f briefing paper
No. 3 / 15 march 2012
Agent Based Models
A New Tool for Economic and Policy Analysis
Mauro Napoletano, Jean-Luc Gaffard et Zakaria Babutsidze
A re current economic models well equipped to provide useful policy prescrip-
tions? Many economists would have certainly answered, “yes” before the recent
Global Recession. This economic crisis has not only demonstrated the importance of
banking and financial markets for the dynamics of real economies. It has also revealed
the inadequacy of the dominant theoretical framework. Standard models have indeed
failed to forecast the advent of the crisis. In addition, they have been unable to indicate
a therapy able to restore economic growth.
Since the onset of the crisis, the discontent towards the dominant approach to
economic modeling has flourished.1 Criticism has been mainly directed towards the
over-simplicity of standard models. Most features that have played a key role in gene-
rating the crisis, such as heterogeneity of agents, markets, and regulatory frameworks,
financial innovation, securitization, are by and large overlooked by standard macro-
models. A second kind of dissatisfaction is related to the hyper-rationality of indivi-
duals in standard models. Real markets (and financial markets in particular), are plenty
of people acting on the basis of overconfidence, heterogeneous beliefs, imperfect
knowledge of the states of the world, and of the consequence of humans’ actions, etc.
These features are not present in standard macro models, which typically build on the
assumption of a representative individual knowing all the characteristics of the
economy and able to replicate whatever human intelligence can do (Leijonhufvud,
1993). A third concern is the assumption of equilibrium. Standard models typically
focus on states of the economy in which all markets clear. In contrast, the crisis has
shown the possibility of situations in which some markets (and the market for labor in
particular) do not clear. Standard models ignore the problems that would result from
reactions of agents to such market disequilibria. They are therefore badly equipped to
study how the economy behaves during crises.
1. Interestingly, this time critiques have not only come from “heterodox” schools of thought. They were also
raised by more orthodox scholars, who made a significant use of the ingredients of standard models in the past
(see e.g. Caballero, 2010, Krugman, 2009, Stiglitz, 2011). Moreover, the discussion about the adequacy of
standard model has spread over the academia and it has also touched policy making authorities (see e.g. Trichet,
2010).
This note will provide a concise account of the current criticism towards standard
models. At the same time, it will describe an alternative class of models, which has
received increased attention in the recent years: Agent Based Models (ABMs). These
new models characterize economic processes as dynamical systems of heterogeneous
agents. The approach followed by ABMs is to a large extent opposite to the standard
one. This is because heterogeneity, bounded rationality of agents and market disequili-
bria are explicitly taken into account. This feature alone makes these models a
promising research tool for understanding the situations and problems emerged with
the recent crisis.
The note is organized as follows. Section 1 briefly reviews the critiques raised
towards standard models. Section 2 is devoted to present Agent Based Models, their
strengths and limits vis-à-vis standard models. Section 3, contains a brief overview of
the recent contributions to the analysis of macroeconomic policies in Agent Based
Models. Finally, Section 4 concludes.
1. The case against standard macroeconomic models
Today most macroeconomic theories are microfounded, that is they are grounded
on an account of what typical agents do and why they do it. Macroeconomic aggre-
gates (such as GDP, unemployment) result from the decisions of the millions of
individuals populating the economy. Thus, it seems quite reasonable to study the
dynamics of those aggregates by starting from the characterization of the behavior of
single agents at the microeconomic level. There is also another important motivation
justifying the use of microfoundations in macroeconomics. The design of policies
grounded on macroeconomic models not derived from first principles at the micro-
level can be seriously flawed. This is because the reaction of agents to a given policy
(e.g. lowering taxes) will change the parameters of the model used when the effects of
the same policy were estimated. Thus without an understanding of such reactions, the
desired results of a policy intervention can be invalidated2.
The type of models that have become the standard today in macroeconomics3 –
namely the Dynamic Stochastic General Equilibrium Models (DSGEs) – faithfully
adheres to the “microfoundations approach”. This approach looks sound and uncon-
troversial, so at first glance no problem arises. However, the worries are related to the
type of microfoundations adopted by DSGE models.4 Let us start by briefly examine
their main building blocks. First, DSGE models explain aggregate regularities in a
“general equilibrium” framework, i.e. assuming that every market clears. Second, the
behaviors of agents populating the economy are approximated via the introduction of
a representative agent (Representative Agent Hypothesis, RAH), i.e. it is assumed that
all agents in the economy are equal with respect to their preferences and characteris-
tics.5 Third, the choices of the representative agent are derived from a constrained
2. This is the essence of the famous « Lucas Critique » to earlier Keynesian general aggregative models, see
Lucas (1976).
