Econophysics: What can physicists

contribute to economics?
Parameswaran Gopikrishnan*, Luis A. Nunes Amaral*,
Yanhui Liu*, Martin Meyer*, Vasiliki Plerou1"*,
Bernd Rosenow**, and H. Eugene Stanley*.
* Center for Polymer Studies1 and Department of Physics,
Boston University, Boston MA 02215, USA.
t Department of Physics, Boston College, Chestnut Hill, MA 02617, USA.
Institut fur Theoretische Physik, Universitat zu Koln, D-50937 Koln, Germany.

Abstract. In recent years, a considerable number of physicists have started applying

physics concepts and methods to understand economic phenomena. The term "Econophysics" is sometimes used to describe this work. Economic fluctuations can have
many repercussions, and understanding fluctuations is a topic that many physicists
have contributed to in recent years. Further, economic systems are examples of complex interacting systems for which a huge amount of data exist and it is possible that
the experience gained by physicists in studying fluctuations in physical systems might
yield new results in economics. Much recent work in econophysics is focussed on understanding the peculiar statistical properties of price fluctuations in financial time series.
In this talk, we discuss three recent results. The first result concerns the probability
distribution of stock price fluctuations. This distribution decreases with increasing
fluctuations with a power-law tail well outside the Levy stable regime and describes
fluctuations that differ by as much as 8 orders of magnitude. Further, this non stable
distribution preserves its functional form for fluctuations on time scales that differ by
3 orders of magnitude, from 1 min up to approximately 10 days. The second result
concerns the accurate quantification of volatility correlations in financial time series.
While price fluctuations themselves have rapidly decaying correlations, the volatility
estimated by using either the absolute value or the square of the price fluctuations
has correlations that decay as a power-law and persist for several months. The third
result bears on the application of random matrix theory to understand the correlations
among price fluctuations of any two different stocks. We compare the statistics of the
cross-correlation matrix constructed from price fluctuations of the leading 1000 stocks
and a matrix with independent random elements, i.e., a random matrix. Contrary to
first expectations, we find little or no deviation from the universal predictions of random matrix theory for all but a few of the largest eigenvalues of the cross-correlation
matrix.

The Center for Polymer Studies is supported by the National Science Foundation.

CP511, Unsolved Problems of Noise and Fluctuations, edited by D. Abbott and L. B. Kish
2000 American Institute of Physics 1-56396-826-6/007$ 17.00

233

Downloaded 13 Jan 2009 to [Link]. Redistribution subject to AIP license or copyright; see [Link]

INTRODUCTION
The analysis of financial data using concepts and methods developed for physical systems has a long tradition [1-4] and has recently attracted the interest of
physicists [5-9]. Possible reasons for this interest include the scientific challenge
of understanding the dynamics of a strongly fluctuating complex system with a
large number of interacting elements. Moreover economic fluctuations could have
many repercussions and understanding fluctuations in physical systems is a topic
where many physicists have contributed. In addition, it is possible that the experience gained by studying fluctuations in physical systems might yield new results
in economics.
One can ask how physicists can contribute to the search for solutions to the puzzles posed by modern economics that economists themselves have not yet solved?
One approachin the spirit of experimental physicsis to begin empirically, with
real data that one can analyze in some detail, but without prior models. In economic systems such as financial markets, one has available a great deal of real data.
Moreover, if one has at one's disposal the tools of statistical physics and the computing power to carry out any number of approaches, this abundance of data is to
great advantage. Thus, for many physicists, studying the economy means studying
a wealth of data on a strongly fluctuating complex system. Indeed, physicists in increasing numbers are finding problems posed by economics sufficiently challenging
to engage their attention [10-26].
Recent studies attempt to uncover and explain the peculiar statistical properties of financial time series such as stock prices, stock market indices or currency
exchange rates. The dynamics of financial markets is difficult to understand not
only because of the complexity of its internal elements but also due to the many
intractable external factors acting on it, which may differ from market to market.
Remarkably, the statistical properties of certain observables appear to be similar
for quite different markets [31], consistent with the possibility that there may exist
"universal" mechanisms.
The most challenging difficulty in the study of financial markets is that the
nature of the interactions between the different elements comprising the system
is unknown, as is the way in which external factors affect it. Therefore, as a
starting point, one may resort to empirical studies to help uncover the regularities or
"empirical laws" that may govern financial markets [27]. The interactions between
the different elements comprising financial markets generate many observables such
as the transaction price, the share volume traded, the trading frequency, and the
values of market indices. Recent empirical studies are based on the analysis of
price fluctuations. This talk reviews recent results on (a) the distribution of stock
price fluctuations and its scaling properties, (b) time-correlations in financial time
series, and (c) correlations among the price fluctuations of different stocks. Space
limitations restrict us to focusing mainly on our group's work; a more balanced

