.2%
Quantitative Finance 0 0.0% 179 3.1% 179 2.9%
Physical Review Letters 5 1.7% 162 2.8% 167 2.7%
Nature 2 0.7% 147 2.5% 149 2.4%
Journal of Finance 2 0.7% 128 2.2% 130 2.1%
American Economic Review 18 6.2% 107 1.8% 125 2.0%
International Journal of Theoretical 0 0.0% 113 1.9% 113 1.8%
and Applied Finance
Econometrica 7 2.4% 101 1.7% 108 1.8%
International Journal of Modern 0 0.0% 107 1.8% 107 1.7%
Physics C
Journal de Physique I 2 0.7% 93 1.6% 95 1.6%
Journal of Business 6 2.1% 85 1.5% 91 1.5%
Journal of Economic Behavior 5 1.7% 84 1.4% 89 1.5%
and Organization
Journal of Political Economy 5 1.7% 73 1.3% 78 1.3%
Quarterly Journal of Economics 10 3.4% 62 1.1% 72 1.2%
Economic Journal 10 3.4% 58 1.0% 68 1.1%

Source: Web of Science.

Taken together, these data confirm that econophysics has developed on the existing
institutional structures of physics, rather than attempting to impose itself inside the
existing field of financial economics. As we have already pointed out, econophysics is
being promoted by “outsiders” to financial economics.

4.3.  ADVANTAGES AND DISADVANTAGES

OF THIS INSTITUTIONAL POSITION
This singular institutional position—​outside financial economics and in the shadow
of physics—​ has structured exchanges between econophysicists and financial
 

91  The Disciplinary Position of Econophysics

economists. While it is not hard to understand that this disciplinary structure makes
dialogue difficult, it has provided a surprisingly fruitful context for scientific innova-
tions. This section will analyze this fertility by investigating to what extent their po-
sition as outsiders has allowed econophysicists to innovate and to contribute to the
understanding of financial markets.

4.3.1.  The Difficult Dialogue between Financial

Economists and Econophysicists
As the previous section showed, papers dedicated to econophysics are mainly
published in physics journals, and their vocabulary, method, and models are
those used in physics. Now, although scientific papers appear contextless, they
are social constructions referring to a disciplinary culture whose knowledge is
founded on the production, reception, and use of texts. In their organization, these
texts share a highly stylized and formal system of presentation that aims to con-
vince readers, who will expect to find this specific system (Knorr-​Cetina 1981,
chap.  5; Gilbert and Mulkay 1984; Bazerman 1988). Consequently, it has been
easier for econophysicists to present their work to physics journals as examples
of modeling exercises analogous to those found in physics than to try to get past
the gatekeepers of financial-​economics journals. This situation was compounded
by the fact, already mentioned, that the conceptual foundations behind the math-
ematical techniques are very different from those found in financial economics.
Beyond this mathematical difference, the position of econophysics in the dis-
ciplinary space also results from four other major obstacles to dialogue between
the two communities. These obstacles, or difficulties, concern the vocabulary used,
the publishing process, the use of data, and the methodology employed.
The first difficulty concerns the vocabulary used. Although econophysics and fi-
nancial economics study the same topics (stock price variations), they differ in the
way they define their mathematical concepts. Econophysics’ distinctive feature is
the use of power laws and stable Lévy processes for modeling stock price variations.
However, the term commonly used in the econophysics literature is simply Lévy pro-
cesses (omitting the qualifier “stable”). This vocabulary is confusing because financial
economists also use Lévy processes in their conditional description of financial data
(­chapter  2).13 However, the two communities do not associate the same statistical
treatment with the label “Lévy processes”: while financial economists use this term
to characterize a conditional implementation (i.e., capturing the variability of a major
trend) of Lévy processes, econophysicists associate the term with an unconditional
use (i.e., describing the whole distribution) of stable Lévy processes (we will explain
this opposition in detail in the following chapter). This confusion in nomenclature
also generates debates on the real contributions of econophysics simply because many
economists tend to consider this field as a pale copy of what Mandelbrot and Fama
tried to do in the 1960s (­chapter 2). In this context, the novelties in the econophysics
approach are not always clear.
 

92  Econophysics and Financial Economics

The second obstacle to dialogue is the different practices in the publication of

knowledge. While economists usually wait several months (sometimes several years!)
for the finalization of the editorial process, physicists consider that, once accepted for
publication, a paper must be published quickly because its analysis and data are sig-
nificant only in the period immediately following completion of the research, and not
several months or years later.14 For physicists, the excessively long editorial process of
economics journals acts as a disincentive to submitting a paper. On the other hand,
financial economists have to face to an amazing number of publications in which it is
not easy to find the information they are looking for. Indeed, financial economists and
econophysicists use different stylistic conventions that make reading laborious for the
other community: both communities use their own classification scheme15 and their
own reference style.16 Beyond the different stylistic conventions, Bazerman (1988)
noted that financial economists and physicists tend to present their scientific writing
differently. A common practice in financial economics is to write a substantial literature
review “demonstrating the incrementalism of this literature” (Bazerman 1988, 274)
in order to emphasize the accumulation of knowledge and the ability of authors to
contribute to a preexisting codified knowledge. In complete contrast, physicists focus
on the practical implications of their articles, mentioning only references that deal
with potential applications.
These dissimilarities between publishing standards used in financial economics
and in physics have helped to keep econophysics publications out of financial eco-
nomic journals. This is consistent with Whitley’s observation that “research which ig-
nores current priorities and approaches and challenges current standards and ideals
is unlikely to be published in academic journals of the discipline” (1986b, 192).
A  reliable measure of rejected submissions is difficult to obtain, but an informal
survey shows that the major actors of econophysics did try to publish in those
mainstream economics journals with little success. We sent a questionnaire to 27
leading econophysicists (included as the key authors mentioned in the bibliomet-
ric analysis presented in the previous section) about the degree of closure of eco-
nomic journals to econophysicists. To the question “Have you submitted a paper to
a ranked journal in economics,” a large majority of authors replied yes. However, as
we saw previously, very few papers are now published in economic journals. Thus,
when authors were asked to give the main reasons for the rejection of their paper,
they replied that referees in economic journals often have difficulties with the topic
or/​and the method used in their paper.17 Although based on a small sample, these
results strongly suggest that economic journals are reluctant to publish papers de-
voted to econophysics. It appears, therefore, that this resistance is what prompted
econophysicists to exclude themselves and move toward journals more open to the
econophysics perspective.
The third obstacle to dialogue is that financial economists and econophysicists di-
verge markedly on the way they use empirical data. Econophysicists claim that their
approach is more neutral (i.e., not based on an a priori model). They explicitly demon-
strate a willingness to develop models that are, on the one hand, more coherent from a
physics point of view, and on the other hand based on “raw observations” of economic
 

93  The Disciplinary Position of Econophysics

systems (Stanley, Gabaix, and Plerou 2008). By “raw observations,” econophysicists

mean nonnormalized data. Financial economics (and financial econometrics) is
mainly based on the Gaussian framework, and when financial economists (financial
econometricians) observe abnormal data (by abnormal data, we mean statistically un-
usual from a Gaussian point of view), they normalize these data. They use data mining
in order to consider that all abnormal data have an expected mean equal to zero. In
other words, financial economists implicitly assume that price changes obey the log-​
normal probability distribution implying that massive fluctuations have a small prob-
ability. According to econophysicists, that perspective leads them to underestimate
the occurrence of financial crashes (Mandelbrot and Hudson 2004, 25). In marked
contrast, econophysicists claim to work directly with observed data, in which there are
no “abnormal data” since, from a physical point of view, extreme events are a specific
configuration of the system (Schinckus 2010b, 2010c). Therefore, econophysicists
consider any normalization as a priori reasoning about the economic phenomena that
they study.
Therefore, although the “empirical data” are the same for financial economists and
for econophysicists (financial quotations in the form of temporal series), the latter are
quick to point to their “direct use of raw data,” criticizing the statistical transforma-
tions performed by financial economists to “normalize” data. Here is Mandelbrot on
this point:

The Gaussian framework being a statistician’s best friend, often, when he must process data
that are obviously not normal, he begins by “normalizing” them … in the same way, it has
been very seriously suggested to me that I normalize price changes. (1997, 142)

McCauley also discusses the normalizing practice used by financial economists,

explaining,

We [econophysicists] do not “massage” the data. Data massaging is both dangerous and mis-
leading. (2006, 8)

This methodological position is widespread among econophysicists, who work in the

spirit of experimental physics. An empirical perspective is also justified, in their view,
by the evolution of financial reality. The computerization of financial markets has led
to better quantification of financial data, which should be studied as an “empirical sci-
ence” (McCauley 2004; Bouchaud 2002). However, this radical viewpoint, espoused
by some econophysicists, has an element of naiveté. No way of collecting data can ever
be totally neutral, in that all data are implicitly or explicitly the result of a selective pro-
cess (Schinckus 2010a; Fayyad 1996; Hastie et al. 2005; Hand 1998). In a sense, any
sampling method is embedded in a theory. By developing only physically plausible
frameworks, econophysicists also appear to have some a priori beliefs about the world
(Schinckus 2013). Although econophysicists and economists may study the same
epistemological concept of “empirical realism,” they do not agree about the definition
of this criterion. In line with the confusion about the term “Lévy processes” evoked
in the previous section, the two communities use the same words but seem to live in
 

94  Econophysics and Financial Economics

a different “conceptual matrix”: while economists use data in the last step of their re-
search process as an empirical justification of their theoretical argument, econophysi-
cists use data as an epistemic starting point—​suggesting the selection of an existing
theoretical framework. Because data are not perceived, used, and presented in the same
way in econophysics and financial economics, they can be a point of contention be-
tween the two communities.
The fourth and last obstacle to dialogue between the two communities concerns
methodology. In this domain, financial economists and econophysicists do not share
the same assumptions about readers’ expectations. Although the empirical dimension
is emphasized in both communities, financial economists take an apriorist approach
(axiomatically justified argumentation), while econophysicists develop an a posteriori
perspective (phenomenological data-​driven models). These methodological diver-
gences are responsible for two main shortcomings of econophysics: a lack of theoret-
ical explanation and a lack of quantitative tests.
Although econophysicists have obtained numerous empirical observations
(­chapter  3), there are no theoretical explanations to support them (Mitzenmacher
2005). Now, from the statistical physics viewpoint, one can consider that there is a
theoretical justification of sorts: the proof that the phenomenon studied is a critical
phenomenon, which justifies the use of the specific models and approach coming from
statistical physics. However, leaving this theoretical argument aside, econophysicists
have produced no theoretical justification to explain why the economic phenomena
studied are governed by a power law.18 The economist Steven Durlauf pointed this out
in 2005:

The empirical literature on scaling laws [i.e. power laws] is difficult to interpret because of the
absence of a compelling set of theoretical models to explain how the laws might come about.
This is very much the case if one examines the efforts by physicists to explain findings of scal-
ing laws in socioeconomic contexts. (Durlauf 2005, F235)

Consequently,

The econophysics approach to economic theory has generally failed to produce models that
are economically insightful.” (Durlauf 2005, F236)

The lack of theoretical explanation constitutes a strong limitation on using econophys-

ics’ methods and models in financial economics. Indeed, financial economists largely
base—​one could even say almost exclusively base—​their work on models with theo-
retical explanations.
Simulations of real phenomena without theory are considered weak results.19 The
Koopmans-​Vining debate at the end the 1940s, which pitted the National Bureau of
Economic Research against the Cowles Commission over the lack of theoretical ex-
planations and the need to link measurement with theory, underlines the importance
of this point.20 More recently, the vector autoregressions (VARs) modeling and the
 

95  The Disciplinary Position of Econophysics

real business cycle (RBC) models have faced the same criticism.21 Some works fo-
cusing on the theoretical foundations of power laws have been initiated over the past
few years. Chapter 5 will analyze them in detail. However, these attempts were mar-
ginal or not sufficiently general to be adopted by all. Some econophysicists today have
a greater awareness of this dearth of theory:

One [problem with the efforts to explain all power laws using the things statistical physi-
cists know] is that (to mangle Kipling) there turn out to be nine and sixty ways of con-
structing power laws, and every single one of them is right, in that it does indeed produce a
power law. Power laws turn out to result from a kind of central limit theorem for multipli-
cative growth processes, an observation which apparently dates back to Herbert Simon,
and which has been rediscovered by a number of physicists (for instance, Sornette). Reed
and Hughes have established an even more deflating explanation… . Now, just because
these simple mechanisms exist, doesn’t mean they explain any particular case, but it does
mean that you can’t legitimately argue “My favorite mechanism produces a power law;
there is a power law here; it is very unlikely there would be a power law if my mechanism
were not at work; therefore, it is reasonable to believe my mechanism is at work here.”
(Deborah Mayo would say that finding a power law does not constitute a severe test of
your hypothesis.) You need to do “differential diagnosis,” by identifying other, non-​power-​
law consequences of your mechanism, which other possible explanations don’t share. This,
we hardly ever do.22

In addition to this lack of theoretical interpretation, there exists another issue re-
lated to techniques of validating statistical analysis. Chapter 3 explained that econo-
physicists base their empirical results on a visual technique for identifying a fit
between a phenomenon and a power law. As the next chapter will explain, until re-
cently there was no quantitative test on the power-​law hypothesis. However, this visual
approach can be considered qualitative testing, and it is extremely problematic for fi-
nancial economists, who are strong defenders of quantitative tests. Durlauf pointed
out the insufficiency of such qualitative tests:

Literature on power and scaling laws has yet to move beyond the development of statistical
measures to the analyses of model comparison and evaluation. In other words, many of the
empirical claims in this literature concerning the presence of a particular law in some data set
fail to address the standard statistical issues of identification and statistical power adequately.
Hence, it is difficult to conclude that the findings in this literature can allow one to infer that
some economic environment is complex. (Durlauf 2005, F232)

In aiming to understand such criticism, we must bear in mind that financial econo-
mists systematically use econometrics for testing their models. Moreover, econo-
metric tests are a criterion of scientific acceptability on which financial economics was
built (­chapter 1). Section 4.3.3 will come back to this point: suffice it to say here that,
at the time financial economists created their own discipline, they strongly rejected the
visual approach used by the chartists and defended quantitative tests.23 Quantitative
 

96╇ Econophysics and Financial Economics

tests were a crucial point in financial economists’ argument that their approach was
based on scientific criteria, while chartism was presented as a practice with no sci-
entific foundation. Considering this methodological position, the visual tests used
by econophysicists are considered to lack credibility, and even scientific foundation.
The lack of quantitative tests makes econophysics literature unacceptable to financial
economists:

The power law/╉scaling literature has yet to develop formal statistical methodologies for
model comparison exercises; until such methods are developed, findings in the econophys-
ics literature are unlikely to persuade economists that scaling laws are empirically important.
(Durlauf 2005, F234)

An important criticism of the visual test used in econophysics was formulated by

LeBaron (2001), who latter showed that simple stochastic volatility models can pro-
duce behaviors similar to those obtained by econophysicists with power laws. Like
econophysicists, LeBaron based his argument on a visual approach. His paper makes
clear that visual inspections of log/╉log probability plots to uncover power laws can
lead to misleading inferences.24 Indeed, a significant disadvantage of the visual ap-
proach is that linearity is not a sufficient condition for having a power-╉law distribu-
tion, because power laws can visually be close to so-╉called exponential laws (Clauset,
Shalizi, and Newman 2009; Newman 2005). Only a large volume of data make it pos-
sible to distinguish between the two types of law (Mitzenmacher 2004).25 The prob-
lems created by the visual approach are also shared by other scientific communities—╉
computer science, for instance (Mitzenmacher 2004)—╉and are acknowledged by
some econophysicists:

Well-╉founded methods for analyzing power-╉law data have not yet taken root in all, or even
most, of these areas and in many cases hypothesised distributions are not tested rigorously
against the data. This naturally leaves open the possibility that apparent power-╉law behavior
is, in some cases at least, the result of wishful thinking … [T]â•„he common practice of identi-
fying and quantifying power-╉law distributions by the approximately straight-╉line behavior of
a histogram on a doubly logarithmic plot should not be trusted: such straight-╉line behavior
is a necessary but by no means sufficient condition for true power-╉law behavior. (Clauset,
Shalizi, and Newman 2009, 691)

In fact, many of the power laws econophysicists been trying to explain are not power
laws at all (Clauset, Shalizi, and Newman 2009). The next chapter will return to this
crucial assertion in detail.

4.3.2.╇ An Opportunity for Scientific Innovations

Although the position of econophysics in the disciplinary space creates a difficult dia-
logue between the two communities, one should not underestimate the impulses this
seam gives to new ideas. It is well known that scientific disciplines are “reputationally
 

97  The Disciplinary Position of Econophysics

controlled systems of knowledge production” that include specific behaviors, in par-

ticular vis-​à-​vis new ideas (Whitley 1986b, 190).

The dominant characteristic of the modern Western sciences as particular kinds of social or-
ganisation which generate new knowledge is their combination of novelty with conformity.
They reward intellectual innovation—​only new knowledge is publishable—​and yet contri-
butions have to conform to collective standards and priorities if they are to be regarded as
competent and scientific. Scientific fields can therefore be regarded as conservative novelty-​
producing systems. (Whitley 1986b, 186–​87)

To publish new ideas, authors have to share the criteria, standards, and objectives of
the scientific community of the specific field. Consequently, new ideas that are not
plainly compatible with the scientific framework will be more frequently developed
outside the field. This is particularly true in economics and physics, which are the
most controlled disciplines and in which scientists share a strong awareness of the
boundaries of their own discipline (Whitley 1986b, 193). Financial economics is an
institutionalized discipline with a strong control over its theoretical goals and models.
In this context, all new contributions have to conform to the criteria of conventional
acceptance—​postulates, beliefs, standards, priorities, and so on—​shared by financial
economists (Whitley 1986a, 1986b), and, consequently, there is a danger that novel-
ties will be kept in the periphery of the discipline. In addition, like economics, financial
economics “has a strong hierarchy of journals, so that researchers seeking the highest
reputations have to adhere to the standards and contribute to the goals of those con-
trolling the most prestigious outlets… . Because of this hierarchy, deviant economists
publishing in new journals can be ignored as not meeting the highest standards of the
discipline” (Whitley 1986b, 192–​93). Because econophysics appears precisely to be a
theoretical innovation that is not in line with the ideals, goals, and standards of finan-
cial economics, its position in the shadow of physics and outside financial economics
has allowed econophysics to develop scientific innovations that would have very little
chance of being developed inside financial economics.
Being outside financial economics, econophysicists can ignore the theoretical con-
straints imposed by the foundations of financial economics. The institutional situation
of econophysics provides a significant advantage for easing scientific innovations. A tell-
ing example is option pricing. Econophysics models of option pricing generally ignore
one of the most important strengths of the Black-​Scholes-​Merton model: the replicat-
ing portfolio reasoning.26 Basically, this notion refers to the possibility of replicating
(in terms of cash flows) the payoff of an option with a portfolio that is constituted by a
combination of the risk-​free asset and the underlying asset. The idea of replicating port-
folio is very important in finance since it leads, by arbitrage reasoning, to obtaining only
one price for the option (Cont and Tankov 2004). The practical usefulness of this ap-
proach is well known in finance (Derman 2009; Gaussel and Legras 1999).27 Because
econophysicists developed their knowledge outside of financial economics, they pro-
posed option-​pricing models partially independent of arbitrage reasoning. Moreover,
the option-​pricing models based on stable Lévy processes pose serious problems for
 

98  Econophysics and Financial Economics

obtaining a replicating portfolio, because in this case the market is incomplete (Ivancevic
2010; Takayasu 2006; Cont and Tankov 2004; Miyahara 2012; Zhang and Han 2013).
The use of these processes is therefore problematic for financial economics since they
are in conflict with the discipline’s probabilistic framework, as defined in the work of
Harrison, Kreps, and Pliska (­chapters 1 and 2). This divergence helps to explain the mar-
ginal use of these processes in financial economics. It is no coincidence that financial
economists and financial mathematicians have developed few models based on stable
Lévy processes since the 1990s (although there exist some generalized Lévy processes
in finance, as mentioned in c­ hapter 2). One must conclude that it is precisely because
econophysicists have developed their work outside the theoretical framework of finan-
cial economics that they can apply such processes more freely. This liberty paved the way
to potential new developments, as we will illustrate in c­ hapter 6.
Equilibrium is another example. There is a fundamental difference between fi-
nancial economics and econophysics concerning financial market equilibrium.
Equilibrium is a key concept in financial economics. While financial economics pro-
vides a less restrictive condition (a no-​arbitrage condition) than the traditional eco-
nomic equilibrium, econophysicists have developed a technical framework without
having given any consideration to this aspect of nonarbitrage or to the theoretical as-
sumptions that could bring the market to its equilibrium. Actually, these notions do
not play a key role in econophysics; they instead appear for econophysicists as a priori
beliefs28 that provide a “standardised approach and a standardised language in which
to explain each conclusion” (Farmer and Geanakoplos 2009, 17). Specifically, econo-
physicists do not reject the concept of equilibrium, but they do not assume a conver-
gence toward such a state. In addition, equilibrium is considered merely a potential state
of the system because “there is no empirical evidence for equilibrium” seen as a final state
of the system (McCauley 2004, 6). Similarly, while they do not reject the condition of
no arbitrage, they are indifferent to this restriction. Such a position is possible because
econophysicists do not have to deal with the scientific framework that rules financial
economics.
A last illustration refers to the way econophysicists handle empirical data, lead-
ing them to introduce models with power laws in order to deal with extreme values
in finance. The vast majority of contributions from econophysicists take a phenom-
enological approach based on a determination of the probability distribution from
stock prices, which are directly observed (­chapter 3). In other words, their models are
derived from empirical data. This approach, econophysicists believe, guarantees the
scientificity of their work by ensuring that the model obtained is as close as possible
to empirical data. From this perspective, econophysicists often present their discipline
as a data-​driven field, echoing a naive form of empiricism in which “rough data” would
constitute the only empirical basis of science. This approach implicitly promotes
a “passivist theory of knowledge” (Lakatos 1978, 20) according to which scientists
merely have to describe phenomena as they appear to them. However, working with
rough data does not prevent one’s disciplinary background from influencing how phe-
nomena are perceived. In the case of econophysics, given that power laws are a key
framework in statistical physics (­chapter 3), it is not surprising that econophysicists
 

99  The Disciplinary Position of Econophysics

systematically identify power laws in their visual analysis of data. Indeed, while econo-
physicists rarely say so, they are expressly looking for a power law, on the basis that
the phenomenon studied is complex and therefore ruled by such a law. This postulate
emerges clearly when one considers the visual tests econophysicists use to validate
observations with a power law (­chapter 3). The power law is postulated first (­chapter
5 will analyze this test in greater depth). In other words, it appears that, despite their
phenomenological claims, econophysicists turn the empirical world into their theo-
retical world in accordance with a “conservative activism” (Lakatos 1978, 20).
In this regard, the major objective of econophysics studies is not to “reveal” the
true statistical distribution that describes the rough data, but rather to determine the
value of the critical exponent of the (expected) power law by calibrating the param-
eters of the law required to fit the data. Such an approach is, therefore, phenomenolog-
ical in a very specific way: the implicit disciplinary assumption that econophysicists
have regarding the identification of statistical laws comes from the hypothesis of the
universality of power laws. To put it in other words, econophysics inductively expects
to identify a power law. Figure 4.4 summarizes this approach based on an observa-
tional postulate.

Empirical Data

Disciplinary expectation: power laws

Calibration

Parameters

Identification of the probability

distribution
Critical exponent

Figure 4.4  Econophysicists’ phenomenological approach

The first step of this approach is the disciplinary expectation (observational pos-
tulate) of a power law that would govern the empirical data. The second step is to
calibrate the parameters of this power law in order to fit the empirical data. Finally,
the third step is to identify the critical exponent of the power law that best allows
econophysicists to specify the category of models they can use to describe the data.
This identification is based on visual tests, as explained in ­chapter 3.
This approach draws little attention in financial economics simply because the re-
sults do not respect two of the major scientific criteria in financial economics: on the
one hand, the postulate that stock prices are not predictable or at least not predictable
enough to make a profit; and, on the other hand, the fact that empirical works must be
 

100  Econophysics and Financial Economics

based on hypotheses that can be assessed by statistical tests. Because of these postu-
lates, financial economists proceed in a very different way than econophysicists. Their
first step is to postulate that financial markets are competitive, agents are rational, and
prices result from market information. These postulates lead economists to formulate
testable hypotheses. The best-​known is the efficient-​market hypothesis, according to
which, when a market is competitive, prices reflect all available (and relevant) infor-
mation and there is no arbitrage opportunity. The third step is to test this hypothesis
with statistical tests. Thus, financial economists will follow a hypothetico-​deductive
method: they proceed by formulating a hypothesis from experiences. This hypothesis
can then be found to be valid or invalid by a statistical test on observable data.29 In
other words, in the hypothetico-​deductive method, models are derived/​deduced
from theory (figure 4.5).

Postulates
Perfect Rationality of
Competitive Markets
Agents

Hypotheses
Efficient market
Equilibrium model
Hypothesis

Statistical tests

Figure 4.5  Financial economists’ hypothetico-​deductive method

This deductive approach is often criticized by econophysicists (Sornette 2014;

McCauley 2006). They decry it as “top-​down” because it is driven not by data but by
a priori theoretical interpretations of the market. The main criticism is the Gaussian
distribution that econophysicists strictly associate with the efficient-​market hypo-
thesis. However, ­chapter 1 explained that this hypothesis has to be separated from the
Gaussian distribution. In addition, ­chapter 2 clarified that statistical tests have forced
financial economists to keep the Gaussian framework until now. Finally, it is worth
mentioning that the problem is not the phenomenological approach. Indeed, some
work in financial economics also follows a phenomenological approach that seems
quite similar to the one used in econophysics. In this category of work, one could
mention the very popular ARCH-​type modeling (­chapter 2).30 However, in this case,
financial economists have to deal with their scientific constraints, which are differ-
ent from those ruling econophysics. Specifically, the hypothetico-​deductive method
influences the phenomenological approach in financial economics. The ARCH-​type
models propose a corrective technique (conditional distribution) combined with a
 

101  The Disciplinary Position of Econophysics

classical Gaussian description (unconditional distribution) in order to capture the oc-

currence of extreme values. On the one hand, ARCH-​type models are methodologi-
cally associated with the efficient-​market hypothesis, because they were developed for
testing the correlations in stock market returns/​prices (and consequently their pre-
dictability). On the other hand, the statistical tests used by financial economists are
the classic ones developed in the Gaussian framework based on asymptotic reasoning
(the central-​limit theorem). Consequently, these empirical works, which, at first sight,
adopt a phenomenological approach, are implicitly built in order to be compatible
with the Gaussian framework. Figure 4.6 summarizes the so-​called phenomenological
approach associated with ARCH-​type models.

Empirical data
Disciplinary expectation: Gaussian law

Calibration

Identification of the unconditional

distribution

Hypothesis

Statistical tests

Figure 4.6  Financial economists’ phenomenological approach

The five steps portrayed in figure 4.6 describe the whole phenomenological approach
observed in the implementation of certain models in financial economics. When fi-
nancial economists try to describe the occurrence of extreme values in empirical data,
they combine an empirical analysis of these data based on the existence of a main
Gaussian trend with more phenomenological techniques relying on ARCH-​type
models to capture the high volatility of this main trend. This technique is usually as-
sociated with a conditional distribution whose calibration allows financial economists
to identify clearly the models (GARCH, EGARCH, etc.) they will implement. The
identification of this conditional distribution remains compatible with the Gaussian
framework, making it possible to save the classical foundations of financial economics.
This conceptual machinery is finally validated through statistical tests.
A comparison between the two phenomenological approaches underlines a major
difference between the two fields. Because of the scientific constraints ruling each
discipline, when financial economists adopt a phenomenological approach, their
modeling differs from that in econophysics. Being outside of financial economics,
econophysicists were not theoretically constrained by the conceptual framework de-
veloped by financial economists. In this open context, they implemented a phenom-
enological approach based on their own disciplinary matrix by focusing on the use of
power laws to characterize the dynamics of financial markets. Because the use of such
 

102  Econophysics and Financial Economics

tools was abandoned in financial economics in the 1970s (­chapter 2), their reintro-
duction in finance is doubtless the major contribution of econophysics. Consequently,
econophysicists can freely investigate financial markets to innovate and develop a new
way of describing financial return/​price fluctuations. The introduction of power laws
is the direct result of this lack of constraints that rule financial economics.
In our view, the position of econophysics in the disciplinary space explains why
econophysicists positioned themselves in theoretical a niche that mathematicians and
economists have barely investigated, or not investigated at all, because of their theoret-
ical constraints. The three previous examples provide a telling illustration of this spe-
cific situation for introducing scientific innovations. As Kuhn (1989) put it, a scientific
framework provides a cognitive foundation for a specific community. That shared way
of thinking defines theoretical constraints without which organized research would
not be possible. An absence of theory would lead to a chaotic situation that would
undermine the confidence of the scientific community, and possibly lead to its disin-
tegration. In other words, the conceptual framework shared by a specific community
allows the development of structured research. However, that framework also defines
the conceptual horizons for that community. Because econophysics emerged in the
shadow of physics, it did not have to deal with the theoretical constraints that deter-
mine the structure of financial economics. That conceptual freedom favored scientific
innovation by paving the way for the development of a new perspective shaping fi-
nancial knowledge from outside the mainstream (Chen, Anderson, and Barker 2008).
This strategy of positioning itself in theoretical niches is not specific to econophysi-
cists. It was exactly the one used by financial economists to introduce innovations in
finance when they created their field in the 1960s and the 1970s, during the period of
high theory of finance.

4.3.3.  A Strategy Already Used by Financial

Economics to Innovate
The parallelism between the creation of financial economics and the birth of econo-
physics provides an interesting framework for analyzing the relation between the two
disciplines and their potential contribution. Although econophysics was created more
than 30 years later than financial economics, the actors involved in the emergence of
these two fields used the same strategies to promote their works and to find a place
in the disciplinary space:  in both cases, a recognized discipline expanded toward a
new field of research whose study had hitherto been dominated by another frame-
work. In the 1960s, economics expanded to the study of financial markets, which at
the time was dominated by the chartism; in the 1990s, statistical physics expanded to
the study of financial markets, which at the time were dominated by current financial
economics. In both cases, the new community was made up of scientists trained out-
side the discipline, and hence outside the mainstream. A  particular colonization of
finance has occurred.
The beginning of this colonization can be detected in the new arrivals’ publica-
tion strategy. In the 1960s, the newcomers took control of the two main journals
 

103  The Disciplinary Position of Econophysics

specializing in finance at the time, the Journal of Business and the Journal of Finance.
The aim was to modify the content of published articles by imposing a more strongly
mathematical content and by using a particular structure: presenting the mathematical
model and then empirical tests. To reinforce the new orientation, these two journals
also published several special issues. Once control over these journals had been estab-
lished, the newcomers developed their own journals, such as the Journal of Financial
and Quantitative Analysis, created in 1965. Similarly, econophysicists chose to publish
and gain acceptance in journals devoted to an existing theoretical field in physics (sta-
tistical physics) rather than create new journals outside an existing scientific space and
hence structure. As explained in section 4.1, they took control of editorial boards (as
in the case of Physica A and the European Journal of Physics B). Once control over these
journals had been established, econophysicists developed their own journals, such as
Quantitative Finance and the Journal of Economic Dynamics and Control. This “coloni-
zation strategy” allowed the adepts of the new approach to bypass the partisans of the
dominant paradigm (and hence of the so-​called mainstream journals), who rejected
these new theoretical developments in which they were not yet proficient. Gradual
recognition of the new discipline subsequently allowed new specialist journals to be
created, making it possible to reach a wider readership (especially in economics).
In aiming to colonize finance, the partisans of the two disciplines used the same
discourse to justify the scientificity of the new approach. In each case, outsiders chal-
lenged the traditional approach by asking its adepts to prove that it was scientific. This
“confrontational” attitude is founded upon the challengers’ contention that the empir-
ical studies, the new mathematics and methodology they use guarantee a scientificity
(i.e., a way of doing science) absent from the traditional approach.31 The challengers
maintain that the scientificity of a theory or a model should determine whether it is
adopted or rejected. During the 1960s and 1970s, financial economists underlined
the importance of the empirical dimension of their research from their very first pub-
lications (Lorie 1965, 3). They saw the testability of their models and theories as a
guarantee of scientificity ( Jovanovic 2008). Consider Fama’s three articles (Fama
1965a, 1965b, 1970). All used the same structure: the first part dealt with theoretical
implications of the random-​walk model and its links with the efficient-​market hypo-
thesis, while the second part presented empirical results that validate the model. This
sequence—​theory, then empirical results—​is today familiar in financial economics.
It constitutes the hypothetico-​deductive method, the scientific method that has been
defended in economics since the middle of the twentieth century. Indeed, in the
1960s, financial economists criticized the chartists for their inability to present their
works with “scientific” arguments, accusing them of using a pure rhetorical justifica-
tion rather than a strong theoretical demonstration of their findings.32 Financial econ-
omists then developed a confrontational approach in their opposition to the chartists.
As an example, James Lorie (1965, 17) taxed the chartists with not taking into account
the tools used in a scientific discipline such as economics. In this debate, financial
economists argued that their approach was based on scientific criteria, while chartism
was based on folklore and had no scientific foundation.33 Consequently, financial eco-
nomics should supplant previous “folkloric” practices, judged to be groundless.
 

104  Econophysics and Financial Economics

Econophysicists have proceeded in like fashion. In their work, they belittle the meth-
odological framework of financial economics, using a familiar vocabulary. They describe
the theoretical developments of financial economics as “inconsistent … and appalling”
(Stanley et al. 1999, 288). Despite being an economist,34 Keen (2003, 108–​9) discred-
its financial economics by highlighting the “superficially appealing” character of its key
concepts and by comparing it to a “tapestry of beliefs.” Marsili and Zhang (1998, 51)
describe financial economics as “anti-​empirical,” while McCauley (2006, 17) does not
shrink from comparing the scientific value of the models of financial economics to that
of cartoons. The vocabulary used is designed to leave the reader in no doubt: “scientific,”
“folklore,” “deplorable,” “superficial,” “sceptical,” “superstition,” “mystic,” “challenge.” All
these wrangling words serve to dramatize a situation in which actors simply hold diver-
gent positions.
Another feature of this colonization strategy is the use of new mathematical tools
combined with the creation of new statistical data, both being considered as a guar-
antee of scientificity. Chapters 1 and 3 showed that the development of modern prob-
ability theory, on the one hand, and the evolution of financial markets, which are
increasingly quantitative (or digitized), on the other, contributed to the emergence
of financial economics and of econophysics. In each case these two factors triggered
the emergence of an alternative approach. In the 1960s, some financial economists
took up random processes at a time when mathematical developments had become
newly accessible to nonmathematicians (­chapter 1). The use or nonuse of these new
tools—​modern probability theory and work on statistical data—​constituted the main
element setting the “new approach” against the “traditional approach” of the time.35
This mathematical evolution went hand in hand with technological developments as
the use of computers became widespread. Computers made it possible to perform
tests on empirical data to assess the methods proposed for earning money on financial
markets, particularly chartist analysis.36 The creation of empirical databases stimulated
the application of mathematical models taken from modern probability theory and
research into stock market variations.
The development of probability theory and finer quantification of financial mar-
kets (thanks to developments in computing) were also triggering factors in the emer-
gence of econophysics. Indeed, since the 1990s, electronic markets have ruled the
financial sphere, while the use of information technologies has grown consistently
in companies. Computerization allowed the use of “high-​frequency data” offering a
more accurate study of the evolution of real-​time data (­chapter 3). Accumulated data
are then stored in the form of time series. While this type of data has been studied
by economists for several decades, the automation of markets has enabled “intraday”
data to provide “three orders of magnitude more data” to be recorded (Stanley et al.
2000, 339). The quantity of data is an important factor at a statistical level because
the larger the sample, the more reliable the identification of probabilistic patterns. In
other words, the development of computers, both in the 1960s and in the 1990s, cre-
ated two favorable decades for the development of probability theory, and c­ hapter 3
introduced the evolution of this theory in the 1990s (with the development of trun-
cation techniques). Quantified information and statistical data possess this property
 

105╇ The Disciplinary Position of Econophysics

of readily moving disciplinary boundaries and show the role of quantification in the
construction of an objective knowledge for an emerging field. Actors involved in the
development of financial economics and those who developed econophysics made
use of technological evolution to remain up-╉to-╉date with computational possibilities.
This parallel between these two disciplines is particularly interesting because this
institutional strategy occurred during the years of high theory of financial economics
(Â�chapter  1). In both cases, discriminating against existing practices, challengers at-
tempted to introduce scientific innovations that could not be developed in the disci-
pline because they were incompatible with the existing theoretical framework.

4.4.╇CONCLUSION
This chapter showed that econophysics is nowadays a fully-╉fledged scientific field with
its own journals, conferences, and education, which standardize and promote the key
concepts associated with this new approach. Econophysics emerged in the shadow
of physics and, until recently, has stayed outside financial economics. Two general
consequences of this disciplinary position were studied. First, this background gener-
ates a difficult dialogue between econophysicists and financial economists. Second,
being outside financial economics, econophysicists have been able to develop scien-
tific innovations that have not been explored by financial economists. We consider
that this issue of innovation is crucial for a future dialogue between the two areas of
knowledge. This chapter pointed out that financial economists followed the same
path, introducing new ideas and models, when they were outside the mainstream of
finance in the 1960s—╉these innovations led to the creation of modern finance theory.
This similarity in the evolution of the two fields suggests that the outsider position
of econophysics might lead to the introduction of their innovations in financial ec-
onomics. This chapter emphasized that some scientific innovations are directly re-
lated to the phenomenological approach (implemented in both econophysics and
financial economics). However, the comparison between the techniques of modeling
implemented in the two fields showed that the current phenomenological approach
(bottom-╉up) provided by econophysicists is quite similar to the one implemented
by financial economists in the 1960s. We suggested that econophysics could evolve
toward a more deductive (top-╉down) method like the one financial economists use.
We can therefore expect that stable Lévy processes and other contributions pushed by
econophysicists will be integrated into a framework common to financial economics
and econophysics. This is the path we will explore in the two last chapters.
 

5
M A J O R C O N T R I BU T I O N S O F EC O N O P H Y S I C S
TO F I N A N C I A L EC O N O M I C S

As explained in the two previous chapters, econophysics made a useful contribution

in the understanding of stock price and return variations. As detailed in c­ hapter 3, this
contribution is mainly based on the use of power-​law distributions to characterize the
occurrence of extreme values.1 While fat-​tailed distributions in finance are supported
by a large number of observations, ­chapter 2 pointed out that power laws pose prob-
lems for financial economics because they invalidate the traditional measurement of
risk given by variance, which is used in all pricing models.
In this challenging context, this chapter scrutinizes the possible contributions of
econophysics to finance by stressing the financial point of view. Our approach differs
from that generally used in the literature dedicated to econophysics, which usually starts
from the viewpoint of econophysics or complexity. Precisely, we aim to go beyond the
disciplinary boundaries. This chapter will discuss four potential uses of econophysics’
results by emphasizing to what extent this field can be useful in trading rooms and for
traders/​financiers. Afterward, this chapter will focus on the contributions of econo-
physics seen from the viewpoint of financial economists. This analysis will show the
surprising proximity between the two fields. In this context, the conditions of a poten-
tial emergence of a common conceptual scheme will be discussed. For this purpose,
recent advances in all key concepts of econophysics (power laws, self-​criticality, etc.)
are analyzed by clarifying what remains to be done to produce integrated models and a
theoretical framework common to econophysics and financial economics.

5.1.  THE USE OF ECONOPHYSICS

IN TRADING ROOMS
Econophysics literature devoted to financial economics contains numerous hetero-
geneous observations, showing that power laws appear to fit most of the relevant fi-
nancial data. Among others, Bassler et al. (2007) demonstrated that the euro-​dollar
exchange rate can be fitted to such distribution in time; with respect to foreign-​ex-
change markets, Seemann et al. (2011) showed that the mean square fluctuation of
increments can be fitted to power-​law scaling in time. Gopikrishnan et al. (2000) de-
scribed the volume of individual transactions on the NYSE with a power-​law distribu-
tion, while Mandelbrot (1997) and Lillo and Mantegna (2004) showed that the cu-
mulative sum of negative returns following a crash can also be approximated with such

106
 

107╇ Major Contributions of Econophysics to Financial Economics

a distribution. In the same vein, Lillo and Mantegna (2004) identified a power-╉law-╉
like relationship between financial prices and market capitalization, whereas Kaizoji
and Kaizoji (2004) used the same relation to describe the occurrence of calm-╉time
intervals in price changes. Mike and Farmer (2008) described the liquidity and the
volatility derived from empirical regularities in trading order flow by using a power
law. Financial autocorrelations have also been analyzed through the lens of this distri-
bution. Studies have shown that autocorrelations of order volume (Farmer and Lillo
2004), of signs of trading orders (Bouchaud, Mezard, and Potters 2002; Potters and
Bouchaud 2003; Ivanov et al. 2004; Farmer et al. 2005), and of liquidity at the best
bid and ask prices (Farmer and Lillo 2004) can all also be modeled with this pattern.
One could extend the list to many other publications. Basically, the point that all these
empirical works have in common is the conceptual role played by power laws, which
appears to be universally associated with different phenomena independently of their
microscopic details. The empirical studies have highlighted statistical properties re-
lated to financial data that financial economists traditionally do not take into account
in their work. As shown in �chapter 4, economics and physics use different methods
of producing scientific knowledge and results. We must bear in mind that for statis-
tical physicists, working on empirical data is the first step in scientific methodology,
the second step being the construction of a theoretical framework.2 This phenome-
nological approach used by econophysicists is, per se, a first significant contribution
of econophysics to finance because it has identified statistical patterns in raw financial
data. This approach should not be considered separately from the fact that the mathe-
matics and statistics on which econophysics models are based are relatively recent and
are still in development.3 In this context, “Rather than investigating the underlying
forces responsible for the universal scaling laws of financial markets, a relatively large
part of the econophysics literature mainly adapts physics tools of analysis to more
practical issues in finance. This line of research is the academic counterpart to the work
of ‘quants’ in the financial industry who mostly have a physics background and are oc-
cupied in large numbers for developing quantitative tools for forecasting, trading and
risk management” (Lux 2009, 14). Moreover, one has to consider that the impact of
econophysics on financial practices is potentially substantial because it involves the
statistical description of financial distributions and thus the statistical characterization
of financial uncertainty. However, this potential impact must be considered by taking
into account several drawbacks that we will discuss in this section.
To date, one can consider five major practical contributions of econophysics to
finance: practical implementations; identification of the common critical exponent;
volatility considered as a whole; uses in a long-╉term perspective; and prediction of
financial crashes and their possible management.

5.1.1.╇ Practical Implications

The practical implications of econophysics and its models for the financial industry
can roughly be associated with three observed trends, (1) hedging, (2) trading and
 

108  Econophysics and Financial Economics

portfolio management, and (3) software, and with one potential issue related to the
future of risk management.
The first issue deals with option pricing and hedging. It is well known that the
statistical description of financial distribution is the major information on which port-
folio managers can base their decisions. Two different statistical descriptions will nec-
essarily generate two different decision schemas. A non-​Gaussian framing of financial
distributions will produce a pricing different from one typically proposed by classical
Gaussian descriptions (McCauley 2004; Dionisio, Menezes, and Mendes 2006), and
consequently a different hedging strategy (Mantegna and Stanley 2000). While early
works in econophysics showed that an optimal hedging strategy appears impossible in
a non-​Gaussian framework (Aurell et al. 1997; Bouchaud and Sornette 1994), more
recent articles have defined mathematical conditions for optimal hedging, in line with
what financial economists mean by “optimal hedging” (McCauley, Gunaratne, and
Bassler 2007; Bucsa et al. 2014). We will illustrate this point in the next chapter.
Although a hedging solution based on econophysics is in its infancy, the pricing
issue has already generated high interest among practitioners, and econophysics-​based
portfolio management leaves room for a theoretical integration of a potential financial
crisis. The use of distributions in which extreme values can occur leads portfolio man-
agers to consider more appropriate track records in the case of large variations. This
probably explains why some banks and investment firms (BNP Paribas, Morningstar
Investment, Ibbotson Associates, etc.) have successfully used models from econophys-
ics (Casey 2013).4 Moreover, some econophysicists have created their own investment
companies that propose econophysics-​based management.5 Although these firms do
not reveal which models they use, physics-​based management is clearly part of their
strategy and advertising. For example, Capital Fund Management’s website states,

Research in the statistical properties of financial instruments and the development of sys-
tematic trading strategies are carried out at CFM by a team of Ph.D.’s, most of them former
physicists from prestigious international institutions.6

The third practical implementation concerns software whose algorithms are

based on models related to econophysics. Among these are ModEco,7 developed by
a retired academic physicist, and Rmetrics,8 developed by the Econophysics Group
at the University of Zurich—​ETH Zurich. It is worth mentioning that the latter is
directly used in university modules developed by this group.9 Better-​known statis-
tical and mathematical software is gradually integrating key econophysics concepts.
Mathematica, for example, has proposed a “stable distributions package” for 2005 (see
Rimmer and Nolan 2005). Aoyama et al. (2011) showed that the statistical software
Stata and SAS can also be used for an econophysical analysis of economic data. We
can also mention Alstott et al. (2014), or the script provided by Aaron Clauset on
his web page,10 to be used with Matlab. The development of computerized solutions
based on econophysics will contribute to widespread use of econophysics models (al-
though customers usually use these computerized solutions as “black boxes”) in fi-
nancial practices.
 

109╇ Major Contributions of Econophysics to Financial Economics

The last practical implementation concerns risk management. The most recent fi-
nancial crisis questioned the ability of risk managers to predict and limit financial catas-
trophes. In this challenging context, calls emerged for a new conceptual framework for
risk management. The objective of risk managers, portfolio managers, and regulators
is to limit the potential damage resulting from crises. Their ability to limit damage de-
pends directly on their understanding of financial uncertainty. By providing a different
statistical characterization of empirical distributions, econophysicists offer a new per-
spective on modeling financial uncertainty. From this perspective, econophysics could
raise questions about the models used in financial regulation, which are mainly based
on the concept of value-╉at-╉risk (VaR—╉introduced by JPMorgan in 1992). VaR, the
most widely used measure of market risk, “refers to the maximum potential loss over a
given period at a certain confidence level” (Bormetti et al. 2007). This measure is very
well known to practitioners since it can also be used to estimate the risk of individual
assets or portfolios. In practice, VaR models are easy to use since they refer to a high
quartile of the loss distribution of a portfolio over a certain time horizon. Therefore,
calculating VaR implies knowledge of the tail behavior of distribution returns. To pro-
mote the diffusion of its variance model, JPMorgan developed a computer program,
RiskMetrics, that is widely used in the financial industry. Some econophysicists (Wang,
Wang, and Si 2012; Bormetti et al. 2007) proposed an empirical comparison between
results produced by RiskMetrics and a VaR model based on non-╉Gaussian distribu-
tions (such as Lévy processes for instance), concluding that the latter provide better
forecasting of extreme variations in financial prices.11 Although in their infancy, these
reflections on non-╉Gaussian descriptions of financial distributions could have a direct
impact on the use of VaR (and consequently on the statistical models used) in financial
regulations and the prudent supervision of financial institutions.
These practices and practical reflections have been implemented at the same time
that academicians have provided practical results in four major domains, which we
will set out in the four next sections.

5.1.2.╇ Identification of a Common Critical Exponent

The first practical result one can find in the literature of econophysics is the identifica-
tion of a common critical exponent. This critical exponent enables modelers to identify
a specific (stable Lévy) stochastic regime for distributions. When the evolution of a
random variable can be characterized by a power law whose critical exponent is lower
than 2, the process is said to be stable, implying that all statistical features of this process
are the same whatever the level of study. As mentioned in Â�chapter 2, statistical charac-
terization can be very useful in finance, where scale refers to differ investment periods
(day, week, month, etc.). Regarding this exponent, Rama Cont et  al. (1997) identi-
fied a critical exponent for characterizing the power-╉law behaviors of financial assets
whose evolution can be described with a stable process. Currently, the identification
of a common critical exponent can be associated with the identification of a category
of phenomena (universality class) for which a unifying mathematical model could be
used. In other terms, phenomena whose dynamics can be described through a power
 

110╇ Econophysics and Financial Economics

law are assumed to have the same statistical features. According to Stanley and Plerou
(2001, 563), the identification of a universality class for financial data was “vital” in
experimental research into critical phenomena in finance, because in this phenome-
nological perspective, identifying a critical exponent will provide a model that can be
tested (see Â�chapter 4).
Nowadays, a consensus has emerged in econophysics literature on the value of the
critical exponent, since stock price variations follow a power law with an exponent close
to 3 (Stanley, Gabaix, and Plerou 2008; Plerou and Stanley 2008; Gopikrishnan et al.
1999; Malevergne, Sornette, and Pisarenko 2005; Plerou et al. 1999). This result also
concerns options, commodities, currencies, and interest rates (Bouchaud and Challet
2014). Figure 5.1 illustrates this result for US stocks, for the “at the money” volatility of
the corresponding options, and for credit default swaps on these same stocks.