3. In what follows we shall use the term “DSGE Model” and “Standard Model”, interchangeably.
4. This discussion of the criticism against standard models extensively builds on Stiglitz (2011). See also
Caballero, (2010), Krugman, (2009), Kirman, (2010), Colander et al, (2009), and Leijonhufvud (2011) for other
recent accounts of the fallacies of DSGE models in macroeconomics.
5. A more refined version of this assumption consist in the statement that only the first moments (i.e. the mean)
of the distributions of agents preferences and characteristics matter at the macro level.
2 briefing paper no. 3 / 15 march 2012
inter-temporal maximization problem, assuming that the agent has rational expecta-
tions (Rational Expectations Hypothesis, REH), i.e. assuming that he knows the true
underlying model of the economy and that he makes an efficient use of all information
available6.
There are a number of reasons to believe that the above features make DSGE
models badly equipped to investigate relevant economic issues, especially those
emerged with the recent crisis. Consider for instance the provision of credit and the
consequences of credit default and bankruptcy, i.e. elements that have been at the
core of the causes of the crisis and of the diffusion of its adverse effects. The investiga-
tion of credit contracts needs to take into account the presence of information
asymmetries between two distinct agents, the borrower and the lender. By assuming a
representative agent the analysis of informational asymmetries is not possible.7
Actually, even the credit markets are not conceivable (who is lending to whom?). As a
consequence, there is no scope for credit default and bankruptcy. Likewise, the event
of bankruptcy implies the distribution of losses among different actors (e.g. in the case
of a bank, between depositors, shareholders and the state). How is it possible to
account for these in a model if the actor is unique? 8
More in general, the recent crisis has shown that distributions matters. One instance
of this is the market for credit, where the distribution of information between
borrowers and lenders plays a key role. However, the crisis and the associated surge in
unemployment also generated sharp inequalities within the population: some indivi-
duals have seen their income falling either because they got unemployed or because of
falling wages in a situation with depressed labor demand. Reduced incomes by a signi-
ficant fraction of the population would normally lead to a fall in aggregate
consumption (especially in presence of frozen credit markets). However, this cannot
happen by assumption in a representative agent framework. Since the agent is at the
same time worker and owner of the companies, what he gets less in terms of wages is
compensated by what he gets in the form of higher profits of the company.9
Finally, many explanations of the emergence of the crisis have emphasized the role
played by the wide inequalities that already existed before the crisis (see e.g. Fitoussi
and Saraceno, 2010). All advanced countries have indeed been characterized by an
increase in inequality in income and wealth since the 80s. In many countries (e.g. the
US), such inequalities have not mapped into similar inequalities in consumption,
thanks to the credit available to poor households (see e.g. Iacoviello, 2005, and
Krueger and Perri, 2006). However, it is precisely the excessive indebtedness of the
latter that ultimately originated the credit defaults triggering the crisis. Nevertheless, it
is not possible to investigate the macroeconomic impact of different levels of inequality
6. Models belonging to the Real Business Cycle Tradition are surely pushing ahead for the most the foregoing
synthesis. Nevertheless, it is also pursued by the core of the competing - and today more popular - New
Keynesian theories of business fluctuations. Indeed, once the various frictions are let away, the New Keynesian
”representative firm” still behaves like in a pure general equilibrium framework.
7. See also Kirman (1992) for a devastating account of the theoretical inconsistencies of the RAH.
8. Very recent works in the DSGE tradition however try to overcome the limits of the REH, and study the role of
credit into models where some agents’ heterogeneity is assumed (see e.g. Iacoviello, 2005, Campbell and
Hercowitz, 2004, Berka and Zimmermann, 2011).
9. Another problem in standard models is represented by the assumption of homothetic preferences, i.e. the
requirement that expenditure shares into different goods stay constant as income increases. In the real world,
preferences are not homothetic. Accordingly, income distribution can influence both the level and the
composition of final demand across different goods.
briefing paper no. 3 / 15 march 2012 3
in DSGE models. By construction the RAH implies that only average income or wealth
matter. The dispersion across individuals plays no role.
One possible objection to the above criticism is that any economic model is a highly
stylized representation of reality and – besides the aforementioned cases – the RAH is
still a useful trick to simplify matters in the analysis of other important issues wherein
micro-heterogeneity is not relevant. Nevertheless, a good deal of recent empirical
research on datasets at the microeconomic level (see e.g. Haltiwanger, 1997; Dosi,
2007, to cite only few of them) has revealed that heterogeneity matters in most cases.
The presence of huge and persistent asymmetries has been found to be widespread
across agents (e.g. among firms in employment, investment and in productivity levels).