234

Downloaded 13 Jan 2009 to [Link]. Redistribution subject to AIP license or copyright; see [Link]

account can be found in two recent books [5,6], other articles in these proceedings,
and two other recent international conferences [7,9]. Recent work in this field
also focuses on applications such as risk control, derivative pricing, and portfolio
selection [28], which shall not be discussed in this talk. The interested reader should
consult, for example, Refs. [5,6,29].

DISTRIBUTION OF PRICE FLUCTUATIONS

The recent availability of "high frequency" data allows one to study economic
time series on a wide range of time scales varying from seconds up to a few months.
For example, our recent work [18,40,41] involves the analysis of the S&P 500 index,
an index of the New York Stock Exchange that consists of 500 companies representative of the US economy. It is a market-value (stock price times number of
shares outstanding) weighted index, with each stock's weight in the index proportionate to its market value [18]. The S&P 500 index is one of the most widely used
benchmarks of U.S. equity performance. We analyzed high frequency data for the
13-year period 1984-96 with a recording frequency of one minute or shorter and the
daily records for the 35-year period 1962-96 (Fig. la).
The S&P 500 index Z(t) from 1962-96 has an overall upward driftinterrupted
by drastic events such as the market crash of October 19, 1987 (Fig. la). One
analyzes the difference in logarithm of the index, often called the return G(t) =
loge Z(t + At) logeZ(t), where At is the time scale investigated (Fig. Ib). One
only counts the number of minutes during the opening hours of the stock market. It

10

CO
C

o
o
IT)

Q.
$

1Q2

A(i f

. x\ /\ <A/^ * G^=ln s(t+M)/s(t)

(a>:

(b) S&P 500 (10min data) ;

^UWI^Y^

3 -is

r LJ'

(c) S&P 500 (monthly data)

0 v^i^
-15

(d) Gaussian noise

0 h<^^^^^

Time

Year

FIGURE 1. (a) The daily records of the S&P 500 index for the 35-year period 1962-96 on a
linear-log scale. Note the large jump which occured during the market crash of October 19, 1987.
Sequence of (b) 10 min returns and (c) 1 month returns of the S&P 500 index, normalized to
unit variance, (d) Sequence of i.i.d. Gaussian random variables with unit variance, which was
proposed by Bachelier as a model for stock returns [1]. For all 3 panels, there are 850 events
i.e., in panel (b) 850 minutes and in panel (c) 850 months. Note that, in contrast to (b) and
(c), there are no large events in (d).

235

Downloaded 13 Jan 2009 to [Link]. Redistribution subject to AIP license or copyright; see [Link]

is apparent from Fig. Ib that when one analyzes returns on short time scales, large
events are much more likely to occur, in contrast to a sequence of Gaussian distributed random numbers of the same variance (Fig. Id). As one analyzes returns
on larger time scales, this difference is apparently much less pronounced (Fig. Ic).
In order to understand this process, one starts by analyzing the probability distribution of returns on a given time scale At, which in our study, varies from 1 min
up to a few months.
The nature of the distribution of price fluctuations in financial time series is a
long standing open problem in finance which dates back to the turn of the century.
In 1900, Bachelier proposed the first model for the stochastic process of returns
an uncorrelated random walk with independent, identically Gaussian distributed
(i.i.d) random variables [1]. This model is natural if one considers the return over
a time scale At to be the result of many independent "shocks", which then lead
by the central limit theorem to a Gaussian distribution of returns [1]. However,
empirical studies [4,18,19] show that the distribution of returns has pronounced
tails in striking contrast to that of a Gaussian. Despite this empirical fact, the
Gaussian assumption for the distribution of returns is widely used in theoretical
finance because of the simplifications it provides in analytical calculation; indeed,
it is one of the assumptions used in the classic Black-Scholes option pricing formula
[30].
In his pioneering analysis of cotton prices, Mandelbrot observed that in addition
to being non-Gaussian, the process of returns shows another interesting property:
"time scaling" that is, the distributions of returns for various choices of At,
ranging from 1 day up to 1 month have similar functional forms [4]. Motivated
by (i) pronounced tails, and (ii) a stable functional form for different time scales,
Mandelbrot [4] proposed that the distribution of returns is consistent with a Levy
stable distribution [2,3].
Conclusive results on the distribution of returns are difficult to obtain, and require a large amount of data to study the rare events that give rise to the tails.
More recently, the availability of high frequency data on financial market indices,
and the advent of improved computing capabilities, has facilitated the probing of
the asymptotic behavior of the distribution. For example, Mantegna and Stanley [18] analyzed approximately 1 million records of the S&P 500 index. They
report that the central part of the distribution of S&P 500 returns appears to be
well fit by a Levy distribution, but the asymptotic behavior of the distribution of
returns shows faster decay than predicted by a Levy distribution. Hence, Ref. [18]
proposed a truncated Levy distributiona Levy distribution in the central part followed by an approximately exponential truncationas a model for the distribution
of returns. The exponential truncation ensures the existence of a finite second moment, and hence the truncated Levy distribution is not a stable distribution [32,33].
The truncated Levy process with i.i.d. random variables has slow convergence to
Gaussian behavior due to the Levy distribution in the center, which could explain
the observed time scaling for a considerable range of time scales [18].
Recent studies [34,35] on considerably larger time series using larger databases