PDF of daily returns

US Stocks, implied vol and CDS
4

–2

CDS
–4 Stocks
Vol
Inverse cubic law x–3
–6

–8
–4 –2 0 2 4

Figure 5.1╇ Distribution for daily returns

Source: Bonart cited from Bouchaud and Challet 2014. Reproduced by permission of the author.

This is a major practical contribution of econophysics to finance. Indeed, having this

parameter fixed, it is much easier to implement financial economics models based on
econophysics results. It also facilitates the creation of statistical tests, as we will explain
in the next chapter.

5.1.3.╇ Volatility Considered as a Whole

The second practical result concerns volatility, which is closely linked with distribu-
tion. Broadly speaking, econophysicists use power laws to describe the evolution of
 

111  Major Contributions of Econophysics to Financial Economics

financial prices in two ways. A first category of studies focuses on modeling volatility
distribution (only the tail part of a distribution is considered in this case) with a spe-
cific (conditional) distribution. A second category of studies use power laws to char-
acterize the entire distribution of financial prices/​returns.
Econophysicists who use power laws to describe the evolution of volatility (i.e., de-
scription of fat tails only), implement the same methodology as financial economists
with the ARCH class of models, which are nowadays the principal econometric used
in financial economics to model financial returns. The key idea behind these models
is that the most recent volatility influences current volatility. As ­chapter 2 explained,
the ARCH class of models assumes that financial returns follow the Gaussian distri-
bution (called unconditional distribution in the ARCH class of models) but charac-
terize the dynamics of the volatility with another distribution, which is not necessarily
Gaussian. The distribution describing the evolution of volatility is called “conditional
distribution.” It is worth mentioning that the conditional and unconditional dimen-
sions are defined in reference to the way we consider the history of the data. An
unconditional study will consider that all past data have the same influence on the
current evolution of the variable; data are considered in their entirety, and no specific
weight is assigned to the various past fluctuations. In contrast, a conditional treatment
will consider recent fluctuations to have a stronger influence on the current evolution
of the variable, and so more weight will be given to recent fluctuations. These time-​
dependent dynamics can be modeled through various potential statistical processes
(Kim et al. 2008), and this variety has generated a huge literature.12 To characterize
the occurrence of extreme values, financial economists use a conditional distribution
with the Gaussian distribution, in which large variations are associated with another
distribution that can take the form of a power law. In this situation, power laws are
considered a corrective tool for modeling important fluctuations that the Gaussian
framework does not capture: the Gaussian (unconditional) distribution can therefore
be combined with a power law (characterizing the conditional distribution) in order
to obtain a better fit with empirical results. The major advantage of a conditional ap-
proach is to capture the time-​dependent dynamics observed in the variability of finan-
cial prices. However, this advantage is weakened by the assumed Gaussian dimension
of unconditional distribution, which underestimates risk even when associated with
a power law for describing the (conditional) evolution of volatility, as explained by
Farmer and Geanakoplos. However, when the model is matched with real data, the
tail exponent is much too large, that is, “The tails of an ARCH process are too thin to
explain the fat tails of prices” (Farmer and Geanakoplos 2009, 17).
Unlike this first category of works, the great majority of studies by econophysicists
do not break down statistical analysis into conditional and unconditional dimensions;
instead, they use power laws to characterize the entire distribution. However, keeping
the previous vocabulary, we can say that they mainly work on unconditional distribu-
tions by considering the empirical time series as a whole. Econophysicists often un-
derline the advantage of dealing with historical distributions by claiming that financial
distributions must be described as they appeared in the past and not as they should
be according to a preexisting theoretical framework (McCauley 2004).13 Beyond
 

112╇ Econophysics and Financial Economics

this positivist justification, the major advantage of focusing on unconditional distri-

butions is that they do not require conditional parameterization (i.e., a definition of
time-╉dependent dynamics), which can generate a particular “subjectivism” (Matei,
Rovira, and Agell 2012, 95) because “conditional evidence is unobserved and there
is no natural and intuitive way to model the conditional heteroskedasticity, so that
each model will try to capture features considered important by the author” (Matei,
Rovira, and Agell 2012, 95). The parameterization of conditional distribution can
open the door to an “adhocity” of statistical treatments whose outcome could result
from a high level of parameterization (McCauley 2006). This argument provides justi-
fication for the unconditional approach embraced by econophysicists; and, as Farmer
and Geneakoplos (2009, 18) explained,

The practical value of the power law for risk control is that it results in more efficient risk es-
timates than extrapolation methods […] [W]â•„hen used to extrapolate risk levels that are not
contained in the sample [i.e., the use of unconditional distribution], they will consistently
underestimate risk. The power law, in contrast, is more parsimonious, and so is more efficient
with limited data. This can result in less biased estimates.

However, it is worth mentioning that what econophysicists view as an advantage is

seen as a drawback by financial economists: although working with an unconditional
approach enables econophysicists to avoid the “ad hoc perspective,” the unconditional
perspective cannot capture short-╉term, time-╉varying volatility. This static disadvan-
tage of an unconditional approach justifies the use of the ARCH class of models in
financial economics.
As another practical contribution of econophysics, we would point out that the
opportunity to consider volatility in its entirety also offers very interesting prac-
tical results since it simplifies the statistical treatment of investment periods. A new
framework based on the statistical property of stability could significantly influence
financial practices. By emphasizing the importance of unconditional distributions,
econophysicists have influenced the way that this level of statistical description of fi-
nancial fluctuations is modeled: Broda et al. (2013), for example, have tried to over-
come the drawbacks evoked in this section by using an ARCH framework within the
unconditional distribution characterized by a power law.

5.1.4.╇ Uses in a Long-╉Term Perspective

The third practical result still concerns volatility, which is a key concept in finance
because it measures risk. Statistical characterization of volatility has generated much
debate, especially regarding what we call “volatility clustering.” The “clustering” of vol-
atility is a stylized fact referring to an observed empirical pattern in the amplitude (vol-
atility) of financial fluctuations whose strong correlation suggests that a big move in a
given time period is likely to be followed by another big move in the next time period
(although the sign of this move, positive or negative, remains unpredictable). The big-
gest hurdle in this time dependence involves the appropriate horizon on which we can
observe statistical dependence.
 

113  Major Contributions of Econophysics to Financial Economics

Some authors (Bollerslev 1986; Engle 1982; Rachev et  al. 2011)  have claimed
that current financial prices are more sensitive to recent fluctuations of the market. In
other words, statistical description of financial returns should take into account this
short-​term influence by proposing a conditional level assigning autocorrelations on
the second moment of the distribution that attribute a greater weight to more recent
returns (Bollerslev 1986). Current volatility is assumed to be more closely correlated
to its more recent variations than to its older ones. From this perspective, ARCH
models are particularly well suited to describing the evolution of short-​run dynamics
of volatility since, because of their definition, these models assign more weight to the
influence of recent fluctuations in describing current volatility. This short-​term per-
spective on financial stocks makes ARCH models useful for traders, whose investment
window is often shorter than three months. In this context, traders often characterize
the dynamics of variation through ARCH models by adjusting a three-​month vola-
tility to the daily volatility of assets. Moreover, the fact that traders rarely work on a
sample older than four years does not favor the use of power laws in trading rooms.14
However, in some specific situations, financial management has to focus on long-​
term issues. Stress tests, for example, are a well-​known analysis to determine the extent
to which a financial instrument (an institution or a portfolio) can deal with an ec-
onomic crisis or the occurrence of extreme situations that could generate a critical
loss. By generating a computerized simulation based on historical distributions, this
methodology gauges how robust a financial asset (institution, portfolio) is in a crisis
situation. In this situation, a long-​term approach is usually implemented since a longer
sample will provide a better estimation of large fluctuations without diminishing the
magnitude of the maximum variation (Matei, Rovira, and Agell 2012). Large disrup-
tive events are not frequent, but they can be captured and approximated through a
long-​term historical (unconditional) analysis (Buchanan 2004). Because extreme
variations involve the vulnerability of financial assets to a crisis situation, the statis-
tical characterization of fat tails is very important. In that perspective, power laws are
the most appropriate statistical framework for extreme value management, whose key
theory (extreme-​value theory) can be seen as “a theoretical background for power law
behavior” (Alfarano and Lux 2010, 2).
The pertinence of power laws for estimating long-​term risk was highlighted by the
dramatic case of Long-​Term Capital Management (LTCM).15 Buchanan (2004, 5)
argued that the collapse of this hedge fund was partly due to inadequate estimation of
long-​term risk, whose historical (unconditional) form took the form of a power law.
In Buchanan’s view,

Fund managers at LTCM were sophisticated enough to be aware that their bell-​curve esti-
mates were probably low, yet they lacked methods for assessing the likelihood of more ex-
treme risks [in the long term]. In September of 1998, “unexpected” volatility in the markets,
set off by a default in Russia’s sovereign debt, led LTCM to lose more than 90 percent of its
value. … A power law is much better than the bell curve at establishing this long term risk.
It is also better at helping managers avoid the painful consequences of “unexpected” fluctua-
tions.” (Buchanan 2004, 5)
 

114  Econophysics and Financial Economics

Although LTCM managers knew their unconditional distributions were not nor-
mally distributed, they applied a mainstream way of managing the fund (Bernstein
2007) for which “they just add a large fudge factor at the end [as proposed by ARCH
class of models]. It would be more appropriate for the empirical reality of market
fluctuations—​as captured by the power law—​to be incorporated in the analysis of
risk” (Buchanan 2004, 6).
In the same vein, Buchanan (2004) also mentioned the importance of unconditional
power laws for estimating financial risk for pension funds and the insurance industry.
The first have to provide stable growth over the long term in order to ensure periodic
payments related to pensions. That financial objective makes pension funds particularly
sensitive to liquidity risk, leading them to deal with a long-​term horizon to predict the
worst loss possible and avoid “liquidity black holes” (Franzen 2010). The insurance in-
dustry is faced with the challenge of integrating extreme variations into their portfolio
management, especially insurance companies involved in contracts related to natural
disasters. “The bell curve does the job with automobile insurance but it fails miserably
when assessing large catastrophic losses due to hurricanes and earthquakes” (Buchanan
2004, 6).
Moreover, there is empirical evidence supporting the existence of a long-​
memory process in the volatility of prices, meaning that there is persistent temporal
dependence between observations widely separated in time. That long-​memory
property is well known and well documented in specialized literature (Ding, Enlge,
and Granger 1993; Andersen and Bollerslev 1997; Breidt, Crato, and de Lima
1998). The origin of the long-​memory process lies in the work of Hurst (1951), a
hydrologist commissioned by the British government to study changes in the level
of the Nile. More specifically, he was in charge of designing dams to manage water
reserves guarding against risk of drought. This task required a deep understanding
of Nile floods, which Hurst described with a temporal series characterizing the
changes in water levels. In his research, he observed that the optimal storage ca-
pacity for water tanks divided by the standard deviation of the river floods was
proportional by following a power law with an exponent of between 0.5 and 1.0.
Moreover, Hurst noticed that the evolution of water levels was not independent in
time: a major recent flood would influence a great number of future floods, which
implies a long-​memory process.
Several long-​memory processes have been observed in finance: volatility for stocks
(Ding, Enlge, and Granger 1993), exchange rates (Lu and Guegan 2011), and trading
volume (Lillo and Mantegna 2004).16 Financial economists acknowledge the exist-
ence of long-​memory processes, which they have usually captured through a myriad
of ARCH-​type models (IGARCH, EGARCH, GARCH, NGARCH, etc.) gradually
moving away from the use of power laws (Lux 2006). However, although ARCH
models provide a good approximation for short-​time dependence, they fail to capture
this long-​memory property (Ding, Enlge, and Granger 1993) and, contrary to models
that explicitly take long memory into account, an ARCH model that fits on one time-
scale does not work well on a different timescale (Farmer and Geanakoplos 2009,
19). The necessity of capturing long-​memory properties from time series, such as
 

115  Major Contributions of Econophysics to Financial Economics

long-​term volatility clustering or management mainly based on the long term, should
encourage the use of power laws in finance.
In order to pinpoint the differences between the ways in which financial econo-
mists and econophysicists deal with fat-​tailed distributions, table 5.1 summarizes the
main methodological differences between them.

Table 5.1  Statistical treatment of fat-​tailed distributions in financial economics and

econophysics

Financial economists Econophysicists

Statistical tool ARCH models Power laws
Analysis Broken down into two levels: One level of
unconditional (Gaussian) analysis: unconditional
distribution and conditional distribution based on
distribution historical data
Unconditional Gaussian Often associated with a power
distribution law, but this is not a necessary
condition
Application To characterize the fat tails of the To describe the whole of the
distribution distribution or to characterize
the fat tails of the distribution
Conditional A variety of distribution depending Often associated with a power
distribution on the type of ARCH model law, but this is not a necessary
condition
Methodology Corrective Descriptive
Necessary Existence of second statistical None
condition moment (because volatility is
associated with this parameter)
Time Short-​memory property (volatility Long-​memory property (Hurst
dependence clustering) exponent)
Time horizon Short term Long term

5.1.5.  Prediction of Financial Crashes

and Their Possible Management
The fourth practical result concerns the prediction of financial crashes and, conse-
quently, their possible management. Some econophysicists have applied their models
to forecast financial crashes, and more specifically to predicting financial-​market
downturns or “changes of regime” (Sornette and Cauwels 2015; Ausloos, Ivanova,
and Vandewalle 2002; Johansen and Sornette 2000; Pandey and Stauffer 2000). This
kind of application calls on the theoretical framework of econophysics regarding crit-
ical phenomena (we discussed in ­chapter 3). In this perspective, financial markets are
characterized by transformations associated with the usual phase transitions observed
in critical phenomena. In this way of conceptualizing crashes, all large transformations
 

116  Econophysics and Financial Economics

detected on the financial market are looked on as a passage from one phase to an-
other. This evolution can be statistically characterized by a power law. In so doing,
econophysicists use one property of the scaling hypothesis according to which there
“is a sort of data collapse, where under appropriate axis normalization, diverse data
‘collapse’ onto a single curve called a scaling function” (Preis and Stanley 2010, 4).
Sornette and Woodard (2010, 119) explained that

according to this “critical” point of view, the specific manner by which prices collapsed is not
the most important problem: a crash occurs because the market has entered an unstable phase
and any small disturbance or process may reveal the existence of the instability… . The col-
lapse is fundamentally due to the unstable position; the instantaneous cause of the collapse
is secondary. In the same vein, the growth of the sensitivity and the growing instability of the
market close to such a critical point might explain why attempts to unravel the proximal origin
of the crash have been so diverse. Essentially, anything would work once the system is ripe.

In this perspective, econophysics describes the occurrence of a financial crash as the

apparition of a critical point/​state of a complex system. This approach paves the way
for new practical tools to detect and manage financial-​market instabilities. A telling
example of this application is the Log-​Periodic Power Law (LPPL),17 which is a log-​
periodic oscillation model for describing the characteristic behavior of a speculative
bubble (and thus for predicting its subsequent crash). This model was originally pro-
posed by Sornette, Johansen, and Bouchaud (1995), Sornette and Johansen (1997,
2001), and Feigenbaum and Freund (1996, 1998). In its simplest form, this model
states that the asset price, p, before a crash evolves according to

{ }
ln[ p(t )] = A + B(t c − t ) 1 + Ccos ω ln(t c − t ) + φ  , 
β
(5.1)

where p(t) is the price index at time t, tc is most probable time of crash (i.e., the crit-
ical time) and quantifies the power-​law acceleration of prices, β is the exponent of the
power-​law growth, ω is the frequency of the fluctuations during the bubble, and the
remaining variables carry no structural interpretation (A, B, C, and φ).18
The LPPL model echoes two concepts well known in financial economics: noise trad-
ers (Black 1986; Grossman 1976; Dow and Gorton 2008; Shleifer and Summers 1990;
Kyle 1985) and mimetic/​herding behaviors (Orléan 1989; Keynes 1936, chap. 12; Orléan
1995; Banerjee 1993; Welch 1992).19 Indeed, according to this LPPL model, two catego-
ries of agents trade on financial markets: rational traders (who share the same characteris-
tics/​preferences) and noise traders (who mimic their network of friends). In this context,
the decisions of the latter depend on the behavior of traders within their network. Financial
crashes occur when mimetic behaviors between noise traders become self-​reinforcing,
leading to a large proportion of noise traders to share the same position. In this situation
the market becomes extremely sensitive to small global disturbances and a crash can occur.
As mentioned by Johansen et al. (1999, 7), “If the tendency for traders to ‘imitate’ their
‘friends’ increases up to a certain point called the ‘critical’ point, many traders may place the
same order (sell) at the same time, thus causing a crash.” This collective phenomenon takes
place when the market is close to its critical point. This situation can be detectable through
 

117  Major Contributions of Econophysics to Financial Economics

the size of the fluctuations: the closer the market is its critical point, the greater the price
fluctuations (see ­chapter 3 for the analogy with critical phenomena in statistical physics).
LPPL-​type modeling has already provided some successful predictions, such as
the fall of the Japan Nikkei index in 1999, the 2007 crash, and some others ( Jiang
et  al. 2010). The Financial Crisis Observatory (FCO), which was created by the
Econophysics Group at the University of Zurich, has developed an indicator of this
type with the aim of monitoring the risk of crises.

5.2.  THE CONTRIBUTIONS OF ECONOPHYSICS

SEEN FROM FINANCIAL ECONOMISTS’ VIEWPOINT
As we have seen, some financial practitioners are using econophysics models, even if
these models are currently less used than those of financial economics. We will turn
now our attention to the theoretical contributions of econophysics models to finan-
cial economics. This section will also underline the major puzzles currently associated
with econophysics, which have to be solved before the development of a potential
common framework between the two fields.
First, let us recall an important finding from the previous chapter: econophysics has
innovated and could provide a useful contribution to the understanding of stock price and
return variations, because it has sprung up outside financial economics. The outsider posi-
tion of econophysicists is also fueled by the fact that “econophysics … rarely postulates new
economic or financial theories, or finds contradictory evidence to existing theories” (2004,
175). Chapter 4 mentioned the remark by the economist Blake LeBaron (2001), who
showed that a number of simple stochastic volatility models can visually produce power
laws and long-​memory effects similar to those that have been reported in econophysics liter-
ature. LeBaron did not call for rejecting econophysics’ results, on the contrary: “It does not
say that power-​law results are wrong. It is only that they should be viewed as less conclusive
than they often are, since there may be many explanations beyond those related to critical
phenomena” (2001, 629). He added that “The search for reliable scaling laws in economics
and finance should continue… . The visual indication of a straight line going through some
points should not be taken on its own as a ‘test for complexity,’ or critical behavior… .
It would be best not to abandon these concepts, but to improve statistical understanding
of both the empirical tests and the theoretical models under consideration” (2001, 630).
In the same vein, the economist Steven Durlauf (2005), who is one of the defenders of
complexity in economics,20 asks econophysicists to consider the results from economics.
Precisely, numerous econophysicists ignore the literature of financial economics and pre-
sent their results as completely new when they are not:

One often finds [in the literature of econophysics] a scolding of the carefully maintained
straw man image of traditional finance. In particular, ignoring decades of work in dozens
of finance journals, it is often claimed that “economists believe that the probability dis-
tribution of stock returns is Gaussian,” a claim that can easily be refuted by a random
consultation of any of the learned journals of this field. In fact, while the (erroneous)
juxtaposition of scaling (physics!) via Normality (economics!) might be interpreted as
an exaggeration for marketing purposes, some of the early econophysics papers even
 

118  Econophysics and Financial Economics

gave the impression that what they attempted was a first quantitative analysis of financial
time series ever. If this was, then, performed on a level of rigor way below established
standards in economics (a revealing example is the analysis of supposed day-​of-​the-​
week effects in high-​frequency returns in Zhang, 1999) it clearly undermined the stand-
ing of econophysicists in the economics community. (Lux 2009, 15)

These remarks corroborate results detailed in ­chapter 4: up to now, models developed

by econophysicists have mainly stayed within the boundaries of statistical physics. It
is important to emphasize that these quotations are more a call for collaboration than
a critique of econophysicists.21 The problem is not econophysics concepts, per se, but
rather the lack of links with the existing knowledge in financial economics. Indeed, as
illustrated in c­ hapter 4, the majority of econophysicists apply concepts and models of
physics as they exist today, ignoring features of financial economics, particularly the
need for quantitative tests validating the power laws and the need to have generative
models that explain the emergence of such patterns.
One must admit that, for a long time, research into power laws has suffered from
these two major weaknesses. On the one hand, there were no statistical tests, the only
tests being based on a visual comparison method (­chapter 3). On the other hand, no
generative models existed for explaining the emergence of power laws. These two ab-
sences are crucial for financial economists because, as c­ hapter  1 explained, statistical
tests and theoretical explanations are the twin foundations of their discipline. Indeed,
from the most common viewpoint in financial economics, a scientific model must not
only reproduce reality but also be validated by econometric tests and by a theoretical
explanation that is compatible with the recognized theoretical framework. Some econo-
physicists do not feel especially concerned by these two conditions because, as explained
in ­chapters 3 and 4, from their scientific perspective, they do not need to meet these
conditions in order to propose a model. By contrast, these two conditions have largely
contributed to the maintenance of the Gaussian framework by financial economists even
when they describe the occurrence of extreme variations (­chapter 2). Consequently, this
methodological gap has strongly supported the misgivings of financial economists about
the potential contribution of econophysics to their field. Up to now, these contributions
are still difficult to value in light of the theoretical mainstream in financial economics.
The gap evoked above is not only due to the methodological dissimilarities be-
tween the two communities. It also results from the current lack of knowledge about
the statistical treatment of power laws. These misgivings were given strong expres-
sion in 2005 by Michael Mitzenmacher, a professor of computer science. In a seminal
paper on power laws, he asserted that the characterization of empirical distributions
by power laws is only a part of the challenge that faces researchers involved in explain-
ing the causes and roles of these laws. More precisely, he pointed out the need for the-
oretical models that could explain them:

While numerous models that yield power law behavior have been suggested and, in fact, the
number of such models continues to grow rapidly, no general mechanisms or approaches
have been suggested that allow one to validate that a suggested model is appropriate… .
[W]‌e have beautiful frameworks, theory, and models—​indeed, we have perhaps far too many
 

119  Major Contributions of Econophysics to Financial Economics

models—​but we have been hesitant in moving to the next steps, which could transform this
promising beginning into a truly remarkable new area of science. (Mitzenmacher 2005, 526)

Mitzenmacher (2005, 526) suggests a sequence of five steps for studies on power laws:

1. Observe: Gather data on the behavior of a system and demonstrate that a power-​

law distribution appears to fit the relevant data.
2. Interpret: Explain the significance of the power-​law behavior to the system.
3. Model: Propose an underlying model that explains the power-​law behavior.
4. Validate: Find data to validate, and if necessary specialize or modify, the model.
5. Control: Use the understanding from the model to control, modify, and improve
the system behavior.

Mitzenmacher’s disclaimer was relevant for econophysics. Like other fields (geog-
raphy, biology, etc.) using power laws in their research, econophysics had not really
been able to go beyond the third step when Mitzenmacher published his article in
2005. Mitzenmacher’s argument is important because, on the one hand, it underlines
that the claims made by economists have an echo in other fields dealing with power
laws; and on the other hand, it paves the way for a potential research agenda that would
ease the collaboration between econophysics and financial economists.
This situation indicates that the approach used by econophysicists is not unknown
to financial economists. Indeed, the models of the ARCH class and of econophysics
deal with data in the same way: both make a calibration in order to estimate the param-
eters of the models (­chapters 2 and 4). This similarity helps to explain why the lack of
interest in econophysics models is not fully comprehensible to econophysicists, and
by contrast why it is so comprehensible to financial economists. From a financial ec-
onomics’ viewpoint there are two nuances in these phenomenological approaches—​
although these might appear very tenuous. First, the ARCH class of models use
statistical tests, while econophysics models use visual tests. Financial economists are
skeptical about such visual tests because they provide no results about the statistical
behavior of the quantities estimated. Second, the ARCH class of models is based on
the Gaussian framework and is associated with the efficient-​market hypothesis by
testing the predictability of stock price/​return variations, both of which are founding
hypotheses of financial economics (­chapter 1). Therefore, most financial economists
using the ARCH class of models consider their models to have foundations from a
theoretical point of view (and not merely from a statistical one).
The next section will show that the situation concerning statistical tests for power
laws has changed very recently; this section will give some details concerning the the-
oretical foundations of the ARCH class of models. It seems that there is a misunder-
standing of the theoretical importance of the Gaussian framework: many econophysics
studies have associated the Gaussian random character of stock market prices/​returns
with the efficient-​market hypothesis, believing that it has been theoretically demon-
strated, as financial economics’ literature suggests. However, as explained in c­ hapters 1
and 2, this association is not theoretically demonstrated:  efficient markets and the
random walk or the martingale hypothesis are two distinct ideas. A random walk or a
 

120  Econophysics and Financial Economics

martingale is neither necessary nor sufficient for an efficient market; in addition, the
efficient-​market hypothesis implies no particular stochastic process. Therefore, it must
be admitted that the methodology used with the ARCH class of models is similar to
that used in econophysics modeling: a purely statistical modeling approach without
theoretical interpretations.22 In addition, it must be clarified that ARCH models “do not
provide an avenue towards an explanation of the empirical regularities” (Lux 2006, 8).
In fact, “Few [economic] models are capable of generating the type of ARCH [class of
models] one sees in the data” (Pagan 1996, 92), leading Adrian Pagan to conclude his
survey on the econometrics of financial markets with the following remarks:

One cannot help but feel that these statistical approaches to the modeling of financial series
have possibly reached the limits of their usefulness… . Ultimately one must pay more at-
tention to whether … it is possible to construct economic models that might be useful in
explaining what is observed. … For many economists … it is desirable to be able to under-
stand the phenomena one is witnessing, and this is generally best done through theoretical
models. Of course this desire does not mean that the search for statistical models which fit
the data should be abandoned. . . . It is this interplay between statistics and theoretical work in
economics which needs to become dominant in financial econometrics in the next decade.
(Pagan 1996, 92)

Since 1996, very little progress has been made toward a higher interplay. As one can
see, the approaches defended in econophysics and in financial econometrics are very
close: they both describe financial data without explaining the emergence of the ob-
served patterns.23 Methodologically speaking, it appears the two fields are using the
same kind of phenomenological approach, leading us to complete our analysis initi-
ated in the previous chapter. Indeed, ­chapter 4 showed that econophysicists follow the
same three first steps as does the hypothetico-​deductive approach used by financial
economists. The two last steps (formulation of hypotheses and validation with statis-
tical tests) do not really appear (yet) in econophysics literature (figure 5.2).

Econophysicists’ Financial economists’ Financial economists’

phenomenological approach phenomenological approach hypothetico-deductive method

Empirical Data Empirical Data

Calibration Calibration

Identification of the Probability distribution

probability distribution Postulates and postulates

Hypothesis Hypothesis

Statistical tests Statistical tests

Figure 5.2  Comparison of the three approaches

 

121  Major Contributions of Econophysics to Financial Economics

However, regarding the evolution of financial economics (­chapters 1 and 2), econo-
physics appears to be at the same stage as financial economics was in the 1960s. At that
time, financial economists tried to identify the most appropriate statistical description
in accordance with the data available at the time and their disciplinary expectations
(i.e., the existence of a Gaussian trend—​see ­chapter 1). In the same vein, since the
1990s, econophysicists have focused on statistical characterization of empirical data
through the data available and their disciplinary expectations (i.e., the existence of a
power law). After calibrating their statistical description, econophysicists can iden-
tify the class of models likely to explain the occurrence of extreme values (McCauley
2006). When econophysicists identify such a class of models, they can provide, on
the one hand, a potential explanation for the emergence of power laws and, on the
other hand, statistical tests to validate their use in finance. Chapters 6 will study this
agenda. In this analysis, econophysics could take the form of a hypothetico-​deductive
approach: testing the class of models identified from empirical data (figure 5.3).

Hypothesis related to
the emergence of
power laws

Statistical tests

Figure 5.3  Toward a potential hypothetico-​deductive approach in econophysics

What econophysicists will suggest to explain the emergence of power laws will act as
future hypotheses, which will have to be tested and validated through statistical tests.
Such an evolution could methodologically ease the development of future collabora-
tions between this field and financial economics.
Such a rapprochement between the two fields could be supported by general devel-
opments we observe in the sciences. Sciences are not static; over the past few decades
disciplinary borders have been changing, and new fields have emerged that are not mono-
disciplinary. If disciplinarity implies a monodiscipline describing a specialized scientific
field, multidisciplinarity (or pluridisciplinarity), interdisciplinarity, and transdisciplinarity
imply a variety of disciplines.24 “Researchers from different fields not only work closely to-
gether on a common problem over an extended period but also create a shared conceptual
model of the problem that integrates and transcends each of their separate disciplinary
perspectives” (Rosenfield 1992, 55). All participants then have common roles and try to
offer a holistic scheme that subordinates disciplines. The ecology of today can be looked
on as an illustration of these changes. Max-​Neef (2005) explained that growth and envi-
ronment were frequently identified as opposites in conventional economics because they
were mainly based on anthropocentric reasoning. By taking into account different fields
(economics, demography, biophysics, etc.), a more biocentric ecology has recently been
 

122  Econophysics and Financial Economics

developed. This ecology has proposed a new framework in order to solve the traditional
opposition between environment and growth. In this perspective, these opposite concepts
can now be seen as complementary in a unified development.
The emergence of econophysics is a clear illustration of this recent evolution.
But, as we have observed, to date no common conceptual scheme integrating and
transcending financial economics and econophysics has emerged. By staying within
their traditional boundaries, (econo)physicists and financial economists do not facil-
itate the creation of a common scientific literature that could be shared by the two
disciplines and could allow the creation of new models or knowledge. Morin (1994)
explained that “the big problem is to find the difficult path of the inter-​relationship
(l’entre-​articulation) between sciences that have not only their own language, but
basic concepts that cannot move from one language to another.” Although an inte-
grated approach requires that disciplines share common features, and in particular a
common conceptual scheme, the problem of language (concordances) must be also
considered. As Klein (1994) explained, “A ‘pidgin’ is an interim tongue, based [on]
partial agreement on the meaning of shared terms… . [An integrated approach] will
begin with a pidgin, with interim agreement on basic concepts and their meanings.”
The concept of a pidgin was introduced by Galison (1997), who called Kuhnian
incommensurability into question by explaining how people from different social
groups can communicate.25 From this perspective, a pidgin can be seen as a means of
communication between two (or more) groups that do not have a shared language.26
Galison (1997, 783)  used the metaphor of “trading zone” (because in situations
of trade, groups speak languages other than that of their home country) to charac-
terize this process of communication between people who do not share the same
language. More specifically, “Two groups can agree on rules of exchange even if they
ascribe utterly different significance to the objects being exchanged” (Galison 1997,
783). As Chrisman (1999) pointed out, the emergence of a pidgin requires specific
conditions: regular contact between the language communities involved, the need
to communicate between these communities, and, last, the absence of a widespread
interlanguage.
All the conditions seem to be met for the emergence of a new pidgin in the case of
econophysics and financial economics. Some theoretical continuities exist between
these two disciplines because Gaussian processes are a particular case of Lévy pro-
cesses. These continuities do not imply a common language, but they do encourage
the potential emergence of a common conceptual scheme based on a new axiomatic,
which would make sense for financial economists and econophysicists as well. In ac-
cordance with the analysis of Morin and Klein, a more integrated approach must take
into account the constraints that both original disciplines have to face: for econophysi-
cists the need for statistical processes to be physically plausible (i.e., in line with the
properties of physical systems) combined with the fit between model and empirical
data; for financial economists, the existence of reliable statistical tests for distribu-
tion describing data on the one hand, and the compatibility between the model and
theoretical framework on the other. In other words, this potential language implying
a common conceptual scheme must result from a double movement:  models from
physics must incorporate the theoretical framework from financial economics and, at
 

123╇ Major Contributions of Econophysics to Financial Economics

the same time, theories and concepts from financial economics must be modified so
that they encompass richer models from physics.
This double movement is a necessary step toward a more integrative approach in
econophysics. This adaptation implies the integration of theoretical constraints ob-
served in each discipline in such a way that the new shared conceptual framework will
make sense in each discipline. As Chrisman (1999, 5) has pointed out, the emergence
of a pidgin can be seen as a new integrated jargon between two disciplines, as op-
posed to a multidisciplinary approach that relies on what Chrisman called “boundary
objects,” which imply an agreement and an “awareness” between the groups involved
through which each can understand that the other may not see things in the same way.
The following section will present recent developments that could pave the way to a
progressive integration of econophysics models into financial economics.

5.3.╇ RECENT DEVELOPMENTS: TOWARD

POSSIBLE INTEGRATION
In recent years, substantial steps toward a common conceptual scheme and the emer-
gence of a new pidgin have been made: econophysicists have developed a number of
collaborations with economists (Gabaix et al. 2000; Farmer and Lux 2008; Farmer
and Foley 2009; Ausloos, Jovanovic, and Schinckus 2016; McCauley et al. 2016
[forthcoming]), economists have provided a conceptual reflection on econophysics
(Keen 2003; Rosser 2010, 2008),27 and financial economists have been taking power
laws into account in their ARCH family models (Broda et al. 2013). In addition, since
Mitzenmacher’s disclaimer in 2005, several authors have done significant work on
providing statistical tests for power laws, and generative models to explain the power-╉
law behaviors of financial-╉asset prices/╉returns.

5.3.1.╇ New Generative Models

The first crucial step in very recent contributions by econophysicists concerns gener-
ative models in order to explain power-╉law behavior in financial data. As previously
mentioned, such models are vital from the perspective of financial economics, because
they shed light on the potential reasons that power laws emerge. In this section, we
present the four main categories of generative models that explain the power-╉law be-
havior of financial data and could pave the way to a potential collaboration between
the two communities. According to these models, power laws would come from

1. A statistical effect (more precisely, from specific conditions on random growth);

2. The heterogeneity of investor sizes;
3. A specific organization (microstructure) of the market;
4. Self-╉organized criticality.

The first two categories of models find the origin of power laws in the time series, while
the last two associate these macro laws with an emerging property resulting from a
statistical expression of a large number of microinteractions. It is worth mentioning
 

124╇ Econophysics and Financial Economics

that the majority of these works (except the first category) have been developed by
theoreticians involved in econophysics. While the first category emerged from strictly
statistical development, the second category refers to a theory proposed by a financier
trained in mathematics (Xavier Gabaix) in collaboration with econophysicists (partic-
ularly Eugene Stanley). In the same vein, the third category of models combines works
proposed by economists with models provided by econophysicists. However, this cat-
egory is a direct application of the renormalization group theory developed in physics,
presented in Â�chapter 3. The last set of models refers to a strictly physical perspective
on power laws, since it is founded on what physicists call the self-╉organized criticality
that we introduced in Â�chapter 3.

5.3.1.1.╇ Random Growth and Power Laws

The first generative model for explaining the distribution of phenomena according to
a power law dates back to Yule (1925), who claimed that the distribution of species
among genera of plants followed a power law; he then proposed a stochastic process
based on proportional random growth, allowing this specific distribution to be gen-
erated. This notion was introduced into economics by Champernowne (1953) and
Simon (1955) and studied in detail by Kesten (1973) to explain the distribution of
city populations. Later, Gabaix (1999) clarified the idea by showing that an approx-
imate power law might emerge from Gibrat’s Law: in this context, the assumption is
that the expected rate of growth of a city and its variance are independent of its size.
Economists have progressively been using models of this kind, for example Gabaix
(1999, 2009)  and Krugman (1991, 1996a, 1996b).28 Although Champernowne
(1953), Simon (1955), and Kesten (1973) did not directly deal with financial data
in their models, they left room for a specific explanation of the rise of power laws in
finance, as explained below.
Roughly speaking, the model proposed by Champernowne (1953) refers to a pop-
ulation Pti of the city i at time t for which a normalized population size is defined as
Pti
Sti = , where Pt is the average population and where Sti is assumed to grow at a rate
Pt
of γ it +1 from time t to t + 1, so that we can write,

Sti +1 = Sti + γ it +1 . (5.2)

Using the seminal model proposed by Yule (1925), Kesten (1973) showed that this
basic equation can generate a power law when the growth rate itself follows a power
law. More precisely, Kesten (1973) associated the probability at time t + 1 to observe
a population S i higher than a specific size x with the distribution function Dt+1 ( x ) ,
which he formalized as follows:
Dt +1 ( x ) = P (Sti +1 > x)

= P (Sti + γ it +1 > x)
 

125  Major Contributions of Econophysics to Financial Economics

 x 
= P  Sti > i  . (5.5)
 γ t +1  

If growth rates in time are assumed to be i.i.d. random variables with a density f ( γ ) ,
the previous equation can be rewritten as

∞
 x
Dt +1 ( x ) = ∫Dt   f ( γ )d γ , (5.6)
0
γ 

which can be reformulated in terms of city size (S),

∞
 S
D ( S ) = ∫D   f ( γ )d γ . (5.7)
0
γ 

Yule’s model (1925), extended in economics by Champernowne (1953) and Kesten

(1973), defines the condition for which this equation can generate a power law: the
growth rate ( γ ) must itself follow a power law. In other words, the expected value of
the growth rate must be equal to 1, which can be expressed as

E  γ a  = 1.  (5.8)

Reed and Hughes (2002) sustained this statistical analysis later and showed that a
power-​law distribution can be obtained when things grow exponentially at random
times. The formalization presented here shows that the condition for which Dt de-
scribes a power law results from a technical constraint in the statistical description of
economic data. In other words, the origin of power laws is not in the observed phe-
nomenon but rather in its statistical characterization independently of the disciplinary
context in which this statistical investigation is made. However, a more economic ex-
planation of why the equation above generates a power law can be derived from Gibrat
(1931), who claimed that frictions preventing cities (or firms) from becoming too
small must be built into the model. Gibrat explained that, in the presence of frictions,
the pure random growth process modeled by Yule (1925) will follow not a power law
but the Gaussian law. In this perspective, the condition under which the above equa-
tion generates a power law requires the addition of a disturbing factor (a positive cost)
characterizing the existence of frictions. As Gabaix (2009) explained, this technical
point developed by Gibrat (1931) paved the way (which was not followed) for an
economic explanation for the emergence of power laws, since this idea of friction can
easily take the form of a positive probability of death or a high cost of expansion that
prevents cities (or firms) from growing.
Gabaix (2009) also explained that the statistical constraint identified by Yule
(1925) opened the door to another unexpected potential explanation for the emer-
gence of power laws: they could result from a specific parameterization of ARCH-​type
 

126╇ Econophysics and Financial Economics

models since they imply random growth. And indeed, in line with what we mentioned
above, ARCH models can generate a power law if they are parameterized with a
random growth rate that follows a power law. Let us illustrate this idea with a classical
application of an ARCH model taking the form

σt2 = ασt2−1ε t2 + β ,  (5.9)

where ε t σt −1 is the return, with ε t independent of σt . Therefore, in line with the Yule-╉
2 2

Kesten model, it is possible to reformulate this ARCH model using St = σt2 , γ t = αε t2 ,

and the β as a constant. From this perspective, ARCH models are consistent with the
random-╉growth model presented previously, which will generate a power law under
either of the following two conditions:

A situation in which β = 0 , implying a classical model of a random-╉growth pro-

•â•‡
cess that will generate a power law when the rate γ t (αε t2 ) follows a power law
•â•‡
A situation in which β (or E (β)) >0, meaning that this parameter characterizes
some frictions in the growth preventing the process from converging toward a
Gaussian distribution

This ARCH origin of power laws is conceptually troubling: ARCH models are sup-
posed to describe the evolution of financial returns through an improved Gaussian
description whose parameterization can generate a power law! While econophysi-
cists see a power law “in the observations,” financial economists create it through
statistical treatment of data. Indeed, this statistically generated power law appears to
be a simulacrum aimed at capturing the occurrence of extreme variations in line with
the methodological framework (based on the importance of the Gaussian uncondi-
tional distribution) used in finance. This first category of models generating power
laws does not really explain the origin of the macro laws, since they appear to result
from another power law characterizing the evolution of the growth rate. As strange
as this might be, we will find this kind of explanation again in the following section.

5.3.1.2.╇ Heterogeneity of Investor Sizes and Power Laws

In their model, Gabaix et  al. (2003) demonstrated that an institutional reason can
explain power-╉law behaviors. These authors detailed their model in a second paper
(Gabaix et al. 2006) directly inspired by existing literature on the behavior of institu-
tional investors. Specifically, they showed that institutional investors’ trades have an
impact on the evolution of financial prices. In this analysis, the volatility of financial
prices is derived from the distribution of the trading volume:

The fat-╉tailed distribution of investor sizes generates a fat-╉tailed distribution of volumes

and returns. When we derive the optimal trading behavior of large institutions, we are able
to replicate the specific values for the power law exponents found in stylized fact (i) [the
 

127  Major Contributions of Econophysics to Financial Economics

observation of a power-​law distribution for trading volume] and (ii) [the observation of a
power law for the financial returns]. (Gabaix et al. 2006, 463)

The model’s starting point is observation of the distribution related to the investors’
size, which takes the form of a power law.29 The fat-​tailed distribution means that we
can observe a big difference between large and small trading institutions, implying
an important heterogeneity of actors (in terms of size). This diversity results in a dis-
persal of trading power in which only bigger institutions will have a real impact on the
market.30 Starting from this conceptual framework, Gabaix et al. (2003, 2006) used
three initial empirical observations. First, the distribution of the size of investors (in
the United States) can be characterized through a power law (with an exponent of 1).
Second, the distribution of the trading value follows a power law (with an exponent
equal to 1.5). Third, the distribution of financial returns can be described with a power
law (whose exponent is equal to 3).
Given these observations, Gabaix et  al. demonstrated that optimal trading be-
havior of large institutions (considering that they are the only elements influencing
market prices) can generate a power law in the distribution of trading volume and fi-
nancial returns. Gabaix et al. (2003, 2006) proposed a model describing a single risky
security with a fixed supply and with the price at time t denoted p(t). At time t, a
large fund gets a signal M about a mispricing (M < 0 is associated with a selling signal,
while M > 0 is a buying one). That signal leads the fund to negotiate with a market
supplier a potential buy on the market. Of course, there is a lag between the negotia-
tion, the transaction, and the impact on the market price. Basically, at t = 1 − 2 ε ( ε is
a small positive number), the fund buys a quantity V of shares at price p + r, where
r is the price concession (or the full price). Afterward, at t = 1 − ε , the transaction is
announced on the market. Consequently, the price jumps, at t = 1 , to p(1) = p + π (V ) ,
where π (V ) is called the permanent impact, while the difference ( ϕ = r − π ) is the
temporary price impact. In this modeling, the generalized evolution of price will take
the form

p(t ) = p + π(V ) + σ [B(t )],  (5.10)

where B is a standard Brownian motion with B(1) = 0. By using this generalized evolu-
tion of price, Gabaix et al. (2006) showed that a power law describing the evolution of
financial returns can result from an optimizing behavior in line with a classical meth-
odology used in financial economics. Specifically, the liquidity supplier and the large
fund will try to maximize their utility function formalized as
δ
U = E[w] − λ [Var ( w )] 2 ,  (5.11)

where λ is a positive factor describing the aversion to risk, and δ is the δ order of
th

risk aversion. W (r ) is the money earned during the trade, which directly depends on
the expected return E[ r ] generated by the transaction. In other words, the optimizing
 

128╇ Econophysics and Financial Economics

process implies a maximization of the expected value of total return given by the trade
(r). This return must take into account a potential uncertainty (denoted C) on the
mispricing signal M. When the mispricing signal results only from noise, this uncer-
tainty parameter is equal to zero (C = 0); otherwise, C is positive when the mispricing
is real. If the fund is assumed to spend S dollars in assets that it buys for a volume of Vt
in a transaction in which it pays a price concession equal to R(Vt ) , the total return of
the fund’s portfolio can be summarized through the equation

Vt (CM t − R (Vt ) + ut )
rt = , (5.12)
S

where ut is the mean zero noise. By simulating data with this model, Gabaix et al.
(2003, 2006) were able to replicate the power law observed for financial returns.
More formally, their model showed that the evolution of financial returns can gen-
erate a power law when the trading volume is used as an explanative variable. To
put it in another way, a classical model maximizing a utility function based on
wealth can generate a power law related to the trade volume under the condition
that the size distribution of traders follows a power law, r ~kV α , k being a posi-
tive parameter. In this theory, power laws in the financial sphere would result from
another power law characterizing the heterogeneity of investors’ size. The origin
of power laws in financial data would be due to another (unexplained) power law
observed in the size distribution of investors. We find here the same kind of argu-
ment as that used in the previous section. Although this theory incorporates some
key optimizing elements used in the financial mainstream, it must be explained
“by a more complete theory of the economy along with the distribution of firm
sizes [investors]” (Lux 2009, 8). Moreover, the theoretical framework proposed by
Gabaix et al. (2003, 2006) appears to be incompatible with the observation made
by Aoki and Yoshikawa (2007) that financial and foreign exchange markets are dis-
connected from their underlying economic fundamentals, since the first usually
demonstrate power-╉law behavior, while the latter are characterized by an exponen-
tial distribution.31

5.3.1.3.╇ Power Laws and Microstructure of the Market

According to the third category of underlying models, the source of power laws in fi-
nance can be found in the organization of the market. More precisely, authors involved
in this area of knowledge design their models by integrating specific microstructure
features of the markets (i.e., bid-╉ask spread, double auction, etc.) related to the ex-
change mechanism, which they sometimes combine with certain behavioral biases
(mimetism, over-╉or underestimation, etc.) observed in the financial markets in order
to replicate the power laws associated with the evolution of financial returns. In this
context, this category of models sheds new light on the trading conditions in which
power laws can emerge.
 

129  Major Contributions of Econophysics to Financial Economics

First of all, we must remember that the microstructural behaviors of financial mar-
kets have been investigated in financial economics in theoretical frameworks other than
the one set out in ­chapters 1 and 2. In the 1980s there emerged two alternative theo-
retical approaches that took as their starting point a questioning of empirical anoma-
lies32 and of the main hypotheses of the dominant framework: behavioral finance and
financial market microstructure. Both directly called upon the informational efficiency
theory, which, as we have seen, was a crucial element in the birth of financial economics.
Although the theory of financial market microstructure has been developing since the
1980s,33 the first works appeared closer to 1970 with an article by Demsetz (1968),
which looked at how to match up buyers and sellers to find a price when orders do not
arrive synchronously. In 1971, Jack Treynor, editor-​in-​chief of Financial Analysts Journal
from 1969 to 1981, published a short article under the pseudonym of Walter Bagehot,
“The Only Game in Town,” in which he analyzed the consequences when traders have
different motivations for trading. Maureen O’Hara, one of the leading authors of this
theoretical trend, defined market microstructure as “the study of the process and out-
comes of exchanging assets under a specific set of rules” (1995). Financial market mi-
crostructure focuses on how specific trading mechanisms and how strategic behaviors
affect the priceformation process. This field deals with issues of market structure and
design, price formation and price discovery, transaction and timing cost, information
and disclosure, and market-​maker and investor behavior. A central idea in the theory of
market microstructure is that asset prices do not fully reflect all available information
even if all participants are rational. Indeed, information may be unequally distributed
between, and differently interpreted by, market participants. This hypothesis stands
in total contradiction to the efficient-​market hypothesis defended by the dominant
paradigm.34
The second alternative approach is behavioral finance. In 1985 Werner F. M. De
Bondt and Richard Thaler published “Does the Stock Market Overreact?,” effectively
marking the start of what has become known as behavioral finance. Behavioral fi-
nance studies the influence of psychology on the behavior of financial practitioners
and the subsequent effect on markets.35 Its theoretical framework is drawn mainly
from behavioral economics. Behavioral economics uses social, cognitive, and emo-
tional factors to understand the economic decisions of economic agents performing
economic functions, and their effects on market prices and resource allocation. It is
primarily concerned with the bounds of rationality in economic agents. The first im-
portant article came from Kahneman36 and Tversky (1979), who used cognitive psy-
chology to explain various divergences of economic decision-​making from neoclas-
sical theory. There exists as yet no unified theory of behavioral finance.37 According to
Schinckus (2009b), however, it is possible to characterize this new school of thought
on the basis of three hypotheses common to all the literature: (1) the existence of
behavioral biases affecting investor behavior; (2) the existence of bias in investors’
perception of the environment that affects their decisions; 3) the existence of system-
atic errors in the processing of information by individuals, which affects the market’s
informational efficiency. The markets are therefore presumed to be informationally
inefficient.
 