In addition, most of this micro-heterogeneity does survive aggregation. It turns out
that, by sticking to the RAH, one loses the ability to investigate important aspects of
reality which map from micro-heterogeneity to macro outcomes.
Besides the RAH, the assumption of rational expectations of agents (REH) constitutes
another major weakness of DSGE models. Rational expectations presume that agents
know the true underlying model of the economy and make an efficient use of the infor-
mation they have. In addition, it requires that all agents must have the same (rational)
expectation about a given variable.10 These assumptions could sound conceptually
reasonable to many. Indeed, why should individuals with a minimum sense of
selfishness lose money by not exploiting as much as possible any piece information
available to them? Nonetheless, a large evidence, especially in (but not limited to)
financial markets (see e.g. Thaler, 2000, Schiller, 2001 to mention only few of them)
has documented how expectations of agents are neither rational nor equal. Agents
persistently make an inefficient use of information in the formation of their beliefs.
More importantly, it is precisely the difference in beliefs that drives most of what
occurs in financial markets (including the emergence of bubbles and crashes, see e.g.
Hommes, 2006 and references therein). Besides this empirical evidence there is
another more conceptual reason, that makes the REH problematic in light of the recent
crisis. As we already mentioned, the REH implies that agents make an efficient use of
the information available to them. However, it is the usefulness of available informa-
tion, which looks severely undermined in the case of the recent crisis. The crisis has
indeed been characterized by a major fall in the overall level of economic activity, not
comparable to the small perturbations characterizing advanced economies before it.
The response of the economic system to such a major shock can be quite different
from the one to small fluctuations. It follows that the information gathered in the
course of small perturbations can be hardly applicable to extrapolate an appropriate
forecast for economic variables during the crisis. As Stiglitz (2011, pp. 602-603),
candidly puts it:
“There hasn’t been a crisis as deep as the current one for three-quarters of a century,
so how can market participants form rational expectations about how modern econo-
mies respond to such a situation, unless they make the leap of faith that responses to a
large crisis are similar to responses to smaller perturbations? […] Describing and analy-
zing the full rational expectations equilibrium is complicated enough; to presume that in
10. Rational expectations are indeed an equilibrium concept applied to the context of agents’ expectations. In
this fashion, rational expectations have also a simplifying purpose similar to the RAH. Indeed, by assuming rational
expectations, heterogeneity in expectations is washed away, which greatly simplify matters in several cases.
4 briefing paper no. 3 / 15 march 2012
a rare event such as the current one that all economic and political actors have quickly
gravitated to this equilibrium is beyond credence.”
Furthermore, recent empirical works (e.g. Canning et al., 1998, Fagiolo et al., 2008,
Castaldi and Dosi, 2009) have revealed that the distribution of aggregate output
growth rates displays flat-tails. In practice, this means that the dynamics of advanced
economies is characterized by small perturbations alternating with extreme growth
events, which are more likely than what would be predicted by a normal distribution. It
follows that extreme fluctuations (such as the recent crisis) are an inherent property of
advanced economies, and thus their importance cannot be downplayed the rank of
“black swan” or “outliers”. Moreover, large fluctuations seem to be more related to
errors and coordination failures of (heterogeneous) agents at the micro level, than to
exogenous shocks at the macro level.11 Models where one representative agent
correctly anticipates the consequence of aggregate shocks (like DSGEs do) can hardly
account for these extreme phenomena.
The general occurrence of extreme fluctuations at the macro level makes questio-
nable another pillar of standard models, namely the hypothesis that all markets clear at
any point in time. Indeed, large negative growth events (such as economic crises) are
typically associated with situations of persistent unemployment, i.e. situations where
one fundamental market – the market for labor – does not clear. By construction,
models based on the market clearing assumption, cannot be used to analyze such
situations. Furthermore, standard models often embed the presumption that the full
employment equilibrium is also stable, i.e. that the economy naturally gravitates
around such a state, being occasionally buffeted by exogenous shocks. This hypothesis
is of little applicability in environments characterized by the presence of extreme fluc-
tuations. Moreover, it can lead to fatal flaws in the design of policies. According to
many (e.g. Stiglitz, 2011, Mishkin, 2011), monetary authorities overlooked the
growing bubble in financial and housing markets partly because Standard Models
where predicting that bubbles couldn’t occur, and that was enough to focus the atten-
tion on inflation control to guarantee efficiency and growth.