236

Downloaded 13 Jan 2009 to [Link]. Redistribution subject to AIP license or copyright; see [Link]

show quite different asymptotic behavior for the distribution of returns. Our recent
work [34] analyzed three different data bases covering securities from the three
major US stock markets. In total, we analyzed approximately 40 million records of
stock prices sampled at 5 min intervals for the 1000 leading US stocks for the 2-year
period 1994-95 and 35 million daily records for 16,000 US stocks for the 35-year
period 1962-96. We study the probability distribution of returns (Fig. 2(a,b,c)) for
individual stocks over a time interval At, where At varies approximately over a
10U
c
~ 10~1

Livy Regime ,
a; 17

Postive taila'3
o Negative tail

(a)
1

10

100

Normalized S&P500 returns

10

100

Normalized S&P500 returns

10

100

Normalized returns
FIGURE 2. (a) Log-log plot of the cumulative distribution of the normalized 1 min returns for
the S&P 500 index. Power-law regression fits in the region 3 < g < 50 yield a = 2.95 0.07
(positive tail), and a = 2.75 0.13 (negative tail). For the region 0.5 < g < 3, regression
fits give a = 1.6 0.1 (positive tail), and a = 1.7 0.1 (negative tail), (b) Log-log plot of
the cumulative distribution of normalized returns of the S&P 500 index. The positive tails are
shown for At = 16, 32,128, 512 mins. Power-law regression fits yield estimates of the asymptotic
power-law exponent a = 2.69 0.04, a = 2.53 0.06, a = 2.83 0.18 and a = 3.39 0.03 for
At = 16,32,128 and 512 mins, respectively, (c) The positive and negative tails of the cumulative
distribution of the normalized returns of the 1000 largest companies in the TAQ database for the
2-year period 1994-1995. The solid line is a power-law regression fit in the region 2 < x < 80.

237

Downloaded 13 Jan 2009 to [Link]. Redistribution subject to AIP license or copyright; see [Link]

factor of 104from 1 min up to more than 1 month. We also conduct a parallel

study of the S&P 500 index.
Our key finding is that the cumulative distribution of returns for both individual
companies (Fig. 2c) and the S&P 500 index (Fig. 2a) can be well described by a
power law asymptotic behavior, characterized by an exponent a ~ 3, well outside
the stable Levy regime 0 < a < 2. Further, it is found that the distribution,
although not a stable distribution, retains its functional form for time scales up to
approximately 16 days for individual stocks and approximately 4 days for the S&P
500 index, Fig. 2b. For larger time scales our results are consistent with breakdown of scaling behavior, i.e., convergence to Gaussian [34]. Similar results have
also been found for currency exchange data [35].