130  Econophysics and Financial Economics

The third category of underlying models from econophysics for explaining power-​
law behaviors of financial data is close to the microstructure behaviors that we find
in financial economics. The approach defended in this category of models can meth-
odologically be broken down into two steps:  (1)  designing the model by defining
specific rules governing agents’ interactions; (2)  simulating the market in order to
replicate (hopefully) the data initially recorded on the financial markets. This method-
ology provides a microsimulation of economic phenomena. It was initiated by Stigler
(1964) and was mainly expanded in the 1990s with the development of what is known
as agent-​based modeling.
Initiated by the Santa Fe Institute in the 1980s, agent-​based modeling was gradu-
ally developed in the 1990s and has now become one of the most widely used tools
to describe dynamic (adaptive) economic systems (Arthur 2005). The methodology
designs sets of abstract algorithms intended to describe the “fundamental behavior”
of agents, formulating it in a computerized language in which agents’ behavioral char-
acteristics are inputs, while outputs are associated with the macro level resulting from
the computerized iterated microinteractions. The microscopic level is characterized
by hypotheses about agents’ behaviors. Agent-​based modeling is also widely used in
economics literature, with authors using it to model many economic phenomena: the
opinion transmission mechanism (Amblard and Deffuant 2004; Guillaume 2004);
the development of industrial networks and the relationship between suppliers and
customers (Epstein 2006; Gilbert 2007; Brenner 2001) the addiction of consumers
to a brand ( Janssen and Jager 1999); the description of secondhand (car) markets
(Izquierdo and Izquierdo 2007), and so on.38 In this section, we mention only agent-​
based models used in finance, and, more specifically, we focus on models in which a
large number of computerized iterations can generate power laws describing the evo-
lution of financial returns.39
Lux and Marchesi (1999, 2000) proposed a model simulating an “artificial finan-
cial market” in which fat tails characterizing large fluctuations (described through
a power law) would result from speculative periods leading to the emergence of a
common opinion among agents, who regularly tend to over and underestimate finan-
cial assets. This agent-​based approach integrating the over/​underestimation bias as a
major interaction rule has been confirmed by Kou et al. (2008) and Kaizoji (2006).
In the same vein, Chen, Lux, and Marchesi (2001) and Lévy et al. (1995, 2000) pro-
posed a model in which traders can switch between a fundamentalist and a chartist
strategy. In this framework, the authors showed that financial returns follow a power
law only when financial prices differed significantly from the evolution of economic
fundamentals (in contrast with the efficient-​market paradigm). Alfarano and Lux
(2005) proposed an agent-​based model in which power laws emerged from interac-
tions between fundamentalists and chartists whose adaptive behaviors are based on
a variant of the herding mechanism. By using only elements from the microstruc-
ture, Wyart et al. (2008) proposed a model in which a power law can result from a
specific relationship between the bid-​ask spread and the volatility. More specifically,
their model generates a power law when these variables are directly related through
the following form:
 

131╇ Major Contributions of Econophysics to Financial Economics

Ask − Bid σ
=k , (5.13)
Price N 

where σ is the daily volatility of the stock, N is the average number of trades for the
stock, and k is a constant. Bouchaud, Mezard, and Potters (2002) and Bouchaud,
Farmer, and Lillo (2009) also defined some conditions in the trade process that can
generate a power law in the evolution of financial returns.
Unlike the models presented in the previous section, these models do not associate
the emergence of power laws in finance with the existence of another unexplained
power law. However, there are gaps in the explanation proposed by these models.
Although they provide particular conditions under which financial distributions can
take the form of a power law, they do not really explain how these conditions might
come about. Models referring to behavioral biases do not explain why the selected
bias would primarily shape the market, while models defining specific conditions in
terms of the relationship between variables involved in the exchange do not explain
why these conditions would exist. It is worth mentioning that the feedback effects
between micro and macro levels lead to complex behaviors that cannot be analytically
studied in terms of renormalization group theory (Sornette 2014).
Although work in this category avoids the argument consisting in justifying the
emergence of a power law through the existence of another power law, it raises an-
other curiosity, since power laws appear as an emergent macroproperty resulting from
a large number of computerized iterations characterizing microinteractions (Lux
2009). By identifying the trading conditions in which a power law can emerge, this
category of models sheds new light on the occurrence of such laws. However, this idea
of a power law as an emergent property still generates debate, since its “microfounda-
tions remain unclear” (Gabaix 2009, 281). Theoreticians involved in self-╉organized
criticality models, which are presented in the following section, have also used this
reference to an emerging mechanism.

5.3.1.4.╇ Self-╉Organized Criticality

The fourth type of generative models is based on “self-╉organized criticality.” Such
models are another direct application in finance of the theoretical framework of
statistical physics that we studied in Â�chapter 3. They are based on the idea that the
phenomena studied by their very nature maintain themselves continuously at their
critical point.
As explained in Â�chapter  3, aggregate physical phenomena can appear that show
macroproperties distinct from the properties associated with their microcomponents.
When this perspective is imported into economics, agents can be considered to be
interacting particles whose adaptive behaviors create different structures (such as mol-
ecules, cells, crystals, etc.). These studies are methodologically in line with the class of
models presented in the previous section, since the calibration of microinteractions
requires assumptions concerning individual behaviors, which can generate an (unex-
pected) emerging macro order. Unlike the studies discussed in the previous section,
 

132  Econophysics and Financial Economics

these works implement an agent-​based modeling combined with noneconomic as-

sumptions to calibrate microinteractions, as explained below.
There is a growing literature combining econophysics and agent-​based mod-
eling. Bak et al. (1997) used a reaction diffusion model to describe the dynamics of
orders. In this model, orders were particles moving along a price line, whose random
collisions were seen as transactions (see also Farmer et al. 2005 for the same kind of
model). Maslov (2000) tried to make the model developed by Bak et al. (1997) more
realistic by adding specific features related to the microstructure (organization) of the
market. In the same vein, Challet and Stinchcombe (2001) improved Maslov’s (2000)
model by considering two particles (ask and bid) that can be characterized through
three potential states: deposition (limit order), annihilation (market order), and evap-
oration (cancellation). Slanina (2001) also proposed a new version of Maslov’s model
in which individual position (order) is not taken into account but rather substituted
by a mean-​field approximation. Some authors used an agent-​based approach to char-
acterize the emergence of nontrivial behavior such as herding: Eguiluz et al. (2000),
Stauffer et al. (1999), and Wang et al. (2005), for example, associate the information-​
dissemination process with a percolation model among traders whose interactions
randomly connected their demand through clusters.40
These works can be methodologically characterized by a noneconomic agent-​
based approach since noneconomic assumptions are initially made or used for the
calibration of microinteractions. In this perspective, econophysicists define algo-
rithmic rules (generating microinteractions) that are calibrated in terms of physi-
cally plausible events. Therefore, agents and their interactions are defined in terms
usually applied to physical systems such as potential states (deposition, cancella-
tion, annihilation, etc.), thermal features (heat release rate, ignition point, etc.),
or magnetic dimensions (magnetic permeability, excitation). In line with the com-
ment made for the previous category of models, the computerized technique used
by econophysicists to generate a power law implies the idea that these macro-​laws
are an emergent property. Although this emerging mechanism appears strange at
first sight, it is not new, given the development of the self-​criticality theory intro-
duced, in physics, by Bak et al. (1987) and applied in finance by Bak et al. (1988).41
According to Bak, the linearity visually identified in the histogram related to the
occurrence of a phenomenon can be interpreted as the expression of the phenom-
enon’s complexity (Bak 1994, 478). Although this theory was developed to describe
the emergence of power laws characterizing the evolution of physical systems, its
theoretical justification is often used by econophysicists to defend the existence of
similar macro laws in the finance area. Concerning the visual linearity observed in a
log-​log scale graph, Bak explained:

This is an example [plot with occurrences of earthquakes] of a scale-​free phenomenon: there

is no answer to the question “how large is a typical earthquake?” Similar behaviour has been
observed elsewhere in Nature… . The fact that large catastrophic events appear at the tails of
regular power-​law distributions indicates that there is “nothing special” about those events,
and that no external cataclysmic mechanism is needed to produce them. (Bak 1994, 478)
 

133  Major Contributions of Econophysics to Financial Economics

The author associated “self-​organized criticality” with “slowly driven dynamic sys-
tems, with many degrees of freedom, [that] naturally self-​organize into a critical state
obeying power-​law statistics” (Bak 1994, 480). More specifically, some physical sys-
tems appear to be ruled by a single macroscopic power law describing the frequency
at which phase transitions occur (Newman 2005). Chapter 3 already introduced this
idea when the foundations of statistical physics were presented. The particularity of
the self-​organized criticality refers to the assumption that certain phenomena main-
tain themselves near a critical state. An illustration of such a situation is a stable heap
of sand to which the addition of one grain generates miniavalanches. At some point,
these minicascades stop, meaning that the heap has integrated the effect of this addi-
tional grain. The sand heap is said to have reached its self-​organized critical state (be-
cause the addition of a new grain of sand would generate the same process). Physicists
talk about a “critical state” because the system organizes itself into a fragile config-
uration resting on a knife-​edge (where the addition of a single sand grain would be
enough to modify the sand heap).
This self-​organized criticality has been extended in economics by Bak et al. (1993),
who proposed a model in which a shock in the supply chain (which acts as an ad-
ditional grain in the sand heap) generates economy-​wide fluctuations (like miniava-
lanches in the sand) until the economy self-​organizes critically (i.e., at a fragile state
that could easily be modified by an additional small shock). In their model, the au-
thors showed that the occurrence of large fluctuations in the economy can statistically
be described through a power law. This conceptual framework can also be applied in
finance. Ponzi and Aizawa (2000), Bartolozzi et al. (2004), and Dupoyet et al. (2011)
proposed models describing financial markets as self-​organized critical systems in
which actors would tend to drive the market to a stable state. Once the market is close
to this stable level, profits become lower and lower until negative variations (returns)
are generated. These variations act as the extra grain of sand, leading to a cascade of
losses, putting some agents out of the market until a situation in which profit for the
remaining actors becomes possible again. In this context, new agents, attracted by the
profit, re-​enter the market, which therefore will tend toward a stable state, and so on.
In this perspective, power laws are presented as fluctuations around the market stable
state, which is looked on as a critical state perennially set on a razor’s edge.
The idea that the phenomenon is always at its critical point allows physicists to give
a meaning to the emergence of power laws in their field. Indeed, for a large number of
self-​organized critical systems, Bak et al. (1987) showed that many extreme variations
(i.e., critical modifications of the system) follow a power-​law distribution.
In contrast with the first two categories of models, which were based on a dynam-
ical mechanism for producing power laws that are stochastic processes and in which
noise was supplied by an external source,

Under appropriate conditions it is … possible to generate power laws from deterministic

dynamics. This occurs when the dynamics has a critical point. This can happen [with] self-​
organized criticality, which keep a system close to a critical point for a range of parameters.
Critical points can amplify noise provided by an external source, but the amplification is
 

134╇ Econophysics and Financial Economics

potentially infinite, so that even an infinitesimal noise source is amplified to macroscopic

proportions. In this case the properties of the resulting noise are independent of the noise
source, and are purely properties of the dynamics. (Farmer and Geanakoplos 2008, 43)

In other words, the emergence of power laws results from the evolution of microcon-
figurations of the system itself and not from an external factor such as the existence of
another power law or frictions in the growth rate.
However, as with the previous category of works, the self-╉organized criticality
theory is silent on how the power laws would emerge. This framework associates
power laws with an emergent statistical macro result. Econophysicists have justi-
fied the existence of power laws in finance through two arguments: (1) microscopi-
cally, the market can be seen as a self-╉organized critical system (i.e., sand heap) and
(2) macroscopically, the statistical characterization of the fluctuations in such a self-╉
organized system appears to be “universally” associated with power laws since “there
are many situations, both in dynamical systems theory and in statistical mechanics, in
which many of the properties of the dynamics around critical points are independent
of the details of the underlying dynamical system” (Farmer and Geanakoplos 2008,
45–╉46).
Although the four categories of works presented in this section do not really ex-
plain the reasons why power laws emerge on the financial markets, they open concep-
tual doors for a potential economic interpretation of these macro laws. The following
section will deal with the second crucial step for the potential development of an inte-
grative collaboration between econophysicists and economics: the creation of statis-
tical tests validating the power laws.

5.3.2.╇ New Quantitative Tests

As Â�chapter 3 explained, the existence of a power law is commonly tested by econo-
physicists through a visual inspection:  the authors plot the data in a double loga-
rithmic scale and attempt to fit a line to part of it. This procedure dates back to Pareto’s
work at the end of the nineteenth century. Unfortunately, this method generates sig-
nificant systematic errors by wrongly attributing power-╉law behaviors to phenomena42
(Clauset, Shalizi, and Newman 2009, Stumpf and Porter 2012, Gillespie 2014). From
a financial-╉economics viewpoint, such visual tests have two major drawbacks. First,
they provide no objective criterion for determining what a “good fit” is. Second, as
already explained in chapters 1
� and 4, financial economists consider only statistical
tests scientific. Therefore, empirical investigations from the literature of econophys-
ics tend to be regarded with suspicion by financial economists. However, it is worth
mentioning that financial economists are not alone in noting the weaknesses of visual
inspection; some econophysicists have also pointed out that “better and more careful
testing is needed, and that too much of data analysis in this area relies on visual inspec-
tion alone” (Farmer and Geanakoplos 2008, 24).
The difficulty of implementing statistical tests dedicated to power laws is not
unrelated to the fact that historically the vast majority of statistical tools have been
 

135  Major Contributions of Econophysics to Financial Economics

developed in the Gaussian framework (due to the asymptotic properties of the

central-​limit theorem), which is not suitable for testing power laws. In other words,
satisfactory statistical tools and methods for testing power laws do not yet exist. This is
a big challenge, and one that very few authors have been working on. Moreover, from
the perspective of financial economics, there are several obstacles to the development
of statistical tests dedicated to power laws. For instance, Broda et al. (2013) explain
that the computation of stable Paretian distribution density is complex. While with
modern computing power parameter estimation is no longer a hindrance, “There still
appears to be no existing method which is both fast and delivers the high accuracy
required for likelihood optimization” (2013, 293). In addition, these authors point
out that “the real problem with the use of the stable-​Paretian, or any skewed, fat-​tailed
distribution for modelling the unconditional distribution of asset returns, is that they
cannot capture the time-​varying volatility so strongly evident in daily and higher fre-
quency returns data” (Broda et al. 2013). Last, according to these authors, the stability
property of the stable distribution and the definition of log returns imply that “the
tail index of the return distribution should remain the same at any frequency, i.e., in-
traday, daily, weekly, monthly, etc. However, it is well known that this is usually not the
case, with, say, daily returns exhibiting a tail index considerably lower than two, but
monthly data exhibiting nearly normal behavior. This occurs because, for such series,
the returns are not i.i.d. stable Paretian, but rather have a distribution such that, via a
central-​limit theorem, their sums approach normality” (Broda et al. 2013).
Although no appropriate statistical tools exist to date, the use by econophysicists
of stable Lévy processes based on power-​law distribution has implicitly generated a
need for statistical tools to test the power-​law behaviors of data. In recent years signif-
icant results have emerged from the rapid expansion of statistical studies on this topic.
The literature has taken two different paths: in one, authors have used a rank-​size rule;
in the other, authors have focused on the size-​frequency relation.
The rank-​size approach starts by transforming the size frequency of the power-​law
distribution, P [X > x] = c x–α​ , by ordering the observations according to rank size:
x(1) ≥ x(2) ≥  ≥ x(r ) ≥  ≥ x(n−1) ≥ x(n) .  (5.15)

Thus, the rank of the observation r is given by r ≈ n * P [X > x(r)] ≈ n * [c x(r)–α​ ].

Considering the logarithms—​ that is, log ( r ) = log (n + c ) − α log ( xr ) which can
take the following form, log ( Rank ) = c − α log (Size ) , where c is a constant—​the
critical exponent α is determined by running the OLS on the log-​log rank-​size re-
gression. Gabaix and Ibragimov (2011) pointed out that the previous OLS log-​log
rank-​size regression is strongly biased for small samples; they demonstrated that
this bias can be reduced by using log ( Rank − 1 / 2 ) rather than log ( Rank ) —​that  is, 
log ( Rank − 1 / 2 ) = c − α log ( Size ) .
Authors who have focused on size frequency have built statistical tests for the
purpose of comparing the power law with other distributions that are very close,
mainly the log-​normal distribution (Eeckhout 2004; Clauset, Shalizi, and Newman
2009; Fujimoto et al. 2011; Malevergne, Pisarenko, and Sornette 2011; Rozenfeld
 

136  Econophysics and Financial Economics

et al. 2011; Goerlich 2013).43 Among these works, the most significant research is
probably by Clauset, Shalizi, and Newman (2009), wherein they presented a set
of statistical techniques that allow the validation of power laws and calculation of
their parameters. “Properly applied, these techniques can provide objective evi-
dence for or against the claim that a particular distribution follows a power law”
(Clauset, Shalizi, and Newman 2009, 692). As explained hereafter, their test follows
three steps.
The first step aims to estimate the two parameters of the power law by using the
method of maximum likelihood:  the lower bound to the power law,44 xmin, and the
critical exponent, α. The necessity for estimating the lower bound comes from the fact
that, in practice, few empirical phenomena obey a power law for all values of x. More
often the power law applies only for values greater than some minimum xmin, leading to
the use of truncated distributions (as explained in ­chapter 3) for capturing statistical
characteristics associated with values < xmin. These authors suggest the following max-
imum likelihood estimator:
−1
 
 n xi 
α ≅ 1 + n  ∑ ln
 .
1 (5.16)
 i=1 x min −  
 2

The maximum likelihood estimators are only guaranteed to be unbiased in the asymp-
totic limit of large sample size, n → ∞. For finite data sets, biases are present, but decay
with the number of observations for any choice of xmin. In order to estimate the critical
exponent, α, they provide a method that estimates xmin by minimizing the “distance”
between the power-​law model and the empirical data using the Kolmogorov-​Smirnov
statistic, which is simply the maximum distance between the cumulative distribution
functions of the data and the fitted model. This first step makes it possible to fit a power-​
law distribution to a given data set and to provide an estimation of the parameters
xmin and α.
The second step aims to determine whether the power law is a plausible fit to the
data. Clauset, Shalizi, and Newman propose a goodness-​of-​fit test, which generates
a p-​value that quantifies this plausibility. Such tests are based on the Kolmogorov-​
Smirnov statistic previously obtained, which enables measurement of the “distance”
between the distribution of the empirical data and those of the hypothesized model.
The p-​value is defined as a fraction of the synthetic distances that are larger than the
empirical distance. If the resulting p-​value is greater than 0.1, the power law is a plau-
sible hypothesis for the data; otherwise it is rejected.
The third and last step aims to compare the power law with alternative hypotheses
via a likelihood-​ratio test. Even if a power law fits the data well, it is still possible that
another distribution, such as an exponential or a log-​normal distribution, might give
as good a fit or better. To eliminate this possibility, Clauset, Shalizi, and Newman sug-
gest using a goodness-​of-​fit test again, calculating a p-​value for a fit to the competing
distribution and comparing it with the p-​value for the power law. For each alternative,
 

137╇ Major Contributions of Econophysics to Financial Economics

if the calculated likelihood ratio is significantly different from zero, then its sign indi-
cates whether the alternative is favored over the power-╉law model or not.
Goerlich (2013) extended the article by Clauset et al. (2009) by proposing an-
other kind of test derived from the Lagrange multiplier principle. He showed that
the statistical power of the Lagrange multiplier test is higher than the bootstrapped
Kolmogorov-╉Smirnov test and needs less computational time. However, according
to Gillespie (2014), Clauset, Shalizi and Newman’s method has three main draw-
backs:  “First, by fitting xmin we are discarding all data below that cut-╉off. Second, it
is unclear how to compare distributions where each distribution has a different xmin.
Third, making future predictions based on the fitted model is not possible since values
less than xmin have not been directly modelled.”
These two kinds of tests provide an important contribution to the creation of an
approach more integrated between financial economics and econophysics, because
they pave the way for testing explicative theoretical models based on the power-╉law
behavior of financial prices/╉returns. However, it is worth emphasizing that the de-
velopment of such tests could directly contribute to the development of an integra-
tive approach between econophysicists and financial economists: this kind of test
can establish the scientific justification for using power laws while paving the way for
an appropriate comparison between these statistical laws and patterns usually used
in finance. This statistical approach is very recent and is not widely disseminated
among econophysicists. Moreover, it is worth mentioning that, to date, statistical tests
of power laws have not yet been used with financial data. (They have been used for
wealth, income, city sizes and firm sizes). Despite their drawbacks and the fact that
further investigation is needed, we can consider that these tests have opened the door
to some research into statistical tests. We can add that while visual tests are the most
common in the econophysics’ literature for testing the power-╉law distribution, some
econophysicists use statistical tests. Among other works we may mention Redelico
et al. (2009) and Gligor and Ausloos (2007), who used Student’s t-╉test; Clippe and
Ausloos (2012) and Mir et al. (2014), who used a chi-╉square test; and Queiros (2005),
Zanin et al. (2012), Theiler et al. (1992), Morales et al. (2013).

5.4.╇CONCLUSION
This chapter analyzed the potential uses of econophysics models by financial econo-
mists. Although econophysics offers a set of practical solutions, its theoretical develop-
ments remain outside of financial economics. In this challenging context, the question
of theoretical integration is crucial. First of all, this chapter studied how econophysics
can be useful in trading rooms and also the drawbacks in its broader use. Beyond these
practical implementations, the theoretical contributions of econophysics were also
investigated from the viewpoint of financial economists. The similarities between the
phenomenological approach used by financial economists and the one implemented
by econophysicists have been presented. Afterward, we discussed to what extent this
methodological situation can play a key role in a potential future rapprochement
 

138  Econophysics and Financial Economics

between the two fields. In this context, we also explained how econophysics’ contri-
butions can become effective. Two conditions have then been highlighted:  (1)  the
elaboration of models explaining the emergence of power laws and (2) the creation of
statistical tests validating (or not) these macro laws. As explained, the latter calls for
further research in statistics.
 

6
TO WA R D A C O M M O N F R A M E W O R K

The two previous chapters identified what remains to be done (from an economist’s
point of view) for econophysicists to contribute significantly to financial economics
(and finance) and to develop a fruitful collaboration between these two fields.
Chapter 5 showed that recent developments in mathematics and statistics have cre-
ated an opportunity for an integration of results from econophysics and financial
economics. It pointed out that three paths are still waiting to be investigated: (1) de-
velopment of a common framework/​vocabulary in order to better compare and in-
tegrate the two approaches; (2) creation of statistical tests to identify and to test
power laws or, at least, to provide statistical tests to compare results from econo-
physics models with those produced by financial models; and, finally, (3) work on
generative models for giving a theoretical explanation of the emergence of power
laws. These three axes of research could be considered as a possible research pro-
gram to be developed in the coming years by those who would like to contribute
to financial theory by developing “bridges” between econophysics and financial
economics.
Of the three paths mentioned above, the creation of statistical tests is the most
crucial step from the point of view of attracting the attention of financial economists
and practitioners. As explained in the previous chapters, financial economists have
founded the scientific dimension of their discipline on the use of statistical tests for
the characterization of empirical data. Consequently, such tests are a necessary condi-
tion for doing research in financial economics. Although several problems still present
obstacles to the development of statistical tests dedicated to power laws, significant re-
sults have emerged in recent years (­chapter 5). Moreover, because of the rapid expan-
sion of statistical studies on this topic, new research may develop in the near future.
Developing such tests is a research program on its own and cannot be undertaken here
without leading us very far from the initial objectives of this book. However, from
our point of view, research of this kind is an important first step in the search for a
common theoretical framework between financial economics and econophysics. If
one wants to create statistical tests for comparing results produced by models from
the two fields, the first question is to know what is to be compared. Although econo-
physics is a vibrant research field that has given rise to an increasing number of models
that describe the evolution of financial returns, the vast majority focus on a specific
statistical sample without offering a generic description of the phenomenon. The lack
of a homogeneous perspective creates obvious difficulties related to the criteria for
choosing one model rather than another in a comparison with key models of financial
139
 

140╇ Econophysics and Financial Economics

economics. Consequently, the existing literature offers a broad catalog of models but
no unified conceptual framework. The standardization of knowledge through such a
common framework is a necessary condition if econophysics is to become a strong
discipline (Kuhn 1962).
This chapter will propose the foundations of a unified framework for the major
models published in econophysics. To this end, it will provide a generalized formula
describing the evolution of financial returns. This formula will be derived from a meta-╉
analysis of models commonly used in econophysics. By proposing a generic formula
characterizing the statistical distributions usually used by econophysicists (Lévy, trun-
cated Lévy, or nonstable Lévy distributions), we are seeking to pave the way toward
a more unified framework for econophysics. This formula will be used to propose an
econophysics option-╉pricing model compatible with the basic concepts used in the
classical Black and Scholes framework.

6.1.╇ PROPOSITIONS FOR A COMMON FRAMEWORK

6.1.1.╇ The Issues for a Common Framework
The literature of econophysics proposes a wide variety of models to describe the
probability distribution of stock market price variations, each using various types
of probability functions. Broadly speaking, and in line with what was mentioned in
Â�chapter 4, every econophysicist has his or her own model. To date, none of these
models is supported by statistical tests acceptable according to the scientific crite-
ria used in financial economics (Â�chapter 5). In addition, while the number of such
models continues to grow, no comparison between them has been proposed in
order to identify a general common mechanism. Another question has to be con-
sidered: while models in econophysics seem to provide significant empirical results,
until now there has been no “crucial experiment or test” to demonstrate that they
successfully predict occurrences that financial economics models do not. By that
measure, there is no theoretical basis for preferring econophysics models, and their
use rather than those from financial economics appears to be a choice more ideo-
logical than scientific.
Most conceptual comparisons between econophysics and financial economics
support this conclusion about the force of ideology, since they suggest that the results
and methodology of the two disciplines are not directly comparable. A recent article by
Sornette (2014) illustrates the point. Adopting a physicists’ viewpoint, Sornette pro-
vided a stimulating review and an interesting discussion of the slow cross-╉fertilization
between physics and financial economics. But despite its well-╉documented argument,
the article perpetuates a widespread misunderstanding that keeps the window of con-
troversy between econophysicists and financial economists wide open. This misun-
derstanding is rooted in the crucial differences, explained by Sornette, between the
modeling methodology used by economists and that used by physicists, which is
broadly summarized by the “difference between empirically founded science and norma-
tive science” (Sornette 2014, 3). Sornette added:
 

141  Toward a Common Framework

The difference between [the model for the best estimate of the fundamental price from
physics] and [the model for the best estimate of the fundamental price from financial eco-
nomics, i.e. efficient-​market theory] is at the core of the difference in the modelling strategies
of economists, [which] can be called top-​down (or from rational expectations and efficient
markets), compared with the bottom-​up or microscopic approach of physicists. (7)

The distinction is frequently found in econophysics literature.1 From the elements dis-
cussed in the previous chapters, we can claim that this distinction between the two
modeling “strategies” is false and that the authors who rely on it are confusing two
levels of analysis—​the methodological aspect and the modeling dimension. More im-
portantly, this confusion provides a good illustration of the nature of the differences
between financial economics and econophysics by highlighting a difficulty in compar-
ing them. Let us explore this point.
In the literature of econophysics, it is commonly asserted that the “modelling strat-
egies of economists can be called top-​down” because a “financial modeller builds a
model or a class of models based on a pillar of standard economic thinking, such as
efficient markets, rational expectations, representative agents” in order to draw “some
prediction that is then tested statistically, often using linear regressions on empirical
data” (Sornette 2014, 5). In this context, rational expectations and the idea that mar-
kets are efficient are assumed. So it appears from this formulation that, methodolog-
ically, economists usually start their analysis with assumptions that they implement
in modeling and from which they deduce conclusions; “top-​down” modeling appears
therefore to be associated with the deductive approach mainly used in economics.
Econophysics is, in contrast, presented as a data-​driven field founded on descriptive
models resulting from observations. First of all, it is worth mentioning that the belief
in “no a priori” is in itself an a priori since it refers to a form of positivism. Second, in
such a comparison, this inductive approach is implicitly considered to be a “bottom-​
up” methodology because it starts with data related to a phenomenon rather than as-
sumptions about it. In this case then, the terms “top-​down” and “bottom-​up” refer to
methodological strategies rather than to modeling methods.
Although the difference between the two dimensions seems subtle, it is impor-
tant:  methodology refers to the conceptual way of dealing with phenomena (i.e.,
quantitatively vs. qualitatively; empirically vs. theoretically, etc.), while modeling
methods concern the kind of computation (and data) used by scientists. Actually, by
contrasting the two fields in this way, authors focus on a very specific way of doing
econophysics and economics (i.e., on a very specific part of the literature related to
these two areas of knowledge). On the one hand, although a part of economics is well
known for its representative-​agent hypothesis, there are works in financial economics
that do not necessarily implement top-​down modeling. Agent-​based modeling, which
has become a common practice in (financial) economics, is directly founded on a
micro-​oriented modeling from which an aggregate (bottom-​up) result is derived. On
the other hand, a large part of the literature of econophysics is dedicated to the phe-
nomenological macrodescription of the evolution of financial prices, making these
studies ineligible for consideration as “bottom-​up” modeling.
 

142  Econophysics and Financial Economics

As mentioned in ­chapters 4 and 5, a number of economists have pointed out this

confused way of contrasting econophysics and economics and stressed the limitations
of doing so. In the words of Lux (2009, 230) we have quoted previously,

One often finds [in the literature of econophysics] a scolding of the carefully maintained
straw man image of traditional finance. In particular, ignoring decades of work in dozens of
finance journals, it is often claimed that “economists believe that the probability distribution
of stock returns is Gaussian,” a claim that can easily be refuted by a random consultation of
any of the learned journals of this field.

In addition, while econophysicists use models based on statistical physics in an at-

tempt to reproduce stylized facts, they also use the fact that the Gaussian law is a foun-
dation of financial economics to argue that economists ignore empirical and stylized
fact. However, in doing so, econophysicists skillfully ignore the fact that economists
identified stylized facts on stock price extreme variations and their leptokurticity sev-
eral decades before the emergence of econophysics (­chapter 2).
The misunderstanding between the two communities is often due to a compar-
ison between two different conceptual levels: it is very common for econophysicists
to compare the results of their statistical models with the theoretical framework of
financial economics. When authors explain that econophysics employs a microscopic
approach, they are implicitly referring to the way of dealing with data (experimental
level), contrasting this with the efficient-​market hypothesis, which involves assump-
tions usually made in economics (theoretical level). In this context, it is a truism to
claim that the experimental level is closer to reality than the theoretical one (whatever
the field). Econophysicists seem to consider that economic reality is a given, while
financial economists directly contribute to the determination of the reality they are
supposed to be studying. Physicists are aware that they have an influence on the meas-
urement of physical phenomena, but they do not influence the nature of the physical
phenomena they study. Financial economics is a little different, since financial reality
is a social reality that must be built using a conceptual framework. Consequently, the
efficient-​market hypothesis can be seen as an idealized framework that has been used
theoretically in the development (computerization of financial markets, elaboration
of a legal framework, etc.) of financial spheres (Schinckus 2008). This clarification is
important because an interdisciplinary comparison makes sense only if one compares
the same conceptual levels.
If one wishes to compare the experimental levels and statistical models used in the
two disciplines, the comparison should be between econophysicists’ models and the
ARCH class of models empirically implemented by financial economists. Surprisingly,
and as ­chapter 5 explained, despite the fact that econophysicists and financial econo-
mists in general do not use the same methodology (the former’s is data-​driven, while
the latter’s starts with assumptions), such a comparison shows that they proceed in
the same way in the implementation of their models: both start with an empirical cal-
ibration of their models in order to simulate features (price variations). The sole dif-
ference lies in the way of modeling extreme values:  econophysicists consider these
 

143╇ Toward a Common Framework

values part of the system, whereas financial economists associate these extreme values
with a statistical distribution characterizing the error terms of a major trend. In other
words, financial economists break down stylized facts into two elements, a general
trend (assumed to be governed by a Brownian uncertainty) and a statistical error term
that follows a conditional distribution for which a calibration is required. For these
conditional distributions, it goes without saying that data-╉driven calibrations (i.e.,
without theoretical justification) are common in financial economics. As explained in
Â�chapter 5, the major difference between the two fields appears in how they calibrate
their models:  calibration in econophysics mainly results from data, whereas most
financial economists consider their models to have theoretical foundations from a fi-
nancial (and not merely from a statistical) point of view.2
In conclusion then, econophysicists and financial economists model in the same
way (calibrating models to fit data), but the latter combine the calibration step with a
specific theoretical framework,3 while the former claim to be more data-╉driven. Rather
than focusing on the differences, we see in these similarities a need to develop statis-
tical tools in order to compare the two ways of modeling and then to create a common
framework. However, before talking about a possible comparison, it is important to
know what has to be compared. While the key models of finance are clearly defined
in the literature, there is no agreement on what would be the core models in econo-
physics. The following section will propose a unified framework, generalizing several
important models identified in the literature of econophysics.

6.1.2.╇ A Generic Model for Econophysics’ Price-╉Return Models

For the purpose of providing a framework to unify econophysics models, this section
will propose a generic formula for econophysics models dedicated to the analysis of
extreme values. This formula will open the door to potential comparisons between
econophysics and key models of the financial mainstream. Such comparisons will
also pave the way for the development of a common framework between the two
fields.
Our generic formula results from a meta-╉analysis of econophysics literature deal-
ing with the statistical description of financial returns, integrating more than 20 of
the key models. In our meta-╉analysis, we focused on the manner of describing the
evolution of financial returns from stocks, and specifically on the treatment of extreme
values, since this aspect is a defining element of econophysics. Our selection of models
refers to a specific statistical methodology: they all provide an unconditional descrip-
tion of the evolution of returns. Tables 6.1 to 6.5, given in the appendix to this chapter,
list the main econophysics articles dedicated to distribution of price returns. From
our statistical analysis of these articles (see tables 6.1 and 6.2 for detailed comments
on each article), we propose the following generalized formula for the probability dis-
tribution function:

P(x) = Cf (x)e − g [h( x )]+d ,  (6.1)

 

144  Econophysics and Financial Economics

where C and d are constants that might have temporal variation. The analytical form of
f(x) is not always known for all the values of x, but it has a power law variation in the
limit of large x (x → ∞):

1 (6.2)
f (x )  b1 + a1 α
for x → ∞. 
x

The parameters a1 and b1 (usually equal to 1) define the shape of the distribution at
large x, and α is the principal exponent of the power law. The function g(x) introduced
in equation (6.1) has the form

g (x) = (a2h(x) + b2 )c2  (6.3)

with two possible forms for h(x): x or log(x). As mentioned in c­ hapter 1, the use of
the log-​normal law in finance was introduced by Osborne (1959) in order to avoid
the theoretical possibility of negative prices. Moreover, this use is also based on the
assumption that rates of return (rather than price changes) are independent random
variables. In equation (6.3), a2, b2, and c2 are parameters that differ from one model to
another, defining the final shape of the distribution function. Finally, our generic prob-
ability distribution function can be expressed as
1 c2
(6.4)
P (x )  C b1 + a1α
e −(a2 x+b2 ) +d
.
x

This metaformula makes it possible to rewrite and to compare the probability distribu-
tion function of price changes used in the main econophysics models identified in the
literature that deal with the statistical description of financial returns. Actually, these
models can be classified in three groups depending on the distribution used: stable
Lévy distribution, truncated stable Lévy distribution, and nonstable Lévy distribution.
As previous chapters explained, stable Lévy distributions were first proposed by
Mandelbrot (1962) and afterward used by the pioneers of econophysics to describe
the tail of the distribution of financial data. Most importantly, for large x values,
stable Lévy distributions are well approximated by a power law as described in e-
quation (6.2), with the exponent α having values between 1 and 2, generally around
1.5. Regarding the equation (6.4), most authors4 use a1 = b1 = 1, with one special
case where a1 = −1 and b1 = 1 (Weron, Mercik, and Weron 1999). The parameter
a2 is nonzero only in three cases (Cont and Bouchaud 2000; Gupta and Campanha
2002; Louzoun and Solomon 2001), and Louzoun and Solomon (2001) set c2 = −1.
The other parameters of equation (6.4) are taken to be zero in three cases out of 20
models studied in our meta-​analysis. The observed nonzero a2 involves the pres-
ence of an exponential term in the distribution function P(x), which is derived by
using models to explain the empirical data—​for example, the generalized Lotka-​
Voltera model (Louzoun and Solomon 2001) and the Percolation model (Cont and
Bouchaud 2000; Gupta and Campanha 2002). But most often the authors focus
 

145  Toward a Common Framework

on calculating the power-​law exponent of the distribution tail in some specific

situations.
As explained in ­chapter 3, the main drawback of stable Lévy distributions is that
they have infinite variance, a situation that in physics cannot be accepted (because
physically plausible measures cannot be infinite). Remember that econophysicists de-
veloped truncated Lévy distributions in order to solve the problem of infinite variance.
This is the second category of models covered by equation (6.4). For these models,
stable Lévy distributions are used with the specific condition that there is a cutoff
length for the price variations above which the distribution function is set to zero,
in the simplest case (Mariani and Liu, 2007), or decreases exponentially (Gupta and
Campanha 2002, 52–​55). Using our generic formula, we obtain truncated distribu-
tions when a1 = b1 = 1 and at least a2 and c2 are nonzero—​we should mention that b2
is nonzero in some articles (Gupta and Campanha 1999; Michael and Johnson 2003;
Gabaix et al. 2007). Only Gunaratne and McCauley (2005) give a distribution func-
tion with the d constant nonzero. For the simply truncated distribution proposed by
Mariani and Liu (2007) one can consider a2 very large (going to ∞) beyond the cutoff
length. Concerning truncated stable distribution, one has nonzero a2 and b2 param-
eters that ensure the stability of the distributions and the finiteness of the variance.
Finally, the last category of models covered by our equation (6.4) is those based on
nonstable Lévy distributions, whose use is motivated by several reasons. For instance,
because some empirical studies of financial markets suggested that stable Lévy distri-
butions could overestimate the presence of large price variations even though they fit
the data better than the Gaussian distribution (Gabaix et al. 2007), some authors have
developed a power-​law variation of P(x) for large x but with the values of the char-
acteristic exponent greater than 2. The parameters of equation (6.4) are in this case
a1 = b1 = 1 (Lux 1996; Gopikrishnan et  al. 1999; Mantegna and Stanley 1994;
Alejandro-​Quiñones et al. 2006) and a2 = b2 = c2 = 0—​except for the paper by Clementi,
Di Matteo, and Gallegati (2006), where a1 = −1, b1 = 1, a2 is nonzero, and c2 = 1. In this
last case, one can use the Focker Planck equation for anomalous diffusion to derive
the probability function.5 A nonstable Lévy distribution is also obtained as a special
case of a Tsallis distribution derived by McCauley and Gunaratne (2003). One should
also note that for large x, a Student distribution, such as used by McCauley (2003),
approaches Lévy distributions. Finally, some authors have proposed exponential dis-
tribution functions in terms of logarithmic price differences for the intraday trading
of bonds and foreign exchange (Gallegati et al. 2006; McCauley 2003; McCauley and
Gunaratne 2003). In terms of price differences, such a distribution would be a power
law as described in equation (6.2) (thus in this case h(x) = log(x)).
When dealing with nonstable distributions, econophysicists consider that a2 and
b2 are always zero. A  telling example of this category of models was developed by
Gopikrishnan et al. (1999, 2000) and Stanley et al. (2008), who showed that the char-
acteristic exponent (α) exhibited in equation (6.4) would be approximately equal to
3 for financial markets. These authors studied the distribution of the fluctuations in
an index (S&P 500) between 1984 and 1996, but also price fluctuations for financial
 

146╇ Econophysics and Financial Economics

stocks (from 1926 to 1996); they concluded that in both cases these distributions can
be described with a power law in line with our equation (6.4), whose characteristic
exponent is asymptotically equal to 3—╉meaning that the greater the volume of data,
the higher the probability of approximating 3. In a such context, our generalized prob-
ability distribution function could be simplified as follows:

1 1
P ( x ) ~C ~C . (6.5)
x 1+ 3
x4 

In their publications, Gopikrishnan et al. (1999, 2000) explained that this asymptotic

observation (i.e., the exponent = 3 for a greater volume of data) could not be captured
by classical financial models such as GARCH, because they deal only with short-╉term
investment horizons; note that what is presented as a drawback for econophysicists is
often seen as an advantage for financial economists, who focus on short-╉term invest-
ment strategies (see Â�chapter 5).

6.1.3.╇ Perspectives on the Generic Model

Our generic formula offers a general equation whose domain of applicability is wider
than the particular models studied in our meta-╉analysis. From this perspective, our for-
mula can be seen as a “conceptual generalization” (i.e., an abstract generalization coming
from an inductive meta-╉analysis of existing models [Tsang and Williams 2012]). This
kind of generalization makes sense only because all models studied in our meta-╉analysis
can be described as a specific case of our generic formula. The development of such a
unified framework raises methodological questions: first, beyond the idea of associat-
ing several models, a generalized formula provides a unified conceptual way of thinking
about the evolution of financial returns. By identifying the key parameters appearing in
every model, this generalization paves the way to a better (re)estimation of the weight
of each variable. In other words, this generalized formula can also contribute to new
potential adjustments in the statistical understanding of financial returns by opening
the door to the development of new tests, for example. As mentioned in Â�chapter 5 and
detailed in the following section, this aspect is a key issue in the future evolution of
econophysics and its potential for recognition by financial economists.
By covering the key models developed in econophysics, our generalized formula
points out the coherent progress that has taken place in this field in recent years, but
it also opens up the possibility of a search for common variables between econophys-
ics and financial economics. As explained in Â�chapter 2, the vast majority of models
used by the financial mainstream are developed in the Gaussian framework. Now,
the Gaussian framework can be seen as a specific case of our generalized framework.
Specifically, the Gaussian distribution, which is simplest case of equation (6.4), can
also be obtained for the following specific parameters:

1 1
c= , a1 = b1 = 0 , b2 = d = 0 , c2 = 2 and a2 = .
2 πσ 2
2σ 2
 

147╇ Toward a Common Framework

In this situation, equation (6.4) can be expressed as the Gaussian distribution:

1 − x 2 ( 2 σ2 )
P(x) = e . (6.6)
2 πσ 2 

This point is important for two reasons: first, this generalization shows that a conceptual
bridge between econophysics and financial economics is possible, implying the need to
develop new non-╉Gaussian statistical tests; second, it opens the door to the identifica-
tion of potential statistical similarities that would go beyond the disciplinary perspec-
tives. In other words, our meta-╉analysis proposes a unified equation that can make sta-
tistical sense for both economists and econophysicists. This methodological implication
is in line with the “transdisciplinary analysis” proposed in our previous work (Jovanovic
and Schinckus 2013) in which we suggested that an integrated conceptual framework
could transcend the disciplinary perspectives as summarized in figure 6.1.

Econophysics

Holistic perspective

Financial Mathematical
economics finance

Figure 6.1╇ Transdisciplinarity (cooperation with integrative collaboration)

This transdisciplinary approach to econophysics is possible only if the scientists involved

in this field (economists and physicists) can develop integrative research. Although some
collaborations exist,6 integrative research of this kind has not yet appeared in the litera-
ture. However, in considering that the statistical description can be similar in financial
economics and physics, we implicitly assume an analogical generalization whose role is to
propose a conceptual framework for unifying a field that has not yet been investigated. In
this perspective, our generalized equation can be seen as a preliminary condition for the
development of more integrative research between economists and physicists.

6.2.╇ APPLICATION TO OPTION PRICING

A first application of our generic formula is to develop an option-╉pricing model based
on the econophysics approach that is compatible with the framework of financial ec-
onomics. The model proposed here must be considered a basic and preliminary ap-
plication, because it does not have stochastic volatility (which is generally present
in contemporary modeling of option pricing in financial economics). Nevertheless,
it shows that a theoretical bridge between econophysics and the original Black and
Scholes model (the nonstochastic volatility model) is possible. Our model shows that,
from a theoretical point of view, an unconditional use of a stable Lévy framework (as
enhanced by econophysicists) can make sense in the financial mainstream.
 

148  Econophysics and Financial Economics

The key element in the Black and Scholes model was the application of the “ar-
bitrage argument,” which is an extension of the economic law of one price in perfect
capital markets.7 This law was popularized by Modigliani and Miller (1958)8 and is
used as the foundation of equilibrium in financial economics. In financial economics’
option-​pricing model, the arbitrage argument ensures that the replicating portfolio
has a unique price. In econophysics’ models, this concern does not exist,9 something
completely unacceptable according to the fundamental intuitions of financial eco-
nomics, in which this arbitrage argument is a necessary condition for the uniqueness
of the solution for any option-​pricing methods.
The model presented in this section must be considered a first attempt to work
within an integrated framework between econophysics and financial economics. More
specifically, we will use our generic formula originating in econophysics to describe
the evolution of underlying returns by showing that this statistical description is also
compatible with the necessary condition for the use of an arbitrage argument (so im-
portant in financial economics). As explained below, we will use our generic formula
in its truncated form in order to avoid the problem of infinite variance (discussed in
­chapter 3). Some authors have worked on this issue: Matacz (2000) and Boyarchenko
and Levendorskii (2002, 2000) offered interesting option-​pricing models based on
an exponential truncated stable Lévy distribution. However, although these authors
gave a procedure for minimizing an estimated risk, they did not define the conditions
for a potential optimal risk hedge. In contrast, McCauley et al. (2007b), for example,
showed that a non-​Gaussian option-​pricing model can provide optimal risk hedging;
but their model focused on general conditions without defining an individual cutoff.
Significantly, this existing literature does not directly address the implications of trun-
cated exponential Lévy models for the optimal option-​hedge ratio problem. However,
this is of interest in an econophysics context as the result of various contributions
stretching back to studies by Mantegna and Stanley in the mid-​1990s that employed
truncation to deal with statistical limitations of “infinite variance.”
Our model aims to define the minimal condition required for an optimal risk hedg-
ing for a particular cutoff based on an exponentially truncated stable Lévy distribution.
In other words, our contribution is to show that a (nonconditional) exponential stable
Lévy description of a financial distribution (i.e., description of the whole distribution, as
usually proposed by econophysicists) is compatible with an admissible hedging strategy
in the sense defined by Harrison and Kreps (1979) and Harrison and Pliska (1981),
who showed that the existence of an equivalent martingale measure is a necessary con-
dition for optimal hedging (see Carr and Madan 2005 for further details on this theme).
It is worth bearing in mind that this probabilistic framework defined by Harrison, Kreps,
and Pliska has progressively become the mainstream in financial economics (­chapter 2).
Section 6.2.1 presents our generalized model in line with the Black and Scholes
framework (based on a martingale measure), and section 6.2.2 defines a specific risk
measure for a truncated stable Lévy version of this model that will be presented in sec-
tion 6.2.3. In line with McCauley et al. (2007a), we use an exponentially truncated stable
Lévy distribution whose statistical conditions are defined to make this model viable
(meeting the necessary condition) in the sense suggested by Harrison and Kreps (1979).
 