2. Agent-Based Models and their structure
A natural way to follow in face of the problems exposed in the previous section
would be departing from the representative agent paradigm, thereby introducing
heterogeneity of agents’ characteristics and behavior, and allowing for markets that do
not clear. All the aforementioned characteristics add new degrees of complexity to
macroeconomic analysis. As eloquently expressed by Tesfatsion (2006) :
“The modeler must now come to grips with challenging issues such as asymmetric
information, strategic interaction, expectation formation on the basis of limited informa-
tion, mutual learning, social norms, transaction costs, externalities, market power,
predation, collusion, and the possibility of coordination failures.”
Exploiting the growing capabilities of computers, Agent Based Models (ABMs)
model economic processes as dynamic systems of heterogeneous interacting agents
(Epstein and Axtell, 1996; Tesfatsion and Judd, 2006). In ABMs repeated interactions
11. According to Amendola and Gaffard (2011) the coordination failures arose in the course of the process of
structural change which has characterized advanced economies are one of deep causes of the recent crisis.
briefing paper no. 3 / 15 march 2012 5
among agents over time induce continuously changing microeconomic patterns, the
aggregation of which generates a dynamics for the macroeconomic variable of interest
(Pyka and Fagiolo, 2005).
Thus, in contrast to Standard Models, ABMs explicitly model agents’ heterogeneity.
Moreover, in ABMs each agent embodies less capability that the one required to let the
entire system work. In other words, in ABMs agents do not know the “model” wherein
they operate and therefore they do not have rational expectations. It is also important
to stress that in ABMs is typically not imposed that interactions must occur at “equili-
brium” (e.g. that transactions must occur only at equilibrium prices). This does not
mean that equilibrium states cannot exist in ABMs and be eventually reached. The
distinctive feature is instead that equilibria are true emergent properties of a far-from-
equilibrium dynamics whose properties are explicitly analyzed and taken into account
in the description of macroeconomic phenomena.12 Moreover, in ABMs agents inte-
ract both through price as well as via non-price variables. Finally, market interactions
are not mediated by a central authority (like in standard models), but it is local and
occurs through a network of existing relationships across agents.13
The above ingredients represent a key advantage of ABMs compared to Standard
Models. For instance, they allow one to explicitly accounts for phenomena such as
asymmetric information between borrowers and lenders, inequality across agents,
persistent unemployment, information diffusion, heterogeneous beliefs and the emer-
gence of bubbles and crashes, financial contagion. As it was already discussed in the
previous section, these are elements that have played a central role in the recent crisis.
ABMs have also another advantage that is more related to policy design, and is
represented by their finer description of the economy compared to DSGE models. In
modern economies very articulated and heterogeneous institutional arrangements
often govern the functioning of key markets (e.g. labor and credit markets). The same
macroeconomic policy can have different effects according to the different institutional
setting in which it is implemented (Stiglitz, 2011). Institutional elements are typically
excluded in DSGE models, whose results are derived in highly stylized frameworks.
Baumol (2000 p. 231) discusses the perils of economic policies derived in such abstract
settings:
“As is suggested by the old Yiddish proverb […], ‘for example’ is not a proof. .[...]
The proper first reply to the proverb is that ‘for example’ can certainly be a disproof.
Illustrations can demonstrate conclusively that widely held beliefs, even some that have
been derived rigorously from (rather oversimplified) models, are just not valid as general
propositions. While things may sometimes work out as is usually believed, it is very
possible and perhaps even very likely that outcomes will be entirely different. Thus the
policy or behavior adopted on the basis of general belief can turn out to be counterpro-
ductive or even highly damaging”
ABMs allow one to flexibly characterize various complex institutional arrangements,
and to study their impact at the micro- and macroeconomic level. In the same respect,
it is also possible to add to the model elements of real economic structures which
12. On these grounds ABMs follow the track initiated by previous disequilibrium models (e.g. Barro and
Grossman, 1976, Amendola and Gaffard, 1998).
13. A more detailed and formal account of the basic structure of ABMs can be found in Pyka and Fagiolo (2005),
Fagiolo and Roventini (2008).
6 briefing paper no. 3 / 15 march 2012
policy-makers are more familiar with, thus improving the guidance of policy-making
applied to particular contexts (see Dawid and Neugart, 2011 for a discussion of this
point).
This finer-grained account of the economy does not come without costs. ABMs are
typically more complex than standard models, and this implies that their analytical
investigation is limited if not impossible. One must therefore employ computer simula-
tions to analyze them. This raises a number of important issues such as: how get rid of
the randomness present in any computer simulation exercise, and how to analyze and
validate the results of the simulations. Another major challenge in ABMs lies in the
choice of the rules agents follow to take their decisions. Indeed once the requirement
of rationality and consistency of agents’ choices (e.g. as it is implied by the REH) is let
away one is left with many degrees of freedom in the design of agents’ behavior. As a
consequence, the model can be hardly falsifiable (a requirement that would be desi-
rable for any theoretical model). Finally, the attempt to provide a finer description of
economic dynamics can easily end up in over-complicated models, wherein causal
relations are hard to grasp. Let us examine each of the above points in detail.