CORRELATIONS IN FINANCIAL TIME SERIES

In addition to the probability distribution, an aspect of equal importance for
the characterization of any stochastic process is the quantification of correlations.
Studies of the autocorrelation function of the returns show exponential decay with
characteristic decay times of only 4 min [36] consistent with the efficient market
hypothesis [37]. This is paradoxical, for in the previous section, we have seen that
the distribution of returns, in spite of being a non-stable distribution, preserves
its shape for a wide range of At. Hence, there has to be some sort of correlations
or dependencies that prevent the central limit theorem to take over sooner and
preserve the scaling behavior.
Indeed, lack of linear correlation does not imply independent returns, since there
may exist higher-order correlations. Recently, Liu and his collaborators found that
the amplitude of the returns, the absolute value or the square closely related
to what is referred to in economics as the volatility [38] shows long-range
correlations [21,23,39-42] with persistence [43] up to several months, Fig. 3(a,b).
They analyzed the correlations in the absolute value of the returns [40,41] of the
SfeP 500 index using traditional correlation function estimates, power spectrum and
the recently-developed detrended fluctuation analysis (DFA). All the three methods
show the existence of power-law correlations with a cross-over at approximately
1.5 days. For the S&P 500 index, DFA estimates for the exponents characterizing
the power law correlations are OL\ 0.66 for short time scales smaller than
1.5 days and a?2 = 0.93 for longer time scales up to a year, Fig. 3b. For individual
companies, the same methods yield OL\ = 0.60 and a<2 0.74, respectively. The
power spectrum gives consistent estimates of the two power-law exponents, Fig. 3a.
The long memory in the amplitude of returns suggests that it is useful to define
a subsidiary process, referred to as the volatility. Volatility of a certain stock
measures how much it is likely to fluctuate. It can also be related to the amount of
information arriving at any time. The volatility can be estimated for example by
the local average of the absolute values or the squares of the returns. In their recent
work on the statistical properties of volatility Liu et. al. [41,42] show that the

238

Downloaded 13 Jan 2009 to [Link]. Redistribution subject to AIP license or copyright; see [Link]

(b)

10 .

one day

(a)

10

A A A A ^ A A A A A A A A .

one day

a2=0.93

|io-

O
0)
Q.

2 io~4

10
P2=0.90

210'

0)

I ,

Q. 10

10
10

10

10

10

10

10

102

103

104

105

t [min]

/[min' ]

FIGURE 3. Plot of (a) the power spectrum S(f) and (b) the detrended fluctuation analysis
F(t) of the absolute values of returns #(), after detrending the daily pattern [40,41] with the
sampling time interval A = 1 min. The lines show the best power law fits (R values are better
than 0.99) above and below the crossover frequency of /x = (1/570) min"1 in (a) and of the
crossover time, tx = 600min in (b). The triangles show the power spectrum and DFA results for
the "control", i.e., shuffled data.

volatility correlations show asymptotic I// behavior [40-42]. Using the same data
bases as above, Liu and his collaborators also study the cumulative distribution of
volatility [40,42] and find that it is consistent with a power-law asymptotic behavior,
characterized by an exponent JJL w 3, just the same as that for the distribution of
returns. For individual companies also, one finds a similar power law asymptotic
behavior [41]. In addition, it is also found that the volatility distribution scales for
a range of time intervals just as the distribution of returns.

CORRELATIONS AMONG DIFFERENT UNITS

Recently, the problem of understanding the correlations among the returns of
different stocks has been addressed by applying methods of random matrix theory
to the cross-correlation matrix [46,47]. Aside from scientific interest, the study of
correlations between the returns of different stocks is also of practical relevance in
quantifying the risk of a given portfolio [28]. Consider, for example, the equaltime correlation of stock returns for a given pair of companies. Since the market
conditions may not be stationary, and the historical records are finite, it is not clear
if a measured correlation of returns of two stocks is just due to "noise" or genuinely
arises from the interactions among the two companies. Moreover, unlike most
physical systems, there is no "algorithm" to calculate the "interaction strength"
between two companies (as there is for, say, two spins in a magnet). The problem is
that although every pair of companies should interact either directly or indirectly,

239

Downloaded 13 Jan 2009 to [Link]. Redistribution subject to AIP license or copyright; see [Link]

the precise nature of interaction is unknown.