149  Toward a Common Framework

6.2.1.  An Econophysics Model Compatible

with the Financial Economics Framework
Let us consider a portfolio made up of a (call) option and a short position on ϕ stocks.
At time t = 0, the value of this portfolio V is V = C − φS , where C is the price of the call
option (with strike price K and expiration T); S is the stock price, and ϕ is the quan-
tity of the stock. The product φS = ∑ φiSi is to be seen as a scalar product with Si the
price for each stock. Initially the stock price is considered to be S0, and the portfolio is
considered to be self-​financing. In other words, the value of the portfolio only changes
with changes in the stock price. In this situation the variation of the portfolio between
time t = 0 and T is given by ∆V = ∆C − φ∆S . The variation of the portfolio due to the
call option is, when continuously discounted at the risk-​free rate r,

e − rT ∆V = e − rT max(S − K , 0) − e − r 0C(S0 , K ,T ) − e − rT φ∆S ,  (6.7)

where the first term is the value of the option at time T and the second term is the pre-
mium paid for the option at time t = 0. Assuming that one is operating in a no-​arbitrage
market, the discounted stochastic price process must be a martingale (Harrison and
Kreps 1979). The option price is
∞

C(S0 , K ,T ) = e − rT E[max(S − K , 0)] = e − rT ∫ (S − K ) f (S)dS. (6.8)

K


Due to the stochastic nature of the price process, risk is inherent in the financial evalu-
ation of options and stocks. Black and Scholes (1973) showed that for the log-​normal
distribution, this risk can be hedged by using an appropriate hedging condition (the so-​
called ϕ hedging) for the financing strategy. But for nonnormal models, the Black and
Scholes procedure for hedging risk no longer works.10 A measure of risk that was also used
in Bouchaud and Sornette (1994) and Aurell et al. (1997) is the variance of the value of
the portfolio V = C − φS . We make the supposition here that this variance is finite.11 Thus:

R = E[∆V 2 ] = E[(max(S − K , 0) − C(S0 , K ,T ) − φ∆S)2 ]. (6.9)

First of all, note that for uncorrelated assets, one has the following expression:
E[(φ∆S)2 ] = φ2 σ 2 = ∑ φi2 σi 2 , where σi is the volatility. However, when there exists a
correlation between the assets, one can write E[(φ∆S)2 ] = ∑ φi2σi2 = 2 ∑ φi φ jσij , where
i,j
σij is the covariance matrix. In a sense, our conceptual model defined in equation (6.9)
is in line with the generalized call-​option pricing formula defined by Tan (2005), in
which S is observed for non-​Gaussian distributions. This approach is well known in
finance and requires a minimization of the risk with respect to the trading strategy:
∞
1 1
φ* = E[(S0 − S)max(S − K , 0)] = 2 ∫ (S0 − S)(S − K ) f (S)dS. (6.10)
σ 2
σ K 
 

150  Econophysics and Financial Economics

This equation is valid for a martingale process S with E[∆S] = 0 , ensuring the necessary
condition for an optimal hedging solution. If there is more than one uncorrelated asset
(stock), the above equation should be applied for each stock individually in order to
obtain the total optimal hedging strategy. The optimal strategy for the ith asset would
be written like the above equation with index i on all variables. For many correlated
assets using E[(φ∆S)2 ] = ∑ φi2σi2 = 2 ∑ φi φ j σij , one finds
i ,j

1  
∞

φi* = 2 ∫
(Si 0 − Si )(Si − K ) f (Si )dSi − ∑ φ j σ ij  , (6.11)
σi  K j  

It is straightforward to observe that in the simplest case of the Gaussian distribution

with log-​returns, the optimal hedging strategy given in equation (6.10) is the same
dC
as the hedging strategy from the Black and Scholes model, that is,   . The min-
 dS 
imal risk R corresponding to the optimal hedging strategy is obtained from equa-
tion (6.9):  R * = RC − φ *2σ 2 for one stock. Note that RC is a risk term not dependent
2
∞
∞ 
on the investment strategy, defined as RC = ∫ (S − K )2 f (S)dS −  ∫ (S − K ) f (S)dS .
K K 
In general cases with many correlated assets, minimal risk is obtained by taking
E[(φ∆S)2 ] = φ2 σ 2 = ∑ φi2 σi2 into account. Bouchaud and Sornette (1994) showed
that R* vanishes when a log-​normal density is used. In other words, equation (6.10)
provides a martingale-​based risk measure, which must be minimized in an incomplete
market structure.

6.2.2.  Hedging Strategy with Exponentially

Truncated Lévy Stable Distribution
As previously mentioned, a stable Lévy framework generates infinite variance, and
hence a noncomputable risk, leading authors to search for a solution. To solve this
problem, this section focuses on a particular cutoff based on an exponentially trun-
cated stable Lévy process for option pricing by determining conditions for which the
risk measure defined in the previous section could be viable in a hedging framework
developed by Harrison and Kreps (1979) (i.e., the risk measure for our model is based
on a martingale). We start our reasoning with our generalized formula for the proba-
bility distribution function (6.4) we defined earlier:

1 −(a2 x+b2 )c2 +d

P (x )  C e . (6.4)
xb1 +a1α

In line with the central-​limit theorem, a stable Lévy regime will converge12 toward
a Gaussian asymptotic distribution after a very high number of variables x. In other
words, there is a cross-​value l after which the stable Lévy process is assumed to switch
 

151  Toward a Common Framework

into the asymptotic (Gaussian) regime. That particular evolution implies a Gaussian
asymptotic regime for x > l and a truncated stable regime for x < l. The former can also
be seen as a specific case of the generic formula (6.4) with the following parameters,
1 1
C= , a1 = b1 = 0, b2 = d = 0, c2 = 2, and a2 = ; the latter can be described
2 πσ 2
2σ 2
through the following specific case where distribution density function for the log re-
turns of underlying options can take the form

e− γ x
f (x) = C α +1 , 
(6.12)
x

where x = log (S/​S0), C > 0, γ ≥ 0, and 0 < α < 2, which is the necessary condition
for a stable Lévy distribution. C can be seen as a measure of the level of activity in
a case where all other parameters are constant (i.e., a parameter of scale). The pa-
rameter γ is the speed at which arrival rates decline with the size of the move (i.e.,
rate of exponential decay). This model accords with studies dealing with exponen-
tial truncation exposed in ­chapter 3, and the formula is a symmetric version of the
so-​called CGMY model (named after the authors of Carr, Geman, Madan, and Yor
2002) and a generalization of the exponentially truncated Lévy stable models pro-
posed by Koponen (1995) and by Boyarchenko and Levendorskii (2002, 2000).
However, while Carr et al. (2002) applied this model in a time-​changed framework,
Koponen (1995) did not apply his model to option pricing, while Boyarchenko
and Levendorskii (2002, 2000) did not seek conditions for a potential risk hedge.
Our objective here is to show that a stable Lévy regime (with no time-​changed dis-
tribution) is theoretically compatible with a key assumption of the financial main-
stream. Consequently, the rest of this chapter will focus only on the stable Lévy
regime.
Because stable Lévy processes generate infinite variance, we use an exponential
truncation implying an exponential decay of the distribution. This restriction means
that the truncated distribution generates finite variations, making possible the esti-
mation of the variance (in the stable Lévy regime), which is given by the following
equation:
l

σ 2 = 2C γ α-2 Γ(2 − α) with Γ(z)=∫ e − t t z −1dt . (6.13)

0


Using the general equation (6.6), we calculate the option price for this model for the
chosen portfolio, by considering the density distribution of stock returns:
∞
e− γx
C = e − rT ∫ (S0 e − x − K )C
ln( K /S0 )
dx.
x α +1 
(6.14)
 

152  Econophysics and Financial Economics

∞ −u ∞
e E (x ) e − xt
Using the result ∫ n
du = nn−1 with En (x) = ∫ n dt in equation (6.14), and express-
x
u x 1
t
ing C as a function of squared volatility, yields:
−α
σ 2 e − rT   K     K    K   
C=  ln  S0 Eα +1 ( γ − 1) ln    − KEα+1  γ ln     .  (6.15)
2 γ Γ(2 − α)   S0  
α −2
   S0     S0   

Given this result, we can estimate the hedging strategy that minimizes risk by using
equation (6.10):
∞
C e−γx
σ 2 ln( K∫/S0 )
φ* = (S − S e x
)(S e x
− K ) dx. (6.16)
x α+1 
0 0 0

However, this optimal hedging can be implemented in the nonasymptotic regime,

implying that the variance (equation (6.13)) is finite only for x < l. By integrating
l

Γ(2 − α) = ∫ e − t t z −1dt
0
, we can detail equation (6.16) as

   K  
(S0 K + S0 )Eα+1 (γ − 1)ln   
2

1   K     S0   
φ = α −2 1−α − l
*
*  ln  .
2 γ l (e − 1)   S0    2   K    K  
 − S0 Eα +1 (γ − 2)ln  S   − S0 KEα +1  γ ln  S   
  0   0 

Although Tan (2005) did not deal with infinite variance processes, we came to the same
conclusions as he did about non-​Gaussian densities, φ , which explicitly depend on
*

(1) higher partial derivatives of the call-​option pricing function toward the price of the
underlying asset; and (2) the value of the cumulants (as they are used in the logarithm of
the characteristic function). Although equation (6.14) could then be further generalized
as proposed by McCauley et al. (2007b) and Tan (2005), generalization would require
specific statistical conditions (defined in this chapter) in order to offer a viable hedg-
ing solution in a stable Lévy framework. Our objective here is to show the theoretical
compatibility between an exponentially truncated stable Lévy framework and the nec-
essary condition for an optimal hedging in the Harrison, Kreps, and Pliska theoretical
framework.
 

153╇ Toward a Common Framework

6.3.╇CONCLUSION
The objective of this final chapter was to develop a unified framework generalizing
major models found in econophysics literature. The identification of such a framework
allowed us to work on the minimal condition under which it could be compatible with
the financial mainstream. This task suggests that a potential fruitful future collabora-
tion between econophysicists and financial economists is possible.
The first step in the elaboration of a conceptual bridge between the two fields was
to show that, in the diversified econophysics literature dealing with extreme values in
econophysics, it was possible to deduce a unified technical framework. We proposed
such a unification in the first section of this chapter by showing that econophysics
models can be seen as a specific derivation of a generalized equation. The second
step was to show that the generalized equation elaborated to describe the key models
developed in econophysics could be compatible with a strictly Gaussian approach.
That was the aim of our second section. Equations (6.4) and (6.6) showed that the
Gaussian framework can be expressed as a specific case of the generalized equation.
This point is important, since it facilitates the potential development of a common
vocabulary between the two communities. While equation (6.4) highlighted the sta-
tistical parameters common both to models used in econophysics and to those used
in finance, the next step will be to give them an economic/╉financial meaning. In a
sense, proposing a generalized equation is a preliminary condition for this kind of
theoretical investigation, which requires a combination of theoretical and technical
knowledge of the key assumptions used by econophysicists and economists, because
the interpretation of the statistical parameters must make sense for both communi-
ties. After defining such a unified equation, we presented it in the light of the financial
mainstream. We used our generic formula originating in econophysics to describe
the evolution of underlying returns by showing that this statistical description is also
compatible with the necessary condition for the use of an arbitrage argument. The
model proposed here must be considered a basic and preliminary application whose
objective is to stress the feasibility of a conceptual framework common to econo-
physics and financial economics. Hopefully, this first attempt will generate further
research on the topic.

APPENDICES: MODELS STUDIED IN OUR SURVEY

  
Table 6.1  Lévy stable –​Gaussian distributions

Reference Author formula Generic formula

1 1 c2
e −(a2 x+b2 ) +d
for x → ∞
2
/( 2 σ2 )
1. Bachelier (1900) P(x) = e−x P(x) ~ C b1 + a1 α
2 πσ 2 x
The first model for the stochastic process of returns
1
C=
2 πσ2
, h(x)=x

a1 = 0, b1 = 0, a2 = 1/​(2σ2)0.5, b2 = 0, c2 = 2, d = 0

1 1 2
 log( x/x0 )−( µ −σ 2 /2 ) /( 2 σ2 ) 1 −(a2 x+b2 )c2 +d
2. Black and Scholes (1973) P(x) = e 
P(x) ~ C e for x → ∞
An option pricing technique –​the Gaussian distribution x 2 πσ 2
x b1 +a1α
1
is one of the principal assumptions C= , h(x)=log(x)
2 πσ2
a1 = 0, b1 = 0, a2 = 1/​(2σ2)0.5, b2 = 0, c2 = 2, d = 0

1 2
/( 2 σ2 ) 1 c2
3. Clark (1973) P(x) = e−x P(x) ~ C b1 + a1 α
e −(a2 x+b2 ) +d
for x → ∞
Mixture of Gaussian distributions 2 πσ 2
x
1
C= , h(x)=x
2 πσ2
a1 = 0, b1 = 0, a2 = 1/​(2σ2)0.5, b2 = 0, c2 = 2, d = 0

 (log(s ′ / s) + BT )2  1 c2
4. Martin Schaden (2002) P(s ′ ,t f ;s ,t i ) = (4 πσ2Ts ′s)−1 exp   P(x) ~ C b1 + a1 α
e −(a2 log( x )+b2 )
+d
for x → ∞
 2 σ 2T  x
s′ –​price at time tf,; s –​price at time t; T=tf–​ti h(x) = log(x)
a1 = 0, b1 = −1, a2 = 1/​(2σ2T)0.5,
b2 = [BT-​log(s)]/​ (2σ2T)0.5, c2 = 2, d = 0
  
Table 6.2  Lévy stable –​Paretian distributions

Reference Author formula Generic formula

1 −(a2 x+b2 )c2 +d
5. Weron et al. (1999) C1(λr)α −1 , 0 ≤ λr ≤ 1 P(x) ~ C e for x → ∞
f (r) =  − ( α/k )−1 x b1 +a1α
They use the conditionally exponential decay model
C2 (λr) , λr ≥ 1
a1 = −​1, b1 = 1, a2 = b2 = c2 = 0
Power-​law distribution
α ≅ 1.09 α ≅ 1.09
Apply this distribution function for modelling daily returns of the
C , C depend upon parameters α, λ and k h(x) = x, d = 0
DJIA and S&P500 financial indices as well as returns of the USD/​DEM 1 2
exchange rate
1 −(a2 x+b2 )c2 +d
6. Louzoun and Solomon (2001) P(xi ) = xi−1−α e −2 a/( Dxi ) P(x) ~ C e for x → ∞
x b1 +a1α
They use the Generalized Lotka Voltera (GLV) model to explain a1 = 1, b1 = 1, a2 = 2a/​D, b2 = 0, c2 = –​1
power law distributions in individual wealth (Pareto laws) and in α = 1.5
α ≅ 1.5
financial market returns h(x) = x, d = 0

1 −(a2 x+b2 )c2 +d

7. Blanck and Solomon (2000) P(w) ∝ w −1−α P(x) ~ C e for x → ∞
x b1 +a1α
They model the distribution of population in cities α ≈ 1.4
a1 = 1, b1 = 1, a2 = b2 = c2 = 0
and that of the wealth
h(x) = x, d = 0
Power law with α ≈ 1.4
α ≅ 1.4

(continued)
  
Table 6.2 (Continued)
Reference Author formula Generic formula
1 −(a2 x+b2 )c2 +d
8. Malcai, Biham et al. (1999) Power law α=1.5 P(x) ~ C e for x → ∞
x b1 +a1α
Power law with α = 1.5 Idem to article 7 above.
Obs. Articles 6, 7, 8 are connected a1 = 1, b1 = 1, a2 = b2 = c2 = 0
h(x) = x, d = 0
α ≅ 1.5

1 c2
9. Scalas and Kim (2007) The characteristic function for symmetric P(x) ~ C b1 + a1 α
e −(a2 x+b2 )
+d
for x → ∞
x
Lévy α stable distribution distributions:
a1 = 1, b1 = 1, a2 = b2 = c2 = 0
Ex: α = 1.57, β = 0.159, γ = 6.76 × 10−​3, δ = 3.5 × 10−​4
General forms for Lévy distributions
(
Lα (k) = exp −a k
α
) h(x) = x, d = 0
1 α ≅ 1.4
The paper illustrates a procedure for fitting financial data with Lévy Lα (x) ~ 1+α for x → ∞
x
α-​stable distributions

10. Mandelbrot (1963b) (

Lα (k) = exp −a k
α
) P(x) ~ C
1 −(a2 x+b2 )c2 +d
x b1 +a1α
e for x → ∞
Application of Lévy distributions for cotton price changes (rather log 1
Lα (x) ~ 1+α for x → ∞ a1 = 1, b1 = 1, a2 = b2 = c2 = 0
differences) in USA (1800 –​1958) x
h(x) = x, d = 0
α = 1.7, β = 0, δ = 0
  
1 c2
11. Skjeltorp (2000) Lévy stable -​α = 1.64 P(x) ~ C b1 + a1 α
e −(a2 x+b2 ) +d
for x → ∞
x
Analyse variations on the Norwegian and USA stock markets
a1 = 1, b1 = 1, a2 = b2 = c2 = 0
Measure the local Hurst exponent H (related to α; α = 1/​H)
h(x) = x, d = 0
Reach the conclusion that Lévy stable distributions describe
α ≅ 1.64
empirical data much better than Gaussian.
H is obtained from log-​log plot of P(Dt) when Dx = 0

1 1
exp( − ε 2 S )
c2
12. Cont and Bouchaud (2000) P(S) ~ P(x) ~ C b1 + a1 α
e −(a2 x+b2 )
+d
for x → ∞
S 5/2 x
Percolation theory—one gets the distribution for the cluster
S—cluster size; ε = 1−​c a1 = 1, b1 = 1, a2 = ε2, b2 = 0, c2 = 1
(of financial operators) size—identic to the distribution of price
for c = 1, one has a pure power law, Lévy α ≅ 1.5
changes according to percolation theory
symmetrical, distribution with α = 3/​2. h(x) = x, d = 0
The probability that financial operators interact between them is
ε =1−​c
c/​N; N is the total number of operators
h(x) = x, d = 0

1 1
13. Bouchaud (2002) exp( − ε 2 S )
c2
P(S) ~ P(x) ~ C b1 + a1 α
e −(a2 x+b2 )
+d
for x → ∞
S 5/2 x
α = 3/​2 a1 = 1, b1 = 1, a2 = ε2, b2 = 0, c2 = 1
h(x) = x, d = 0
α ≅ 1.5
  
Table 6.3  Lévy non-​stable distributions

Reference Author formula Generic formula

P( g i (t ) > g ) ∝ g − α 1 −(a2 x+b2 )c2 +d

14. Gopikrishnan, Mayer, Amaral, Stanley (1998) P(x) ~ C e for x → ∞
α≈3 x b1 +a1α
Power law (asymptotic) distribution
a1 = 1, b1= 1, a2 = b2 = c2 = 0
α = 3
h(x) = x, d = 0
α≅3

1 −(a2 x+b2 )c2 +d

15. Lux (1996) See comments P(x) ~ C e for x → ∞
x b1 +a1α
Examination of German stocks with stable Lévy -​Power law –​Lévy non-​stable
a1 = 1, b1= 1, a2 = b2 = c2 = 0
distributions
h(x) = x, d = 0
A detailed analysis leads to conclusion that Lévy
α >2
stable is not such a good fit –​it is assumed that the
real distribution is not stable

1 1 −(a2 x+b2 )c2 +d

16. Gopikrishnan, Plerou, Amaral, Meyer, P( g > x) ∝ P(x) ~ C e for x → ∞
xα x b1 +a1α
Stanley (1999)
G− G a1 = 1, b1= 1, a2 = b2 = c2 = 0
They analyse the values of the S&P500 index g is a normalized return : g = T

v h(x) = x, d = 0
ν2 = G 2 T − G T volatility
2 α≅3
α ≈ 3 for 3 < g < 50
α ≈ 1.6 for 3 < g < 50
  
−u 1 −(a2 x+b2 )c2 +d
17. Alejandro-​Quinones, Bassler, Field, McCauley, F(u) = C exp  (ε u + 1)α −1 P(x) ~ C e for x → ∞
 D0 ε  x b1 +a1α
Nicol et al. (2006)
x a1 = −​1, b1 = 1, a2 = 1/​(D0ε), b2 = 0, c2 = 1
Theoretical model that uses Focker Plank equation
for D = D0 (1 + ε u ) , u = , x price return h(x) = x, d = 0
(describes anomalous diffusion) to derive the t
probability function C
F(u) = , D = D0 (1 + εu )
2

Non constant diffusion coefficient (1 + εu 2 )1+β

One approaches a power law for high u and a
quadratic variation for D

1 1 c2
18. Clementi, Di Matteo and Gallegati (2006) P(i > x) ∝ P(x) ~ C e −(a2 x+b2 ) +d
for x → ∞
They estimate the power law tail exponent (α) xα xb1 + a1 α

α ~ 2.3 –​Australia a1 = 1, b1 = 1, a2 = b2 = c2 = 0
using Hill estimator, for the personal income in
α ~ 2.5 –​Italy h(x) = x, d = 0
Australia and Italy
α ≅ 2.5
  
Table 6.4  Lévy truncated distributions

Reference Author formula Generic formula

1 −(a2 x+b2 )c2 +d
19. Mantegna and Stanley (1994). 0 , x>l P(x) ~ C e for x → ∞
 x b1 +a1α
Truncated Lévy distribution, α=1.5 P(x) = cPl (x), − l < x < l ,
a1 = 1, b1 = 1, a2 = b2 = c2 = 0
Lévy distribution normalized by a 0 , x<l
 α ≅ 1.5 for –​l<x<l.
constant. Pl(x) –​Lévy function h(x) = x, d = 0
Not stable, has finite variance.
P(x)= 0 for all other values of x, a2~∞

1 c2
20. Mariani and Liu (2007). They use the characteristic function in simulations P(x) ~ C b1 + a1 α
e −(a2 x+b2 )
+d
for x → ∞
x
Study of market indices from Brazil, -​Probably same kind of equation as at number 21
a1 = 1, b1 = 1, a2 = 1/​k, b2 = −l/​k, c2 = 1
Mexico, Argentine
h(x) = x, d = 0
Exponentially truncated Lévy
α <2 for |x|>l
distribution α <2

1 c2
21. Gupta and Campanha (1999)  cLα (x , ∆t ), −lC ≤ x ≤ lC P(x) ~ C b1 + a1 α
e −(a2 x+b2 )
+d
for x → ∞
 x
They propose a gradually truncated β
P(x) =    x − lC   a1 = 1, b1 = 1, a2 = 1/​k, b2 = −​l/​k, c2 = β~0.6
Lévy distribution. CL α (x , ∆t )exp  −    , x 〉lC
   k   h(x) = x, d = 0
k, β are constants α =1.2 for |x|>l
α ≅ 1.2, β ≈ 0.6 pour S&P500
  
1 1
P(S) ~ 5/2 exp( − ε 2 S )
c2
22. Cont and Bouchaud (2000). P(x) ~ C b1 + a1 α
e −(a2 x+b2 )
+d
for x → ∞
S x
Percolation theory –​one gets
S –​cluster size; ε = 1−​c a1 = 1, b1 = 1, a2 = −​ε2, b2 = 0, c2 = 1
the distribution for the cluster
for c < 1, one has an exponentially truncated Lévy distribution h(x) = x, d = 0
(of financial operators) size –​identic
α ≅ 1.5
to the distribution of price changes
ε = 1−​c
The probability that operators interact
between them is c/​N; N is the total
number of operators

1 1 −(a2 x+b2 )c2 +d

23. Bouchaud (2002) P(S) ~ exp( − ε 2 S ) P(x) ~ C e for x → ∞
S 5/2 x b1 +a1α
See discussion of results of
a1 = 1, b1 = 1, a2 = −ε2, b2 = c2 = 0
number 22
h(x) = x, d = 0
α ≅ 1.5
  
Table 6.5  Other Distributions

Reference Author formula Generic formula

{ }
1
1 1
1 + β(t )(q − 1)[ x − x (t )]
2 1−q c2
24. Michael and Johnson (2003) P(x ,t ) = P(x) ~ C b1 +a1 α
e −(a2 x+b2 ) +d
for x → ∞
Z(t ) x
Tsalis power law distribution −2 a1 = 1, b1 = 1, a2 = b2 = c2 = 0
The value of q smaller than 5/​3 allows a finite variance P(x ,t ) = x q−1 = x −(α +1) for x large h(x) = x, d = 0
Can be classified as Lévy stable q Tsallis parameter, (1.64 for S&P500) –​good fit on the α ~ 2.15
empirical data
for q = 5/​3 one gets Gaussian;
for 1 < q ≤ 5/​3 –​Lévy non stable;
for 5/​3 < q < 3 –​Lévy stable.

C(α) 1 −(a2 x+b2 )c2 +d

25. Blattberg and Gonedes (1974) St α (x) = P(x) ~ C e for x → ∞
( A x 2 )(1+α )/2
+ 2
xb1 +a1α
Student-​t-​distribution
a1 = 1, b1 = 1, a2 = b2 = c2 = 0
St α (x) ~ x −(1+α ) for x → ∞
h(x) = x, d = 0
behaves like Lévy distribution for large x
α -​parameter

26. Richmond (2001) P(φ) =

1
1−
J { }
exp ( 2bφ − cφ2 ) / D P(x) ~ C
1 −(a2 x+b2 )c2 +d
xb1 +a1α
e for x → ∞
Power law distribution Zφ D
h(x) = x, d = -​b2/​(cD)
φ is related to price return
a2 = c1/​2, b2 = −​b/​c1/​2, c2 = 2, a1, b1, and
b, c, D, J parameters
α - parameters
  
ν2 − ν(x −R∆t ) 1 −(a2 x+b2 )c2 +d
27. McCauley and Gunaratne (2005) f (x , t ) = e , x > R ∆t P(x) ~ C e for x → ∞
γ +ν xb1 +a1α
Empirical Exponential Distribution function for
γ 2 γ (x −R∆t ) a1 ≠ 0, b1 = 0, a2 = 0, b2 = 0, c2 = 0
intraday trading of bonds and foreign exchange, f (x , t ) = e , x < R ∆t
γ +ν h(x) = log(x), d = 0
written in terms of returns
v , γ ∝ ∆t
x = ln(p(t)/​p(t0)).

ν2 − ν(x −R∆t ) 1 c2
28. McCauley (2003) f (x , t ) = e , x > R ∆t P(x) ~ C b1 +a1 α
e −(a2 x+b2 )
+d
for x → ∞
γ +ν x
Empirical Exponential Distribution function for
γ 2 γ (x −R∆t ) a1 ≠ 0, b1 = 0, a2 = 0, b2 = 0, c2 = 0
intraday trading of bonds and foreign exchange, f (x , t ) = e , x < R ∆t
γ +ν h(x) = log(x), d = 0
written in terms of returns
v , γ ∝ ∆t
x = ln(p(t) /​p(t0)).

ν2 − ν(x −R∆t ) 1 c2
29. Gunaratne and McCauley (2005) f (x , t ) = e , x > R ∆t P(x) ~ C b1 +a1 α
e −(a2 x+b2 )
+d
for x → ∞
γ +ν x
Empirical Exponential Distribution function for
γ 2 γ (x −R∆t ) a1 ≠ 0, b1 = 0, a2 = 0, b2 = 0, c2 = 0
intraday trading of bonds and foreign exchange, written f (x ,t ) = e , x < R ∆t
γ +ν h(x) = log(x), d = 0
in terms of returns
v , γ ∝ ∆t
x = ln(p(t) /​p(t0)).
 

C O N C LU S I O N
W H AT K I N D O F F U T U R E L I E S I N STO R E F O R
ECO N O P H Y S I CS ?

This book has investigated the development of econophysics and its potential implica-
tions for financial economics. Our project was to analyze the issues and contributions
of these two disciplines, focusing mainly on extreme values on stock prices/​returns
and a common vocabulary and perspective. As we have explained, in the context of
the difficult dialogue between the two communities, there is now a pressing need to
adopt a homogenous presentation. This opens a door to a real comparison between
the contributions of the two disciplines; it paves the way for a profitable dialogue be-
tween the two fields; and it also offers conceptual tools to surmount barriers that cur-
rently limit potential collaborations between the two communities. For the purpose of
providing a homogenous perspective, we have identified the disciplinary constraints
ruling each discipline and then suggested some paths for econophysicists and financial
economists to overcome the current limitations of their collaboration.
Throughout the book, we have studied econophysics’ contributions in the light
of the evolution of financial economics. In addition, all our analyses have taken the
standpoint of financial economics. In so doing, we have expressed the dissimilarities
in terms of vocabulary and methods of modeling between the two fields in a common
terminology. We have sought to provide financial economists with a clear introduc-
tion to econophysics, its current issues, its major challenges, and its possible future
developments in relation to financial economics. We have shown that econophysics
literature holds a twofold interest for financial economists: first, this discipline pro-
vides alternative tools for analytical characterization of financial uncertainty (through
a specific statistical treatment of long-​term samples). Second, by defending a strict
phenomenological method, econophysics renews the analysis of empirical data and
practical implications. Beyond the interest for financial economists, this book also
concerns econophysicists. Indeed, it gives an opportunity on the one hand to under-
stand the reasons why financial economists have been unable to use econophysics
models in their current form; and on the other hand, to identify the current challenges
econophysics has to solve in order to be accepted in financial economics and to pro-
vide more efficient models for analyzing financial markets.
In taking stock of the current situation of econophysics and in identifying poten-
tial paths for further investigation, this book is a first step toward more integrative re-
search between econophysics and financial economics. Collaborative research will

164
 

165 Conclusion

not be easy: after two decades, econophysics is a well-​established and recognized sci-

entific field analyzing numerous topics in finance. However, it still lacks recognition
among financial economists. In this challenging context, this book has clearly identi-
fied a future research agenda for improving collaboration between financial econo-
mists and econophysicists. As we have shown, three research axes are still waiting to
be investigated: (1) the development of a common framework/​vocabulary in order to
better compare and integrate the two approaches; (2) the creation of statistical tests to
measure results from econophysics models with those produced by financial models;
and finally (3) the elaboration of generative models to provide a theoretical explana-
tion for the emergence of power laws. This research agenda can be considered the key
challenge that econophysics has to face in the future; more generally, it can be seen as
what Lakatos (1978, 135) called a “positive heuristic,” “a set of suggestions or hints
on how to change, develop the ‘variants’ of the research programme.” In other words,
such a research agenda, which is a body of beliefs, suggests revising modeling methods
without abandoning the core assumptions that have developed in both disciplines.
In accordance with this idea, it is crucial for econophysicists to show to what extent
their works improve the statistical description used by financial economists, since the
Gaussian framework is a specific case of power laws.
We have illustrated this possible extension by proposing a first unified econophys-
ics framework to characterize the evolution of financial prices while keeping the key
characteristics of the Gaussian approach. This unified framework can be seen as the
first step in building conceptual bridges between the two fields. Given the diversified
literature dealing with extreme values in econophysics, our analysis has led us to put
forward a unified technical nomenclature for characterizing econophysics models.
In our opinion, this is an important contribution of this book. First, our generalized
econophysics framework describes the evolution of financial prices in a way that is
compatible with financial economics, easing the potential development of a common
vocabulary that makes sense to both communities. Second, it facilitates future
comparisons between models developed in financial economics and those used in
econophysics in a common framework. Third, it constitutes a first step through the de-
velopment of statistical tests applicable to both sets of models. It is worth mentioning
that the need for statistical tests is crucial for financial economists, but also for econo-
physicists: as Ausloos has claimed, “The econophysics approach should be taken with
caution, indeed. The robustness and soundness of models are fundamental questions.
Models should be predictive, and should be tested” (2013, 109).
We believe that the creation of statistical tests is the most crucial step for attracting
the attention of financial economists and practitioners. From a theoretical perspective,
statistical tests will allow systematic comparison of the models used by econophysicists
with those used by financial economists. From a practical perspective, statistical tests
will make it possible to measure the efficiency of each category of models and to im-
prove them. Statistical comparison between models from the two fields is relevant be-
cause the two areas of knowledge deal with the same financial phenomena, both using
a frequentist approach1. More specifically, key models in financial economics and in
econophysics usually consider empirical data as a random sample from a hypothetical
 

166  Econophysics and Financial Economics

infinite population in which frequencies are associated with probabilities. In this per-
spective, the question of comparison becomes inevitable, but econophysicists and fi-
nancial economists work in different statistical frameworks, and at present there are
no uniform statistical tests allowing one to choose between a model from a strictly
non-​Gaussian framework and one developed in the Gaussian framework.
Given the current situation, two future research areas can be investigated: on the
one hand, development of Bayesian tests in order to compare Gaussian-​based models
with non-​Gaussian ones and, on the other hand, development of (frequentist) statis-
tical tests for identifying power laws. A Bayesian approach uses the Bayes factor for
comparing two models using the ratio of the marginal likelihood of data used by the
models. The advantage of this testing approach lies in the fact that it is independent of
the statistical distribution of data. In other words, it offers a statistical framework for
comparing Gaussian and non-​Gaussian models. Conversely, the second possible area
of investigation involves developing a new frequentist testing approach implying the
possibility of comparing non-​Gaussian models with the Gaussian one.
One of our reasons for developing a generalized formula was precisely to create
room for the development of statistical tests. Indeed, in proposing a generalized for-
mula, we have solved the problem of the lack of unified non-​Gaussian statistical de-
scription of financial returns, thus justifying the need to develop non-​Gaussian statis-
tical tests. Moreover, our formula shows that econophysics models, though varied, are
technically compatible and can be presented through a unified and coherent frame-
work. This can then be used as a theoretical benchmark for a comparison with tra-
ditional Gaussian descriptions (GARCH-​type models) used in financial economics.
Statistically speaking, the fact that the Gaussian framework can be expressed as a
specific case of the generalized equation shows that a comparison between econo-
physics models and financial ones makes sense. In our opinion, a Bayesian comparison
could provide an interesting conceptual tool in order to go beyond the differences in
terms of modeling since it would compare models based on a conditional distribu-
tion (GARCH approach) and models based on an unconditional description of return
(econophysics perspective).
The last point on the research agenda is the development of generative econophys-
ics models to explain the emergence of power laws in financial data. The main idea of
these models is to go beyond the mere statistical description given by current power-​
laws model. As we saw, Gabaix et al. (2003) showed that institutional investors’ trades
have an impact on the emergence of a power law in the evolution of financial prices.
Although some technical models explaining the emergence of a power law in statistical
data exist, this area of investigation is still in its infancy concerning the economic in-
terpretation of these factors. Future collaboration between econophysics and financial
economics must pay more attention to this question.
To conclude, although the suggested agenda raises a number of questions and chal-
lenges, it creates many research opportunities by improving collaboration between fi-
nancial economists and econophysicists.
 

N OT E S

Introduction

1. The literature has expanded greatly since the early 2000s (Bouchaud, Mezard, and Potters
2002; Potters and Bouchaud 2003; McCauley 2009; Gabaix 2009; Lux 2009; McCauley,
Gunaratne, and Bassler 2007; Sornette 2014; Bouchaud 2002; McCauley 2006; Stanley and
Plerou 2001; Durlauf 2005, 2012; Keen 2003; Chen and Li 2012; Ausloos 2001; Chakrabarti
and Chakraborti 2010; Farmer and Lux 2008; Carbone, Kaniadakis, and Scarfone 2007;
Ausloos 2013; Jovanovic and Schinckus 2016; Schinckus 2010a, 2010b).
2. Those who are interested in such presentations can refer to the literature (Bouchaud and
Potters 2000; Cai, Lax, and Xu 2006; Chakrabarti, Chakraborti, and Chatterjee 2006;
Malevergne and Sornette 2005; Mantegna and Stanley 2000; Roehner 2002; Savoiu 2013;
Sornette 2003, 2006; Voit 2005; Bouchaud and Potters 2003; McCauley 2004; Richmond,
Mimkes, and Hutzler 2013; Takayasu 2002; Slanina 2014; Takayasu, Watanabe, and
Takayasu 2010; Paul and Baschnagel 2013).

Chapter 1

1. In this book, we use the term “stock market variations” to cover fluctuations in both the
prices and the returns of securities.
2. It is worth mentioning that this statement is true for any science. Physics, for instance, is
based on Euclidean geometry (or since the beginning of the twentieth century, non-╉Euclidian
geometrics, such as quantic physics). Euclidean geometry is founded on five axioms or
postulates, that is, five propositions that are accepted without proof. One of these postulates,
for example, states that a straight line segment can be drawn joining any two points. By
changing these postulates, it has been possible to create non-╉Euclidean geometries, which
enable the creation of other mathematics.
3. Jules Regnault (1834–╉1894) came from modest beginnings but died a millionaire, a fact
probably not unconnected with the model he proposed for determining stock market
variations. A biography of Regnault can be found in Jovanovic 2004, 2006a, 2016. His work
is analyzed by Jovanovic (2006a, 2000, 2002, 2016) and by Jovanovic and Le Gall (2001).
4. On Quételet, see Hankins 1908; Porter 1986; or Donnelly 2015.
5. Regnault never explicitly named the normal law:  the term only appeared in 1877 with
Wilhelm Lexis (Armatte 1991).
6. The central-╉limit theorem, along with the law of large numbers, is one of the most important
results of probability theory. This theorem is crucial because it states that the sum of many
independent random variables with finite variance will tend to the normal distribution.
This phenomenon was first observed by Gauss. Proof of the theorem was provided by de

167
 

168 Notes

Moivre and Laplace and published in 1738 in the second edition of The Doctrine of Chances
by Abraham de Moivre. It was subsequently generalized by Gnedenko and Kolmogorov in
1954. Let us remind readers that the central-​limit theorem states that the average of many
independent and identically distributed random variables with finite variance tends toward
a normal distribution irrespective of the distribution followed by the original random
variables.
  7. The term “random” first appeared in its statistical meaning in 1898 (Frankfurter and McGoun
1999, 166–​67), and the term “random walk” was first used in 1905 by Karl Pearson (1905a,
1905b).
  8. This term is explicitly used by Regnault (1863, 40).
  9. Mathematically, his model is of the following type: Pt +1 = Pt + ε t +1 , where ε = {ε t ,t ∈N } is
white noise and Pt+1 the price of the bond at time t. As a result, the expectation of profit
between two periods is nil, E ( Pt +1 − Pt ) = 0  .
10. These recordings were a way for the French government to control government bond prices.
France looks like an exception, because in other countries, official price data recording
started later (Preda 2007, 48). For instance, “The Wall Street Journal began publishing
closing quotations only in 1868. The NYSE got an official quotation list on February 28,
1872. ‘Official’ did not mean, however, that the NYSE guaranteed price data. In the 1860s
in London, only the published quotations of consols were closing prices” (Preda 2006,
760). By contrast, continuous data began to be organized with the introduction in 1923
of the Trans-​Lux Movie Ticker (Preda 2006). Of course, technology was not available for
recording these continuous flows of data.
11. It is worth mentioning that Regnault’s ideas were diffused and used both during his lifetime
and also after ( Jovanovic 2016).
12. Bachelier’s research program and his work are presented by Courtault et  al. (2002),
Jovanovic (2000, 2012), Davis and Etheridge (2006), and Ben-​El-​Mechaiekh and Dimand
(2006, 2008).
13. Here we preserve Bachelier’s notation, which we have reproduced in contemporary
language.
14. We should point out that equation 1.2 is not, strictly speaking, the law of probability of
a Brownian movement, but that of a Brownian movement multiplied by the standard
deviation, σ , which here is equal to 2 πk .
15. “Primes” (i.e., premiums) are derivative contracts similar to options but with some
differences, particularly in the evolution of their prices (Courtadon 1982; Cuoco and
Barone 1989; Jovanovic 2016). A “prime” is a conditional forward contract that allows the
holder to buy (or sell) an underlying asset at the strike price at the liquidation date of the
primes. To buy this contract, traders had to pay a fixed forfeit called “prime” (i.e. premium).
The expression “dont” served to indicate the amount to be forfeited. One traded Primes from
France to Central Europe (Weber 2009, 455).
16. Note that Bachelier did not evoke this second type of derivative contract, which did not exist
on the French stock market at the time. For this second type of contract, his calculations
were purely theoretical mathematical explorations with no empirical basis.
17. See Cramer 1983 for a presentation of the context and contributions of this period.
18. In 1906 Andrei Markov introduced Markov chains for the purpose of generalizing the law of
large numbers for a series of interdependent experiments.
19. Smoluchowski (1906) described Brownian motion as a limit of random walks.
 

169  Notes

20. Wiener (1923) carried out the first rigorous mathematical study of Brownian motion and
proved its existence.
21. One of the difficulties in reading Bachelier stemmed from the language he used, which was
not that of pure mathematics but that of mathematical physics: “In fact, the mathematicians
of the 30s who read Bachelier felt that his proofs are not rigorous and they are right, because
he uses the language of a physicist who shows the way and provides formulas. But again,
there is a difference between using that language and making mistakes. Bachelier’s arguments
and formulas are correct and often display extreme originality and mathematical richness”
(Taqqu 2001, 23).
22. One of the major contributions of Kolmogorov’s 1931 article was to make rigorous the
move from discrete to continuous schemes, a development that is a direct continuation
of Bachelier’s work. Moreover, Bachelier also influenced Khinchine (1933), with whom
Kolmogorov worked.
23. The two main schools studying probability theory prior to the 1940s were the French and the
Russian. From the 1940s onward, a number of important papers were developed in Japan,
influenced by Kiyosi Itô’s work in particular. On Japanese contributions, see Watanabe 2009.
24. On the Cowles Commission and its links with financial econometrics, see Morgan 1990;
Mirowski 2002; Christ 1994; and Dimand 2009. See also Hendry and Morgan 1995 on the
foundations of econometric analysis.
25. Working was professor of economics and statistics at Stanford University’s Food Research
Institute. He was never a member of the Cowles Commission, but took part in its summer
conferences.
26. A  Tippett table is “random number table,” used to produce series of random numbers.
There are several such tables, the first being created by the English statistician Leonard
Henry Caleb Tippett in 1927. It is made up of 10,400 four-​digit numbers extracted from
nineteenth-​century British census records.
27. Slutsky published his 1927 article in Russian; it was translated into English in 1937.
Barnett (2011) provides a detailed presentation of Slutsky’s contributions from the
perspective of the history of economics.
28. For example, Brown, Goetzmann, and Kumar (1998) redid Cowles’s calculations,
disregarding these ambiguous forecasts, and obtained the opposite result: William Peter
Hamilton’s recommendations made it possible to achieve a better performance than
the market. For further details on this point, see also Dimand 2009 and Dimand and
Veloce 2010.
29. Note that Cowles and Jones (1937) compared the observed distribution of stock prices with
the normal distribution to determine the possibility of making a profit.
30. In his article, Roberts used the same graphical methodology as Slutsky and Working
to demonstrate the random character of stock price variations. He did not analyze the
distribution of prices.
31. The work of Regnault and Bronzin seems unknown to American writers, and almost none
of the American writers working in financial econometrics prior to 1955 were aware of the
work of Bachelier. Two exceptions were Samuelson, who in the 1930s learned of Bachelier’s
work, and Arne Fisher, who in 1922 suggested applying Bachelier’s formulas to securities
( Jovanovic 2012).
32. On the emergence of financial economics see Jovanovic 2008, 2009a, 2009b; MacKenzie
2006; and Whitley 1986a.
 

170 Notes

33. For a retrospective on Markowitz, see Rubinstein 2002 and Markowitz 1999. On Roy’s
contribution, see Sullivan 2011.
34. As previously mentioned, diversification strategy was already used at the end of the nineteenth
century (Edlinger and Parent 2014; Rutterford and Sotiropoulos 2015). Moreover, the
relationship between risk and return had already been emphasized by Williams (1938),
although he did not provide an operational definition of this link.
35. This theorem can actually be thought of as an extension of the “separation theorem” originally
developed by Irving Fisher (1930). For an introduction to the work of Fisher, see Dimand
and Geanakoplos 2005. For a retrospective look at the Modigliani and Miller model, see
Miller 1988 and Rubinstein 2003.
36. Formally, these definitions take place in a probability space (Ω, F, P) with filtration
(Φ n )0≤n≤ N . A  filtration is an ascending—​ or descending—​ family of Φi tribes:
Φ1 ⊂ Φ 2 ⊂ Φ 3  ⊂ Φ n−1 ⊂ Φ n , a tribe being a family of parts, verifying stability hypotheses,
of all states of nature, Ω. The Φt tribe is a list of the events of which one can say at date t
whether they have occurred or not. It translates all information known on date t.
37. Note that if stock exchange prices follow a martingale, the expectation of profit, y, between
two consecutive periods is nil, E( yt +1 / Φ t ) = 0 , taking into account information Φt. In other
words, this is a fair game, as is the random walk also. Following Samuelson’s and Mandelbrot’s
articles, random movements of stock market prices were represented using martingales.
38. Markowitz (1952, 1955), for instance, was the first scholar to apply the expected-​utility
theory in financial (portfolio) management.
39. See Mackenzie 2006, 72–​73; Whitley 1986a, 1986b; Fourcade and Khurana 2009; and
Bernstein 1992.
40. The same issues were raised in training sessions given by Financial Analysts Seminar, one of
the leading professional organizations connected with financial markets (Kennedy 1966).
41. It is worth mentioning that the CAPM is often presented as a logical extension of the
portfolio theory, developed by Markowitz (1952), based on expected-​utility theory.
42. See, for instance, Cohen and Pogue 1967.
43. As explained by Jovanovic (2008), this polarization of results largely stems from the
theoretical frameworks propounded at MIT (Keynesian) and the Graduate School of
Business at the University of Chicago (monetarist).
44. Cowles and Jones (1937) had obtained a statistical dependence on monthly or weekly
averages of daily prices. Working (1960) explained that in this case it is possible to obtain
a degree of dependency because statistical analyses based on average prices can introduce
artificial correlations between consecutive variations.
45. In Fama’s thesis, this equilibrium value is the fundamental—​or intrinsic—​value of a security.
The signification of this value is unimportant: it may be the equilibrium value determined by
a general equilibrium model, or a convention shared by “sophisticated traders” (Fama 1965a,
36 n. 3). Fama later dropped the reference to a convention.
46. This is the most commonly accepted definition of efficiency, which Fama proposed in his
1970 paper: “A market in which prices always ‘fully reflect’ available information is called
‘efficient’ ” (1970, 383).
47. Fama acknowledged the difficulties involved in this joint test in a report on efficiency
published in 1991 (Fama 1991, 1575–​76).
48. Nearly all direct contributors to this hard core have received the Nobel Prize in
Economics: Markowitz, Sharpe, and Miller were joint winners in 1990; Merton and Scholes
 

171  Notes

received the award jointly in 1997. Although the contributions of Black (1938–​1995)
were explicitly recognized, he was not a named recipient, as the prize cannot be awarded
posthumously. Fama was awarded in 2013.
49. See Mehrling 2005 on Fischer Black, and MacKenzie 2006 on the influence of this model.
50. Merton (1973) would later show that use of the CAPM was unnecessary.
51. A security is said to be contingent if its realization depends on states of another factor (price
of another asset, climatic conditions, etc.).
52. Elimination of risk is a misnomer, because there is always a modeling risk related to, among
other things, the choice of the process or distribution to model stock market variations.
53. In an economy having T periods, the existence of a complete system of markets allows agents
from the initial moment to make intertemporal choices for all present and future prices at
determined prices that are known to all. The organization of such a system of markets appears
very complicated, since it would require a market for every good at every period. Arrow
(1953) showed that a complete system of markets can be replaced by a financial market
through which the assets exchanged allow agents to transfer their revenue independently
in each state (Arrow’s equivalency theorem). Subsequently Ross (1976b) showed that
options can complete incomplete markets and thus lend greater verisimilitude to Arrow-​
Debreu general equilibrium. Bayeux-Besnainou and Rochet (1996) extended Ross’s work
to a multiperiod model. However, it was the work of Harrison, Kreps, and Pliska that would
provide the rigorous mathematical framework for this intuition.
54. John McQuown at Wells Fargo and Rex Sinquefield at American National Bank in
Chicago established the first Standard and Poor’s Composite Index Funds in 1973
(http://​www.ifa.com).
55. These are the PriceWaterhouseCooper and BGI report (1998), 25 Years of Indexing, and the
PriceWaterhouseCooper and BDM Alliance report (1999), Investment Style and Its Growing
Role in Packaged Investment Products.
56. This influence is also found in the Basel II Accords (Pillar 3 explicitly refers to efficiency).
57. Specifically, these authors demonstrate two fundamental theorems. Th. 1: A market is
arbitrage-​free (efficient) if there exists at least one martingale measure.
This means that in a market free of arbitrage the stochastic price process for financial
assets must have at least one martingale measure—​under which the expected value of the
stochastic price does not change in time. Th. 2: A market is complete if there is a unique
martingale measure for the financial assets.
This theorem gives the conditions for a market to be complete: the stochastic price must
have the martingale property and there must be only one martingale measure. Thus in order
to price a financial asset (such as options) one must find the unique martingale probability
measure and then use the martingale property for the stochastic price.
58. In this case there is more than one probability measure that causes the stochastic price
process to be a martingale, and, consequently, this is an arbitrage-​free market.
For an incomplete market it can be shown that the price of a contingent security in a
market that does not allow arbitrage is found in an interval of values, xmin ≤ EQ [βT X ] ≤ xmax ,
with X the contingent security and β the discount process (βT = e − rT ) . The existence of many
possible prices is equivalent with the existence of a risk imbalance between the buyer and
the seller of the option. In this case, to have a unique price for the options, we have to add a
minimization procedure for the risk that leads to some mathematical conditions in the final
option formula.
 

172╇Notes

59. We might add that the continuity that characterizes Brownian motion is also a vital element
of the Black and Scholes model.
60. http://╉post.nyssa.org/╉nyssa-╉news/╉ 2010/╉10/╉in- ╉defense-╉of-╉the-╉quant-╉ii-╉the-╉ups-╉ and-╉
downs-╉of-╉the-╉normal-╉distribution.html.