In some ways, the problem of interpreting the correlations between individual
stock-returns is reminiscent of the difficulties experienced by physicists in the fifties,
in interpreting the spectra of complex nuclei. Large amounts of spectroscopic data
on the energy levels were becoming available but were too complex to be explained
by model calculations because the exact nature of the interactions were unknown.
Random matrix theory (RMT) was developed in this context, to deal with the
statistics of energy levels of complex quantum systems [48-50]. With the minimal assumption of a random Hamiltonian, given by a real symmetric matrix with
independent random elements, a series of remarkable predictions were made and
successfully tested on the spectra of complex nuclei [48]. RMT predictions represent an average over all possible interactions [49]. Deviations from the universal
predictions of RMT identify system-specific, non-random properties of the system
under consideration, providing clues about the underlying interactions [50]
Recently, Plerou and her collaborators analyzed the cross-correlation matrix C
= dj = (GiGj) (Gi)(Gj)/ai<jj of the returns at 30-minute intervals of the largest
1000 US stocks for the 2-year period 1994-95. They analyze the statistical properties of C by applying techniques of random matrix theory (RMT) [46,47]. First, they
test the eigenvalue statistics of the cross-correlation matrix for universal properties
of real symmetric random matrices such as the Wigner surmise for the eigenvalue
spacing distribution and eigenvalue correlations. Remarkably, they find that eigenvalue statistics of the correlation matrix agree well with the universal predictions
of random matrix theory for real symmetric random matrices, in contrast to our
naive expectations for a strongly interacting system.
Deviations from RMT predictions represent genuine correlations. In order to
investigate deviations, we compute the distribution of the eigenvalues of the C and
compare with the prediction [46] for uncorrelated time series [51]. We find that the
statistics of all but a few of the largest eigenvalues in the spectrum of C agree with
the predictions of random matrix theory, but there are deviations for a few of the
largest eigenvalues [46,47], The deviations of the largest few eigenvalues from the
random matrix result are also found when one analyzes the distribution of eigenvector components. Specifically, the largest eigenvalue which deviates significantly (25
times larger than random matrix bound) has almost all components participating
equally and thus represents the correlations that pervade through the entire market. This result is in agreement with the results of Laloux and collaborators [46]
for the eigenvalue distribution of C on a daily time scale.

CONCLUSIONS AND OPEN QUESTIONS

Econophysics is a field wherein questions are as difficult to pose as answers. The
empirical results shown above clearly beckon explanation. For example, in first
two sections, we have looked mainly at two empirical results: (i) the distribution
of fluctuations, which shows a power law behavior well outside the stable Levy

240

Downloaded 13 Jan 2009 to [Link]. Redistribution subject to AIP license or copyright; see [Link]

regime, and yet preserves its shapescalesfor a range of time scales and (ii) the
long range correlations in the amplitude of price fluctuations. How are the two
related ?
Previous explanations of scaling relied on Levy stable [4] and exponentiallytruncated Levy processes [6,18]. However, the empirical data that we analyze
are not consistent with either of these two processes. In order to confirm that
the scaling is not due to a stable distribution, one can randomize the time series
of 1 min returns, thereby creating a new time series which contains statisticallyindependent returns. By adding up n consecutive returns of the shuffled series, one
can construct the nmin returns. Both the distribution and its moments show a
rapid convergence to Gaussian behavior with increasing n, showing that the time
dependencies, specifically volatility correlations are intimately connected to the
observed scaling behavior [34].
Using the statistical properties summarized above, can we attempt to deduce a
statistical description of the process which gives rise to this output? For example,
the standard ARCH model [27,44] reproduces the power-law distribution of returns;
however it assumes finite memory on past events and hence is not consistent with
long-range correlations in volatility. On the other hand, the distribution of volatility and that of returns which have similar asymptotic behavior, however support
the central ARCH hypothesis that g(t) = tv(t), where e is an i.i.d. Gaussian
random variable independent of the volatility v ( t ] , and g(t) denotes the returns.
A consistent statistical description may involve extending the traditional ARCH
model to include long-range volatility correlations [45].
A more fundamental question would be to understand the above results starting
from a microscopic setting. Researchers have also studied microscopic models that
might give rise to the empirically observed statistical properties of returns [5,10].
For example, Lux and Marchesi [10] recently simulated a microscopic model of
financial markets with two types of traders, what they refer to as 'fundamentalist'
and 'noise' traders. Their results reproduce the power-law tail for the distribution
of returns and also the long range correlations in volatility.
In the last section, we found evidence for different modes of correlations between
different companies. For example, the largest eigenvalue of the cross-correlation
matrix showed correlations that pervade the entire market. Could it be that the
above observed scaling properties are related to how correlations propogate from
one unit to the other such as occur in critical phenomena? Researchers have studied
economic data from the physics perspective of a complex system with each unit
depending on the other. Specifically, the possibility that all the companies in a
given economy might interact, more or less, like a spin glass. In a spin glass, each
spin interacts with every other spinbut not with the same coupling and not even
with the same sign. For example, if the stock price of a given business firm A
decrease by, e.g., 10% , this will have an impact in the economy. Some of these will
be favorablefirm B, which competes with A, may experience an increase in market
share. Others will be negativeservice industries that provide personal services for
firm A employees may experience a drop-off in sales as employee salaries will surely

241

Downloaded 13 Jan 2009 to [Link]. Redistribution subject to AIP license or copyright; see [Link]

decline. There must be positive and negative correlations for almost any economic
change. Can we view the economy as a complicated spin glass?
To approch this problem, M. H. R. Stanley and M. A. Salinger first located and
secured a databasecalled COMPUSTATthat lists the annual sales of every firm
in the United States. With this information, M. H. R. Stanley and co-workers calculated histograms of how firm sizes change from one year to the next [52]. They
find that the distribution of growth rates of firm sales has the same functional

form regardless of industry or market capitalization. Moreover, the width of these

distributions decrease with increasing sales as a power-law with an exponent ap-

proximately 1/6. Recently, similar statistical properties were found for the GDP of
countries [53] and for university research fundings [54]. Hence, it is not impossible
to imagine that there are some very general principles of complex organizations
at work here, because similar empirical laws appear to hold for data on a range of

systems that at first sight might not seem to be so closely related. Buldyrev models
this firm structure as an approximate Cayley tree, in which each subunit of a firm
reacts to its directives from above with a certain probability distribution [55]. More
recently, Amaral et al. [56] have proposed a microscopic model that reproduces both
the exponent and the distribution function. Takayasu and Okuyama [57] extended

the empirical results to a wide range of countries.

We conclude by thanking all our collaborators and colleagues from whom we
learned a great deal. These include the researchers and faculty visitors to our
research group with whom we have enjoyed the pleasure of scientific collaboration. Those whose research provided the basis of this short report include: S. V.

Buldyrev, D. Canning, P. Cizeau, X. Gabaix, S. Havlin, P. Ch. Ivanov, R. N. Mantegna, C.-K. Peng, M. A. Salinger, and M. H. R. Stanley. We also thank M.
Barthelemy, J.-P. Bouchaud, D. Sornette, D. Stauffer, S. Solomon, and J. Voit for
helpful dicussions and comments.

REFERENCES
1. L. Bachelier, Ann. Sci. Ecole Norm. Sup. 3, 21 (1900).
2. V. Pareto, Cours d'Economie Politique, Lausanne and Paris, 1897.
3. P. Levy, Theorie de I'Addition des Variables Aleatoires, Gauthier-Villars, Paris,
1937.
4. B. B. Mandelbrot, J. Business 36, 294 (1963).
5. J. P. Bouchaud and M. Potters, Theorie des Risques Financieres, Alea-Saclay, Eyrolles, 1998.
6. R. N. Mantegna and H. E. Stanley, An Introduction to Econophysics: Correlations
and Complexity in Finance, Cambridge University Press, Cambridge 1999.
7. I. Kondor and J. Kertesz (eds) Econophysics: An Emerging Science, Kluwer, Dordrecht, 1999.
8. R. N. Mantegna (ed), Proceedings of the International Workshop on Econophysics
Statistical Finance, Physica A [special issue] 269, 1 (1999).

242

Downloaded 13 Jan 2009 to [Link]. Redistribution subject to AIP license or copyright; see [Link]