Chapter 2

1  ( x − x0 )2 
╇ 1. The expression is P ( x ) = exp −  , defined for −∞ < x < +∞ . Therefore, the
√ 2 πσ2  2σ2 
deviations from the mean larger than a few standard deviations are rare for the Gaussian law,
as one can see on Â�figure 2.3.
╇ 2. Olivier pointed this out in his thesis (1926, 81), which was one of the first (if not the first)
exhaustive analysis on time-╉series analysis applied to economic phenomena: “Regardless of
the importance of the problem of the dispersion of prices and their distribution around their
mean—╉and for my part I believe the problem to be fundamental—╉hitherto it has been only
very insufficiently studied. This is explained by the very small number of prices tracked by
financial and economic newspapers that publish price index numbers.”
╇ 3. Cowles and Jones (1937) were the only authors to compare the distribution and cumulative
frequency of observed series of stock price variations with those of the random series. They
used them in order to determine the distribution of the expected net gain (Cowles and Jones
1937, 293).
╇4. Armatte (1995), Boumans (2007), Friedman (2009), Le Gall (1994, 2006), Morgan
(1990, 2012), and Morgan and Klein (2001) provided careful analysis of this development.
However, we should point out that index numbers did not exist before that time, because
goods were not sufficiently standardized and quantifiable to guarantee reliable data.
╇5. Barometers being defined as “time-╉series representations of cyclical activity” (Morgan
1990, 57).
╇ 6. See, for instance, Friedman 2009 on this barometer. We can also mention Wesley Clair
Mitchell at Columbia and Irving Fisher at Yale, who also promoted rigorous empirical
approaches for understanding economic fluctuations.
╇ 7. Fisher (1911) published the first systematic research on the mathematical formula that can
be applied to price indices calculus.
╇ 8. Mitchell (1915) can be considered as the first major study, because he provided a theoretical
and a practical analysis of index numbers.
╇ 9. As we explained in Â�chapter 1, Osborne made the same observation in 1959 and also suggested
using the logarithm of the price: finding that the distribution of price changes was not normal,
he suggested the use of logarithms of stock prices, which did exhibit the normal distribution.
By proceeding in this way, these authors sought to return to the normal distribution so that
they could apply the results from statistics and the probability theory that we know today.

10. The cotton price time series was the most complete, and daily quotations were available.
11. We can mention, for instance, Eugene Fama, Benjamin King, and Arnold Moore at Chicago;
and Walter Barney, John Bauer, Sidney Levine, William Steiger, and Richard Kruizenga
at MIT.
12. For instance, if X1 and X2 are independent random variables that are normally distributed
with the respective parameters (µ1 + σ1 ) and (µ1 + σ 2 ), then their sum, X1 + X2, is also a
normal random variable with parameters (µ1 + µ 2 , σ12 + σ22 ).
13. Since there is no closed-╉form formula for densities, a Lévy distribution is often described
by its characteristic. See Nolan 2005 and Samorodnitsky and Taqqu 1994 for different equivalent
definitions of stable Lévy distributions.
 

173  Notes

14. Only three parameters appear in equation (2.1) because the fourth one, δ, is included in
the set ℜ.
15. See Nolan 2009 for the demonstration.
16. Paretian law was the first statistical regularity associated with leptokurtic distribution used
in science. Pareto used it in his Cours d’économie politique (1896–​97) to characterize the
distribution of income and wealth.
17. “Scale invariance,” a term that Mandelbrot held dear, is in a sense the geometric translation
of the stability of the distribution of a stochastic process in probability theory.
18. The general central-​limit theorem claims that a sum of many independent and identically
distributed random variables with power-​law distributions decreasing through a Paretian
1
law, as α+1 where 0  < α < 2, will tend to be distributed according to a small attractor
x
distribution (i.e., an asymptotic limit for statistical distribution). When α = 2, we have the
Gaussian case of central-​limit theorem where the variance is finite and for which the attractor
distribution is normal, whereas a 0 < α < 2 process converges instead toward stable Lévy
distribution, as we will explain in this chapter. In other words, the sum of random variables
according to a Lévy law, distributed independently and identically, converges toward a stable
Lévy law having the same parameters.
19. This speculative approach is identical, for example, to that which led to the discovery of the
Higgs boson, whose existence was first theoretically assumed (and demonstrated) before
being empirically observed.
20. When α = 1, there is no impact of an increasing diversification on the scale factor, and when
α < 1, the scale factor increases in case of increasing diversification (Fama 1965b, 412).
21. Student distribution arises when estimating the mean of a normally distributed population
in situations where the sample size is small and population standard deviation is unknown.
This distribution is symmetric and bell-​shaped, like the normal distribution, but has heavier
tails. It approaches the normal distribution when the number of degrees of freedom grows.
22. Obviously, stability concerns Gaussian processes too.
23. Fama and Roll (1971) showed that a stable nonnormal process cannot be obtained from a
mixture of (stable) normal processes.
24. It is worth mentioning that these empirical series were based on relatively small samples
in comparison with the data available nowadays in finance. This point is quite important
because the length of price series can reduce the scope of the results. For instance, the
occurrence of a power law only makes sense on very large samples and working with a small
sample directly influences the statistical description that will emerge from the data.
25. Note that the technique used by Fama and Roll (1968) was generalized by McCulloch (1986),
who provided an estimation for all four parameters with no restriction on the symmetry of
distribution. Basically, McCulloch (1986) used the information given by the empirical
distributions’ quantiles, from which he computed the four parameters of the process.
26. The method based on iterated regressions developed by Press (1972) was generalized by Arad
(1980) and Koutrouvelis (1980). Also worthy of mention is the Paulson-​Holcomb-​Leitch
(1975) method, which minimizes the difference between the theoretical characteristic function
and a polynomial extrapolation of the empirical characteristic function. For further information
about these methods, see Embrechts, Klüppelberg, and Mikosch 1997b and Nolan 2009.
27. This relationship between risk and return was given shape in Markowitz’s seminal portfolio
theory (1952, 1959).
28. For instance, Blume (1968) and Officer (1972).
 

174╇Notes

29. Since the only statistical condition for describing this jump is the provision of a finite mean
in order to ensure the finiteness of variability (in line with the mean variance approach).
30. “The random-╉walk model, however, even in modified logarithmic form, has been found
inadequate on the basis that the tails of the distribution of price changes (or their logarithms)
appear to be too long, based on sample evidence, to be accounted for by the appropriate
normal distribution. One effort at compensation engendered the variant of the random-╉walk
model due to Cootner (1962), which added reflecting barriers to the Markov (random-╉walk)
process. Mandelbrot (1962) proposed a model for the behaviour of security-╉price changes that
generated keen interest in the finance area, in the mathematics of stable laws” (Press 1967, 318).
31. See Cont and Tankov 2004 for an overview of this literature.
32. “One-╉block reasoning” is not the only difference between the jump-╉diffusion processes
presented above and pure-╉jump processes. Indeed, while the first can use stable Lévy processes
to characterize the jump part of the model, the latter deal only with nonstable Lévy processes
(i.e., with a characteristic exponent alpha > 2), allowing them to value all statistical moments.
33. See Cont and Tankov 2004 for further details on this literature.
34. This time-╉dependence dynamic is defined by the modeler.
35. See Francq and Zakoian 2010; Bauwens et al. 2006; Tim 2010; and Pagan 1996 for further
details on these categories of models.

Chapter 3

╇ 1. We have borrowed the term from Le Gall (2002, 5), who provides an excellent introduction
to the analysis of methodology transfer between physical sciences and economics.
╇ 2. Note that the explanations about statistical physics are borrowed from an article by Richard
Fitzpatrick (2012) available on the Web (farside.ph.utexas.edu/╉teaching/╉sm1/╉statmech.pdf).
╇ 3. As Fitzpatrick (2012) noticed, to solve a system with 6 × 1023 particles exactly, we would
have to write down 1,024 coupled equations of motion, with the same number of initial
conditions, and then try to resolve the system.
╇ 4. In particular the central-╉limit theorem.
╇ 5. For instance, as Fitzpatrick (2012) commented, the familiar equation of state of an ideal
gas, PV = nRT, is actually a statistical result. In other words, it relates the average pressure
(P) and the average volume (V) to the average temperature (T) through the number (n)
of particles in the gas. “Actually, it is virtually impossible to measure the pressure, volume,
or temperature of a gas to such accuracy, so most people just forget about the fact that the
above expression is a statistical result, and treat it as a law of physics interrelating the actual
pressure, volume, and temperature of an ideal gas” (Fitzpatrick 2012, 6).
╇ 6. As Lesne and Laguës (2011, 3)  point out, the study of critical phenomena was initiated
by Cagnard de Latour in 1822 and then boosted with the work of Thomas Andrews from
1867 onward. In 1869 Andrews observed a spectacular opalescence near the critical point
of carbon dioxide. However, “The 1960s saw the emergence of a new general approach to
critical phenomena, with the postulation of the so-╉called scaling laws, algebraic relations
holding between the critical exponents for a given system” (Hughes 1999, 111).
╇ 7. It is important to mention that the concept of equilibrium can be associated with the notion
of phase in an “all other things being equal” analogy. Indeed, in the case of continuous
variations of pressure/╉temperature, the phase is progressively moving toward a critical state,
implying that it cannot be associated with a static equilibrium.
 

175  Notes

  8. Lesne and Laguës (2011) and Lesne (1998) provide an extremely clear and exhaustive
presentation of renormalization methods. These papers make a very good introduction to
intuitions and formalisms. Stanley (1999) provides a short presentation. See also Wilson
1993; Jona-​Lasinio 2001; Calvo et al. 2010; and Stanley 1971 for further details.
  9. Sornette (2006, chap. 2) provides a detailed presentation of this method.
10. For more details, see Samorodnitsky and Taqqu 1994; and Lesne 1998.
11. For instance, it exists in the work of Euclid and Galileo.
12. To understand the importance of this approach, one has to keep in mind that the macroscopic
level is directly observable—​for instance, a table—​but the microscopic level—​the molecules
that constitute the table—​is not directly observable (one needs a tool, as a microscope).
13. It is worth mentioning that large variation correlation length appears to be ruled by a power
law, as the next section will detail.
14. Proceeding by analogies is very common in sciences, and particularly in economics. For
instance, Henri Lefevre used analogies with the human body to analyze financial markets
( Jovanovic 2006). Cohen (1993) provides a good analysis of the use of analogy, homology,
and metaphor in interactions between the natural sciences and the social sciences.
15. The physicist Serge Galam initiated this discipline in papers published between 1980
and 1983, and proposed the term sociophysics in 1982 (Chakrabarti, Chakraborti, and
Chatterjee 2006; Stauffer 2005, 2007; Galam 2008). Săvoiu and Iorga-​Simăn (2013) give
some historical perspective on sociophysics.
16. For instance, corporate revenue (Okuyama, Takayasu, and Takayasu 1999), the emergence
of money (Shinohara and Gunji 2001), and global demand (Donangelo and Sneppen 2000).
17. In 1977, the Toronto Stock Exchange became the first stock exchange to be fully automated.
Then, the Paris Stock Exchange (now Euronext) imported Toronto’s system and became fully
automated at the end of the 1980s. These changes occurred for NASDAQ between 1994 and
2004, and later for the NYSE in 2006 with the introduction of the NYSE hybrid market. The
Tokyo Stock Exchange switched to electronic trading for all transactions in 1999.
18. The term “high-​frequency data” refers to the use of “intraday” data, meaning that price
changes can be recorded at every transaction on the market.
19. Today there is a large literature on this subject, in particular with the theory of financial
market microstructure, which focuses on how specific trading mechanisms and how
strategic comportments affect prices. Maureen O’Hara (1995), one of the leading lights of
this theoretical trend, provides a good introduction to this field.
20. As explained in ­chapter 2, a process is said to be a Lévy process when it has (1) independent
increments; (2)  stationary increments, and (3)  a continuous probability function.
Lévy processes include a large category of statistical distributions (Poisson, Gaussian,
Gamma, etc.).
21. There are some exceptions, particularly in the most recent works (Nakao 2000). We will
come back to these recent works in ­chapter 5.
22. Let us mention the so-​ called Kosterlitz-​Thoules transition, which is an exception
characterizing a transformation from a disordered vortex fluid state with equal number
of vortices to an ordered molecular like state composed by pairs of vortices with different
polarities. We thank Marcel Ausloos for this precision. For further details on this point, see
Hadzibabic et al. 2006.
23. Shalizi’s notebook, http://​bactra.org/​notebooks/​power-​laws.html.
24. Precisely, ξ(T ) ∝ T − Tc
−υ
 .
 

176 Notes

25. Precisely, at the critical point T = TC the correlation length diverges. There is no typical size.
−r −r
r 1 1
Hence, at the critical point: ξ ( T ) → ∞ ; → 0 ; e ξ(T) → 1  ; α e ξ(T) → α  .
ξ( T) r r
26. The correlation length is finite. The exponential term “wins” over the power-​law term,
since it decreases more rapidly as the distance r increases. Hence the correlation function is
described by an exponential function.
27. Shalizi’s notebooks are available at http://​bactra.org/​notebooks/​. One can also read Jona-​
Lasinio 2001 for further details.
28. The first studies on statistical scale invariance came from Kolmogorov (1942, 1941), who studied
data related to phenomena associated with turbulence. His research on turbulence also led him to
introduce power laws (and their scaling property) into physics at the same period. These concepts
progressively became widespread in the discipline (Hughes 1999). In the 1960s, scholars such
as Domb and Hunter (1965), Widom (1965a, 1965b), Fisher (1967), and Kadanoff (1966)
established some of the most important theoretical results on scaling properties, which
contributed to the crucial developments that have occurred in physics since the 1970s.
29. We use here the easiest case of scaling invariance (i.e., scaling factor  =  1.1). For more
information about more complex scaling properties, see Hartemink et al. 2001.
30. The negative sign is not really important, since it depends on the choice of axes for the
histogram. We base our presentation on Li et al. 2005, which provides a clear mathematical
presentation of power laws. Newman (2005) also provides a clear and exhaustive
mathematical analysis (included the determination of the moments).
31. Stable Lévy processes can be characterized through a specific power law exhibiting an
independence of increments (a defining element of Lévy processes). While all power laws
with 0 < μ ≤ 2 are said to be stable, only those with 0 < μ < 2 are associated with a distribution
whose variance is infinite, since an exponent equal to 2 is a Gaussian distribution (and
therefore a finite variance). In other words, a Gaussian process can be looked on as a
specific case of stable Lévy processes or, to put it in another way, a Gaussian distribution
can mathematically be expressed as a stable power law with an exponent equal to 2 (Nolan
2009). It is worth mentioning that independence of increments is not a required condition
for power laws, which can also describe dependent increments (the reason that power laws
are also used for characterizing long memory). In other words, statistical stability implies
scaling properties, while the reverse is not true (Schoutens 2003).
32. By attractor, we mean an asymptotically approached set of points in phase space, which is
invariant under dynamics.
33. Note that this phenomenological approach is much less common in economics. On the
phenomenological approach in physics, see Cartwright 1983.
34. Mitzenmacher (2004) and Simkin and Roychowdhury (2011) are the best-​documented
articles.
35. In addition to the variation of the characteristic exponent (α), econophysicists also try to
fit the data by adjusting the constant C in the equation ln P[r > x] = −α ln x. The variation
of this parameter C explains why the curves do not begin at the same point in the fi­ gure 3.7.
It is worth mentioning that, because power laws are scale invariant, the positive constant C,
associated with the scale of analysis, does not play a key role in the statistical characterization
of data.
36. It is worth mentioning that Vitanov et al. (2014), like many others, showed that this is not
always true.
 

177  Notes

37. According to Rybski (2013), Berry and Okulicz-​Kozaryn (2012) distinguish three periods
in research on city-​size distributions. The first period was initiated by Auerbach’s empirical
discovery. Then Zipf ’s work (1949) was the starting point of the second period that gave
rise to empirical studies for estimating the exponent for city distributions and cross-​national
analyses. During this second period, an article by Simon (1955) proposed a statistical
process for explaining these observations. He suggested that the probability of a city growing
by a unit is proportional to its size, which is known today as preferential attachment. Finally,
the lack of an intuitive theoretical model led to the third phase. Despite a variety of early
modeling approaches from various disciplines, Berry attributes the first economic theory
addressing city-​size distributions to Gabaix (1999).
38. In 2002 the journal Glottometrics dedicated a special issue edited by Gunther Altam to
Zipf ’s work.
39. See Dubkov et al. (2008) for an astonishing list of stable Lévy-​type phenomena in physical,
biological, chemical, and social systems.
40. Lévy (2003) and Klass et  al. (2006) confirmed Pareto’s results by showing that wealth
and income distribution can both statistically be characterized by a power law. Amaral
et al. (1997) explained that annual growth rates of US manufacturing companies can also
be described through a power law, while Axtell (2001), Luttmer (2007), and Gabaix and
Landier (2008) claimed that this statistical framework can also be used to characterize the
evolution of companies’ size as a variable of their assets, market capitalization, or number
of employees. These “size models” have since been applied for describing the evolution of
cities’ size (Cordoba 2008; Eeckhout 2004; Gabaix 1999; Krugman 1996). In the same vein,
Lux (1996) and Gabaix et al. (2007) observed that the large fluctuations on the financial
markets can be captured through a power law. We can also read Gabaix’s (2009) survey on
power laws in economics and finance.
41. The two next chapters will discuss this point and some limits of this approach.
42. See Ausloos 2014a for a detailed analysis of these analytical tools and Ausloos 2014b for an
application of these techniques.
43. See Ding 1983 for further details on the use of asymptotic (theoretical) statistics for
describing physical (empirical) systems.
44. This idea of truncation dated back to the 1700s, with the famous St. Petersburg paradox
(Csorgor and Simons 1983). Indeed, many works have considered this idea as empirical
evidence because all physical systems are finite.
45. The finite configuration of all samples implies that there is necessarily an upper bound
rank making the statistical description of these samples possible. However, as previously
mentioned, statistical physicists wanted to combine the asymptotic properties of power laws
with the finite dimension of empirical samples. That is the reason why some physicists went
beyond the use of ranks in the treatment of power laws (Mantegna 1991; Mantegna and
Stanley 1994). On this point, see also Ausloos 2014a.
46. It is worth mentioning that Mandelbrot (1963) already mentioned the possibility of using
truncation techniques in the 1960s at low rank scale.
47. See Figueiredo et al. 2007 for a statistical explanation of this slow convergence.
48. However, knowing that critical phenomena have a continuous and abrupt change, we can
suppose that their choice is not independent of this consideration.
49. It is worth mentioning that the first exponentially truncation technique was introduced in
physics by Koponen (1995). However, this author did not justify his paper using a physically
 

178╇Notes

plausible argument, as was the case in the literature after the critique made by Gupta and
Campanha (1999).
50. The truncation technique based on a gradual cutoff can be considered a specific case of the
exponential technique (Matsushita, Rathie, and Da Silva 2003).
51. These constants are usually derived from physical theories (Gupta and Campanha 2002),
thus offering a variety of potential definitions for the truncation function.
52. It is worth mentioning that a variety of exponential truncation functions have been developed
in the specialized literature (Schinckus 2013). Two factors can explain the diversity of these
functions: first, the collection of physical theories that can be used to define the form of the
function; and second, the will to develop more data-╉driven truncations (Matsushita, Rathie,
and Da Silva 2003).
53. These techniques have been presented as a potential solution for power laws based on
high-╉frequency data. For low-╉frequency items (i.e., not financial data), some physicists
implemented alternative solutions by using other statistical law (the Lavalette law, for
instance). See Ausloos 2015; Cerqueti and Ausloos 2015; and Tripp and Feitelson 2001.

Chapter 4

1. Actually, Stanley was the first scholar to propose the term “econophysics” during a conference
dedicated to “physics of economics” organized in Kolkata (India) in 1995 (Chakrabarti and
Chakraborti 2010).
2. According to Standler (2009), the end of this kind of bubble can partly be explained by a
generational shift in the administration: senior officers close to retirement who favored funding
of specific scientific research are no longer able to insist on this generous financial support.
3. See http://╉phys.uh.edu/╉research/╉econophysics/╉index.php.
4. See http://╉www.tcd.ie/╉Physics/╉people/╉Peter.Richmond/╉Econophysics/╉Position.html and
http://╉www.itp.phys.ethz.ch/╉research/╉comp/╉econophys for examples. For further infor�
mation on these programs, see Kutner and Grech 2008 and the website of these universities.
5. http://╉www3.unifr.ch\econophysics.
6. It earned its official recognition in the Physics and Astrophysics Classification Scheme
(PACS):  since 2003, econophysics has been an official subcategory of physics under the
code 89.65 Gh.
7. The sample is composed of Eugene Stanley, Rosario Mantegna, Joseph McCauley, Jean-╉
Philippe Bouchaud, Mauro Gallegati, Benoît Mandelbrot, Didier Sornette, Thomas Lux,
Bikas Chakrabarti, and Doyne Farmer. Moreover, given the usual practice of citations, other
important authors have been retrieved through the analysis of the cited references in these
papers as well as in the papers citing those source papers. A  group of 242 source papers
covering the domain of econophysics and the papers that cite them over the period 1980–╉
2008 were identified in order to analyze the evolution of the field. Starting with these core
papers, which construct the population of researchers, 1,817 other papers that cited the
source articles have been identified.
8. These papers were mainly written by Thomas Lux and Mauro Gallegati and dealt with
macroeconomics (Gallegati 1990; Lux 1992a, 1992b; Gallegati 1994)  or the history of
economics (Gallegati and Dardi 1992).
9. His research focuses partly on complexity in economics, a topic that may cause him to be
more open to the approach proposed by econophysicists.
 

179  Notes

10. The data on the cited journals come from the “Journal of Citation Report 2008” published
by Thomson Reuters and part of the Web of Knowledge.
11. The first is a physicist and the second an economist, and both were in our source authors.
12. Following Backhouse (2004, 265), we distinguish “orthodox dissenters” from “heterodox
dissenters”; the latter reject the mainstream theory and aim at profoundly changing
conventional ideas, while the former are critical but work within mainstream economics.
13. Chapter  2 explained that financial economists describe the occurrence of extreme values
through the evolution of the Gaussian trend (unconditional distribution) corrected by a
conditional capturing the large variations.
14. Eugene Stanley, who is often presented as the father of econophysics, told us privately that
after more than six years (!) he decided to cancel his submission to the American Economic
Review—​although it is a top 10 journal in economics.
15. While economists use the JEL (Journal of Economic Literature) classification, physicists
organize their knowledge through their PACS (Physics and Astrophysics Classification
Scheme) under which econophysics has its own code (89.65 Gh).
16. Economists usually employ stylistic conventions defined by the University of Chicago
Press or the Harvard citation style, where references are listed in alphabetical order, while
physicists adopt the conventions used by the American Institute of Physics, where references
are listed in the order in which they appear in the text.
17. They had to choose from five reasons for having been rejected and were invited to comment
on their choices: (1) the topic of the paper; (2) the assumptions used in the paper; (3) the
method used in the paper; (4) the results of the paper; or (5) another reason.
18. This situation is not specific to economics. It also existed in the other fields into which
statistical physicists have extended their models and methods (Mitzenmacher 2005).
19. This is one of the main reasons for having the efficient-​market theory depict its weak
connection with the random character of stock price variations, as we saw in c­ hapter 1.
20. Mirowski (1989) gives a good overview on this controversy.
21. The VAR approach has been developed by Christopher Sims (1980a, 1980b). VAR models are
a set of related linear difference equations in which each variable is in turn explained by its own
lagged values, plus current and past values of the remaining n − 1 variables (Christiano 2012;
Stock and Watson 2001). Numerous economists have critiqued these models because they
do not shed any light on the underlying structure of the phenomena or the economy studied
(Chari, Kehoe, and McGrattan 2008, 2009; Christiano 2012). Similarly, the RBC approach is
based on calibration. It has been developed by Finn E. Kydland and Edward C. Prescott (1982).
These macroeconomic models study business cycle fluctuations as the result of exogenous
shocks. Their model is based on a calibration approach, and it is validated if the simulations
provided by the model fit with empirical observations. Numerous economists have critiqued
this method (Eichenbaum 1996; Gregory and Smith 1995; Hansen and Heckman 1996;
Hendry 1995; Hoover 1995; Quah 1995; Sims 1996; Wickens 1995). De Vroey and Malgrange
(2007) offer a presentation of this model and its influence in macroeconomics literature.
22. Shalizi’s notebook, http://​bactra.org/​notebooks/​power-​laws.html.
23. Chartism (also called “technical analysis”) is a financial practice based on the observation of
the historical evolution of assets’ prices. More precisely, actors try to identify visual patterns
that help them give a meaning to the financial reality.
 

180╇Notes

24. Stanley and Plerou (2001) replied to LeBaron’s critic. Although their reply provided a
technical answer—╉particularly the limited number of data used by LeBaron—╉it underlines
the difficulties in the expectations of the two communities.
25. Chapter 2 also emphasized that econophysicists can find this linearity by treating statistically
the potential inflection points that could appear in the visual analysis of data.
26. We can mention Bouchaud and Sornette (1994) and Bouchaud and Potters (2003, 2000),
who proposed a first approximation to the European call option formula that is equivalent to
the option-╉pricing formula obtained in financial economics under the risk-╉neutral approach.
However, their arguments are different from those used in financial economics.
27. It is worth mentioning that most practitioners have developed their own formulas that are far
from Black-╉Scholes-╉Merton model (Haug and Taleb 2011).
28. When econophysicists deal with equilibrium, they rather use a “statistical equilibrium”
coming from a statistical mechanism (i.e., a reconciliation between a mechanism and
thermodynamics). See Bouchaud 2002.
See Schinckus 2010a, 2010c for further information about the importance of equilibrium
in econophysics.
29. It is worth mentioning that the hypothetico-╉deductive method is considered the major
scientific method by many authors in philosophy of science (for instance, Popper 1959).
When (financial) economics emerged as discipline, economists integrated hypothetico-╉
deductive reasoning as a scientific foundation:  the assumption of the perfectly rational
agent allows economists to deduce implications in terms of individual behaviors, while
the hypothesis of representative agent offers them the ability to generalize microbehaviors
at a macroeconomic level. Even recent developments such as behavioral finance kept this
deductive method by giving a generalization of the perfect rational agent (Schinckus 2009).
30. ARCH-╉type models include ARCH, GARCH, NGARCH, EGARCH, and similar models.
31. See, for instance, Lorie 1966, 107.
32. In epistemological terms, this opposition between early financial economists and chartists
shaded the classical opposition between deduction (used by financial economists) and
induction (used by chartists). See Jovanovic 2008 for further details on this opposition.
33. See, for instance, Cootner 1964, 1; Fama 1965b, 59; Fisher and Lorie 1964, 1–╉2; and Archer
1968, 231–╉32.
34. With Rosser (2006, 2008), Keen (2003) is one of the rare breed of economists who have
engaged in a dialogue with econophysicists.
35. See, for instance, Weston 1967, 539.
36. In this respect, Rosenfeld (1957, 52)  proved to be visionary when he suggested using
computers for testing theories on a large sample.

Chapter 5

1. One should point out that, more recently, econophysics models have less frequently used
power laws; this represents a new diversification of econophysics models that will be
discussed in this chapter.
2. As detailed in chapter 3,
� theoretical investigation in econophysics is considered only after
having observed patterns in the empirical results.
3. Mandelbrot (1962, 6) had already underlined the need for a phenomenological approach in
his early works applied to finance. From this perspective Mandelbrot’s agenda (building new
mathematical models, new statistical tools, etc.) was very ambitious for the 1960s.
 

181  Notes

4. By “success,” Casey (2013) meant that these models were able to detect an overestimation of
the market (in the credit default swap bubble, for example).
5. Examples are Jean-​Philippe Bouchaud and Mark Potters, who created Capital Fund
Management, and Tobias Preis, who created Artemis Capital Asset Management.
6. https://​www.cfm.fr/​en/​.
7. http://​modeco-​software.webs.com/​econophysics.htm.
8. https://​www.rmetrics.org/​.
9. https://​www.rmetrics.org/​sites/​default/​files/​2013-​VorlesungSyllabus.pdf.
10. http://​tuvalu.santafe.edu/​~aaronc/​powerlaws/​.
11. Cornelis (2005) and Guégan and Zhao (2014) point out that extreme events lead to the
failure of VaR.
12. Pagan (1996) offers a very clear and useful perspective on econometric models applied to
financial markets. See also Bauwens et al. 2006 and Francq and Zakoian 2010 for further
details on these categories of models.
13. Considering the time series as a whole is the most common approach in natural sciences and
is associated with scaling laws.
14. We thank Nicolas Gaussel for helpful discussion on this topic.
15. LTCM was a hedge-​fund management firm that utilized absolute-​return trading strategies
combined with high financial leverage. The firm was founded in 1994 and collapsed in 1998.
Members of its board of directors included Myron S. Scholes and Robert C. Merton. Initially
successful with extremely high annualized return in the first years (21 percent, 41 percent,
and 43 percent after fees), in 1998 it lost $4.6 billion in less than four months following the
Asian and the Russian financial crisis. See Dunbar 2000 and Lowenstein 2000.
16. Harrison (1998) showed that the characteristics of eighteenth-​century financial-​asset
returns are the same as those of the twentieth century: “The distribution of price changes
now and then both exhibit the same patterns or regularities. In particular, the distribution of
price changes is leptokurtic, and fluctuations in variance are persistent” (1998, 55). In other
words, these regularities are stable.
17. Fantazzini and Geraskin (2011) provide a clear presentation of LPPL models.
18. A > 0 is the value of ln p(tc) at the critical time, B < 0 is the increase in ln p(t) over the time
unit before the crash if C were to be close to zero, C ≠ 0 is the proportional magnitude of the
oscillations around the exponential growth, 0 < β < 1 should be positive to ensure a finite
price at the critical time, while 0 < δ < 2π is a phase parameter.
19. The opinions of other participants influence each participant. It is the well-​known beauty
contest described by Keynes in c­ hapter 12 of the General Theory, in which judges picked
whom they thought other judges would pick, rather than whom they considered to be the
most beautiful.
20. Durlauf set out his position more clearly in a later paper (2012, 14).
21. Let us mention, for instance, that Durlauf (Arthur, Durlauf, and Lane 1997; Blume and
Durlauf 2006) was involved in the meetings organized by the Santa Fe Institute dedicated to
the application of physics to economics, while Lux regularly published articles dealing with
econophysics (Lux 2009).
22. In the same vein as Mitzenmacher, a theoretical interpretation is considered here in the sense
of explaining the significance of a mathematical model.
23. However, as explained in c­hapter  2, the characterization of these statistical patterns is
developed in different conceptual framework (i.e., the Gaussian framework for financial
economists and the power-​law perspective for econophysicists).
 

182 Notes

24. See Jovanovic and Schinckus 2013 for a detailed discussion of this point in connection with
econophysics and financial economics.
25. Galison (1997) explained how engineers collaborated with physicists to develop particle
detectors and radar.
26. A Creole (e.g., Chavacano in the Philippines, Krio in Sierra Leone, and Tok in Papua New
Guinea) is often presented as an example of a pidgin because it results from a mix of regional
languages; see Todd 1990.
27. Note also special issues of economic journals, such as the Journal of Economic Dynamics and
Control dedicated to the “Application of Physics to Economics and Finance” published in
2008, and the issue of the International Review of Financial Analysis titled “Contributions of
Econophysics to Finance,” published in 2016.
28. Brakman et al. (1999) extended Krugman’s (1991) model by introducing negative externalities.
29. In contrast, Farmer et al. (2004) have shown that large price changes in response to large
orders are very rare. See also Chiarella, Iori, and Perello 2009 for a more recent model
showing that large price changes are likely to be generated by the presence of large gaps in
the book of orders.
30. It is worth mentioning that this hypothesis is similar to those of Fama (1965) when he
defined and demonstrated the efficiency of financial markets for the first time.
31. From an economic perspective, the difference observed between the distributions
characterizing the evolution of financial variables (returns, foreign exchange) and those
describing economic fundamentals could result from the higher liquidity of the former. See
Aoki and Yoshikawa 2007 for more information on this subject.
32. A  number of empirical studies very soon contradicted the conclusions of the theoretical
framework built during the 1960s and the 1970s (see c­ hapter 1). These empirical studies gave
birth to what is known as the “anomalies literature,” which has become extensive and well
organized since the 1980s. Schwert (2003) provides a fairly exhaustive review of anomalies.
33. The term “market microstructure” was coined by Mark Garman (1976), who studied
order flux dynamics (the dealer must set a price so as to not run out of stock or cash). For a
presentation of the discipline, see O’Hara 1995; Madhavan 2000; and Biais et al. 2005.
34. The first generation of market microstructure literature has shown that trades have both a
transitory and a permanent impact on prices (Biais, Glosten, and Spatt 2005). For instance,
Copeland and Galai (1983) showed that a dealer who cannot distinguish between informed
and uninformed investors will always set a positive spread to compensate for the expected
loss that he will incur if there is a positive probability of some investors being informed.
Kyle (1985) suggests that informed dealers can develop strategic behavior to profit from
their information by concealing their orders among those of noninformed dealers. While
informed dealers thus maximize their own profits on the basis of the information they hold,
their behavior restricts dissemination of the information. O’Hara (2003) presents another
example of results that contradict the dominant paradigm. In this article, she shows that
if information is asymmetrically distributed, and if those who do not have information
know that others know more, contrary to the suggestions of the CAPM, we will not get an
equilibrium where everyone holds the market portfolio.
35. See Schinckus 2009a, 2009b for a presentation of this school and its positioning vis-​à-​vis the
dominant paradigm.
36. In 2002 Daniel Kahneman received the Nobel Prize in Economics for his work on the
integration of psychology with economics.
 

183╇ Notes

37. Note that Shefrin (2002) made a first attempt to unify the theory.
38. Agent-╉based modeling is a computational method applied in so many fields (Epstein
2006)  that it is not possible to number them in this chapter. The agent-╉based approach
appeared in the 1990s as a new tool for empirical research in many fields, including
economics (Axtell 1999), voting behavior (Asselain 1985), military tactics (Ilachinski
2000), organizational behavior (Prietula, Carley, and Gasse 1998), epidemics (Epstein and
Axtell 1996), and traffic-╉congestion patterns (Rasmussen and Nagel, 1994). For a detailed
literature review on the topic, see Epstein 2006 or more recently Cristelli 2014.
39. See LeBaron 2006 for details on agent-╉based modeling used in economics.
40. Note that this agent-╉based econophysics is not limited to financial issues, since Pickhardt and
Seibold (2011), for example, explained that income-╉tax evasion dynamics can be modeled
through an “agent-╉based econophysics model” based on the Ising model of ferromagnetism,
while Donangelo and Sneppen (2000) and Shinohara and Gunji (2001) approached the
emergence of money through studying the dynamics of exchange in a system composed of
many interacting and learning agents. Focardi et al. (2002) and Chiarella and Iori (2002)
also provided an Ising-╉type model with interactions between nearest neighbors.
41. Bak became an external member of the Santa Fe Institute, where he found the perfect
environment to promote his theory of criticality, which gradually spread to several
disciplinary contexts in the 1990s (Frigg 2003). The Santa Fe Institute was founded in 1984
to conduct theoretical research outside the traditional disciplinary boundaries by basing
it on interdisciplinarity. Its original mission was to disseminate complexity theory (also
called complex systems). This institute plays a key role in econophysics due to a fruitful
collaboration between economists and physicists, among them some Nobel laureates such
as Phil Anderson and Kenneth Arrow. For further details, see Schinckus 2016.
42. The visual tests make it difficult to distinguish between power-╉law, log-╉normal, and
exponential distributions. See Clauset et al. 2009 for further details on this point.
43. The difficulty of distinguishing between log-╉normal and Pareto tails has been widely
commented on in the literature (Embrechts, Klüppelberg, and Mikosch 1997; Bee,
Riccaboni, and Schiavo 2011).
44. This bound comes from the size of the sample that leads the probability density to diverge
when x tends toward zero.

Chapter 6

1. Stanley et al. 1999, 157; Challet, Marsili, and Cheng Zhang 2005, 14; Bouchaud and Potters
2003; McCauley 2004; Bouchaud and Challet 2014; McCauley 2006. See also Rickles 2008
and Schinckus 2010, who discussed this point.
2. Although econophysicists (McCauley 2006, Sornette 2014)  criticize this theoretical
dependence of the modeling step, it is worth mentioning that physics also provides telling
examples in which a theoretical framework is accepted while the empirical results are wholly
incompatible with this framework. One could mention the recent example of the Higgs
boson. The conceptual existence of the Higgs boson predated its observation, meaning that
its theoretical framework was assumed for a number of years without the particle being
observed. In the same vein, string theory is an elegant mathematical framework, empirical/╉
concrete evidence of which is still debated. These are not unique examples:  “There are
plenty of physicists who appear to be unperturbed about working in a manner detached
 

184╇Notes

from experiment: quantum gravity, for example. Here, the characteristic scales are utterly
inaccessible, there is no experimental basis, and yet the problem occupies the finest minds in
physics” (Rickles 2008, 14).
3. By being derived from a theoretical framework setting up the initial calibration, the “model
becomes an a priori hypothesis about real phenomena” (Haavelmo 1944, 8).
4. See, for instance, Mandelbrot 1963; Malcai, Biham, and Solomon 1999; Blanck and
Solomon 2000; Skjeltorp 2000; Cont and Bouchaud 2000; Louzoun and Solomon 2001;
Gupta and Campanha 2002; and Scalas and Kim 2007.
5. Derived distributions like that obtained by Clementi, Di Matteo, and Gallegati (2006) are
in general not Lévy-╉like distributions, but they approach a Lévy distribution for a very large
number of data.
6. Bouchaud, Mezard, and Potters 2002; Potters and Bouchaud 2003; McCauley 2009; Gabaix
2009; Lux 2009; McCauley, Gunaratne, and Bassler 2007b; Sornette 2014; Bouchaud 2002;
McCauley 2006; Stanley and Plerou 2001; Durlauf 2005, 2012; Keen 2003; Chen and Li
2012; Ausloos 2001; Chakrabarti and Chakraborti 2010; Farmer and Lux 2008; Carbone,
Kaniadakis, and Scarfone 2007; Ausloos 2013; Jovanovic and Schinckus 2016.
7. The law of one price suggests that the forces of competition will ensure that any given
commodity will be sold at the same price.
8. Although Modigliani and Miller were not the first to apply the arbitrage proof in finance
(Rubinstein 2003), their article led to its popularity for two reasons:  (1)  their article
was one of the first to use modern probability theory to analyze a financial problem; and
(2) the authors were members of strong academic departments (MIT and the University of
Chicago).
9. Except in Bouchaud and Potters 1994. They reach a first approximation to the European
call option formula that is quite equivalent with the option-╉pricing formula obtained in
mathematical finance under the risk-╉neutral approach, but they use arguments that are
somehow different from those used in mathematical finance.
10. The reason is due to the incompleteness of markets (Ivancevic 2010; Takayasu 2006; Cont
and Tankov 2004; Miyahara 2012; Zhang and Han 2013). See also Â�chapters 1 and 4.
11. Because we work unconditionally on the whole distribution (and not only on its fat-╉tail
part), we need not define a short-╉term time dependence for the variance as is usually done in
conditional methodology, such as ARCH-╉type models.
12. It is worth mentioning that this convergence of x to a Gaussian regime is extremely slow due
to the stable property of the Lévy distribution (Mantegna and Stanley 1994). The crossover
value Nc can be derived by using the Berry-╉Esseen theorem (Shlesinger 1995) or by using a
method based on the probability of x returning to the origin (Mantegna and Stanley 1994)—╉
both approaches provide a crossover value equal to Nc ~ c –╉α l –╉α, where c is the scale factor and
l the cross-╉value at which the regime will switch from a stable Lévy to a Gaussian one.

Conclusion

1. Although the Bayesian framework is also implemented in finance, it is not the statistical
approach used by the mainstream. For further details on this point, see Rachev et al. 2008.
 

REFERENCES

Abergel, Frédéric, Hideaki Aoyama, Bikas K. Chakrabarti, Anirban Chakraborti, and Asim
Ghosh, eds. 2014. Econophysics of Agent-​Based Models. Cham: Springer.
Alastair, Bruce, and David Wallace. 1989. “Critical phenomena:  Universal physics at large
length scales.” In The New Physics, edited by Paul Davies, 109–​52. Cambridge: Cambridge
University Press.
Alejandro-​Quiñones, Ángel L., Kevin E. Bassler, Michael Field, Joseph L. McCauley, Matthew
Nicol, Ilya Timofeyev, Andrew Török, and Gemunu H. Gunaratne. 2006. “A theory of fluc-
tuations in stock prices.” Physica A 363 (2): 383–​92.
Alexander, Sidney S. 1961. “Price movements in speculative markets: Trends or random walk.”
Industrial Management Review 2: 7–​26.
Alfarano, Simone, and Thomas Lux. 2005. “A noise trader model as a generator of apparent fi-
nancial power laws and long memory.” Macroeconomic Dynamics 11: 80–​101.
Alfarano, Simone, and Thomas Lux. 2010. “Extreme value theory as a theoretical background
for power law behavior.” MPRA Paper No. 24718. August 30.
Alstott, Jeff, Ed Bullmore, and Dietmar Plenz. 2014. “Powerlaw: A Python package for analysis
of heavy-​tailed distributions.” PloS One 9 (1): e85777.
Altam, Gunther, ed. 2002. Special issue dedicated to Zipf law. Glottometrics 3.
Amaral, Luis A. Nunes, Sergey V. Buldyrev, Shlomo Havlin, Heiko Leschhorn, Philipp Maass,
Michael A. Salinger, H. Eugene Stanley, and Michael H. R. Stanley. 1997. “Scaling behavior
in economics: I. Empirical results for company growth.” Journal de Physique 7: 621–​33.
Amblard, Frédéric, and Guillaume Deffuant. 2004. “The role of network topology on extremism
propagation with the relative agreement opinion dynamics.” Physica A 343: 725–​38.
Anagnostidis, Panagiotis, and Christos J. Emmanouilides. 2015. “Nonlinearity in high-​
frequency stock returns: Evidence from the Athens Stock Exchange.” Physica A: Statistical
Mechanics and Its Applications 421: 473–​87.
Andersen, Torben G., and Tim Bollerslev. 1997. “Intraday Periodicity and Volatility Persistence
in Financial Markets.” Journal of Empirical Finance 4: 115–​58.
Annaert, Jan, Frans Buelens, and Angelo Riva. 2015. “Financial history databases: Old data, new
issues, new insights.” Paper presented at Financial History Workshop, Judge Business School,
Cambridge, July 23–​24.
Aoki, Masanao, and Hiroshi Yoshikawa. 2007. A Stochastic Approach to Macroeconomics and
Financial Markets. Cambridge: Cambridge University Press.
Aoyama, Hideaki, Yoshi Fujiwara, Yuichi Ikeda, Hiroshi Iyetomi, and Wataru Souma. 2011.
Econophysics and Companies:  Statistical Life and Death in Complex Business Networks.
Cambridge: Cambridge University Press.
Arad, Ruth W. 1980. “Parameter estimation for symmetric stable distribution.” International
Economic Review 21 (1): 209–​20.

185
 

186 References

Arbulu, Pedro. 1998. “La Bourse de Paris au XIXe siècle:  L’exemple d’un marché émergent
devenu efficient.” Revue d’Economie Financière 49: 213–​49.
Archer, Stephen H. 1968. “Introduction.” Journal of Financial and Quantitative Analysis 3
(3): 231–​33.
Armatte, Michel. 1991. “Théorie des erreurs, moyenne et loi ‘normale.’” In Moyenne, milieu,
centre:  Histoires et usages, edited by Jacqueline Feldman, Gérard Lagneau, and Benjamin
Matalon, 63–​84. Paris: Editions EHESS.
Armatte, Michel. 1992. “Conjonctions, conjoncture, et conjecture: Les baromètres économiques
(1885–​1930).” Histoire et Mesure 7 (1–​2): 99–​149.
Armatte, Michel. 1995. “Histoire du modèle linéaire:  Formes et usages en statistique et en
econométrie jusqu’en 1945.” Doctoral dissertation, EHESS.
Armatte, Michel. 2003. “Cycles and Barometers:  Historical insights into the relationship be-
tween an object and its measurement.” In Papers and Proceedings of the Colloquium on the
History of Business-​Cycle Analysis, edited by Dominique Ladiray, 45–​74. Luxembourg: Office
for Official Publication of the European Communities.
Arrow, Kenneth Joseph. 1953. Le rôle des valeurs boursières pour la répartition la meilleure des
risques: (The role of securities in the optimal allocation of risk bearing). Cowles Commission for
Research in Economics. Chicago: University of Chicago.
Arrow, Kenneth Joseph, and Gérard Debreu. 1954. “Existence of an equilibrium for a competi-
tive economy.” Econometrica 22 (3): 265–​90.
Arthur, W. Brian. 2005. “Out-​of-​equilibrium economics and agent-​based modeling.” Santa Fe
Institute Working Paper No. 2005-​09-​03.
Arthur, W. Brian, Steven N. Durlauf, and David A. Lane, eds. 1997. The Economy as an Evolving
Complex System II. Reading, MA: Addison-​Wesley.
Asselain, Jean-​Claude. 1985. Histoire économique: De la révolution industrielle à la première guerre
mondiale. Paris: Presses de la Fondation Nationale des Sciences Politiques & Dalloz.
Auerbach, Felix. 1913. “Das Gesetz der Bevölkerungskonzentration.” Petermanns Geographische
Mitteilungen 59: 73–​76.
Aurell, Erik, Jean-​Philippe Bouchaud, Marc Potters, and Karol Zyczkowski. 1997. “Option pric-
ing and hedging beyond Black and Scholes.” Journal de Physique IV 3 (3): 2–​11.
Ausloos, Marcel. 2001. “Editorial of the Special issue on econophysics.” European Physical
Journal B 20 (4): 471.
Ausloos, Marcel. 2013. “Econophysics: Comments on a few applications, successes, methods
and models.” IIM Kozhikode Society & Management Review 2 (2): 101–​15.
Ausloos, Marcel. 2014a. “Toward fits to scaling-​like data, but with inflection points and general-
ized Lavalette function.” Journal of Applied Quantitative Methods 9 (2): 1.
Ausloos, Marcel. 2014b. “Two-​exponent Lavalette function: A generalization for the case of ad-
herents to a religious movement.” Physical Review E: Statistical, Nonlinear, and Soft Matter
Physics 89 (6): 062803.
Ausloos, Marcel. 2015. “France new regions planning? Better order or more disorder?” Entropy
17 (8): 5695–​710.
Ausloos, Marcel, Kristinka Ivanova, and Nicolas Vandewalle. 2002. “Crashes: Symptoms, diag-
noses and remedies.” In Empirical Sciences of Financial Fluctuations: The Advent of Econophysics,
edited by Hideki Takayasu, 62–​76. Berlin: Springer Verlag.
Ausloos, Marcel, Franck Jovanovic, and Christophe Schinckus. 2016. On the “usual” misunder-
standing between econophysics and finance: Some clarifications on modelling approaches
and efficient market hypothesis. International Review of Financial Analysis 47: 7–14.
 

187  References

Axtell, Robert L. 1999. “The emergence of firms in a population of agents: Local increasing re-
turns, unstable Nash equilibria, and power law size distributions.” Santa Fe Institute Working
Paper No. 99-​03-​019.
Axtell, Robert L. 2001. “Zipf distribution of U.S. firm sizes.” Science in Context 293: 1818–​20.
Bachelier, Louis. 1900. “Théorie de la spéculation.” Annales de l’Ecole Normale Supérieure, 3rd
ser. 17 ( January): 21–​86. Reprint, 1995, J. Gabay, Paris.
Bachelier, Louis. 1901. “Théorie mathématique du jeu.” Annales de l’Ecole Normale Supérieure,
3rd ser. 18 ( January): 77–​119. Reprint, 1995, J. Gabay, Paris.
Bachelier, Louis. 1910. “Les probabilités à plusieurs variables.” Annales Scientifiques de l’École
Normale Supérieure 339–​60.
Bachelier, Louis. 1912. Calcul des probabilités. Paris: Gauthier-​Villars.
Backhouse, Roger E. 2004. “A suggestion for clarifying the study of dissent in economics.”
Journal of the History of Economic Thought 26 (2): 261–​71.
Bagehot, Walter (pseudonym used by Jack Treynor). 1971. “The only game in town.” Financial
Analysts Journal 8: 31–​53.
Bak, Per. 1994. “Introduction to self-​criticality.” In Complexity: Metaphors, Models, and Reality,
edited by G. Cowan, D. Pines, and D. Meltzer, 476–​82. Santa Fe: Santa Fe Institute.
Bak, Per, Kan Chen, José Sheinkman, and Michael Woodford. 1993. “Aggregate fluctuations
from independent sectorial shocks: Self-​organized criticality in a model of production and
inventory dynamics.” Ricerche Economische 47 (1): 3–​30.
Bak, Per, Maya Paczuski, and Martin Shubik. 1997. “Price variations in a stock market with
many agents.” Physica A 246 (3–​4): 430–​53.
Bak, Per, Chao Tang, and Kurt Wiesenfeld. 1987. “Self-​organized criticality: An explanation of
1/​f noise.” Physical Review Letters 59 (4): 381–​84.
Bak, Per, Chao Tang, and Kurt Wiesenfeld. 1988. “Self-​organized criticality.” Physical Review A
38 (1): 364–​74.
Banerjee, Abhijit V. 1993. “The economics of rumors.” Review of Economic Studies 60 (2):
309–​27.
Banz, Rolf W. 1981. “The relationship between return and market value of common stocks.”
Journal of Financial Economics 9 (1): 3–​18.
Banzhaf, H. Spencer. 2001. “Quantifying the qualitative:  Quality-​adjusted price indexes in
the United States, 1915–​61.” In History of Political Economy, annual supplement, The Age
of Economic Measurement, edited by Judy L. Klein and Mary S. Morgan, 345–​70. Durham,
NC: Duke University Press.
Barbut, Marc. 2003. “Homme moyen ou homme extrême? De Vilfredo Pareto (1896) à Paul
Lévy (1936) en passant par Maurice Fréchet et quelques autres.” Journal de la Société Française
de la Statistique 144 (1–​2): 113–​33.
Barnett, Vincent. 2011. E. E.  Slutsky as Economist and Mathematician:  Crossing the Limits of
Knowledge. London: Routledge.
Bartolozzi, Marco, and Anthony William Thomas. 2004. “Stochastic cellular automata model for
stock market dynamics.” Physical Review E 4: 46–​69.
Bassler, Kevin E., Joseph L. McCauley, and Gemunu H. Gunaratne. 2007. “Nonstationary incre-
ments, scaling distributions, and variable diffusion processes in financial markets.” Proceedings
of the National Academy of Sciences of the United States of America 104 (44):  17287–​290.
doi: 10.1073/​pnas.0708664104.
Batterman, Robert W. 2002. The Devil in the Details:  Asymptotic Reasoning in Explanation,
Reduction and Emergence. New York: Oxford University Press.
 