9. K. B. Lauritsen (ed), Application of Physics in Financial Analysis, Int. J. Theor.

Appl. Finance [special issue] xx (1999).
10. T. Lux and M. Marchesi, Nature 297, 498 (1999); T. Lux, J. Econ. Behav. Organizat.
33, 143 (1998) ; J. Econ. Dyn. Control 22, 1 (1997); Appl Econ. Lett. 3, 701 (1996).
11. M. Levy, H. Levy and S. Solomon, Economics Letters 45, 103 (1994); M. Levy and
S. Solomon, Int. J. Mod. Phys (77, 65 (1996).
12. J.-P. Bouchaud and R. Cont, Eur. Phys. J. B 6, 543 (1998); R. Cont and J.-P.
Bouchaud, cond-mat/9712318..
13. M. Potters, R. Cont, and J.-P. Bouchaud, Europhys. Lett. 41, 239 (1998).
14. J.-P. Bouchaud and D. Sornette, J. Phys. /(France) 4, 863 (1994); D. Sornette, A.
Johansen, and J.-P. Bouchaud, J. Phys. /(France) 6, 167 (1996); D. Sornette and A.
Johansen, Physica A 261, 581 (1998); D. Sornette, Physica A 256, 251 (1998); A.
Johansen and D. Sornette, Risk I , 91 (1999), cond-mat/9901035, cond-mat/9811292.
15. D. Stauffer, Ann. Phys.-Berlin 7, 529 (1998); D. Stauffer and T. J. P. Penna, Physica
A 256, 284 (1998); D. Chowdhury and D. Stauffer, Eur. Phys. J. B 8, 477 (1999);
I. Chang and D. Stauffer, Physica A 264 (1999) 1; D. Stauffer and T. J. P. Penna,
Physica A 256, 284 (1998); D. Stauffer, P. M. C. de Oliveria and A. T. Bernardes,
Int. J. Theor. Appl. Finance 2, 83 (1999).
16. M. Marsili and Y.-C. Zhang, Phys. Rev. Lett. 80, 2741 (1998); G. Caldarelli, M.
Marsili, and Y.-C. Zhang, Europhys. Lett. 40, 479 (1997); S. Galluccio, G. Calderelli,
M. Marsili, and Y.-C. Zhang, Physica A 245, 423 (1997); M. Marsili, S. Maslov, and
Y.-C. Zhang, Physica A 253, 403 (1998); S. Maslov and Y.-C. Zhang, Physica A
262, 232 (1999); D. Challet and Y. C. Zhang, Physica A 256, 514 (1998).
17. P. Bak, M. Paczuski and M. Shubik, Physica A 246, 430 (1997).
18. R. N. Mantegna and H. E. Stanley, Nature 376, 46 (1995).
19. S. Ghashghaie, W. Breymann, J. Peinke, P. Talkner, and Y. Dodge, Nature 381, 767
(1996); see also R. N. Mantegna and H. E. Stanley, Nature 383, 587 (1996); Physica
A 239, 255 (1997).
20. A. Arneodo, J.-F. Muzy and D. Sornette, Eur. Phys. J. B 2, 277 (1998).
21. M. Pasquini and M. Serva, cond-mat/9810232; cond-mat/9903334.
22. M. M. Dacorogna, U. A. Muller, R. J. Nagler, R. B. Olsen and O. V. Pictet, J.
International Money and Finance 12, 413 (1993); see also the Olsen website at
[Link] for related papers.
23. R. Cont, Statistical Finance: Empirical study and theoretical modeling of price
variations in financial markets, PhD thesis, Universite de Paris XI, 1998; (condmat/9705075); R. Cont, M. Potters, and J.-P. Bouchaud, in Scale Invariance and
Beyond, edited by B. Dubrulle, F. Graner, and D. Sornette, Springer, Berlin, 1997.
24. N. Vandewalle and M. Ausloos, Int. J. Mod. Phys. (79, 711 (1998); Eur. Phys. J. B
4, 257 (1998); N. Vandewalle, P. Boveroux, A. Minguet, and M. Ausloos, Physica A
255, 201 (1998).
25. H. Takayasu, A. H. Sato, and M. Takayasu, Phys. Rev. Lett. 79, 966 (1997); H.
Takayasu, H. Miura, T. Hirabayashi, and K. Hamada, Physica A 184, 127 (1992).
26. K. N. Ilinski and A. S. Stepanenko, cond-mat/9806138; cond-mat/9902046;
K. N. Ilinski, cond-mat/9903142.
27. A. Pagan, J. Empirical Finance 3, 15 (1996).

243

Downloaded 13 Jan 2009 to [Link]. Redistribution subject to AIP license or copyright; see [Link]

28. G. W. Kirn and H. M. Markowitz, J. Portfolio Management 16, 45 (1989); E. J. Elton

and M. J. Gruber, Modern Portfolio Theory and Investment Analysis, J. Wiley, New
York, 1995.
29. J.-P. Bouchaud, cond-mat/9806101; D. F. Wang, cond-mat/9807066, condmat/9809045; J.-P. Bouchaud, G. lori, D. Sornette, Risk 93, 61 (March 1996);
M. Rosa-Clot, S. Taddei, cond-mat/9901277, cond-mat/9901279; B. E. Baaquie,
cond-mat/9708178; R. Cont, cond-mat/9902018.
30. F. Black and M. Scholes, J. Political Economy 81, 637 (1973).
31. R. N. Mantegna, Physica A 179, 232 (1991); J. Skjeltrop, University of Oslo Thesis,
Oslo, Norway, 1998.
32. R. N. Mantegna and H. E. Stanley, Phys. Rev. Lett. 73, 2946 (1994).
33. I. Koponen, Phys. Rev. 52, 1197 (1995).
34. P. Gopikrishnan, M. Meyer, L. A. N. Amaral and H. E. Stanley, Eur. Phys. J. B,
3 139 (1998), cond-mat/9803374; P. Gopikrishnan, V. Plerou, L. A. N. Amaral,
M. Meyer and H. E. Stanley, cond-mat/9905305, Phys. Rev. E., submitted.
35. M. M. Dacorogna, U. A. Muller, O. V. Pictet, and C.G. de Vries (Olsen group,
[Link], 199x).
36. Z. Ding, C. W. J. Granger and R. F. Engle, J. Empirical Finance 1, 83 (1983); For
foreign exchange markets the decay time is even smaller [22].
37. E.-F. Fama, J. Finance 25, 383 (1970).
38. T. Bollerslev, R. Y. Chou, and K. F. Kroner, J. Econometrics 52, 5 (1992); G. W.
Schwert, J. Finance 44, 1115 (1989); A. R. Gallant, P. E. Rossi and G. Tauchen,
Review of Financial Studies 5, 199 (1992); B. Le Baron, J. Business 65, 199 (1992);
K. Chan, K. C. Chan and G. A. Karolyi, Review of Financial Studies 4, 657 (1991).
39. C. W. J. Granger and Z. Ding, J. Econometrics 73, 61 (1996).
40. Y. Liu, P. Cizeau, M. Meyer, C.-K. Peng, and H. E. Stanley, Physica A 245, 437

(1997).
41. Y. Liu, P. Gopikrishnan, P. Cizeau, M. Meyer, C.-K. Peng and H. E. Stanley, Phys.
Rev. 60, xxx (1999), cond-mat/9903369.
42. P. Cizeau, Y. Liu, M. Meyer, C.-K. Peng, and H. E. Stanley, Physica A 245, 441

(1997).
43.
44.
45.
46.

47.
48.
49.
50.

J. Beran, Statistics for Long-Memory Processes (Chapman & Hall, NY, 1994).
R. F. Engle, Econometrica 50, 987 (1982).
R. Bailie, J. Econometrics 17, 223 (1991) .
S. Galluccio, J.-P. Bouchaud, and M. Potters, Physica A 259, 449 (1998); L. Laloux,
P. Cizeau, J.-P. Bouchaud and M. Potters, cond-mat/9810255; Risk 3, 69 (March,
1999).
V. Plerou, P. Gopikrishnan, B. Rosenow, L. A. N. Amaral, and H. E. Stanley, condmat/9903369.
E. P. Wigner, Ann. Math. 53, 36 (1951); E. P. Wigner, in Conference on Neutron
Physics by Time-of-flight (Gatlinburg, Tennessee, 1956).
F. J. Dyson, J. Math. Phys. 3, 140 (1962); F. J. Dyson and M. L. Mehta, J. Math.
Phys. 4, 701 (1963); M. L. Mehta and F. J. Dyson, J. Math. Phys. 4, 713 (1963).
T. Guhr, A. Miiller-Groeling, and H. A. Weidenmiiller, Phys. Rep. 299, 190 (1998);
M. L. Mehta, Random Matrices, Academic Press, Boston, 1991.

244

Downloaded 13 Jan 2009 to [Link]. Redistribution subject to AIP license or copyright; see [Link]

51. A. M. Sengupta and P. P. Mitra, cond-mat/9709283.

52. M. H. R. Stanley, L. A. N. Amaral, S. V. Buldyrev, S. Havlin, H. Leschhorn, P.
Maass, M. A. Salinger, and H. E. Stanley, Nature 379, 804 (1996); L. A. N. Amaral,
S. V. Buldyrev, S. Havlin, H. Leschhorn, P. Maass, M. A. Salinger, H. E. Stanley,
and M. H. R. Stanley, J. Phys. /(France) 7, 621 (1997).
53. Y. Lee, L. A. N. Amaral, D. Canning, M. Meyer, and H. E. Stanley, Phys. Rev. Lett.
81, 3275 (1998).
54. V. Plerou, L. A. N. Amaral, P. Gopikrishnan, M. Meyer, and H. E. Stanley, Nature,
in press, cond-mat/9906229.
55. S. V. Buldyrev, L. A. N. Amaral, S. Havlin, H. Leschhorn, P. Maass, M. A. Salinger,
H. E. Stanley and M. H. R. Stanley, J. Phys. /(France) 7, 635 (1997).
56. L. A. N. Amaral, S. V. Buldyrev, S. Havlin, M. A. Salinger and H. E. Stanley, Phys.
Rev. Lett. 80, 1385 (1998).
57. H. Takayasu and K. Okuyama, Fractals 6, 67 (1998).

245

Downloaded 13 Jan 2009 to [Link]. Redistribution subject to AIP license or copyright; see [Link]