188 References

Bauwens, Luc, Sébastien Laurent, and Jeroen V.  K. Rombouts. 2006. “Multivariate Garch
models: A survey.” Journal of Applied Econometrics 21 (1): 79–​109.
Bayeux-Besnainou, Isabelle, and Jean-Charles Rochet. 1996. “Dynamic Spanning: Are Options
an Appropriate Instrument?” Mathematical Finance, 6 (1): 1–16.
Bazerman, Charles. 1988. Shaping Written Knowledge: The Genre and Activity of the Experimental
Articles in Science. Madison: University of Wisconsin Press.
Bee, Marco, Massimo Riccaboni, and Stefano Schiavo. 2011. “Pareto versus lognormal: A max-
imum entropy test.” Physical Review E 84 (2): 026104.
Ben-​El-​Mechaiekh, Hichem, and Robert W. Dimand. 2006. “Louis Bachelier.” In Pioneers of
Financial Economics, vol. 1: Contributions Prior to Irving Fisher, edited by Geoffrey Poitras,
225–​37. Cheltenham: Edward Elgar.
Ben-​El-​Mechaiekh, Hichem, and Robert W. Dimand. 2008. “Louis Bachelier’s 1938 volume on
the calculus of speculation: Efficient markets and mathematical finance in Bachelier’s later
work.” Working paper.
Bernstein, Peter L. 1992. Capital Ideas:  The Improbable Origins of Modern Wall Street.
New York: Free Press; Toronto: Maxwell Macmillan Canada.
Bernstein, Peter L. 2007. Capital Ideas Evolving. Hoboken, NJ: John Wiley & Sons.
Berry, Brian J. L., and Adam Okulicz-​Kozaryn. 2012. “The city size distribution debate: Resolution
for US urban regions and megalopolitan areas.” Cities 29 (supplement 1): S17–​S23.
Biais, Bruno, Larry Glosten, and Chester Spatt. 2005. “Market microstructure: A survey of micro-
foundations, empirical results, and policy implications.” Journal of Financial Markets 8 (2): 217–​64.
Binney, James, N. J. Dowrick, Andrew J. Fisher, and Mark E. J. Newman. 1992. The Theory of
Critical Phenomena: An Introduction to the Renormalization Group. Oxford: Clarendon Press.
Black, Fischer. 1971a. “Toward a fully automated stock exchange (part 1).” Financial Analysts
Journal 27 (4): 28–​35, 44.
Black, Fischer. 1971b. “Toward a fully automated stock exchange (part 2).” Financial Analysts
Journal 27 (6): 24–​28, 86–​87.
Black, Fischer. 1986. “Noise.” Journal of Finance 41 (3): 529–​43.
Black, Fischer, and Myron Scholes. 1973. “The pricing of options and corporate liabilities.”
Journal of Political Economy 81 (3): 637–​54.
Blank, Aharon, and Sorin Solomon. 2000. “Power laws in cities population, financial markets
and internet sites (scaling in systems with a variable number of components).” Physica A 287
(1–​2): 279–​88.
Blattberg, Robert, and Nicholas Gonedes. 1974. “A comparison of the stable and Student dis-
tributions as statistical models for stock prices: Reply.” Journal of Business 47 (2): 244–​80.
Blattberg, Robert, and Thomas Sargent. 1971. “Regression with non-​Gaussian stable distur-
bances: Some sampling results.” Econometrica 39 (3): 501–​10.
Blume, Lawrence E., and Steven N. Durlauf, eds. 2006. The Economy as an Evolving Complex
System III: Current Perspectives and Future Directions. New York: Oxford University Press.
Blume, Marshall. 1968. “The assessment of portfolio performance: An application of portfolio
theory.” Doctoral dissertation, University of Chicago.
Bollerslev, Tim. 1986. “Generalized autoregressive conditional heteroskedasticity.” Journal of
Econometrics 31: 307–​27.
Bollerslev, Tim. 2010. “Glossary to ARCH (GARCH).” In Volatility and Time Series
Econometrics: Essays in Honour of Robert F. Engle, edited by Tim Bollerslev, Jeffrey R. Russell,
and Mark W. Watson, 137–​64. New York: Oxford University Press.
 

189  References

Borak, Szymon, Wolfgang Hardle, and Rafal Weron. 2005. “Stable distributions.” SFB Working
Paper, Humboldt University Berlin.
Bormetti, Giacomo, Enrica Cisana, Guido Montagna, and Oreste Nicrosini. 2007. “A non-​
Gaussian approach to risk measures.” Physica A 376: 532–​42.
Bouchaud, Jean-​Philippe. 2001. “Power laws in economics and finance: Some ideas from phys-
ics.” Quantitative Finance 1 (1): 105–​12.
Bouchaud, Jean-​ Philippe. 2002. “An Introduction to statistical finance.” Physica A 313
(1): 238–​51.
Bouchaud, Jean-​Philippe, and Damien Challet. 2014. “Behavioral finance and financial mar-
kets: Arbitrage techniques, exuberant behaviors and volatility.” Opinion et débats 7: 20–​35.
Bouchaud, Jean-​Philippe, Doyne Farmer, and Fabrizio Lillo. 2009. “How markets slowly digest
changes in supply and demand.” In Handbook of Financial Markets: Dynamics and Evolution,
edited by Thorsten Hens and Klaus Reiner Schenk-​Hoppe, 57–​160. Amsterdam:  North-​
Holland, Elsevier.
Bouchaud, Jean-​Philippe, Marc Mezard, and Marc Potters. 2002. “Statistical properties of stock
order books: Empirical results and models.” Quantitative Finance 2: 251–​56.
Bouchaud, Jean-​Philippe, and Marc Potters. 2000. Theory of Financial Risks:  From Statistical
Physics to Risk Management. Cambridge: Cambridge University Press.
Bouchaud, Jean-​Philippe, and Marc Potters. 2003. Theory of Financial Risk and Derivate Pricing.
Cambridge: Cambridge University Press.
Bouchaud, Jean-​ Philippe, and Didier Sornette. 1994. “The Black-​ Scholes option pricing
problem in mathematical finance: Generalisation and extension to a large class of stochastic
processes.” Journal de Physique I 4: 863–​81.
Boumans, Marcel, ed. 2007. Measurement in Economics: A Handbook. London: Elsevier.
Bouzoubaa, Mohamed, and Adel Osseiran. 2010. Exotic Options and Hybrids:  A  Guide to
Structuring, Pricing and Trading. London: John Wiley & Sons.
Bowley, Arthur Lyon. 1933. “The action of economic forces in producing frequency distri-
butions of income, prices, and other phenomena: A suggestion for study.” Econometrica 1
(4): 358–​72.
Boyarchenko, Svetlana I., and Sergei Z. Levendorskiĭ. 2000. “Option pricing for truncated Lévy
processes.” International Journal of Theoretical and Applied Finance 3 (3): 549–​52.
Boyarchenko, Svetlana I., and Sergei Z. Levendorskiĭ. 2002. “Barrier options and touch-​and-​out
options under regular Lévy processes of exponential type.” Annals of Applied Probability 12
(4): 1261–​98.
Brada, Josef, Harry Ernst, and John van Tassel. 1966. “The distribution of stock price differ-
ences: Gaussian after all?” Operations Research 14 (2): 334–​40.
Brakman, Steven, Harry Garretsen, Charles Van Marrewijk, and Marianne van den Berg. 1999.
“The return of Zipf: Towards a further understanding of the rank-​size distribution.” Journal
of Regional Science 39: 183–​213.
Breidt, F. Jay, Nuno Crato, and Pedro de Lima. 1998. “On the detection and estimation of long
memory in stochastic volatility.” Journal of Econometrics 83: 325–​48.
Brenner, Menachem. 1974. “On the stability of distribution of the market component in stock
price changes.” Journal of Financial and Quantitative Analysis 9: 945–​61.
Brenner, Thomas. 2001. “Self-​organisation, local symbiosis of firms and the life cycle of lo-
calised industrial clusters.” Papers on Economics and Evolution, Max Planck Institute of
Economics, Jena.
 

190 References

Breton, Yves. 1991. “Les économistes français et les questions de méthode.” In L’Economie
Politique en France au XIXe Siècle, edited by Yves Breton and Michel Lutfalla, 389–​419.
Paris: Economica.
Broda, Simon A., Markus Haas, Jochen Krause, Marc S. Paolella, and Sven C. Steude. 2013.
“Stable mixture GARCH models.” Journal of Econometrics 172 (2): 292–​306.
Brody, Samuel. 1945. Bioenergetics and Growth. New York: Reinhold.
Bronzin, Vinzenz. 1908 “Theory of Premium Contracts.” In Vinzenz Bronzin's Option Pricing
Models: Exposition and Appraisal, edited by Wolfgang Hafner and Heinz Zimmermann 2009.
Berlin: Springer, 113–​200.
Brown, Stephen J., William N. Goetzmann, and Alok Kumar. 1998. “The Dow theory: William
Peter Hamilton’s track record reconsidered.” Journal of Finance 53 (4): 1311–​33.
Buchanan, Mark. 2004. “Power law and the new science of complexity management.” Strategy
+ Business 34: 104–​7.
Bucsa, Gigel, Emmanuel Haven, Franck Jovanovic, and Christophe Schinckus. 2014. “The prac-
tical use of the general formula of the optimal hedge ratio in option pricing: An example of
the generalized model of an exponentially truncated Levy stable distribution.” Theoretical
Economics Letters 4 (9): 760–​66.
Cai, Wei, Melvin Lax, and Min Xu. 2006. Random Processes in Physics and Finance. Oxford: Oxford
University Press.
Calvo, Iván, Juan C. Cuchí, José G. Esteve, and Fernando Falceto. 2010. “Generalized
central limit theorem and renormalization group.” Journal of Statistical Physics 141:
409–​21.
Carbone, Anna, Giorgio Kaniadakis, and Antonio Maria Scarfone. 2007. “Where do we stand
on econophysics.” Physica A 382: 11–​14.
Carr, Peter, Hélyette Geman, Dilip Madan, and Marc Yor. 2002. “The fine structure of asset re-
turns: An empirical investigation.” Journal of Business 75: 305–​32.
Carr, Peter, and Dilip B. Madan. 2005. “A note on sufficient conditions for no arbitrage.” Finance
Research Letters 2 (3): 125–​30.
Cartwright, Nancy. 1983. How the Laws of Physics Lie. Oxford: Oxford University Press.
Casey, Michael. 2013. “Move over economists, markets need physicists.” Market Watch
Magazine, July 10.
Cassidy, David. 2011. A Short History of Physics in the American Century. Cambridge: Harvard
University Press.
Cerqueti, Roy, and Marcel Ausloos. 2015. “Evidence of economic regularities and disparities of
Italian regions from aggregated tax income size data.” Physica A: Statistical Mechanics and Its
Applications 421: 187–​207.
Chaigneau, Nicolas, and Philippe Le Gall. 1998. “The French connection: The pioneering e-
conometrics of Marcel Lenoir (1913).” In European Economists of the Early 20th Century, vol.
1: Studies of Neglected Thinkers of Belgium, France, the Netherlands and Scandinavia, edited by
Warren J. Samuels, 163–​89. Cheltenham: Edward Elgar.
Chakrabarti, Bikas K., and Anirban Chakraborti. 2010. “Fifteen years of econophysics research.”
Science and Culture 76: 9–​10.
Chakrabarti, Bikas K., Anirban Chakraborti, and Arnab Chatterjee. 2006. Econophysics and
Sociophysics: Trends and Perspectives. Weinheim: Wiley-​VCH.
Chakraborti, Anirban, Ioane Muni Toke, Marco Patriarca, and Frédéric Abergel. 2011a.
“Econophysics review: I. Empirical facts.” Quantitative Finance 11 (7): 991–​1012.
 

191  References

Chakraborti, Anirban, Ioane Muni Toke, Marco Patriarca, and Frédéric Abergel. 2011b.
“Econophysics review: II. Agent-​based models.” Quantitative Finance 11 (7): 1013–​41.
Challet, Damien, Mateo Marsili, and Yi Cheng Zhang. 2005. Minority Games: Interacting Agents
in Financial Markets. Oxford: Oxford University Press.
Challet, Damien, and Robin Stinchcombe. 2001. “Analyzing and modeling 1+1d markets.”
Physica A 300 (1–​2): 285–​99.
Champernowne, David Gawen. 1953. “A model of income distribution.” Economic Journal
63: 318–​51.
Chancelier, Éric. 2006a. “L’analyse des baromètres économiques de Persons et
Wagemann:  Instrument de prévision—​instrument de théorisation, 1919–​1932.” Revue
d’Économie Politique 116 (5): 613–​32.
Chancelier, Éric. 2006b. “Les premiers baromètres économiques américains (1990–​1919).”
Revue d’Histoire des Sciences Humaines 15 (2): 135–​55.
Chane-​Alune, Elena. 2006. “Accounting standardization and governance structures.” Working
Paper No. 0609, University of Liège.
Chari, V. V., Patrick J. Kehoe, and Ellen R. McGrattan. 2008. “Are structural VARs with long-​run
restrictions useful in developing business cycle theory?” Journal of Monetary Economics 55
(8): 1337–​52.
Chari, V. V., Patrick J. Kehoe, and Ellen R. McGrattan. 2009. “New Keynesian models: Not yet
useful for policy analysis.” American Economic Journal: Macroeconomics 1 (1): 242–​66.
Chen, Shu-​heng, and Sai-​ping Li. 2012. “Econophysics:  Bridges over a turbulent current.”
International Review of Financial Analysis 23: 1–​10. doi: 10.1016/​j.irfa.2011.07.001.
Chen, Shu-​Heng, Thomas Lux, and Michele Marchesi. 2001. “Testing for non-​linear structure in
an artificial financial market.” Journal of Economic Behavior and Organization 46 (3): 327–​42.
Chen, X., H. Anderson, and P. Barker. 2008. “Kuhn’s theory of scientific revolutions and cogni-
tive psychology.” Philosophical Psychology 11 (1): 5–​28.
Chiarella, Carl, and Giulia Iori. 2002. “A Simulation analysis of the microstructure of double
auction markets.” Quantitative Finance 2 (5): 346–​53.
Chiarella, Carl, Giulia Iori, and Josep Perello. 2009. “The impact of heterogeneous trading rules on
the limit order book and order flows.” Journal of Economic Dynamics and Control 33: 525–​37.
Chrisman, Nicholas. 1999. “Trading Zones or Boundary Objects: Understanding Incomplete
Translations of Technical Expertise.” San Diego.
Christ, Carl F. 1994. “The Cowles Commission’s contributions to econometrics at Chicago,
1939–​1955.” Journal of Economic Literature 32 (March): 30–​59.
Christiano, Lawrence J. 2012. “Christopher A. Sims and vector autoregressions.” Scandinavian
Journal of Economics 114 (4): 1082–​104.
Clark, Peter K. 1973. “A Subordinated stochastic process with finite variance for speculative
prices.” Econometrica 41 (1): 135–​55.
Clauset, Aaron. 2011. Inference, Models and Simulation for Complex Systems. Lecture 3. in, Santa
Fe, http://​tuvalu.santafe.edu/​~aaronc/​courses/​7000/​csci7000-​001_​2011_​L3.pdf
Clauset, Aaron, Cosma Rohilla Shalizi, and Mark Newman. 2009. “Power-​law distributions in
empirical data.” SIAM Review 51 (4): 661–​703.
Clementi, Fabio, Tiziana Di Matteo, and Mauro Gallegati. 2006. “The power-​law tail exponent
of income distributions.” Physica A 370 (1): 49–​53.
Clippe, Paulette, and Marcel Ausloos. 2012. “Benford’s law and Theil transform of financial
data.” Physica A: Statistical Mechanics and Its Applications 391 (24): 6556.
 

192 References

Cohen, I. Bernard. 1993. “Analogy, homology and metaphor in the interactions between
the natural sciences and the social sciences, especially economics.” In Non-​natural Social
Science:  Reflecting on the Enterprise of “More Heat Than Light”, edited by Neil De Marchi,
7–44. Durham, NC: Duke University Press.
Cohen, Kalman J., and Jerry A. Pogue. 1967. “An empirique evaluation of alternative portfolio-​
selection models.” Journal of Business 40 (2): 166–​93.
Condon, E. U. 1928. “Statistics of vocabulary.” Science 67 (1733): 300.
Cont, Rama, and Jean-​Philippe Bouchaud. 2000. “Herd behavior and aggregate fluctuations in
financial markets.” Macroeconomic Dynamics 4 (2): 170–​96.
Cont, Rama, Marc Potters, and Jean-​Philippe Bouchaud. 1997. “Scaling in stock market
data: Stable laws and beyond.” In Scale Invariance and Beyond: Les Houches Workshop, March
10–​14, 1997, edited by B. Dubrulle, F. Graner, and D. Sornette, 75–​85. New York: Springer.
Cont, Rama, and Peter Tankov. 2004a. Financial Modelling with Jump Processes. London: Chapman
& Hall/​CRC.
Cont, Rama, and Peter Tankov. 2004b. “Non-​parametric calibration of jump-​diffusion option
pricing models.” Journal of Computational Finance 7 (3): 1–​49.
Cootner, Paul H. 1962. “Stock prices: Random vs. systematic changes.” Industrial Management
Review 3 (2): 24.
Cootner, Paul H. 1964. The Random Character of Stock Market Prices. Cambridge, MA: 
MIT Press.
Copeland, Thomas E., and Dan Galai. 1983. “Information effects and the bid-​ask spread.” Journal
of Finance 38 (5): 1457–​69.
Cordoba, J. 2008. “On the distribution of city sizes.” Journal of Urban Economics 63: 177–​97.
Cornelis, A. Los. 2005. “Why VAR fails: Long memory and extreme events in financial markets.”
Journal of Financial Economics 3 (3): 19–​36.
Courtadon, Georges. 1982. “A note on the premium market of the Paris stock exchange.” Journal
of Banking and Finance 6 (4): 561–​65.
Courtault, Jean-​Michel, Youri Kabanov, Bernard Bru, Pierre Crépel, Isabelle Lebon, and Arnaud
Le Marchand. 2002. “Louis Bachelier on the centenary of Théorie de la spéculation.” In Louis
Bachelier: Aux origines de la finance mathématique, edited by Jean-​Michel Courtault and Youri
Kabanov, 5–​86. Besançon: Presses Universitaires Franc-​Comtoises.
Cover, John H. 1937. “Some investigations in the Sampling and Distribution of Retail Prices.”
Econometrica 5 (3): 263–​79.
Cowles, Alfred. 1933. “Can stock market forecasters forecast?” Econometrica 1 (3): 309–​24.
Cowles, Alfred. 1944. “Stock market forecasting.” Econometrica 12 (3/​4): 206–​14.
Cowles, Alfred. 1960. “A revision of previous conclusions regarding stock price behavior.”
Econometrica 28 (4): 909–​15.
Cowles, Alfred, and Herbert E. Jones. 1937. “Some a posteriori probabilities in stock market
action.” Econometrica 5 (3): 280–​94.
Cox, John C., and Stephen Ross. 1976. “The valuation of options for alternative stochastic pro-
cesses.” Journal of Financial Economics 3 (1-​2): 145–​66.
Cramer, Harald. 1983. “Probabilité mathématique et inférence statistique: Quelques souvenirs
personnels sur une importante étape du progrès scientifique.” Revue de Statistique Appliquée
31 (3): 5–​15.
Cristelli, Matthieu. 2014. Complexity in Financial Markets:  Modeling Psychological Behavior in
Agent-​Based Models and Order Book Models. Milan: Springer.
 

193  References

Csorgor, S., and G. Simons. 1983. “On the Steinhpaus’s resolution of the St-​Petersburg Paradox.”
Probability and Mathematical Statistics 14: 157–​71.
Cuoco, Domenico, and Eugenio Barone. 1989. “The Italian market for ‘premium’ contracts: An
application of option pricing theory.” Journal of Banking and Finance 13 (4): 709–​45.
Daemen, Job. 2010. “Ketchup economics:  What is finance about?” Presented at the 14th
Annual Conference of the European Society for the History of Economic Thought,
Amsterdam, March.
Davis, Mark, and Alison Etheridge. 2006. Louis Bachelier’s Theory of Speculation. Princeton,
NJ: Princeton University Press.
De Bondt, Werner F. M., and Richard Thaler. 1985. “Does the stock market overreact?” Journal
of Finance 40 (3): 793–​805.
De Meyer, Bernard, and Hadiza Moussa Saley. 2003. “On the strategic origin of Brownian
motion in finance.” International Journal of Game Theory 31: 285–​319.
De Vroey, Michel, and Pierre Malgrange. 2007. “Théorie et modélisation macro-​économiques,
d’hier à aujourd’hui.” Revue Française D’économie 21 (3): 3–​38.
Debreu, Gérard. 1959. Theory of Value. New Haven: Yale University Press.
Demsetz, Harold. 1968. “The cost of transacting.” Quarterly Journal of Economics 82: 33–​53.
Derman, Emanuel. 2001. “A guide for the perplexed quant.” Quantitative Finance 1 (5): 476–​80.
Derman, Emanuel. 2009. “Models.” Financial Analysts Journal 65 (1): 28–​33
Diewert, W. Erwin. 1993. “The early history of price index research.” In Essays in Index Number
Theory, vol. 1, edited by W. Erwin Diewert and Alice O. Nakamura, 33–​65. Amsterdam: North
Holland.
Dimand, Robert W. 2007. “Irving Fisher and financial economics: The equity premium puzzle,
the predictability of stock prices, and intertemporal allocation under risk.” Journal of the
History of Economic Thought 29 (2): 153–​66. doi: 10.1080/​10427710701335885.
Dimand, Robert W. 2009. “The Cowles Commission and Foundation on the functioning of fi-
nancial markets from Irving Fisher and Alfred Cowles to Harry Markowitz and James Tobin.”
Revue d’Histoire des Sciences Humaines 20: 109–​30.
Dimand, Robert W., and John Geanakoplos, eds. 2005. Celebrating Irving Fisher: The Legacy of a
Great Economist. Malden, MA: Blackwell.
Dimand, Robert W., and William Veloce. 2010. “Alfred Cowles and Robert Rhea on the predict-
ability of stock prices.” Journal of Business Inquiry 9 (1): 56–​64.
Ding, E-​Jiang. 1983. “Asymptotic properties of the Markovian master equations for multi-​
stationary systems.” Physica A 119: 317–​26.
Ding, Zhuanxin, Robert F. Enlge, and Clive W. J. Granger. 1993. “A long memory property of
stock market returns and a new model.” Journal of Empirical Finance 1 (1): 83–​106.
Dionisio, Andreia, Rui Menezes, and Diana A. Mendes. 2006. “An econophysics approach to
analyse uncertainty in financial markets:  An application to the Portuguese stock market.”
European Physical Journal B 50 (1): 161–​64.
Domb, Cyril, and D. Hunter. 1965. “On the critical behaviour of ferromagnets.” Proceedings of
the Physical Society 86: 1147.
Donangelo, Raul, and Kim Sneppen. 2000. “Self-​organization of value and demand.” Physica A
276: 572–​80.
Donnelly, Kevin. 2015. Adolphe Quetelet, Social Physics and the Average Men of Science, 1796–​
1874. New York: Routledge.
Doob, Joseph L. 1953. Stochastic Process. New York: John Wiley & Sons.
 

194 References

Dow, James, and Gary Gorton. 2008. “Noise traders.” In The New Palgrave:  A  Dictionary of
Economics, 2nd ed., edited by Steven N. Durlauf and Lawrence E. Blume. New York: Palgrave
Macmillan.
Dubkov, Alexander A., Bernardo Spagnolo, and Vladimir V. Uchaikin. 2008. “Lévy flight
superdiffusion:  An introduction.” International Journal of Bifurcation and Chaos 18
(9): 2649–​72.
Dunbar, Frederick C., and Dana Heller. 2006. “Fraud on the market meets behavioral finance.”
Delaware Journal of Corporate Law 31 (2): 455–​532.
Dunbar, Nicholas. 2000. Inventing Money: The Story of Long-​Term Capital Management and the
Legends behind It. New York: Wiley.
Dupoyet, Brice, Rudolf H. Fiebig, and David P. Musrgrove. 2011. “Replicating financial market
dynamics with a simple self-​organized critical lattice model.” Physica A 390: 3120–​35.
Durlauf, Steven N. 2005. “Complexity and empirical economics.” Economic Journal
115: F225–​F243.
Durlauf, Steven N. 2012. “Complexity, economics, and public policy.” Politics Philosophy
Economics 11 (1): 45–​75.
Eberlein, Ernst, and Ullrich Keller. 1995. “Hyperbolic distributions in finance.” Bernouilli
1: 281–​99.
Edlinger, Cécile, and Antoine Parent. 2014. “The beginnings of a ‘common-​sense’ approach to
portfolio theory by nineteenth-​century French financial analysts Paul Leroy-​Beaulieu and
Alfred Neymarck.” Journal of the History of Economic Thought 36 (1): 23–​44.
Eeckhout, Jan. 2004. “Gibrat’s law for (all) cities.” American Economic Review 94 (1): 1429–​51.
Eguíluz, Victor M., and Martin G. Zimmermann. 2000. “Transmission of information and herd
behavior: An application to financial markets.” Physical Review Letters 85 (26): 5659–​62.
Eichenbaum, Martin. 1996. “Some comments on the role of econometrics in economic theory.”
Economic Perspectives 20 (1): 22.
Einstein, Albert. 1905. “Uber die von der molekularkinetischen Theorie der Wärme geforderte
Bewegung von in ruhenden Flüssigkeiten suspendierten Teilchen.” Annalen der Physik
17: 549–​60.
Embrechts, Paul, Claudia Klüppelberg, and Thomas Mikosch. 1997a. Modelling Extremal
Events: For Insurance and Finance. New York: Springer.
Embrechts, Paul, Claudia Klüppelberg, and Thomas Mikosch. 1997b. Modelling Extreme Values.
Berlin: Springer.
Engle, Robert F., and Jeffrey R. Russell. 2004. “Analysis of high frequency financial data.” In
Handbook of Financial Econometrics: Tools and Techniques, edited by Yacine Ait-​Sahalia and
Lars Hansen, 383–​426. Amsterdam: Elsevier.
Engle, Robert F. 1982. “Autoregressive conditional heteroskedasticity with estimates of the var-
iance of United Kingdom inflation.” Econometrica 50: 987–​1007.
Epstein, Joshua M. 2006. Generative Social Science:  Studies in Agent-​Based Computational
Modeling. Princeton, NJ: Princeton University Press.
Epstein, Joshua M., and Robert L. Axtell. 1996. Growing Artificial Societies: Social Science from the
Bottom Up. Cambridge, MA: MIT Press.
Estoup, Jean-​Baptiste. 1916. Gammes sténographique. Paris: Institut Sténographique.
Fama, Eugene F. 1963. “Mandelbrot and the stable Paretian hypothesis.” Journal of Business 36
(4): 420–​29.
Fama, Eugene F. 1965a. “The behavior of stock-​market prices.” Journal of Business 38 (1): 34–​105.
 

195  References

Fama, Eugene F. 1965b. “Portfolio analysis in a stable Paretian market.” Management Science
Series A: Sciences 11 (3): 404–​19.
Fama, Eugene F. 1965c. “Random walks in stock-​market prices.” Financial Analysts Journal 21
(5): 55–​59.
Fama, Eugene F. 1970. “Efficient capital markets:  A  review of theory and empirical work.”
Journal of Finance 25 (2): 383–​417.
Fama, Eugene F. 1976a. “Efficient capital markets: Reply.” Journal of Finance 31 (1): 143–​45.
Fama, Eugene F. 1976b. Foundations of Finance:  Portfolio Decisions and Securities Prices.
New York: Basic Books.
Fama, Eugene F. 1991. “Efficient capital markets: II.” Journal of Finance 46 (5): 1575–​617.
Fama, Eugene F. 2008. “Interview with Professor Eugene Fama by Professor Richard Roll.”
August 15.
Fama, Eugene F., Lawrence Fisher, Michael C. Jensen, and Richard Roll. 1969. “The Adjustment
of Stock Prices to New Information.” International Economic Review 10 (1): 1–​21.
Fama, Eugene F., and Kenneth French. 2004. “Capital asset pricing model:  Theory and evi-
dence.” Journal of Economic Perspective 18 (3): 25–​36.
Fama, Eugene F., and Richard Roll. 1968. “Some properties of symmetric stable distributions.”
Journal of the American Statistical Association 63 (323): 817–​36.
Fama, Eugene F., and Richard Roll. 1971. “Parameter estimates for symmetric stable distribu-
tions.” Journal of the American Statistical Association 66 (334): 331–​38.
Farmer, J. Doyne, and Duncan Foley. 2009. “The economy needs agent-​based modelling.”
Nature 460: 685–​86.
Farmer, J. Doyne, and John Geanakoplos. 2008. “Power laws in economics and elsewhere.”
Working paper.
Farmer, J. Doyne, and John Geanakoplos. 2009. “The Virtues and Vices of Equilibrium, and the
Future of Financial Economics.” Working paper, Cowles Foundation.
Farmer, J. Doyne, Laszlo Gillemot, Fabrizio Lillo, Szabols Mike, and Anindya Sen. 2004. “What
really causes large price changes?” Quantitative Finance 4: 383–​97.
Farmer, J. Doyne, and Fabrizio Lillo. 2004. “On the origin of power law tails in price fluctua-
tions.” Quantitative Finance 4 (1): 7–​11.
Farmer, J. Doyne, and Thomas Lux. 2008. “Introduction to special issue on ‘Applications of
Statistical Physics in Economics and Finance.’” Journal of Economic Dynamics and Control 32
(1): 1–​6.
Farmer, J. Doyne, Paolo Patelli, Ilija I. Zovko, and Kenneth J. Arrow. 2005. “The predictive power
of zero intelligence in financial markets.” Proceedings of the National Academy of Sciences of the
United States of America 102 (6): 2254–​59.
Fayyad, U. 1996. “Data mining and knowledge discovery: Making sense out of data.” Intelligent
Systems 11 (5): 20–​25.
Feigenbaum, James A., and Peter G. O. Freund. 1996. “Discrete scale invariance in stock markets
before crashes.” International Journal of Modern Physics B 10 (27): 3737–​45.
Feigenbaum, James A., and Peter G. O. Freund. 1998. “Discrete scale invariance and the ‘second
Black Monday.’” Modern Physics Letters B 12 (3): 57–​60.
Feller, William. 1957. An Introduction to Probability Theory and Its Applications. Vol. 1.
New York: Wiley.
Feller, William. 1971. An Introduction to Probability Theory and Its Applications, Vol. 2.
New York: Wiley.
 

196 References

Figueiredo, Annibal, Raul Matsushita, Sergio Da Silva, Maurizio Serva, Ganghi Viswanathan,
Cesar Nasciment, and Iram Gleria. 2007. “The Lévy sections theorem:  An application to
econophysics.” Physica A 386 (2): 756–​59.
Fisher, Irving. 1911. The Purchasing Power of Money. London: Macmillan.
Fisher, Irving. 1922. The Making of Index Numbers: A Study of Their Varieties, Tests, and Reliability.
Boston: Houghton Mifflin.
Fisher, Irving. 1930. The Theory of Interest as Determined by Impatience to Spend Income and
Opportunity to Invest It. New York: Macmillan.
Fisher, Lawrence, and James H. Lorie. 1964. “Rates of return on investments in common stocks.”
Journal of Business 37 (1): 1–​21.
Fisher, Mark. 1967. “The theory of equilibrium critical phenomena.” Reports on Progress in
Physics 30: 615–​730.
Fitzpatrick, Richard. 2012. “Thermodynamics and statistical mechanics: An intermediate level
course.”
Focardi, Sergio, Silvano Cincotti, and Michele Marchesi. 2002. “Self-​organization and market
crashes.” Journal of Economic Behavior and Organization 49 (2): 241–​67.
Fourcade, Marion, and Rakesh Khurana. 2009. “From social control to financial economics: The
linked ecologies of economics and business in twentieth-​century America.” Presented to the
Annual Meeting of the SASE, Paris.
Francq, Christian, and Jean-​Michel Zakoian. 2010. GARCH Models:  Structure, Statistical
Inference and Financial Applications. London: Wiley.
Frankfurter, George M., and Elton G. McGoun. 1999. “Ideology and the theory of financial ec-
onomics.” Journal of Economic Behavior and Organization 39: 159–​77.
Franzen, Dorothee. 2010. “Managing investment risk in defined benefit pension funds.” OECD
Working Papers on Insurance and Private Pensions.
Friedman, Walter A. 2009. “The Harvard Economic Service and the problems of forecasting.”
History of Political Economy 41 (1): 57–​88.
Frigg, R. 2003. “Self-​organised criticality: What it is and what it isn’t.” Studies in the History and
Philosophy of Science Part A 34 (3): 613–​32.
Fujimoto, Shouji, Atushi Ishikawa, Takayuki Mizuno, and Tsutomu Watanabe. 2011. “A new
method for measuring tail exponents of firm size distributions.” Economics 5 (2011–​20): 1.
doi: 10.5018/​economics-​ejournal.ja.2011-​20.
Gabaix, Xavier. 1999. “Zipf ’s law for cities:  An explanation.” Quarterly Journal of Economics
114: 739–​67.
Gabaix, Xavier. 2009. “Power laws in economics and finance.” Annual Review of Economics 1: 255–​93.
Gabaix, Xavier, Parameswaran Gopikrishnan, Vasiliki Plerou, and H. Eugene Stanley. 2000.
“Statistical properties of share volume traded in financial markets.” Physical Review E 62
(4): R4493–​R4496.
Gabaix, Xavier, Parameswaran Gopikrishnan, Vasiliki Plerou, and H. Eugene Stanley. 2003. “A
theory of power law distributions in financial market fluctuations.” Nature 423: 267–​70.
Gabaix, Xavier, Parameswaran Gopikrishnan, Vasiliki Plerou, and H. Eugene Stanley. 2006.
“Institutional investors and stock market volatility.” Quarterly Journal of Economics 121:
461–​504.
Gabaix, Xavier, Parameswaran Gopikrishnan, Vasiliki Plerou, and H. Eugene Stanley. 2007.
“A unified econophysics explanation for the power-​law exponents of stock market activity.”
Physica A 382: 81–​88.
 

197  References

Gabaix, Xavier, and Rustam Ibragimov. 2011. “Rank—​1 /​2:  A  simple way to improve
the OLS estimation of tail exponents.” Journal of Business and Economic Statistics 29
(1): 24–​39.
Gabaix, Xavier, and A. Landier. 2008. “Why has CEO pay increased so much?” Quarterly Journal
of Economics 123: 49–​100.
Galam, Serge. 2004. “Sociophysics: A personal testimony.” Physica A 336 (2): 49–​55.
Galam, Serge. 2008. “Sociophysics: A review of Galam models.” arXiv:0803.1800v1.
Galam, Serge, Yuval Gefen, and Yonathan Shapir. 1982. “Sociophysics: A mean behavior model
for the process of strike.” Journal of Mathematical Sociology 9 (2): 1–​13.
Galison, Peter. 1997. Image and logic: A Material Culture of Microphysics. Chicago: University of
Chicago Press.
Gallais-​Hamonno, Georges, ed. 2007. Le marché financier français au XIXe siècle. Vol. 2.
Paris: Publications de la Sorbonne.
Gallegati, Mauro. 1990. “Financial instability, income distribution and the stock market.” Journal
of Post Keynesian Economics 12 (3): 356–​74.
Gallegati, Mauro. 1994. “Composition effect and economic fluctuations.” Economics Letters 44
(1–​2): 123–​26.
Gallegati, Mauro, and Marco Dardi. 1992. “Alfred Marshall on speculation.” History of Political
Economy 24 (3): 571–​94.
Gallegati, Mauro, Steve Keen, Thomas Lux, and Paul Ormerod. 2006. “Worrying trends in
econophysics.” Physica A 370 (1): 1–​6.
Garman, Mark. 1976. “Market microstructure.” Journal of Financial Economics 3: 257–​75.
Gaussel, Nicolas, and Jérôme Legras. 1999. “Black-​Scholes ... what’s next?” Quants 35: 1–​33.
Geman, Hélyette. 2002. “Pure jump Lévy processes for asset price modelling.” Journal of Banking
and Finance 26 (7): 1297–​316.
Geraskin, Petr, and Dean Fantazzini. 2011. “Everything you always wanted to know about log-​
periodic power laws for bubble modeling but were afraid to ask.” The European Journal of
Finance 19 (5): 1–​26.
Gibrat, Robert. 1931. Les inégalités économiques. Paris: Librairie du Recueil.
Gilbert, G. Nigel. 2007. Agent Based Models. London: Sage.
Gilbert, G. Nigel, and Michael Mulkay. 1984. Opening Pandora’s Box. New  York:  Cambridge
University Press.
Gillespie, Colin S. 2014. “A complete data framework for fitting power law distributions.”
arXiv:1408.1554.
Gillet, Philippe. 1999. L’efficience des marchés financiers. Paris: Economica.
Gingras, Yves, and Christophe Schinckus. 2012. “Institutionalization of econophysics in the
shadow of physics.” Journal of the History of Economic Thought 34 (1): 109–​30.
Gligor, Mircea, and Marcel Ausloos. 2007. “Cluster structure of EU-​15 countries derived from
the correlation matrix analysis of macroeconomic index fluctuations.” European Physical
Journal B 57 (2): 139–​46.
Gnedenko, Boris Vladimirovich, and Andrei N. Kolmogorov. 1954. Limit Distributions for Sums
of Independent Random Variables. Cambridge, MA: Addison-​Wesley.
Godfrey, Michael D., Clive W. J. Granger, and Oskar Morgenstern. 1964. “The random-​walk hy-
pothesis of stock market behavior.” Kyklos 17: 1–​29.
Goerlich, Francisco J. 2013. “A simple and efficient test for the Pareto law.” Empirical Economics
45 (3): 1367–​81.
 

198 References

Gopikrishnan, Parameswaran, Martin Meyer, Luis A. Nunes Amaral, and H. Eugene Stanley.
1998. “Inverse cubic law for the probability distribution of stock price variations.” European
Physical Journal B 3: 139–​40.
Gopikrishnan, Parameswaran, Vasiliki Plerou, Xavier Gabaix, and H. Eugene Stanley. 2000.
“Statistical properties of share volume traded in financial markets.” Physical Review E
62: 4493–​96.
Gopikrishnan, Parameswaran, Vasiliki Plerou, Yanhui Liu, Luis A. Nunes Amaral, Xavier
Gabaix, and H. Eugene Stanley. 2000. “Scaling and correlation in financial time series.”
Physica A:  Statistical Mechanics and its Applications 287 (3):  362–​ 73. doi:  10.1016/​
S0378-​4371(00)00375-​7.
Gopikrishnan, Parameswaran, Vasiliki Plerou, Luıs A. Nunes Amaral, Martin Meyer, and H.
Eugene Stanley. 1999. “Scaling of the distribution of fluctuations of financial market indices.”
Physics Review E 60 (5): 5305–​16.
Granger, Clive W.  J., and Oskar Morgenstern. 1963. “Spectral analysis of New  York Stock
Exchange prices.” Kyklos 16: 1–​27.
Gregory, Allan W., and Gregor W. Smith. 1995. “Business cycle theory and econometrics.”
Economic Journal 105 (433): 1597–​608.
Grossman, Sanford J. 1976. “On the efficiency of competitive stock markets where traders have
diverse information.” Journal of Finance 31 (2): 573–​85.
Grossman, Sanford J., and Joseph E. Stiglitz. 1980. “The impossibility of informationally effi-
cient markets.” American Economic Review 70 (3): 393–​407.
Guégan, Dominique, and Xin Zhao. 2014. “Alternative modeling for long term risk.” Quantitative
Finance 14 (12): 2237–​53.
Guillaume, Deffuant. 2004. “Comparing extremism propagation patterns in continuous opinion
models.” Journal of Artificial Societies and Social Simulation 9 (3): 8–​14.
Gunaratne, Gemunu H., and Joseph L. McCauley. 2005. “A theory for fluctuations in stock
prices and valuation of their options.” Proceedings of the SPIE 5848: 131.
Gupta, Hari, and José Campanha. 1999. “The gradually truncated Lévy flight for systems with
power-​law distributions.” Physica A 268: 231–​39.
Gupta, Hari, and José Campanha. 2002. “Tsallis statistics and gradually truncated Lévy
flight: Distribution of an economic index.” Physica A 309 (4): 381–​87.
Haavelmo, Trygve. 1944. “The probability approach in econometrics.” Econometrica 12
(Supplement): iii–​vi + 1–​115.
Hacking, Ian. 1990. The Taming of Chance. New York: Cambridge University Press.
Hadzibabic, Zoran, Peter Krüger, Marc Cheneau, Baptiste Battelier, and Jean Dalibard.
2006. “Berezinskii-​Kosterlitz-​Thouless crossover in a trapped atomic gas.” Nature
441: 1118–​21.
Hafner, Wolfgang, and Heinz Zimmermann, eds. 2009. Vinzenz Bronzin’s Option Pricing
Models: Exposition and Appraisal. Berlin: Springer.
Hammer, Howard M., and Ronald X. Groeber. 2007. “Efficient market hypothesis and class
action securities regulation.” International Journal of Business Research 1: 1–​14.
Hand, D. 1998. “Data mining: Statistics and more?” American Statistician 52 (2): 112–​18.
Hankins, Frank Hamilton. 1908. “Adolphe Quetelet as statistician.” Doctoral dissertation,
Columbia University.
Hansen, Lars Peter, and James J. Heckman. 1996. “The empirical foundations of calibration.”
Journal of Economic Perspectives 10 (1): 87–​104.
 

199  References

Harrison, J. Michael, and David M. Kreps. 1979. “Martingales and arbitrage in multiperiod secu-
rities markets.” Journal of Economic Theory 20 (3): 381–​408.
Harrison, J. Michael, and Stanley R. Pliska. 1981. “Martingales and stochastic integrals in the
theory of continuous trading.” Stochastic Processes and Their Applications 11 (3): 215–​60.
Harrison, Paul. 1998. “Similarities in the distribution of stock market price changes between the
eighteenth and twentieth centuries.” Journal of Business 71 (1): 55–​79. doi: 10.1086/​209736.
Hartemink, Alexander J., David K. Gifford, Tommi S. Jaakkola, and Richard A. Young. 2001.
“Maximum likelihood estimation of optimal scaling factors for expression array normaliza-
tion.” Proceedings of SPIE 4266: 132–​40.
Hastie, Trevor, Robert Tibshirani, Jerome Friedman, and James Franklin. 2005. “The element
of statistical learning: Data mining, inference and prediction.” Mathematical Intelligencer 27
(2): 83–​85.
Haug, Espen Gaarder, and Nassim Nicholas Taleb. 2011. “Option traders use very sophisticated
heuristics, never the Black-​Scholes-​Merton formula.” Journal of Economic Behaviour and
Organization 77 (2): 97–​106.
Hautcœur, Pierre-​Cyrille, and Angelo Riva. 2012. “The Paris financial market in the nineteenth
century: Complementarities and competition in microstructures.” Economic History Review
65 (4): 1326–​53.
Hendry, David F. 1995. “Econometrics and business cycle empirics.” Economic Journal 105
(433): 1622–​36.
Hendry, David F., and Mary S. Morgan, eds. 1995. The Foundations of Econometric Analysis.
New York: Cambridge University Press.
Hoover, Kevin D. 1995. “Facts and artifacts: Calibration and the empirical assessmen.” Oxford
Economic Papers 47 (1): 24.
Houthakker, Hendrik S. 1961. “Systematic and random elements in short-​term price move-
ments.” American Economic Review 51 (2): 164–​72.
Huber, Michel. 1937. “Quarante années de la statistique générale de la France 1896–​1936.”
Journal de la Société Statistique de Paris 78: 179–​214.
Hughes, Robert I. G. 1999. “The Ising model, computer simulation, and universal physics.” In
Models as Mediators: Perspectives on Natural and Social Science, edited by Mary S. Morgan and
Margaret Morrison, 97–​145. Cambridge: Cambridge University Press.
Hurst, Harold Edwin. 1951. “Long-​term storage capacity of reservoirs.” Transactions of the
American Society of Civil Engineers 116: 770–​99.
Ilachinski, Andrew 2000. “Irreducible semi-​ autonomous adaptive combat (ISAAC):  An
artificial-​life approach to land warfare.” Military Operations Research 5 (3): 29–​46.
Ingrao, Bruna, and Israel Giorgio. 1990. The Invisible Hand: Economic Equilibrium in the History
of Science. Cambridge, MA: MIT Press.
Israel, Giorgio. 1996. La mathématisation du réel:  Essai sur la modélisation mathématique.
Paris: Le Seuil.
Ivancevic, Vladimir G. 2010. “Adaptive-​wave alternative for the Black-​Scholes option pricing
model.” Cognitive Computation 2 (1): 17–​30.
Ivanov, Plamen Ch, Ainslie Yuen, Boris Podobnik, and Youngki Lee. 2004. “Common scaling
patterns in intertrade times of U. S. stocks.” Physical Review E: Statistical, Nonlinear, and Soft
Matter Physics 69 (5.2): 056107. doi: 10.1103/​PhysRevE.69.056107.
Izquierdo, Segismundo S., and Luis R. Izquierdo. 2007. “The impact of quality uncertainty without
asymmetric information on market efficiency.” Journal of Business Research 60 (8): 858–​67.
 

200 References

Jensen, Michael C., ed. 1972. Studies in the Theory of Capital Markets. New York: Praeger.
Jensen, Michael C. 1978. “Some anomalous evidence regarding market efficiency.” Journal of
Financial Economics 6 (2–​3): 95–​101.
Jiang, Zhi-​Qiang, Wei-​Xing Zhou, Didier Sornette, Ryan Woodard, Ken Bastiaensen, and Peter
Cauwels. 2010. “Bubble diagnosis and prediction of the 2005–​2007 and 2008–​2009 Chinese
stock market bubbles.” Journal of Economic Behavior and Organization 74 (3): 149–​62.
Johansen, Anders, and Didier Sornette. 2000. “The Nasdaq crash of April 2000: Yet another ex-
ample of log-​periodicity in a speculative bubble ending in a crash.” European Physical Journal
B 17 (2): 319–​28.
Johansen, Anders, Didier Sornette, and Olivier Ledoit. 1999. “Predicting financial crashes using
discrete scale invariance.” Journal of Risk 1: 5–​32.
Jona-​Lasinio, Giovanni. 2001. “Renormalization group and probability theory.” Physics Reports
352: 439–​58.
Jovanovic, Franck. 2000. “L’origine de la théorie financière:  Une réévaluation de l’apport de
Louis Bachelier.” Revue d’Economie Politique 110 (3): 395–​418.
Jovanovic, Franck. 2001. “Pourquoi l’hypothèse de marche aléatoire en théorie financière? Les
raisons historiques d’un choix éthique.” Revue d’Economie Financière 61: 203–​11.
Jovanovic, Franck. 2002. “Le modèle de marche aléatoire dans la théorie financière quantita-
tive.” Doctoral dissertation, University of Paris 1, Panthéon-​Sorbonne.
Jovanovic, Franck. 2004. “Eléments biographiques inédits sur Jules Regnault (1834–​1894),
inventeur du modèle de marche aléatoire pour représenter les variations boursières.” Revue
d’Histoire des Sciences Humaines 11: 215–​30.
Jovanovic, Franck. 2006a. “Economic instruments and theory in the construction of Henri
Lefèvre’s ‘science of the stock market’.” In Pioneers of Financial Economics, vol. 1, edited by
Geoffrey Poitras, 169–​90. Cheltenham: Edward Elgar.
Jovanovic, Franck. 2006b. “A nineteenth-​century random walk: Jules Regnault and the origins
of scientific financial economics.” In Pioneers of Financial Economics, vol. 1, edited by Geoffrey
Poitras, 191–​222. Cheltenham: Edward Elgar.
Jovanovic, Franck. 2006c. “Was there a ‘vernacular science of financial markets’ in France during
the nineteenth century? A  comment on Preda’s ‘Informative Prices, Rational Investors.’”
History of Political Economy 38 (3): 531–​46.
Jovanovic, Franck. 2008. “The construction of the canonical history of financial economics.”
History of Political Economy 40 (2): 213–​42.
Jovanovic, Franck, ed. 2009a. “L’institutionnalisation de l’économie financière: Perspectives his-
toriques.” Special issue of Revue d’Histoire des Sciences Humaines 20.
Jovanovic, Franck. 2009b. “Le modèle de marche aléatoire dans l’économie financière de 1863 à
1976.” Revue d’Histoire des Sciences Humaines 20: 81–​108.
Jovanovic, Franck. 2012. “Bachelier:  Not the forgotten forerunner he has been depicted as.”
European Journal for the History of Economic Thought 19 (3): 431–​451.
Jovanovic, Franck. 2016. “An introduction to the calculation of the chance:  Jules Regnault,
his book and its influence.” In The Calculation of Chance and Philosophy of the Stock Market,
edited by William Goetzmann, forthcoming. New Haven: Yale University Press.
Jovanovic, Franck, Stelios Andreadakis, and Christophe Schinckus. 2016. “Efficient market hy-
pothesis and fraud on the market theory: A new perspective for class actions.” Research in
International Business and Finance 38 (7): 177–​90.
 

201  References

Jovanovic, Franck, and Philippe Le Gall. 2001a. “Does God practice a random walk? The ‘finan-
cial physics’ of a 19th century forerunner, Jules Regnault.” European Journal for the History of
Economic Thought 8 (3): 323–​62.
Jovanovic, Franck, and Philippe Le Gall. 2001b. “March to numbers:  The statistical style of
Lucien March.” In History of Political Economy, annual supplement:  The Age of Economic
Measurement, edited by Judy L. Klein and Mary S. Morgan, 86–​100.
Jovanovic, Franck, and Christophe Schinckus. 2013. “Towards a transdisciplinary econophys-
ics.” Journal of Economic Methodology 20 (2): 164–​83.
Jovanovic, Franck, and Christophe Schinckus. 2016. “Breaking down the barriers between
econophysics and financial economics.” International Review of Financial Analysis. Online.
Kadanoff, Leo. 1966. “Scaling laws for Ising models near Tc.” Physics 2: 263–​72.
Kahneman, Daniel, and Amos Tversky. 1979. “Prospect theory: An analysis of decision under
risk.” Econometrica 47 (2): 263–​91.
Kaiser, David. 2012. “Booms, busts, and the world of ideas: Enrollment pressures and the chal-
lenge of specialization.” Osiris 27: 276–​302.
Kaizoji, Taisei 2006. “On Stock-​Price Fluctuations in the Periods of Booms and Stagnations.”
In Proceedings of the Econophysics-​Kolkata II Series, edited by Arnab Chatterjee and Bikas K.
Chakrabarti, 3–​12. Tokyo: Springer.
Kaizoji, Taisei, and Michiyo Kaizoji. 2004. “Power law for the calm-​time interval of price
changes.” Physica A 336 (3): 563–​70.
Keen, Steve. 2003. “Standing on the toes of pygmies: Why econophysics must be careful of the
economic foundations on which it builds.” Physica A 324 (1): 108–​16.
Kendall, Maurice George. 1953. “The analysis of economic time-​series. Part I: Prices.” Journal of
the Royal Statistical Society 116: 11–​25.
Kennedy, Robert E. 1966. “Financial analysts seminar.” Financial Analysts Journal 22 (6): 8–​9.
Kesten, Harry. 1973. “Random difference equations and renewal theory for products of random
matrices.” Acta Mathematica 131 (1): 207–​48.
Keynes, John Maynard. 1936. The General Theory of Employment, Interest and Money.
London: Macmillan.
Khinchine, A. Ya. 1933. Asymptotische Gesetze der Warsheinlichekeitsrechnung. Berlin: Springer.
Kim, Young, Svetlozar Rachev, Michele Bianchi, and Frank J. Fabozzi. 2008. “Financial market
models with Lévy processes and time-​varying volatility.” Journal of Banking and Finance 32
(7): 1363–​78.
King, Benjamin. 1964. “The latent statistical structure of security price changes.” Doctoral dis-
sertation, Graduate School of Business, University of Chicago.
Klass, Oren S., Ofer Biham, Moshe Levy, Ofer Malcai, and Sorin Solomon. 2006. “The Forbes
400 and the Pareto wealth distribution.” Economics Letters 90: 90–​95.
Kleiber, Max. 1932. “Body size and metabolism.” Hilgardia 6: 315–​51.
Klein, Julie Thompson. 1994. “Notes toward a social epistemology of transdisciplinarity.”
Paper presented to the First World Congress of Transdisciplinarity, Convento da Arrábida,
Portugal.
Knorr-​Cetina, Karin. 1981. The Manufacture of Knowledge: An Essay on the Constructivist and
Contextual Nature of Science. Elmsford, NY: Pergamon.
Kolmogorov, Andrei N. 1931. “Über die analytischen Methoden in der Wahrscheinlich-​
keitsrechnung.” Mathematische Annalen 104: 415–​58.
 

202 References

Kolmogorov, Andrei N. 1941. “Dissipation of energy in isotropic turbulence.” Doklady Akademii

Nauk SSSR 32: 19–​21.
Kolmogorov, Andrei N. 1942. “Equations of turbulent motion in an incompressible fluid.”
Izvestiya Akademii Nauk SSSR Seriya Fizika 6: 56–​58.
Koponen, Ismo. 1995. “Analytic approach to the problem of convergence of truncated Lévy
flights towards the Gaussian stochastic process.” Physical Review E 52 (1): 1197–​99.
Kou, Xiaodong, Lin Yang, and Lin Cai. 2008. “Artificial urban planning: Application of MAS.”
Urban Planning Education 1: 349–​53.
Koutrovelis, Ioannis. 1980. “Regression-​type estimation of the parameters of stable laws.”
Journal of American Statistical Association 75: 918–​28.
Kreps, David M. 1981. “Arbitrage and equilibrium in economies with infinitely many commodi-
ties.” Journal of Mathematical Economics 8: 15–​35.
Krugman, Paul. 1991. Geography and Trade. Cambridge, MA: MIT Press.
Krugman, Paul. 1996a. “Confronting the mystery of urban hierarchy.” Journal of the Japanese and
International Economies 10: 399–​418.
Krugman, Paul. 1996b. The Self-​Organizing Economy. Cambridge, MA: Blackwell.
Kuhn, Thomas. 1962. The Structure of Scientific Revolutions. Chicago: University of Chicago Press.
Kuhn, Thomas. 1989. “Possible worlds in history of science.” In Possible Worlds in Humanities,
Arts and Sciences, edited by S. Allen, 9–​32. New York: de Gruyter.
Kutner, Ryszard, and Dariusz Grech. 2008. “Report on foundation and organization of econo-
physics graduate courses at Faculty of Physics of University of Warsaw and Department of
Physics and Astronomy of the Wrocław University.” Acta Physica Polonica A 114 (3): 637–​47.
Kydland, Finn E., and Edward C. Prescott. 1982. “Time to build and aggregate fluctuations.”
Econometrica 50 (6): 1345–​70.
Kyle, Albert S. 1985. “Continuous auctions and insider trading.” Econometrica 53 (6): 1315–​35.
Lakatos, Imre. 1978. Philosophical Papers, vol. 1:  The Methodology of Scientific Research
Programmes. Cambridge: Cambridge University Press.
Larson, Arnold B. 1960. “Measurement of random process in futures prices.” Food Research
Institute Studies 1 (3): 313–​24.
Le Gall, Philippe. 1994. “Histoire de l’econométrie, 1914–​1944: L’erosion du déterminisme.”
Doctoral dissertation, University of Paris 1, Panthéon-​Sorbonne.
Le Gall, Philippe. 1999. “A world ruled by Venus: On Henry L. Moore’s transfer of periodogram
analysis from physics to economics.” History of Political Economy 31 (4): 723–​52.
Le Gall, Philippe. 2002. “Les représentations du monde et les pensées analogiques des
economètres:  Un siècle de modélisation en perspective.” Revue d’Histoire des Sciences
Humaines 6: 39–​64.
Le Gall, Philippe. 2006. From Nature to Models:  An Archaeology of Econometrics in FRANCE,
1830–​1930. London: Routledge.
LeBaron, Blake. 2001. “Stochastic volatility as a simple generator of power laws and long
memory.” Quantitative Finance 1 (6): 621–​31.
LeBaron, Blake. 2006. “Agent-​based computational finance.” In Handbook of Computational
Economics, vol. 2:  Agent-​Based Computational Economics, edited by Leigh Tesfatsion and
Kenneth L. Judd, 1187–​233. Amsterdam: North-​Holland.
Lesne, Annick. 1998. Renormalization Methods:  Critical Phenomena, Chaos, Fractal Structures.
Chichester: John Wiley & Sons.
 

203  References

Lesne, Annick, and Michel Laguës. 2011. Scale Invariance: From Phase Transitions to Turbulence.
Berlin: Springer.
Levy, M. 2003. “Are rich people smarter?” Journal of Economic Theory 110: 42–​64.
Levy, Moshe H., Haim Levy, and Sorin Solomon. 1995. “Microscopic simulation of the stock
market: The effect of microscopic diversity.” Journal de Physique I 5 (8): 1087–​107.
Levy, Moshe H., Haim Levy, and Sorin Solomon. 2000. Microscopic Simulation of Financial
Markets: From Investor Behaviour to Market Phenomena. San Diego: Academic Press.
Lévy, Paul. 1924. “Théorie des erreurs: La loi de Gauss et les lois exceptionnelles.” Bulletin de la
SMF 2 (1): 49–​85.
Li, Lun, David Alderson, John C. Doyle, and Walter Willinger. 2005. “Towards a theory of scale-​
free graphs: Definition, properties, and implications.” Internet Mathematics 2 (4): 431–​523.
Li, Sai-​Ping, and Shu-​Heng Chen, eds. 2012. “Complexity and non-​linearities in financial
markets: Perspectives from econophysics.” Special issue of International Review of Financial
Analysis 23.
Lillo, Fabrizio, and Rosario N. Mantegna. 2004. “Dynamics of a financial market index after a
crash.” Physica A 338 (1–​2): 125–​34.
Lintner, John. 1965a. “Security prices, risk and maximal gains from diversification.” Journal of
Finance 20 (4): 587–​615.
Lintner, John. 1965b. “The valuation of risk assets and the selection of risky investments in stock
portfolios and capital budgets.” Review of Economic Statistics 47 (1): 13–​37.
Lorie, James Hirsch. 1965. “Controversies on the stock market.” Selected Papers, Graduate
School of Business of the University of Chicago.
Lorie, James Hirsch. 1966. “Some comments on recent quantitative and formal research on the
stock market.” Journal of Business 39 (1.2: Supplement on Security Prices): 107–​10.
Lotka, Alfred J. 1926. “The frequency distribution of scientific productivity.” Journal of the
Washington Academy of Sciences 16 (12): 317–​24.
Louçã, Francisco. 2007. The Years of High Econometrics. A Short History of the Generation That
Reinvented Economics. London: Routledge.
Louzoun, Yoram, and Sorin Solomon. 2001. “Volatility driven markets in a generalized Lotka
Voltera formalism.” Physica A 302 (1): 220–​33.
Lowenstein, Roger. 2000. When Genius Failed:  The Rise and Fall of Long-​ Term Capital
Management. New York: Random House.
Lu, Zhiping, and Dominique Guegan. 2011. “Testing unit roots and long range dependence of
foreign exchange.” Journal of Time Series Analysis 32 (6): 631–​38.
Lucas, Robert E., Jr. 1978. “Asset Prices in an Exchange Economy.” Econometrica 46 (6): 1429–45.
Luttmer, E. 2007. “Selection, growth, and the size distribution of firms.” Quarterly Journal of
Economics 122: 1103–​44.
Lux, Thomas. 1992a. “A note on the stability of endogenous cycles in Diamond’s model of
search and barter.” Journal of Economics 56 (2): 185–​96.
Lux, Thomas. 1992b. “The sequential trading approach to disequilibrium dynamics.” Jahrbücher
für Nationaloekonomie und Statistik 209 (1): 47–​59.
Lux, Thomas. 1996. “The stable Paretian hypothesis and the frequency of large returns: An ex-
amination of major German stocks.” Applied Financial Economics 6: 463–​75.
Lux, Thomas. 2006. “Financial power laws:  Empirical evidence, models, and mechanism.”
Working paper No12 from Christian-​Albrechts-​University of Kiel, Department of Economics.
 

204 References

Lux, Thomas. 2009. “Applications of statistical physics in finance and economics.” In Handbook
of Research on Complexity, edited by Barkley Rosser, 213–​58. Cheltenham: Edward Elgar.
Lux, Thomas, and Michele Marchesi. 1999. “Scaling and criticality in a stochastic multi-​agent
model of a financial market.” Nature 397: 498–​500.
Lux, Thomas, and Michele Marchesi. 2000. “Volatility clustering in financial markets: A micro-
simulation of interacting agents.” International Journal of Theoretical and Applied Finance 3
(4): 675–​702.
Maas, Harro. 2005. William Stanley Jevons and the Making of Modern Economics.
New York: Cambridge University Press.
Macintosh, Norman B. 2003. “From rationality to hyperreality: Paradigm poker.” International
Review of Financial Analysis 12 (4): 453–​65.
Macintosh, Norman B., Teri Shearer, Daniel B. Thornton, and Michael Welker. 2000.
“Accounting as simulacrum and hyperreality:  Perspectives on income and capital.”
Accounting, Organizations and Society 25 (1): 13–​50.
MacKenzie, Donald A. 2003. “An equation and its worlds:  Bricolage, exemplars, disu-
nity and performativity in financial economics.” Paper presented to “Inside Financial
Markets: Knowledge and Interaction Patterns in Global Markets,” Konstanz, May.
MacKenzie, Donald A. 2006. An Engine, Not a Camera: How Financial Models Shape Markets.
Cambridge, MA: MIT Press.
MacKenzie, Donald A. 2007. “The emergence of option pricing theory.” In Pioneers of Financial
Economics, vol. 2:  Twentieth Century Contributions, edited by Geoffrey Poitras and Franck
Jovanovic, 170–​91. Cheltenham: Edward Elgar.
MacKenzie, Donald A., and Yuval Millo. 2009. “The usefulness of inaccurate models: Towards
an understanding of the emergence of financial risk management.” Accounting, Organizations
and Society 34: 638–​53.
Madan, Dilip B., Peter P. Carr, and Éric C. Chang. 1998. “The variance gamma process and
option pricing.” European Finance Review 2 (1): 79–​105.
Madhavan, Ananth. 2000. “Market microstructure:  A  survey.” Journal of Financial Markets 3
(3): 205–​58.
Malcai, Ofer, Ofer Biham, and Sorin Solomon. 1999. “Power-​law distributions and Lévy-​stable
intermittent fluctuations in stochastic systems of many autocatalytic elements.” Physical
Review E 60: 1299–​303.
Malevergne, Yannick, Vladilen Pisarenko, and Didier Sornette. 2011. “Testing the Pareto against
the lognormal distributions with the uniformly most powerful unbiased test applied to the
distribution of cities.” Physical Review E 83 (3): 036111.
Malevergne, Yannick, and Didier Sornette. 2005. Extreme Financial Risks: From Dependence to
Risk Management. Vol. 1. Berlin: Springer.
Malevergne, Yannick, Didier Sornette, and Vladilen Pisarenko. 2005. “Empirical distributions of
stock returns: Between the stretched exponential and the power law?” Quantitative Finance
5 (4): 379–​401.
Malkiel, Burton G. 1992. “Efficient market hypothesis.” In The New Palgrave Dictionary of Money
and Finance, edited by Peter Newman, Murray Milgate and John Eatwell. London: Macmillan.
Mandelbrot, Benoit. 1957. “Application of thermodynamical methods in communication
theory and in econometrics.” Institut Mathématique de l’Université de Lille.
Mandelbrot, Benoit. 1960. “The Pareto-​Lévy law and the distribution of income.” International
Economic Review 1: 79–​106.
 

205  References

Mandelbrot, Benoit. 1962a. “Sur certains prix spéculatifs: Faits empiriques et modèle basé sur
les processus stables additifs non gaussiens de Paul Lévy.” Comptes Rendus de l’Académie des
Sciences (Meeting of June 4): 3968–​70.
Mandelbrot, Benoit. 1962b. “The variation of certain speculative prices.” IBM Report NC-​87.
Mandelbrot, Benoit. 1963a. “New Methods in Statistical Economics.” Journal of Political
Economy 71 (5): 421–​40.
Mandelbrot, Benoit. 1963b. “The variation of certain speculative prices.” Journal of Business 36
(4): 394–​419.
Mandelbrot, Benoit. 1966a. “Forecasts of future prices, unbiased markets, and “martingale”
models.” Journal of Business 39 (1.2): 242–​55.
Mandelbrot, Benoit. 1966b. “Seminar on the Analysis of Security Prices” held November 12-​13,
1966 at the Graduate School of Business of the University of Chicago.
Mandelbrot, Benoit. 1997. Fractals, hasard et finance. Paris: Flammarion.
Mandelbrot, Benoit, and Richard L. Hudson. 2004. The (Mis)Behavior of Markets:  A  Fractal
View of Risk, Ruin, and Reward. London: Profile Books.
Mantegna, Rosario N. 1991. “Levy Walks and enhanced diffusion in Milan stock exchange.”
Physica A 179 (1): 232–​42.
Mantegna, Rosario N., and H. Eugene Stanley. 1994. “Stochastic process with ultra-​slow con-
vergence to a Gaussian: The truncated Lévy flight.” Physical Review Letters 73 (22): 2946–​49.
Mantegna, Rosario N., and H. Eugene Stanley. 2000. An Introduction to Econophysics: Correlations
and Complexity in Finance. New York: Cambridge University Press.
March, Lucien. 1921. “Les modes de mesure du mouvement général des prix.” Metron 1
(4): 57–​91.
Mardia, Kanti V., and Peter E. Jupp. 2000. Directional Statistics. Chichester: Wiley.
Mariani, Maria Cristina, and Y. Liu. 2007. “Normalized truncated Lévy walks applied to the
study of financial indices.” Physica A 377: 590–​98.
Markowitz, Harry M. 1952. “Portfolio selection.” Journal of Finance 7 (1): 77–​91.
Markowitz, Harry M. 1955. “Portfolio selection.” MA thesis, University of Chicago.
Markowitz, Harry M. 1959. Portfolio Selection:  Efficient Diversification of Investments.
New York: Wiley.
Markowitz, Harry M. 1999. “The early history of portfolio theory:  1600–​1960.” Financial
Analysts Journal 55 (4): 5–​16.
Marsili, Mateo, and Yi-​Chen Zhang. 1998. “Econophysics: What can physicists contribute to
economics?” Physics Review Letters 80 (1): 80–​85.
Maslov, Sergei. 2000. “Simple model of a limit order-​driven market.” Physica A 278 (3–​4): 571–​78.
Matacz, Andrew. 2000. “Financial modeling and option theory with the truncated Levy pro-
cess.” International Journal of Theoretical and Applied Finance 3 (1): 142–​60.
Matei, Marius, Xari Rovira, and Nuria Agell. 2012. “Bivariate volatility modeling for stocks and
portfolios.” Working paper, ESADE Business School, Ramon Llull University.
Matsushita, Raul, Pushpa Rathie, and Sergio Da Silva. 2003. “Exponentially damped Lévy
flight.” Physica A 326 (1): 544–​55.
Max-​Neef, Manfred. 2005. “Foundations of transdisciplinarity.” Ecological Economics 53: 5–​16.
McCauley, Joseph L. 2003. “Thermodynamics analogies in economics and finance: Instability of
markets.” Physica A 329: 199–​212.
McCauley, Joseph L. 2004. Dynamics of Markets: Econophysics and Finance. Cambridge: Cambridge
University Press.
 

206 References

McCauley, Joseph L. 2006. “Response to ‘Worrying Trends in Econophysics.’” Physica A 371

(1): 601–​9.
McCauley, Joseph L. 2009. “ARCH and GARCH models vs. martingale volatility of finance
market returns.” International Review of Financial Analysis 18 (4): 151–​53.
McCauley, Joseph L., and Gemunu H. Gunaratne. 2003. “An empirical model of volatility of
returns and option pricing.” Physica A 329: 178–​98.
McCauley, Joseph L., Gemunu H. Gunaratne, and Kevin E. Bassler. 2007a. “Hurst exponents,
Markov processes, and fractional Brownian motion.” Physica A: Statistical Mechanics and Its
Applications 379 (1): 1–​9.
McCauley, Joseph L., Gemunu H. Gunaratne, and Kevin E. Bassler. 2007b. “Martingale option
pricing.” Physica A 380: 351–​56.
McCauley, Joseph L., Bertrand M. Roehner, H. Eugene Stanley, and Christophe Schinckus, eds.
2016 (forthcoming). “Econophysics, where we are where we go.” Special issue of International
Review of Financial Analysis.
McCulloch, Hu. 1986. “Simple consistent estimators of stable distribution parameters.”
Communications in Statistics: Simulation and Computation 15 (4): 1109–​36.
McGoun, Elton G. 1997. “Hyperreal finance.” Critical Perspectives on Accounting 8 (1–​2): 97–​122.
McNees, Stephen. 1979. “The forecasting record for the 1970s.” New England Economic Review,
September–​October, 33–​53.
Mehrling, Perry. 2005. Fischer Black and the Revolutionary Idea of Finance. Hoboken, NJ: John
Wiley & Sons.
Ménard, Claude. 1981. “La machine et le coeur: Essai sur les analogies dans le raisonnement
économique.” In Analogie et connaissance, edited by André Lichnerowicz, François Perroux,
and Gilbert Gadoffre, 137–​61. Paris: Éditions Maloine.
Ménard, Claude. 1987. “Why was there no probabilistic revolution in economic thought?”
In The Probabilistic Revolution, vol. 2:  Ideas in the Sciences, edited by Lorenz Krüger, Gerd
Gigerenzer, and Mary S. Morgan, 139–​46. Cambridge, MA: MIT Press.
Merton, Robert C. 1976. “Option pricing when underlying stock returns are discontinuous.”
Journal of Financial Economics 3 (1-​2): 125–​44.
Merton, Robert C. 1998. “Applications of option-​pricing theory:  Twenty-​five years later.”
American Economic Review 88 (3): 323–​49.
Meyer, Paul-​André. 2009. “Stochastic processes from 1950 to the present.” Journal Electronique
d’Histoire des Probabilités et de la Statistique 5 (1): 1–​42.
Miburn, J. Alex. 2008. “The Relationship between fair value, market value, and efficient mar-
kets.” Accounting Perspectives 7 (4): 293–​316.
Michael, Fredrick, and M. D. Johnson. 2003. “Financial market dynamics.” Physica A
320: 525–​34.
Mike, Szabolcs, and J. Doyne Farmer. 2008. “An empirical behavioral model of liquidity and
volatility.” Journal of Economic Dynamics and Control 32 (1):  200–​234. doi:  10.1016/​
j.jedc.2007.01.025.
Miller, Merton H. 1988. “The Modigliani-​Miller propositions after thirty years.” Journal of
Economic Perspectives 2 (4): 99–​120.
Millo, Yuval, and Christophe Schinckus. 2016. “A nuanced perspective on episteme and techne in
finance.” International Review of Financial Analysis 46: 124–​30.
Mills, Frederick Cecil. 1927. The Behavior of Prices. New York: National Bureau of Economic
Research.
 

207  References

Mir, Tariq Ahmad, Marcel Ausloos, and Roy Cerqueti. 2014. “Benford’s law predicted digit dis-
tribution of aggregated income taxes: The surprising conformity of Italian cities and regions.”
European Physical Journal B 87 (11): 1–​8.
Mirowski, Philip. 1989a. “The measurement without theory controversy.” Oeconomia 11: 65–​87.
Mirowski, Philip. 1989b. More Heat Than Light: Economics as Social Physics, Physics as Nature’s
Economics. New York: Cambridge University Press.
Mirowski, Philip. 2002. Machine Dreams. Cambridge: Cambridge University Press.
Miskiewicz, Janusz. 2010. “Econophysics in Poland.” Science and Culture 76: 395–​98.
Mitchell, Melanie. 2009. Complexity: A Guided Tour. New York: Oxford University Press.
Mitchell, Wesley C. 1915. “The making and using of index numbers.” Bulletin of the United States
Bureau of Labor Statistics 173: 5–​114.
Mitzenmacher, Michael. 2004. “A brief history of generative models for power law and lognor-
mal distributions.” Internet Mathematics 1 (2): 226–​51.
Mitzenmacher, Michael. 2005. “Editorial: The future of power law research.” Internet Mathematics
2 (4): 525–​34.
Miyahara, Yoshio. 2012. Option Pricing in Incomplete Markets: Modeling Based on Geometric Lévy
Processes and Minimal Entropy Martingale Measures. London: Imperial College Press.
Modigliani, Franco, and Merton H. Miller. 1958. “The cost of capital, corporation finance and
the theory of investment.” American Economic Review 48 (3): 261–​97.
Moore, Arnold B. 1964. “Some characteristics of changes in common stock prices.” In The
Random Character of Stock Market Prices, edited by Paul H. Cootner, 139–​61. Cambridge,
MA: MIT Press.
Moore, Henry Ludwell. 1917. Forecasting the Yield and the Price of Cotton. New  York:
Macmillan.
Morales, Raffaello, Tiziana Di Matteo, and Tomaso Aste. 2013. “Non-​stationary multifractal-
ity in stock returns.” Physica A:  Statistical Mechanics and Its Applications 392 (24):  6470.
doi: 10.1016/​j.physa.2013.08.037.
Morgan, Mary S. 1990. The History of Econometric Ideas:  Historical Perspectives on Modern
Economics. Cambridge: Cambridge University Press.
Morgan, Mary S., and Judy L. Klein, eds. 2001. The Age of Economic Measurement. Durham,
NC: Duke University Press.
Morgan, Mary S., and Tarja Knuuttila. 2012. “Models and modelling in economics.” In Handbook
of the Philosophy of Science, vol. 13: Philosophy of Economics, edited by Uskali Mäki, 49–​87.
Amsterdam: Elsevier.
Morin, Edgar. 1994. “Sur l’interdisciplinarité.” Paper presented to the First World Congress of
Transdisciplinarity, Convento da Arrábida, Portugal.
Mossin, Jan. 1966. “Equilibrium in a Capital Asset Market.” Econometrica 34 (4): 768–​83.
Nadeau, Robert. 2003. “Thomas Kuhn ou l’apogée de la philosophie historique des sciences.” In
Actes du colloque du Centre Culturel International de Cerisy-​la-​Salle sur “Cent ans de philosophie
américaine”, edited by Jean-​Pierre Cometti and Claudine Tiercelin, 273–​97. Pau: P.U. Pau et
Pays de l’Adour –​ Quad.
Nakao, Hiroya. 2000. “Multi-​scaling properties of truncated Levy flights.” Physics Letters A
266: 282–​89.
Newman, Mark. 2005. “Power laws, Pareto distributions and Zipf ’s law.” Contemporary Physics
46 (5): 323–​51.
Niederhoffer, Victor. 1965. “Clustering of stock prices.” Operations Research 13 (2): 258–​65.
 

208 References

Nolan, John P. 2005. “Modeling financial data with stable distributions.” Working paper,
American University.
Nolan, John P. 2009. “Stable distributions:  Models for heavy tailed data.” Working paper,
American University.
O’ Hara, Maureen Patricia. 1995. Market Microstructure Theory. Cambridge, MA: Blackwell.
O’ Hara, Maureen Patricia. 2003. “Presidential address: Liquidity and price discovery.” Journal
of Finance 58 (3): 1335–​54.
Officer, Robert Rupert. 1972. “The distribution of stock returns.” Journal of American Statistical
Association 340 (67): 807–​12.
Okuyama, K., Misako Takayasu, and Hideki Takayasu. 1999. “Zipf ’s law in income distribution
of companies.” Physica A 269 (1): 125–​31.
Olivier, Maurice. 1926. Les nombres indices de la variation des prix. Paris: Marcel Giard.
Orléan, André. 1989. “Mimetic contagion and speculative bubbles.” Theory and Decision 27
(1–​2): 63–​92.
Orléan, André. 1995. “Bayesian interactions and collective dynamics of opinion: Herd behavior
and mimetic contagion.” Journal of Economic Behavior and Organization 28 (2): 257–​74.
Osborne, Maury F. M. 1959a. “Brownian motion in the stock market.” Operations Research 7
(2): 145–​73.
Osborne, Maury F.  M. 1959b. “Reply to ‘Comments on “Brownian Motion in the Stock
Market.’ ” Operations Research 7 (6): 807–​11.
Osborne, Maury F.  M. 1962. “Periodic structure in the Brownian motion of stock prices.”
Operations Research 10 (3): 345–​79.
Pagan, Adrian R. 1996. “The econometrics of financial markets.” Journal of Empirical Finance 3
(1): 15–​102.
Pandey, Ras B., and Dietrich Stauffer. 2000. “Search for log-​periodic oscillations in stock market
simulations.” International Journal of Theoretical and Applied Finance 3 (3): 479–​82.
Pareto, Vilfredo. 1897. Cours d’economie politique. Geneva: Librairie Droz.
Paul, Wolfgang, and Jörg Baschnagel. 2013. Stochastic Processes:  From Physics to Finance.
Berlin: Springer.
Paulson, Alan, Edgar Holcomb, and Robert Leitch. 1975. “The estimation of the parameters of
the stable laws.” Biometrika 62: 163–​70.
Pearson, Karl. 1905a. “The problem of the random walk.” Nature 72 (1865): 294.
Pearson, Karl. 1905b. “The problem of the random walk (answer).” Nature 72 (1867): 342.
Pickhardt, Michael, and Goetz Seibold. 2011. “Income tax evasion dynamics: Evidence from an
agent-​based econophysics model.” Working paper, University of Cottbus.
Pieters, Riek, and Baumgartner Hans. 2002. “Who talks to whom? Intra-​and interdisciplinary
communication of economics journals.” Journal of Economic Literature 40 (2): 483–​509.
Plerou, Vasiliki, Parameswaran Gopikrishnan, L. A. Nunes Amaral, Martin Meyer, and H.
Eugene Stanley. 1999. “Scaling of the distribution of price fluctuations of individual compa-
nies.” Physical Review E: Statistical Physics, Plasmas, Fluids, and Related Interdisciplinary Topics
60 (6.A): 6519–​29.
Plerou, Vasiliki, and H. Eugene Stanley. 2008. “Stock return distributions: Tests of scaling and
universality from three distinct stock markets.” Physical Review E: Statistical, Nonlinear, and
Soft Matter Physics 77 (3.2): 037101.
Poitras, Geoffrey. 2009. “From Antwerp to Chicago: The history of exchange traded derivative
security contracts.” Revue d’Histoire des Sciences Humaines 20: 11–​50.
 

209  References

Poitras, Geoffrey, and Franck Jovanovic, eds. 2007. Pioneers of Financial Economics, vol.
2: Twentieth-​Century Contributions. Northampton, MA: Edward Elgar.
Poitras, Geoffrey, and Franck Jovanovic. 2010. “Pioneers of Financial Economics:  Das Adam
Smith Irrelevanzproblem?” History of Economics Review 51 (Winter): 43–​64.
Ponzi, Adam, and Yoji Aizawa. 2000. “Evolutionary financial market models.” Physica A: 507–​23.
Popper, Karl Raimund. 1959. The Logic of Scientific Discovery. New York: Basic Books.
Porter, Theodore M. 1986. The Rise of Statistical Thinking, 1820–​1900. Princeton, NJ: Princeton
University Press.
Potters, Marc, and Jean-​Philippe Bouchaud. 2003. “More statistical properties of stock order
books and price impact.” Physica A 324: 133–​40.
Preda, Alex. 2001. “The rise of the popular investor:  Financial knowledge and investing in
England and France, 1840–​1880.” Sociological Quarterly 42 (2): 205–​32.
Preda, Alex. 2004. “Informative prices, rational investors: The emergence of the random walk
hypothesis and the 19th century ‘science of financial investments.’” History of Political
Economy 36 (2): 351–​86.
Preda, Alex. 2006. “Socio-​technical agency in financial markets: The case of the stock ticker.”
Social Studies of Science 36 (5): 753–​82.
Preda, Alex. 2007. “Where do analysts come from? The case of financial chartism.” Sociological
Review 55 (s2): 40–​64.
Preis, Tobias, and H. Eugene Stanley. 2010. “Trend switching processes in financial markets.” In
Econophysics Approaches to Large-​Scale Business Data and Financial Crisis, edited by Misako
Takayasu, Tsutomu Watanabe, and Hideki Takayasu, 3–​26. Tokyo: Springer Japan.
Press, S. James. 1967. “A compound events model for security prices.” Journal of Business 40
(3): 317–​35.
Press, S. James. 1972. “Estimation in univariate and multivariate stable distributions.” Journal of
American Statistical Association 67: 842–​46.
Prietula, Michael, Kathleen Carley, and Les Gasse, eds. 1998. Simulating
Organizations: Computational Models of Institutions and Groups. Cambridge, MA: MIT Press.
Quah, Danny T. 1995. “Business cycle empirics: Calibration and estimation.” Economic Journal
105 (433): 1594–​96.
Queiros, Silvio M. Duarte, and Constantino Tsallis. 2005. “Bridging the ARCH model for fi-
nance and nonextensive entropy.” Europhysics Letters 69:  893–​99. doi:  10.1209/​epl/​
i2004-​10436-​6.
Quételet, Adolphe. 1848. Du système social et des lois qui le régissent. Paris: Guillaumin et Cie.
Rachev, Svetlozar T., John S.  J. Hsu, Biliana S. Bagasheva, and Frank J. Fabozzi. 2008.
“Bayesian methods in finance.” In The Oxford Handbook of Bayesian Econometrics, edited
by John Geweke, Gary Koop, and Herman van Dijk, 439–​512. New  York:  Oxford
University Press.
Rachev, Svetlozar T., Young Shin Kim, Michele L. Bianchi, and Frank J. Fabozzi. 2011. Financial
Models with Levy Processes and Volatility Clustering. Hoboken, NJ: John Wiley & Sons.
Redelico, Francisco O., Araceli N. Proto, and Marcel Ausloos. 2009. “Hierarchical structures
in the gross domestic product per capita fluctuation in Latin American countries.” Physica
A: Statistical Mechanics and its Applications 388 (17): 3527–​35.
Reed, William J., and Barry D. Hughes. 2002. “From gene families and genera to incomes
and internet file sizes:  Why power laws are so common in nature.” Physical Review E
66: 067103-​1–​067103-​4.
 

210 References

Regnault, Jules. 1863. Calcul des chances et philosophie de la bourse. Paris: Mallet-​Bachelier and
Castel.
Renfro, Charles G. 2004. “The early development of econometric modeling languages.” In
Computational Econometrics: Its Impact on the Development of Quantitative Economics, edited
by Charles G. Renfro, 145–​66. Amsterdam: IOS Press.
Renfro, Charles G. 2009. The Practice of Econometric Theory: An Examination of the Characteristics
of Econometric Computation. Heidelberg: Springer.
Richmond, P. 2001. “Power law distributions and dynamic behaviour of stock markets.”
European Physical Journal B 20 (4): 523–​26.
Richmond, Peter, Jurgen Mimkes, and Stefan Hutzler. 2013. Econophysics and Physical Economics.
Oxford: Oxford University Press.
Rickles, Dean. 2007. “Econophysics for philosophers.” Studies in History and Philosophy of
Science 38 (4): 948–​78.
Rickles, Dean. 2008. “Econophysics and the complexity of the financial markets.” In Handbook
of the Philosophy of Science, vol. 10: Philosophy of Complex Systems, edited by John Collier and
Cliff Hooker, 531–​65. New York: North Holland Elsevier Editions.
Rimmer, Robert H., and John P. Nolan. 2005. “Stable distributions in Mathematica.” Mathematica
Journal 9 (4): 776–​89.
Roberts, Harry V. 1959. “Stock-​market ‘patterns’ and financial analysis: Methodological sugges-
tions.” Journal of Finance 14 (1): 1–​10.
Robert, Shiller. 1981. “Do Stock Prices Move Too Much to be Justified by Subsequent Changes
in Dividends.” American Economic Review 3: 421–36.
Roehner, Bertrand M. 2002. Patterns of Speculation:  A  Study in Observational Econophysics.
Cambridge: Cambridge University Press.
Rosenfeld, Lawrence. 1957. “Electronic computers and their place in securities analyses.”
Analysts Journal 13 (1): 51–​53.
Rosenfield, Patricia L. 1992. “The potential of transdisciplinary research for sustaining and
extending linkages between the health and social sciences.” Social Science and Medicine 35
(11): 1343–​57.
Ross, Stephen A. 1976a. “The arbitrage theory of capital asset pricing.” Journal of Economic
Theory 13 (3): 341–​60.
Ross, Stephen A. 1976b. “Options and efficiency.” Quarterly Journal of Economics 90 (1): 75–​89.
Ross, Stephen A. 1977. “Return, risk and arbitrage.” In Risk and Return in Finance, edited by
Irwin Friend and James L. Bicksler, 189–​217. Cambridge: Ballinger.
Rosser, Barkley. 2006. “The nature and future of econophysics.” In Econophysics of Stock and Other
Markets, edited by Arnab Chatterjee and Bikas K. Chakrabarti, 225–​34. Milan: Springer.
Rosser, Barkley. 2008. “Econophysics and economics complexity.” Advances in Complex Systems
11 (5): 745–​60.
Rosser, Barkley. 2010. “Is a transdisciplinary perspective on economic complexity possible?”
Journal of Economic Behavior and Organization 75 (1): 3–​11.
Roy, A. D. 1952. “Safety first and the holding of assets.” Econometrica 20 (3): 431–​49.
Rozenfeld, Hernán D., Diego Rybski, Xavier Gabaix, and Hernán A. Makse. 2011. “The Area
and Population of Cities:  New Insights from a Different Perspective on Cities.” American
Economic Review 101 (5): 2205–​25.
Rubinstein, Mark. 2002. “Markowitz’s ‘Portfolio Selection’: A fifty-​year retrospective.” Journal of
Finance 62 (3): 1041–​45.
 

211  References

Rubinstein, Mark. 2003. “Great moments in financial economics:  II. Modigliani-​Miller the-
orem.” Journal of Investment Management 1 (2): 7–​13.
Rutterford, Janette, and Dimitris Sotiropoulos. 2015. “Like the horses in a race’: Financial diver-
sification before modern portfolio theory.” Paper presented to the 19th Annual Conference
of the European Society for the History of Economic Thought (ESHET), Rome.
Rybski, Diego. 2013. “Auerbach’s legacy.” Environment and Planning A 45 (6): 1266–​68.
Samorodnitsky, Gennady, and Murad Taqqu. 1994. Stable Non-​Gaussian Random Processes.
New York: Chapman and Hall.
Samuelson, Paul A. 1965a. “Proof that properly anticipated prices fluctuate randomly.” Industrial
Management Review 6 (2): 41–​49.
Samuelson, Paul A. 1965b. “Rational theory of warrant pricing.” Industrial Management Review
6 (2): 13–​40.
Samuelson, Paul A. 1967. “Efficient portfolio selection for Pareto-​Lévy Investments.” Journal of
Financial and Quantitative Analysis 2 (2): 107–​22.
Savoiu, Gheorghe. 2013. Econophysics: Background and Applications in Economics, Finance, and
Sociophysics. Boston: Academic Press.
Săvoiu, Gheorghe G., and Ion Iorga-​ Simăn. 2013. “Sociophysics:  A  new science or a
new domain for physicists in a modern university?” In Econophysics:  Background and
Applications in Economics, Finance, and Sociophysics, edited by Gheorghe G. Săvoiu, 149–​66.
Oxford: Academic Press.
Scalas, Enrico, and Kyungsik Kim. 2007. “The art of fitting financial time series with Lévy stable
distributions.” Journal of the Korean Physical Society 50 (1): 105–​11.
Schabas, Margaret. 1990. A World Ruled by Number:  William Stanley Jevons and the Rise of
Mathematical Economics. Princeton, NJ: Princeton University Press.
Schaden, Martin. 2002. “Quantum finance.” Physica A 316: 511–​38.
Schinckus, Christophe. 2008. “The financial simulacrum.” Journal of Socio-​ economics 73
(3): 1076–​89.
Schinckus, Christophe. 2009a. “La diversification théorique en finance de marché:  Vers
de nouvelles perspectives de l’incertitude.” Doctoral dissertation, University of Paris 1,
Panthéon-​Sorbonne.
Schinckus, Christophe. 2009b. “La finance comportementale ou le développement d’un nou-
veau paradigme.” Revue d’Histoire des Sciences Humaines 20: 131–​57.
Schinckus, Christophe. 2010a. “Econophysics and economics:  Sister disciplines?” American
Journal of Physics 78 (1): 325–​27.
Schinckus, Christophe. 2010b. “Is econophysics a new discipline? A neo-​positivist argument.”
Physica A 389: 3814–​21.
Schinckus, Christophe. 2010c. “A reply to comment on econophysics and economics: Sister dis-
ciplines?” American Journal of Physics 78 (8): 789–​91.
Schinckus, Christophe. 2012a. “Financial economics and non-​representative art.” Journal of
Interdisciplinary Economics 24 (1): 77–​97.
Schinckus, Christophe. 2012b. “Statistical econophysics and agent-​ based econophysics.”
Quantitative Finance 12 (8): 1189–​92.
Schinckus, Christophe. 2013. “How do econophysicists make stable Levy processes physically
plausible.” Brazilian Journal of Physics 43 (4): 281–​93.
Schinckus, Christophe. 2017. “Econophysics and complexity studies.” Doctoral dissertation,
University of Cambridge.
 

212 References

Schoutens, Wim. 2003. Lévy Processes in Finance. New York: Wiley Finance.

Schwert, G. William. 2003. “Anomalies and market efficiency.” In Handbook of the Economics of
Finance, edited by George M. Constantinides, Milton Harris, and René M. Stulz, 939–​74.
Boston: Elsevier.
Seemann, Lars, Joseph L. McCauley, and Gemunu H. Gunaratne. 2011. “Intraday volatility and
scaling in high frequency foreign exchange markets.” International Review of Financial Analysis
20 (3): 121–​26. doi: 10.1016/​j.irfa.2011.02.008.
Shafer, Glenn, and Vladimir Vovk. 2001. Probability and Finance:  It’s Only a Game!
New York: Wiley.
Shafer, Glenn, and Vladimir Vovk. 2006. “The Sources of Kolmogorov’s Grundbegriffe.”
Statistical Science 21: 70–​98.
Sharpe, William F. 1963. “A simplified model for portfolio analysis.” Management Science
9: 277–​93.
Sharpe, William F. 1964. “Capital asset prices: A theory of market equilibrium under conditions
of risk.” Journal of Finance 19 (3): 425–​42.
Shefrin, Hersh. 2002. Beyond Greed and Fear: Understanding Behavioral Finance and the Psychology
of Investing. New York: Oxford University Press.
Shinohara, Shuji, and Yukio Pegio Gunji. 2001. “Emergence and collapse of money through rec-
iprocity.” Physica A: Applied Mathematics and Computation 117 (1): 131–​50.
Shleifer, Andrei, and Lawrence H. Summers. 1990. “The noise trader approach to finance.”
Journal of Economic Perspectives 4 (2): 19–​33.
Shlesinger, Michael F. 1995. “Comment on ‘Stochastic Process with Ultraslow Convergence to a
Gaussian: The Truncated Lévy Flight.’” Physical Review Letters 74 (24): 4959.
Simkin, Mikhail V., and Vwani P. Roychowdhury. 2011. “Re-​inventing Willis.” Physics Reports
502: 1–​35.
Simon, Herbert A. 1955. “On a class of skew distribution functions.” Biometrika 42
(3–​4): 425–​40.
Sims, Christopher A. 1980a. “Comparison of interwar and postwar business cycles: Monetarism
reconsidered.” American Economic Review 70 (2): 250.
Sims, Christopher A. 1980b. “Macroeconomics and reality.” Econometrica 48 (1): 1–​48.
Sims, Christopher A. 1996. “Macroeconomics and methodology.” Journal of Economic
Perspectives 10 (1): 105–​20.
Skjeltorp, Johannes Atle. 2000. “Scaling in the Norwegian stock market.” Physica A 283
(3): 486–​528.
Slanina, Frantisek. 2014. Essentials of Econophysics Modelling. Oxford:  Oxford University
Press.
Slanina, František. 2001. “Mean-​field approximation for a limit order driven market model.”
Physical Review E 64 (5): 056136-​1–​056136-​5.
Slutsky, Eugen. 1937. “The summation of random causes as the source of cyclic processes.”
Econometrica 5 (2): 105–​46.
Sornette, Didier. 2003. Why Stock Markets Crash: Critical Events in Complex Financial Systems.
Princeton, NJ: Princeton University Press.
Sornette, Didier. 2006. Critical Phenomena in Natural Sciences: Chaos, Fractals, Self-​Organization,
and Disorder. Concepts and Tools. 2nd ed. Berlin: Springer.
Sornette, Didier. 2014. “Physics and financial economics (1776-​2014): Puzzles, Ising and agent-​
based models.” Reports on progress in physics 77 (6): 062001–​28.
 

213  References

Sornette, Didier, and Peter Cauwels. 2015. “Financial bubbles: Mechanisms and diagnostics.”
Review of Behavioral Economics 2 (3): 279–​305.
Sornette, Didier, and Anders Johansen. 1997. “Large financial crashes.” Physica A 245
(1): 411–​22.
Sornette, Didier, and Anders Johansen. 2001. “Significance of log-​periodic precursors to finan-
cial crashes.” Quantitative Finance 1 (4): 452–​71.
Sornette, Didier, Anders Johansen, and Jean-​Philippe Bouchaud. 1995. “Stock market crashes,
precursors and replicas.” Journal de Physique I 6 (1): 167–​75.
Sornette, Didier, and Ryan Woodard. 2010. “Financial bubbles, real estate bubbles, derivative
bubbles, and the financial and economic crisis.” In Econophysics Approaches to Large-​Scale
Business Data and Financial Crisis, edited by Misako Takayasu, Tsutomu Watanabe, and
Hideki Takayasu, 101–​48. Tokyo: Springer Japan.
Sprenkle, Case M. 1961. “Warrant prices as indicators of expectations and preferences.” Yale
Economic Essays 1 (2): 178–​231.
Sprenkle, Case M. 1964. “Some evidence on the profitability of trading in put and call options.”
In The Random Character of Stock Market Prices, edited by Paul H. Cootner. Cambridge,
MA: MIT Press.
Sprowls, Clay. 1963. “Computer education in the business curriculum.” Journal of Business 36
(1): 91–​96.
Standler, Ronald B. 2009. “Funding of basic research in physical sciences in the USA.”
Working paper.
Stanley, H. Eugene. 1971. Introduction to Phase Transitions and Critical Phenomena.
New York: Oxford University Press.
Stanley, H. Eugene. 1999. “Scaling, universality, and renormalization: Three pillars of modern
critical phenomena.” Reviews of Modern Physics 71 (2): S358–​S366.
Stanley, H. Eugene, Viktor Afanasyev, Luis A. Nunes Amaral, Serguey Buldyrev, Albert Goldberger,
Steve Havlin, Harry Leschhorn, Peter Mass, Rosario N. Mantegna, Chung Kang Peng, Paul Prince,
Andrew Salinger, and Karthik Viswanathan. 1996. “Anomalous fluctuations in the dynamics of
complex systems: From DNA and physiology to econophysics.” Physica A 224 (1): 302–​21.
Stanley, H. Eugene, Luis A. Nunes Amaral, D. Canning, Parameswaran Gopikrishnan, Youngki
Lee, and Yanhui Liu. 1999. “Econophysics: Can physicists contribute to the science of eco-
nomics?” Physica A 269 (1): 156–​69.
Stanley, H. Eugene, Xavier Gabaix, and Vasiliki Plerou. 2008. “A statistical physics view of finan-
cial fluctuations: Evidence for scaling universality.” Physica A 387 (1): 3967–​81.
Stanley, H. Eugene, L. A. Nunes Amaral, Parameswaran Gopikrishnan, Yanhui Liu, Vasiliki
Plerou, and Bernd Rosenow. 2000. “Econophysics: What can physicists contribute to eco-
nomics?” International Journal of Theoretical and Applied Finance 3 (3): 335–​46.
Stanley, H. Eugene, and Vasiliki Plerou. 2001. “Scaling and universality in economics: Empirical
results and theoretical interpretation.” Quantitative Finance 1 (6): 563–​67.
Stauffer, D. 2007. “Opinion dynamics and sociophysics.” arXiv:0705.0891.
Stauffer, Dietrich. 2004. “Introduction to statistical physics outside physics.” Physica A 336: 1–​5.
Stauffer, Dietrich. 2005. “Sociophysics simulations II:  Opinion dynamics.” arXiv:physics/​
0503115.
Stauffer, Dietrich, and Didier Sornette. 1999. “Self-​organized percolation model for stock
market fluctuations.” Physica A 271: 496–​506.
 

214 References

Steiger, William Lee. 1963. “Non-​randomness in the stock market: A new test on an existent
hypothesis.” MS thesis, School of Industrial Management, MIT.
Stigler, George J. 1964. “A theory of oligopoly.” Journal of Political Economy 72: 44–​61.
Stock, James H., and Mark W. Watson. 2001. “Vector autoregressions.” Journal of Economic
Perspectives 15 (4): 101–​15.
Stumpf, Michael P. H., and Mason A. Porter. 2012. “Critical truths about power laws.” Science
335 (6069): 665–​66.
Sullivan, Edward J. 2011. “A. D. Roy: The forgotten father of portfolio theory.” In Research in the
History of Economic Thought and Methodology, edited by Jeff E. Biddle and Ross B. Emmett,
73–​82. Greenwich, CT: JAI Press.
Takayasu, Hideki. 2002. Empirical Science of Financial Fluctuations: The Advent of Econophysics.
Tokyo: Springer Japan.
Takayasu, Hideki, ed. 2006. Practical Fruits of Econophysics. Tokyo: Springer Japan.
Takayasu, Misako, Tsutomu Watanabe, and Hideki Takayasu, eds. 2010. Econophysics Approaches
to Large-​Scale Business Data and Financial Crisis. Tokyo: Springer Japan.
Tan, A. 2005. “Long memory stochastic volatility and a risk minimization approach for derivative
pricing and hedging.” Doctoral dissertation, School of Mathematics, University of Manchester.
Taqqu, Murad S. 2001. “Bachelier and his times: A conversation with Bernard Bru.” Finance and
Stochastics 5 (1): 3–​32.
Teichmoeller, John. 1971. “A note on the distribution of stock price changes.” Journal of the
American Statistical Association 66: 282–​84.
Theiler, James, Stephen Eubank, André Longtin, Bryan Galdrikian, and J. Doyne Farmer. 1992.
“Testing for nonlinearity in time series: The method of surrogate data.” Physica D: Nonlinear
Phenomena 58 (1): 77–​94.
Todd, Loreto. 1990. Pidgins and Creoles. London: Routledge.
Treynor, Jack L. 1961. “Toward a theory of market value of risky assets.” Available at SSRN.
Tripp, Omer, and Dror Feitelson. 2001. “Zipf ’s Law Revisited.” Working paper, School of
Engineering and Computer Science, Jerusalem.
Tusset, Gianfranco. 2010. “Going back to the origins of econophysics: The Paretian conception
of heterogeneity.” Paper presented to the 14th ESHET Conference, Amsterdam.
Upton, David, and Donald Shannon. 1979. “The stable Paretian distribution, subordinated
stochastic processes, and asymptotic lognormality:  An empirical investigation.” Journal of
Finance 34: 1031–​39.
van der Vaart, A. W. 2000. Asymptotic Statistics. Cambridge: Cambridge University Press.
Vitanov, Nikolay K., and Zlatinka I. Dimitrova. 2014. Bulgarian Cities and the New Economic
Geography. Sofia: Vanio Nedkov.
Voit, Johannes. 2005. Statistical Mechanics of Financial Markets. 3rd ed. Berlin: Springer.
Von Plato, Jan. 1994. Creating Modern Probability:  Its Mathematics, Physics, and Philosophy in
Historical Perspective. New York: Cambridge University Press.
von Smoluchowski, Marian. 1906. “Zur kinetischen Theorie der Brownschen Molekularbe­
wegung und der Suspensionen.” Annalen der Physik 21: 756–​80.
Walter, Christian. 2005. “La gestion indicielle et la théorie des moyennes.” Revue d’Économie
Financière 79: 113–​36.
Wang, Jie, Chun-​Xia Yang, Pei-​Ling Zhou, Ying-​Di Jin, Tao Zhou, and Bing-​Hong Wang. 2005.
“Evolutionary percolation model of stock market with variable agent number.” Physica A
354: 505–​17.
 

215  References

Wang, Yuling, Jing Wang, and Fengshan Si. 2012. “Conditional value at risk of stock return
under stable distribution.” Lecture Notes in Information Technology 19: 141–​45.
Watanabe, Shinzo. 2009. “The Japanese contributions to martingales.” Journal Electronique
d’Histoire des Probabilités et de la Statistique 5 (1): 1–​13.
Weber, Ernst Juerg. 2009. “A short history of derivative security markets.” In Vinzenz Bronzin’s
Option Pricing Models, edited by Wolfgang Hafner and Heinz Zimmermann, 431–​66.
Berlin: Springer.
Weintraub, Robert E. 1963. “On speculative prices and random walks:  A  denial.” Journal of
Finance 18 (1): 59–​66.
Welch, Ivo. 1992. “Sequential sales, learning and cascades.” Journal of Finance 47 (2): 695–​732.
Weron, Aleksander, Szymon Mercik, and Rafal Weron. 1999. “Origins of the scaling behaviour
in the dynamics of financial data.” Physica A 264: 562–​69.
Weston, J. Fred 1967. “The state of the finance field.” Journal of Finance 22 (4): 539–​40.
Whitley, Richard Drummond. 1986a. “The rise of modern finance theory:  Its characteristics
as a scientific field and connection to the changing structure of capital markets.” In Research
in the History of Economic Thought and Methodology, edited by Warren J. Samuels, 147–​78.
Stanford, CA: JAI Press.
Whitley, Richard Drummond. 1986b. “The structure and context of economics as a scientific
field.” In Research in the History of Economic Thought and Methodology, edited by Waren J.
Samuels, 179–​209. Stanford, CA: JAI Press.
Wickens, Michael R. 1995. “Real business cycle analysis: A needed revolution in macroecono-
metrics.” Economic Journal 105 (433): 1637–​48.
Widom, Benjamin. 1965a. “Equation of state in the neighborhood of the critical point.” Journal
of Chemical Physics 43: 3898–​905.
Widom, Benjamin. 1965b. “Surface tension and molecular correlations near the critical point.”
Journal of Chemical Physics 43: 3892–​97.
Wiener, Norbert. 1923. “Differential-​space.” Journal of Mathematics and Physics 2: 131–​74.
Williams, John Burr. 1938. The Theory of Investment Value. Cambridge, MA:  Harvard
University Press.
Willis, J. C. 1922. Age and Area:  A  Study in Geographical Distribution and Origin of Species.
Cambridge: Cambridge University press.
Willis, J. C., and G. Udny Yule. 1922. “Some statistics of evolution and geographical distribution
in plants and animals, and their significance.” Nature 109: 177–​79.
Wilson, Kenneth G. 1993. “The renormalization group and critical phenomena.” In Nobel
Lectures in Physics (1981–​1990), edited by Gösta Ekspong, 102–​32. London:  World
Scientific.
Working, Holbrook. 1934. “A random-​difference series for use in the analysis of time series.”
Journal of the American Statistical Association 29: 11–​24.
Working, Holbrook. 1949. “The investigation of economic expectations.” American Economic
Review 39 (3): 150–​66.
Working, Holbrook. 1953. “Futures trading and hedging.” American Economic Review 43
(3): 314–​43.
Working, Holbrook. 1956. “New ideas and methods for price research.” Journal of Farm
Economics 38: 1427–​36.
Working, Holbrook. 1958. “A theory of anticipatory prices.” American Economic Review 48
(2): 188–​99.
 

216 References

Working, Holbrook. 1960. “Note on the correlation of first differences of averages in a random
chain.” Econometrica 28 (4): 916–​18.
Working, Holbrook. 1961. “New concepts concerning futures markets and prices.” American
Economic Review 51 (2): 160–​63.
Wyart, M., Jean-​Philippe Bouchaud, J. Kockelkoren, Marc Potters, and M. Vettorazo. 2008.
“Relation between bid-​ask spread, impact and volatility in order-​driven markets.” Quantitative
Finance 8: 41–​57.
Yule, G. Udny. 1925. “A mathematical theory of evolution, based on the conclusions of Dr. J. C.
Willis, F.R.S.” Philosophical Transactions of the Royal Society B 213 (402–​10): 21–​87.
Zanin, Massimiliano, Luciano Zunino, Osvaldo A. Rosso, and David Papo. 2012. “Permutation
entropy and its main biomedical and econophysics applications:  A  review.” Entropy 14
(12): 1553–​77. doi: 10.3390/​e14081553.
Zhang, Qiang, and Jiguang Han. 2013. “Option pricing in incomplete markets.” Applied
Mathematics Letters 26 (10): 975.
Zipf, George K. 1935. The Psychobiology of Language. New York: Houghton Mifflin.
Zipf, George K. 1949. Human Behaviour and the Principle of Least Effort. Reading,
MA: Addison-​Wesley.
 

INDEX

abnormal data, 92–​93 ARCH. See autoregressive conditional

accountants, 22–​23 heteroskedasticity
ad hoc perspective, 112 Armatte, Michel, 172n4
adhocity, 112 Arrow, Kenneth, 21, 171n53, 183n41
agent-​based modeling, 130–​31, 183n38 Arrow-​Debreu general-​equilibrium model,
Aizawa, Yoji, 133 21, 171n53
Alexander, Sidney, 14–​15, 36 Asian financial crisis, 181n15
Alfarano, Simone, 130 attractor, 176n32
Alstott, Jeff, 108 Auerbach, Felix, 70, 177n37
Altam, Gunther, 177n38 Aurell, Erik, 149
Amaral, Luis, 177n40 Ausloos, Marcel, xii, 137, 165, 175n22
American Economic Review, 179n14 autoregressive conditional heteroskedasticity
American Finance Association, 12 (ARCH), 113, 114
American Institute of Physics,  Gaussian framework as compatible with,
179n16 100, 119–​20
American probability school, 7 overview, 41, 45–​48, 46f
Amoroso, Luigi, 31 power law origin, 126
Anderson, Phil, 183n41 unconditional approach justifying, 112
Andrews, Thomas, 174n6 volatility and, 111
anomalies literature, 182n32 average, 2–​3
Aoki, Masanao, 128 Axtell, Robert, 177n40
Aoyama, Hideaki, 108
APFA. See Applications of Physics in Bachelier, Louis, 9, 169n31
Financial Analysis Brownian motion and, 4–​6
“Application of Physics to Economics and derivative contract evoked by, 168n16
Finance,” 182n27 economists discovering, 10
Applications of Physics in Financial Analysis normal distribution used by, 29
(APFA), 82 physicist’s language used by, 169n21
“Applications of Statistical Physics in Backhouse, Roger, 88, 179n12
Economics and Finance” (Farmer and Bagehot, Walter. See Treynor, Jack
Lux), 88 Bak, Per, 132–​33, 183n41
APT. See arbitrage pricing theory Barnett, Vincent, 169n27
Arad, Ruth, 173n26 barometers, 30, 172n5, 172n6
arbitrage, 3, 171n57, 171n58, 184n8 Bartolozzi, Marco, 133
arbitrage argument, 148 Basel II Accords, 171n56
arbitrage pricing theory (APT), 20 Bassler, Kevin, 106
arbitrage proof argument, 13 Batterman, Robert, 52f

217
 

218 Index

Bayesian approach, 166, 184n1 capital-​asset pricing model (CAPM), 13, 20,

Bayeux-Besnainou, Isabelle, 171n53 21, 170n41, 182n34
Bazerman, Charles, 91–​92 Capital Fund Management (CFM), 108
beauty contest, 181n19 CAPM. See capital-​asset pricing model
behavioral finance, 129, 180n29 carbon dioxide, 174n6
behaviors, 130–​31 Carr, Peter, 151
bell curve, 114 Casey, Michael, 181n4
Bernoulli, Daniel, 5–​6 CBOE. See Chicago Board Options
Berry, Brian, 177n37 Exchange
Binney, James, 56f Center for Polymer Studies, 80
Black, Fischer, 20–​21, 44–​45, 60, 149–​50 Center for Research in Security Prices
Black and Scholes model, 22, (CRSP), 14, 15
147–​52, 172n59 central-​limit theorem, 36, 67, 167n6, 174n4
Black-​Scholes-​Merton model, 21, 97, 180n27 CFM. See Capital Fund Management
Blattberg, Robert, 38 Challet, Damien, 110f, 132
Bollerslev, Tim, 47 Champernowne, David, 124–​25
Bonart, Julius, 110f Chane-​Alune, Elena, 22
bond prices, 168n10 changes of regime, 115
bookkeeping machines, 14 Chapman-​Kolmogorov-​Smoluchowski
Borak, Szymon, 41 equation, 4–​5
Bouchaud, Jean-​Philippe, 110f, 116, 149, 150 Chakrabarti, Bikas K., 84, 178n7
European call option formula and, chartism, 179n23
180n26, 184n9 chartists, 103, 180n32
on power laws, 63, 131 Chen, Shu-​Heng, 130
Boumans, Marcel, 172n4 Chiarella, Carl, 182n29, 183n40
bound, 183n44 Chicago Board Options Exchange
bound ranks, 177n45 (CBOE), 22
boundary objects, 123 Chrisman, Nicholas, 122, 123
Bowley, Arthur, 11, 30 city sizes, 70, 177n37, 177n40
Boyarchenko, Svetlana, 148, 151 Clark, Peter, 45
Brada, Josef, 32, 38 Clauset, Aaron, 70, 108, 136–​37
Brakman, Steven, 182n28 Clementi, Fabio, 145, 184n5
Broda, Simon, 112, 135 Clippe, Paulette, 137
Brody, Samuel, 70 cognitive psychology, 129
Bronzin, Vinzenz, 5–​6, 9, 29, 169n31 Cohen, Kalman, 175n14
Brown, Stephen, 169n28 collaborative research, 164–​65
Brownian motion, 24, 45, 168n14, 168n19, colonization strategy, 102–​4
169n20, 172n59 Columbia University, 172n6
Bachelier and, 4–​6 commodity, 184n7
Kendall on, 8–​9 common framework
Buchanan, Mark, 113–​14 issues for, 140–​43
models, 154t–​163t
Calculation of Chances and Philosophy of the option pricing application, 147–​52
Stock Exchange (Calcul des chances et overview, 139–​40
philosophie de la bourse) (Regnault), 2–​4 propositions for, 140–​47, 147f
Campanha, José, 75–​76, 178n49 competition, 184n7
 

219  Index

complexity, 179n9, 183n41 intraday, 60–​61, 61f, 104, 175n18

complex systems, xii–​xiii as nonnormalized, 92–​93
computerization, of science, 60–​62, 61f rough, 98
computers, 13–​15, 32, 180n36 databases, 13–​15
conceptual generalization, 146 Davis, Mark, 5
condensed-​matter physics, 79–​80 dealers, 182n34
conditional distribution, 47–​48, 100, 111–​12 De Bondt, Werner, 129
Condon, E. U., 70 Debreu, Gérard, 21
conservative activism, 98 deduction, 180n32
constants, 158n51, 176n35 De Meyer, Bernard, 18
Cont, Rama, 109 de Moivre, Abraham, 167n6
continuous-​time probability theory, 6 Demsetz, Harold, 129
“Contributions of Econophysics to dependency, 170n44
Finance,” 182n27 derivative contract, 168n16
Cootner, Paul, 13, 14–​15, 20–​21, 174n30 derived distributions, 184n5
Copeland, Thomas, 182n34 Derman, Emanuel, xii
Cornelis, A. Los, 181n11 developments, 122–​37
corporate revenue, 175n16 deviations, 3
correlation lengths, 56, 63, 176n25, 176n26 De Vroey, Michel, 179n21
cotton prices, 35, 172n10 differential diagnosis, 95
Cournot, Antoine, 50 Di Matteo, Tiziana, 145, 184n5
Cours d’économie politique (Pareto), 173n16 distributions, 39–​40, 182n31, 184n5,
Cowles, Alfred, 7–​8, 13, 15, 29, 169n28, 184n11. See also conditional
169n29, 170n44, 172n3 distribution; non-​Gaussian
Cowles Commission, 7–​8, 94, 169n25 distribution; power-​law distributions;
Cox, John, 45 unconditional distributions
Creole, 182n26 diversification, 173n20
criteria of conventional acceptance, 97 diversification strategy, 170n34
critical exponent, 54 The Doctrine of Chances (de Moivre), 167n6
as common, 109–​10, 110f “Does the Stock Market Overreact?”
identification of, 109–​10, 110f (De Bondt and Thaler), 129
critical opalescence, 52 Domb, Cyril, 176n28
critical phenomena, 51–​58, 52f, 54f, 55f, 56f, Donangelo, Raul, 183n40
57f, 177n48 Doob, Joseph, 7, 10
critical point, 116–​17, 174n6, 176n25 downward, 3
critical state, 133 Dubkov, Alexander, 177n39
critical time, 181n18 Dupoyet, Brice, 133
criticality, theory of, 183n41 Durlauf, Steven, 94, 95, 117–​18, 181n21
CRSP. See Center for Research in
Security Prices econometrics, emergence of, 7–​9
cutoff, 74–​76, 75f economic barometers, 29–​30
economic fundamentals, 182n31
data, 31–​32 economic measurement, age of, 29–​30
as abnormal, 92–​93 Economic Office of Harvard, 30
as empirical, 13–​15, 165–​66 economics, 50, 179n9, 181n21, 182n36
high-​frequency, 104, 175n18, 178n53 Economics, 89
 

220 Index

economists, 10, 179n15, 179n16, 180n29, Eguiluz, Victor, 132

183n41. See also financial economists eighteenth-​century financial-​asset
econophysicists, 120–​21, 179n9 returns, 181n16
equilibrium dealt with by, 180n28 Einstein, Albert, 6
financial economists’ dialogue with, 78, electron. See Ising model
90–​105, 99f, 100f, 101f, 180n34 empirical characteristic function, 173n26
unconditional distributions focused on empirical data, 13–​15, 165–​66
by, 111–​12 empirical evidences, 26–​32, 26f, 27f, 28f
econophysics, ix–​xiv, 178n1, 180n2, 181n21 empirical realism, 93
articles, 81–​82, 81f empirical studies, 182n32
definition, 59–​60, 79, 88–​89 engineers, 182n25
disciplinary position of, 78–​105, 80f, Engle, Robert, 45, 47
81f, 84t, 85f, 86t, 87t, 89t, 90t, 99f, l’entre-​articulation. See inter-​relationship
100f, 101f environment, 122
in disciplinary space, 83–​90, 84t, 85f, 86t, equilibrium, 97–​98, 148, 170n45,
87t, 89t, 90t 175n7, 180n28
financial economics contributed to by, Ernst, Harry, 32, 38
106–​38, 110f, 115t, 121f, 164–​66 ESHIA. See Society for Economic Science
from financial economists’ viewpoint, with Heterogeneous Interacting Agents
117–​23, 121f Estoup, Jean-​Baptiste, 70
generic model, 143–​53, 154t–​163t Etheridge, Alison, 5
hypothetico-​deductive approach in, 121f ETH Zurich. See Econophysics Group at the
institutionalization of, 80–​83, 81f, 90–​104, University of Zurich
99f, 100f, 101f Euclidean geometry, 167n2
in physics’ shadow, 84–​88, 84t, 85f, 86t, European Association of Physicists, 82
87t, 178n6 European call option formula,
position of, 83–​90, 84t, 85f, 86t, 87t, 180n26, 184n9
89t, 90t European Physical Journal B, 86–​88, 87t
practical implications, 107–​9 expected-​utility theory, 170n38, 170n41
price-​return models of, 143–​53, 154t–​163t exponential distributions, 183n42
Stanley, Eugene, on, 59, 79, 81, 82, 178n1 exponentially truncated Lévy stable
textbooks on, 82 distribution, 150–​52
trading rooms using, 106–​17, 110f, 115t exponential technique, 178n50
Econophysics Colloquium, 82–​83 exponential truncation functions, 178n52
Econophysics Group at the University of extreme events, 181n11
Zurich, 108, 117 extreme-​value analysis
econophysics models, 108, 149–​50, 180n1 power-​law distributions for, 62–​72,
efficiency, 170n46, 170n47 69f, 71f
efficient-​market hypothesis, 13, 15–​20, tools for, 49–​77, 52f, 54f, 55f, 56f, 57f, 61f,
21–​24, 99 69f, 71f, 75f
efficient markets extreme values, 179n13
definition, 18–​19, 170n46, alternative paths, 41–​48, 42f, 43f, 46f
171n57, 171n58 financial economists on, 37–​48, 42f,
Fama demonstrating, 13, 16–​18, 182n30 43f, 46f
stock price variations as connected Gaussian framework, 25–​48, 26f, 27f, 28f,
to, 179n19 42f, 43f, 46f
 

221  Index

mathematical treatments of, 37–​48, 42f, as scientific discipline, 9–​15

43f, 46f strategy used by, 102–​4
in stock price variations, 26–​29, 26f, financial economists, x–​xiv,
27f, 28f 179n13, 180n32
extreme variations, 33–​37 econophysicists’ dialogue with, 78, 90–​105,
99f, 100f, 101f, 180n34
Fama, Eugene, 23, 44–​4 5, 59, 91, econophysics contributions seen from
170n45, 170n46, 170n47, viewpoint of, 117–​23, 121f
173n23, 173n25 on extreme values, 37–​48, 42f, 43f, 46f
efficient markets demonstrated by, 13, Pareto-​Lévy distributions and, 33–​41
16–​18, 182n30 financial innovations, opportunities for
on Pareto-​Lévy processes, 36–​37, 38 econophysics articles, 81–​82, 81f
on random-​walk model, 103 overview, 78–​105, 80f, 81f, 84t, 85f, 86t,
on stable Lévy processes, 39–​40 87t, 89t, 90t, 99f, 100f, 101f
on statistical tests, 19 physics PhDs, 79–​80, 80f
Farmer, J. Doyne, 88, 107, 111, 112, 182n29 financial management, 170n38
FASB. See Financial Accounting financial variables, 182n31
Standards Board finite configuration, 177n45
fat-​tailed distributions, 115t finite mean, 174n29
FCO. See Financial Crisis Observatory finiteness of variability, 174n29
Feigenbaum, James, 116 Fisher, Arne, 169n31
Feller, William, 7, 10, 34 Fisher, Irving, 30, 170n35, 172n6, 172n7
filtration, 170n36 Fisher, Lawrence, 14
finance, 2–​7, 11–​13, 184n9. See also Fisher, Mark, 176n28
behavioral finance Fitzpatrick, Richard, 50–​51, 174n3, 174n5
finance models, Gaussian framework fluid, temperature-​pressure phase diagram
and, 20–​23 for, 52f
Financial Accounting Standards Board Focardi, Sergio, 183n40
(FASB), 22–​23 Food Research Institute, 169n25
Financial Analysts Seminar, 170n40 forward contract, 168n15
financial-​asset returns, 181n16 free competitive equilibrium, 16
financial crashes, 115–​17 French 3 percent bond, 3
Financial Crisis Observatory (FCO), 117 French barometer, 30
financial distributions, as power law, 131 French engineers (ingénieurs
financial econometrics. See econometrics économistes), 30–​31
financial economics. See also Gaussian Freund, Peter, 116
distribution Friedman, Milton, 10, 172n4, 172n6
creation of, 9–​24 fund management, 21–​22
early roots of, 2–​9
econophysics contributing to, 106–​38, Gabaix, Xavier, 135, 177n37
110f, 115t, 121f, 164–​66 on NYSE, 71, 71f
econophysics model compatible on power laws, 68–​70, 69f, 124,
with, 149–​50 125–​28, 177n40
Gaussian distribution influencing, 2–​24 Galai, Dan, 182n34
innovation of, 102–​4 Galam, Serge, 52, 175n15
power laws’ links with, 65–​67 Galison, Peter, 122, 182n25
 

222 Index

Gallegati, Mauro, 145, 178n8, 184n5 Gopikrishnan, Parameswaran, 69f,

GARCH. See generalized autoregressive 106, 145–​46
conditional heteroscedasticity Graduate School of Business at the
Garman, Mark, 182n33 University of Chicago, 14, 32, 170n43
Gauss, Carl, 167n6 Granger, Clive, 14, 38
Gaussian distribution, 26–​29, 27f, 28f Grech, Dariusz, 79
obtaining, 146–​47, 147f Groeber, Ronald, 22
role of, 1–​24 Grossman, Sanford, 19
Gaussian framework, 20–​24, 165–​66. growth, 122
See also autoregressive conditional Guégan, Dominique, 181n11
heteroskedasticity Gunaratne, Gemunu, 145
ARCH as compatible with, 100, 119–​20 Gunji, Yukio, 183n40
extreme values, 25–​48, 26f, 27f, 28f, 42f, Gupta, Hari, 75–​76, 178n49
43f, 46f
as generalized framework case, Hamilton, William Peter, 169n28
146–​47, 147f Hammer, Howard, 22
Gaussian law, 172n1 Hang Seng Index, 68–​70, 69f
Gaussian processes, 173n22 Harrison, Michael J., 23, 148–​52, 171n53
Gaussian regime, 184n12 Harrison, Paul, 181n16
Geanakoplos, John, 111, 112 Harvard University, 30, 35, 179n16
The General Theory of Employment, Interest hedging, 107, 108
and Money (Keynes), 181n19 heterodox dissenters, 179n12
generalized autoregressive conditional Higgs boson, 173n19, 183n2
heteroscedasticity (GARCH), 47 high-​frequency data, 104, 175n18, 178n53
generalized central-​limit theorem, Houthakker, Hendrik, 14–​15, 32, 35
67, 173n18 Hughes, Barry, 94–​95, 125
generative models Hunter, D., 176n28
as new, 123–​34 Hurst, Harold, 114
overview, 123–​34 hyperasymptotic approach, 5
power laws, 124–​31 hypothetico-​deductive method, 99, 103,
random growth, 124–​26 121f, 180n29
self-​organized criticality, 131–​34
generic model, xi–​xii, 143–​53, 154t–​163t IBM. See International Business Machines
Gibrat, Robert, 125 Ibraginov, Rustam, 135
Gibrat’s Law, 124 ideal gas, 174n5
Gillespie, Colin, 137 incompleteness of markets, 184n10
Giorgio, Israel, 50 independent random variables, 172n12
Gligor, Mircea, 137 index numbers, 172n8
global demand, 175n16 induction, 180n32
Glottometrics, 177n38 infinite variance, 35–​36, 37–​38, 72–​76,
Gnedenko, Boris, 36, 167n6 75f, 148–​50
Godfrey, Michael, 38 inflection points, 180n25
Goerlich, Francisco, 137 informed dealers, 182n34
Goetzmann, William, 169n28 ingénieurs économistes. See French engineers
Gonedes, Nicholas, 38 Ingrao, Bruna, 50
goodness-​of-​fit test, 136–​37 innovation, of financial economics, 102–​4
 

223  Index

institutional investors, 166 Jovanovic, Franck, 30–​31, 32, 170n43

insurance industry, 114 JPMorgan, 109
integrative research, 164–​65 jump-​diffusion, 41–​45, 42f, 43f, 174n32
interdisciplinarity, 121–​22, 183n41 jump-​process models, 45
International Accounting Standards
Board, 22–​23 Kadanoff, Leo, 53, 176n28
International Business Machines (IBM), Kahneman, Daniel, 129, 182n36
26, 26f Kaiser, David, 79, 80f
International Conference on Kaizoji, Michio, 107
Econophysics, 83 Kaizoji, Taisei, 107, 130
International Review of Financial Analysis, Keen, Steve, 103, 180n34
89, 182n27 Kendall, Maurice, 8–​9, 31, 38
International Society for the Advancement of Kesten, Harry, 124–​28
Economic Theory in its Relation with Keynes, John, 181n19
Statistics and Mathematics, 7–​8 Khinchine, Alexandre, 169n22
inter-​relationship King, Benjamin, 14
(l’entre-​articulation), 122–​37 Klass, Oren, 177n40
intraday data, 60–​61, 61f, 104, 175n18 Kleiber, Max, 70
Introduction to Econophysics (Mantegna and Klein, Julie, 29, 122–​23, 172n4
Stanley), 82 Kolmogorov, Andrei, 6–​7, 36, 167n6,
investor sizes, heterogeneity of, 126–​29 169n22, 176n28
Iori, Giulia, 182n29, 183n40 Koopmans-​Vining debate, 94
Ising model, 55–​58, 55f, 56f, 57f, 183n40 Koponen, Ismo, 151, 178n49
Israel, Giorgio, 58 Kosterlitz-​Thoules transition, 175n22
iterated regressions, 173n26 Kou, Xiodong, 130
Ito, Kiyosi, 169n23 Koutrouvelis, Ioannis, 173n26
Kreps, David, 23, 148–​52, 171n53
JEIC. See Journal of Economic Interaction and Krugman, Paul, 124, 182n28
Coordination Kuhn, Thomas, 101
JEL. See Journal of Economic Literature Kumar, Alok, 169n28
Jensen, Michael, 18, 20, 23 Kutner, Ryszard, 79
Jevons, Frank, 50 Kydland, Finn E., 179n21
Johansen, Anders, 116–​17 Kyle, Albert, 182n34
Jones, Herbert, 8, 29,
169n29, 170n44, 172n3 Lagrange multiplier test, 137
Journal of Business, 12, 102 Laguës, Michel, 67, 174n6
Journal of Economic Behavior and Lakatos, Imre, 165
Organization, 88 Landier, Augustin, 177n40
Journal of Economic Dynamics and Control, 82, Laplace, Pierre-​Simon, 167n6
88, 102, 182n27 Larson, Arnold, 32
Journal of Economic Interaction and Latour, Cagnard de, 174n6
Coordination (JEIC), 82, 88 law of large numbers, 168n18
Journal of Economic Literature (JEL), 179n15 law of one price, 23, 184n7
Journal of Finance, 12, 102 LeBaron, Blake, 95–​96, 117, 180n24
Journal of Financial and Quantitative Lefevre, Henri, 175n14
Analysis, 102 Le Gall, Philippe, 30–​31, 172n4
 

224 Index

Lenoir, Marcel, 31 stable Lévy processes worked on by, 33–​37

leptokurtic distributions on stable Paretian distributions, 39
Gaussian distribution compared to, 26–​29, on statistical tests, 19
27f, 28f Mantegna, Rosario, 59, 74, 79, 82,
overview, 26–​29, 27f, 28f, 30–​31, 32 106–​7, 148
LeRoy, Paul, 17–​18 March, Lucien, 30, 31
Lesne, Annick, 67, 174n6 Marchesi, Michele, 130
Levendorskii, Sergei, 148, 151 Mariani, Maria, 145
Lévy, Paul, 6, 33–​37, 130, 177n40 market microstructure, 182n34
Lévy processes, x–​xi, 91, 175n20, 184n5. See Garman coining, 182n33
also stable Lévy processes power laws and, 128–​31
Lexis, Wilhelm, 167n5 markets. See also efficient market
Li, Lun, 176n30 as complete, 171n57
likelihood-​ratio test, 136–​37 incompleteness of, 184n10
Lillo, Fabrizio, 106–​7, 131 as informationally inefficient, 129
Lintner, John, 20 models detecting overestimation of, 181n4
liquidity black holes, 114 system of, 171n53
Liu, Yanhui, 145 Markov, Andrei, 168n18
log-​log scale graph, visual linearity observed Markov chains, 168n18
in, 132–​33 Markov process, 174n30
log-​normal distributions, 183n42 Markowitz, Harry, 14, 36–​37. See also
log-​normal tails, 183n43 portfolio choice theory
Log-​Periodic Power Law (LPPL), 116–​17 CAPM influenced by, 20, 170n41
long-​memory processes, 114 expected-​utility theory applied by,
Long-​Term Capital Management (LTCM), 170n38, 170n41
78, 113–​14, 181n15 Markowitz mean-​variance portfolio
Lorie, James, 14, 103 optimization model, 20, 24
Lotka, Alfred, 70 Marsili, Mateo, 103
LPPL. See Log-​Periodic Power Law martingale model, 11, 17–​18, 45, 170n37,
LTCM. See Long-​Term Capital Management 171n57, 171n58
Lucas, Robert E., 18 Maslov, Sergei, 132
Luttmer, Erzo, 177n40 Massachusetts Institute of Technology
Lux, Thomas, 88, 130, 177n40, (MIT), 32, 170n43
178n8, 181n21 Matacz, Andrew, 148
Mathematica, 108
Maas, Harro, 50 mathematical analogies, 58–​59
Macroeconomic Dynamics, 88 mathematical finance, 184n9
macroscopic levels, 51–​58, 52f, 54f, 55f, 56f, mathematical physics, 4–​5
57f, 175n12 mathematical treatments, of extreme values,
Malgrange, Pierre, 179n21 37–​48, 42f, 43f, 46f
Malkiel, Burton, 18–​19, 23 Matlab, 108
Mandelbrot, Benoit, 41, 70–​71, 91, 93, maximum likelihood estimators, 136
106–​7, 170n37, 173n17, 174n30, Max-​Neef, Manfred, 122
177n46, 180n3. See also FamaEugene; McCauley, Joseph, 82, 93, 103, 145,
martingale model 148, 152
on stability, 66–​67 McCulloch, Hu, 173n25
 

225  Index

McNees, Stephen, 45 National Association of Securities

Ménard, Claude, 50 Dealers Automated Quotations
Merton, Robert, 44–​45, 181n15 (NASDAQ), 175n17
Mezard, Marc, 131 National Bureau of Economic Research, 94
Miburn, J. Alex., 22–​23 negative externalities, 182n28
microscopic levels. See also renormalization negative sign, 176n30
group theory New York Stock Exchange (NYSE), 31,
as observable, 175n12 168n10, 175n17. See also Center for
overview, 51–​58, 52f, 54f, 55f, 56f, 57f Research in Security Prices
Mike, Szabolcs, 107 IBM returns on, 26, 26f
Miller, Merton, 10, 12, 13, 20, 148, 184n8 visual linearity, 71, 71f
Mills, Frederick, 30 Newman, Mark, 136–​37, 176n30
Mir, Tariq, 137 Niederhoffer, Victor, 14–​15
Mirowski, Philip, 50 Nikkei, 68–​70, 69f
MIT. See Massachusetts Institute of Nikkei Econophysics Research Workshop
Technology and Symposium, 82–​83
Mitchell, Wesley, 30, 172n6, 172n8 no-​arbitrage principle, 23
Mitzenmacher, Michael, 118–​19, Nobel Prize in Economics, 170n48, 182n36
176n34, 181n22 Nobel Prize in Physics, 51–​58, 52f, 54f, 55f,
ModEco, 108 56f, 57f
modeling, xi, 2–​4 noise traders, 116–​17
modeling risk, 171n52 Nolan, John, 34
models, 20–​23, 148, 166, 181n4, 184n3. non-​Gaussian distribution, 42, 42f,
See also autoregressive conditional 109, 149–​50
heteroskedasticity; Black and Scholes non-​Gaussian framework, 40
model; generic model; Ising model; non-​Gaussian models, 166
Markowitz mean-​variance portfolio non-​Gaussian option-​pricing model, 148
optimization model; martingale model; non-​Gaussian stable Lévy processes, 72–​73.
price-​return models; random-​walk See also truncation
model; real business cycle models noninformed dealers, 182n34
percolation, 132 nonnormalized data, 92–​93
size, 177n40 nonstable Lévy distribution, 144, 145
stochastic, 2–​4 normal distribution, 2–​4, 7–​8, 24, 29,
two-​block, 44–​45 169n29, 172n9
VARs, 94 normal law, 167n5
modern probability theory, 6–​7, 10–​11 normal processes, 173n23
Modigliani, Franco, 10, 13, 148, 184n8 normal random variable, 172n12
money, 175n16 NYSE. See New York Stock Exchange
Moore, Arnold, 14
Moore, Henry, 30 Officer, Robert, 38, 41
Morales, Raffaello, 137 O’Hara, Maureen, 129, 175n19, 182n34
Morgan, Mary, 29, 172n4 Okulicz-​Kozaryn, Adam, 177n37
Morgenstern, Oskar, 14, 38 Olivier, Maurice, 31, 172n2
Morin, Edgar, 122–​23 one-​block reasoning, 174n32
Mossin, Jan, 20 “The Only Game in Town” (Treynor), 129
multidisciplinarity, 121–​22 optimal hedging, 108
 

226 Index

option pricing, common framework’s scaling properties contributing to, 176n28

application to, 147–​52 Physics and Astrophysics Classification
option-​pricing formula, 180n26, 184n9 Scheme (PACS), 81, 86, 178n6, 179n15
option-​pricing model, 148. See also Black and physics PhDs, 79–​80, 80f
Scholes model Pickhardt, Michael, 183n40
orthodox dissenter journals, 88, 179n12 pidgin, 122–​37, 182n26
Osborne, Maury, 10–​11, 14, 144, 172n9 Plerou, Vasiliki, 67–​68, 110, 180n24
Pliska, Stanley, 23, 148, 152, 171n53
PACS. See Physics and Astrophysics Ponzi, Adam, 133
Classification Scheme portfolio choice theory, 10, 170n41, 174n27
Pagan, Adrian, 66, 120, 181n12 portfolio management, 108, 170n38
papers, 178n7, 178n8, 179n17 positive heuristic, 165
parameters, 173n14 postulates, 167n2
Paretian distributions, 39–​40 Potters, Marc, 131, 180n26, 184n9
Paretian law, 173n16 power-​law distributions, xiii, 183n42
Paretian simulation, 28f, 29 as extreme value analysis tool, 62–​72,
Pareto, Vilfredo, 31, 68, 70, 134, 69f, 71f
173n16, 177n40 as phenomenological law, 67–​72, 69f, 71f
Pareto-​Lévy distributions, 33–​41 power laws, xi, 94–​96, 98, 111, 116, 118–​19,
Pareto tails, 183n43 165–​66, 176n30, 177n40. See also
Paris Stock Exchange, 2–​3, 175n17 visual tests
particle detectors, 182n25 ARCH origin of, 126
particles. See Ising model Bouchaud on, 63, 131
Paulson-​Holcomb-​Leitch method, 173n26 econophysics models using, 180n1
Pearson, Karl, 168n7 as emergent property, 131
pension funds, 114 financial distributions as, 131
percolation model, 132 financial economics links of, 65–​67
Perello, Josep, 182n29 Gabaix on, 68–​70, 69f, 124,
perfect rational agent, 180n29 125–​28, 177n40
Persons, Warren, 30 generative models, 124–​26
phase transitions, 51–​58, 52f, 54f, 55f, high-​frequency data, 178n53
56f, 57f investor sizes and, 126–​29
phenomenological approach, 176n33, 180n3 market microstructure and, 128–​31
phenomenological law, power-​law physicists making possible, 72–​76, 75f
distribution as, 67–​72, 69f, 71f ranks used in, 177n45
Physica A, 81, 86–​88, 87t risk estimation, 113–​14
physical theories, 158n51 self-​organized criticality theory on, 134
physicists, 182n25, 183n2. See also Physics stable Lévy processes characterized
and Astrophysics Classification Scheme by, 176n31
economists collaborating with, 183n41 statistical physics role of, 63–​65
power laws made possible by, 72–​76, 75f stochastic processes as linked with, 66
second bubble, 79–​80, 80f, 178n2 truncation of, 72–​76, 75f
physics, 50, 79–​80, 167n2, 181n21. See also validation of, 136–​37
statistical physics variation correlation length as, 175n13
econophysics in shadow of, 84–​88, 84t, Prescott, Edward, 179n21
85f, 86t, 87t, 178n6 Press, S. James., 40, 42–​45, 43f, 173n26
 

227  Index

price indices calculus, 172n7 risk, 22, 171n52

price-​return models, 143–​53, 154t–​163t power laws’ pertinence to
prices, 23, 168n10, 173n24, 181n16, 182n29, estimating, 113–​14
182n34. See also law of one price; return’s relationship with,
volatility 170n34, 174n27
pricing, 108 risk management, 109
primes, 168n15 RiskMetrics, 109
probability space, 170n36 Rmetrics, 108
probability theory, 169n23 Roberts, Harry, 9, 12–​13, 169n30
profit, 169n29 Rochet, Jean-Charles, 171n53
psychology, 182n36 Roehner, Bertrand, 82
pure-​jump processes, 174n32 Roll, Richard, 40, 173n23, 173n25
Rosenfeld, Lawrence, 180n36
qualitative tests, 95 Ross, Stephen, 20, 45, 171n53
Quantitative Finance, 82, 88, 102 Rosser, J. Barkley, 88, 180n34
quantitative tests, 95, 134–​37 rough data, 98
quants, ix–​x , 107 Roy, A. D., 10
Quételet, Adolphe, 2–​3 Roychowdhury, Vwani, 176n34
Quieros, Silvio, 137 Russian financial crisis, 181n15
Rybski, Diego, 177n37
radar, 182n25
random growth, generative models, 124–​26 St. Petersburg paradox, 177n44
random number table, 169n26 Saley, Hadiza, 18
random size, 34 samples, 177n45, 180n36
random-​walk model, 7–​9, 12–​13, Samuelson, Paul, 11, 18, 37, 45,
103, 174n30 169n31, 170n37
random walks, ix, 3–​4, 168n7, Santa Fe Institute, xii–​xiii, 80, 130,
168n19, 170n37 181n21, 183n41
rational traders, 116–​17 SAS. See Statistical Analysis System
raw observations, 92–​93 scale factor, 37, 173n20
RBC models. See real business cycle models scale invariance,
Reagan administration, 79 67–​68, 176n28, 176n29. See also
real business cycle (RBC) models, critical phenomena
94, 179n21 definition, 52, 64, 173n17
Redelico, Francisco, 137 overview, 64
Reed, William, 94–​95, 125 scaling hypothesis, 116
Regnault, Jules, 6, 9, 167n3, 168n11, 169n31. scaling laws, 174n6, 181n13
See also random walks scaling properties, 176n28, 176n29
Gaussian distribution introduced by, 29 Schabas, Margaret, 50
normal law named by, 167n5 Schinckus, Christophe, 84, 129
stochastic modeling and, 2–​4 Scholes, Myron, 20–​21, 44–​45,
renormalization group theory 149–​50, 181n15
overview, 51–​58, 52f, 54f, 55f, 56f, 57f Schwert, G. William, 182n32
stochastic process application, 54f sciences
replicating portfolio, 97 analogies in, 175n14
returns, 65, 170n34, 174n27 computerization of, 60–​62, 61f
 

228 Index

scientific innovations, opportunity for, 96–​102, distribution; truncated stable Lévy

99f, 100f, 101f distribution
scientificity, 102–​3 Fama on, 39–​40
securities, 169n31, 170n45. See also Mandelbrot on, 33–​37
random walk non-​Gaussian stable Lévy processes
Seemann, Lars, 106 and, 72–​73
Seibold, Goetz, 183n40 power law characterizing, 176n31
self-​criticality theory, 132–​34 stable nonnormal process, 173n23
self-​organized criticality, 131–​34 stable Paretian distributions, 39–​40
separation theorem, 170n35 Standard & Poor’s 500 Index, 21–​22
SGF. See Statistique Générale de Standler, Ronald, 178n2
la France Stanford University, 169n25
Shalizi, Rohilla, 136–​37 Stanley, Eugene, 74, 145– 46, 148, 179n14,
Shannon, Donald, 38 180n24
Sharpe, William, 20 on econophysics, 59, 79, 81, 82, 178n1
Shefrin, Hersh, 183n37 on scaling invariance, 67–​68
Shiller, Robert 48 on universality class, 110
Shinohara, Shuji, 183n40 “The State of the Finance Field” (Weston), 12
short-​term time dependence, 184n11 Statistical Analysis System (SAS), 108
short-​term valuations, 3 statistical dependence, 170n44
Simkin, Mikhail, 176n34 statistical econophysics, xiii
Simon, Herbert, 94, 124, 177n37 statistical equilibrium, 180n28
Sims, Christopher, 179n21 statistical patterns, 181n23
single-​factor model, 20 statistical physics, 80, 179n18
size models, 177n40 application, 58–​60
skewness, 34 borders, 49–​77, 52f, 54f, 55f, 56f, 57f, 61f,
Slanina, František, 132 69f, 71f, 75f
Slutsky, Eugen, 8, 169n27, 169n30 golden age, 50–​58, 52f,
Sneppen, Kim, 183n40 54f, 55f, 56f, 57f
Society for Economic Science with methods, 58–​60
Heterogeneous Interacting Agents power laws’ role in, 63–​65
(ESHIA), 83 statistical tests, 19
sociophysics, 175n15. See also Statistique Générale de la France
econophysics (SDF), 30–​31
software, 108 Stauffer, Dietrich, xii, 132
sophisticated traders, 16–​17, 170n45 Steiger, William, 14–​15
Sornette, Didier, 66, 116, 149, 150, 180n26 Stigler, George, 130
source papers, 178n7 Stiglitz, Joseph, 19
Sprenkle, Case, 21, 32 Stinchcombe, Robin, 132
Sputnik era, 79 Stirling, James, 4
square root of time, law of, 3 stochastic modeling, 2–​4
stability, 66–​67, 173n22 stochastic processes, 171n57, 171n58
stable distributions package, 108 efficient-​market hypothesis and, 15–​20
stable Lévy processes, x–​xi, 38, 41, 49, 53–​55, power laws as linked with, 66
54f, 62, 66–​67, 74–​76, 75f, 91, 144–​145, stock market, 2–​4, 167n1. See also
173n3, 184n12. See also nonstable Lévy random walks
 

229  Index

stock market crash, 1929, 7–​8 overview, 72–​76, 75f

stock prices, 169n29, 172n9 of power law, 72–​76
stock price variations, 2–​4, 26–​29, 26f, 27f, Tversky, Amos, 129
28f, 169n30, 172n3, 179n19 two-​block model, 44–​45
stress tests, 113
string theory, 183n2 unconditional distributions, 47–​48, 100,
student distribution, 173n21 111–​12, 114, 179n13
subjectivism, 112 universality classes, 58, 64–​65,
system of markets, 171n53 68, 109–​10
University of Budapest, 82
Tan, Abby, 149–​50, 152 University of Chicago, 14, 32,
technical analysis, 179n23 170n43, 179n16
technical formations, 9 University of Houston, 83
temperature-​pressure phase University of Silesia, 83
diagram, 52f University of Warsaw, 83
Texas Instruments, 22 University of Wroclaw, 83
textbooks, on econophysics, 82 Upton, David, 38
Thaler, Richard, 129 upward, 3
Theiler, James, 137
theorems, 171n57 value-​at-​risk (VAR), 109, 179n21, 181n11
theoretical characteristic van der Vaart, Aad W., 73
function, 173n26 Vanguard 500 Index fund, 21–​22
Théorie de la spéculation (Bachelier), 4–​5 van Tassel, John, 32, 38
“Théorie mathématique du jeu” VAR. See value-​at-​risk
(Bachelier), 5 variability, 174n29
time, square root of, 3 variance, 184n11
time-​dependence dynamic, 174n34 variation correlation length, 175n13
time series, 181n13 vector autoregressions (VARs)
Tippett, Leonard, 169n26 modeling, 94
Tippett table, 169n26 visual linearity, 132–​33
Toronto Stock Exchange, 175n17 visual tests, 119, 134–​35, 183n42
traders, 16–​17, 116–​17, 132, 170n45 Vitanov, Nikolay, 177n36
trades, 182n34 volatility, 110–​15, 115t
trading, 107–​8 volatility clustering, 112
trading rooms, econophysics used by, 106–​17, von Smoluchowski, Marian, 6, 168n19
110f, 115t
transdisciplinarity, 121–​22, 147f Wall Street Journal, 168n10
transdisciplinary analysis, 147 Wang, Bing-​Hong, 132
Trans-​Lux Movie Ticker, 168n10 Weintraub, Robert, 14–​15
trends, 14–​15 Weston, Paul, 12
Treynor, Jack, 20, 129 Whitley, Richard, 92
tribes, 170n36 Widom, Benjamin, 53, 176n28
truncated stable Lévy distribution, 144, Wiener, Norbert, 6, 169n20
145, 148–​52 Wiener process. See Brownian motion
truncation, 158n51, 177n44, 177n46, Williams, John, 170n34
178n50, 178n52 Willis, J. C., 70
 

230 Index

Wilson, Kenneth Yale University, 172n6

Nobel Prize in Physics received by, 51–​58, Yoshikawa, Hiroshi, 128
52f, 54f, 55f, 56f, 57f Yule, George Udny, 70, 124, 125–​26
overview, 51–​58, 52f, 54f, 55f, 56f, 57f
Woodard, Ryan, 116 Zanin, Massimiliano, 137
Working, Holbrook, 169n25, 169n30, 170n44 Zhang, Yi-​Chen, 103
random-​walk model research of, 7–​8, Zhao, Xin, 181n11
9, 12–​13 Zipf, George, 70, 177n37, 177n38
on trends, 15 Zipf ’s Law, 70
Wyart, Matthieu, 130–​31