First, the randomness present in any economic series artificially generated through
simulations can be easily circumvented through the application of standard Monte-
Carlo procedures, already extensively applied in econometrics as well as in the analysis
of standard models (e.g. Real Business Cycle models).14 The problem of the specifica-
tion of agents' behavioral rules is prima facie more complicated in the absence of some
general principles underpinning agents' behavior (e.g. the one of perfect rationality).
Two complementary solutions to this problem have gained momentum in the litera-
ture. The first one consists in designing agents' behavioral rules building on the
evidence available from psychology experimental studies on human decision-making.
This is also the way already paved by all theoretical studies which go under the label of
“behavioral economics” (see e.g. Akerlof, 2002, Camerer et al., 2004, and reference
therein). The other approach (see e.g. Dosi et al, 2010, Dawid et al, 2011) consists in
selecting only those behavioral rules that are able to reproduce an ensemble of stylized
facts at different levels of aggregation. More precisely, the model (i.e. the set of beha-
vioral rules and their interaction structure) is “validated” only if it is able to jointly
generate statistical properties that are characteristic of empirical data both at the macro
level as well as at a a “meso” level of aggregation, i.e. at a level of aggregation that is
intermediate between the micro and the macro, (i.e. involving industries or set of
households). Requiring that a model is coherent with some statistical properties at the
macro level should not surprise anyone, as it looks a necessary premise of any scientifi-
cally sound theoretical exercise. The novelty amounts to requiring that the model is
also able to reproduce many properties that are observed at an intermediate level of
aggregation (e.g. concerning distributions of firms characteristics, such as firm size and
growth rates). Empirical studies (e.g. those mentioned in Section 2) have indeed
produced a rich ensemble of stylized facts at the “meso” level. Imposing that such
stylized facts must be reproduced by the model is a way of putting “empirical disci-
pline” in the design of the model15.
14. Practically speaking, many artificial series for a given macroeconomic variable are generated by holding fixed
the set of parameters governing the behavior of agents and statistics that are relevant for the issue under analysis
(e.g. average growth or volatility) are computed on each series. Each statistics is then averaged out in order to
remove randomness.
briefing paper no. 3 / 15 march 2012 7
One critique against ABMs, concerns the excessive number of parameters which
enter in the specification of agents' behavioral rules. Many parameters in a model give
a lot of freedom to the modeler, because in the end any possible stylized fact can be
matched with the appropriate set of values for the parameters. Such models are there-
fore not falsifiable and should therefore be discarded. This criticism is similar to the one
against basic polynomial data-fitting exercises: any set of data can be fitted with a
polynomial of appropriate degree. In the case of ABMs it is possible to reply to this
critique in two ways. First, it is true that ABMs in general feature more parameters than
standard models. However, in ABMs it is not a matter of reproducing just one stylized
fact but many at once! Indeed, the number of stylized facts that an ABM tries to repro-
duce (both at macro and meso level) is typically much larger than in a Standard Model,
and this already puts a lot of constraints on the set of parameters' values that can be
selected. Second, differently from polynomial data-fitting exercises, in ABMs it is
required that parameters' values must be economically meaningful. Consider for
example, the case of a parameter's value in a microeconomic investment decision rule
allowing the replication of a stylized fact at the macroeconomic level but implying a
crazy investment behavior at the firm level. This value would not be selected, precisely
because it fails the requirement of being meaningful at the microeconomic level.
Finally, the critique that the complicated structure of ABMs can blur the causal
mechanisms generating results seems to us a quite general remark applicable to any
model, rather than a specific and unavoidable fallacy of ABMs. The “art” of economic
modeling always involves a trade-off between realism and analytical simplicity and
clarity. ABMs certainly privilege the former. Nevertheless, ABMs still are (and must
remain) a stylized representation of economic reality. It is the therefore the ability of
the modeler to design and analyze the model in such a way that the greater
complexity involved by greater realism does not jeopardize the understanding of the
mechanisms underlying certain economic processes.
It must also be stressed that - even in very complicated ABMs - causal mechanisms
can be detected through counterfactual analyses. More precisely, the structure of
ABMs often allow one to control the presence of some dynamics in the model (through
an appropriate setting of the parameters), and to test how results are different when
such dynamics are switched off/on. One can of course disapprove this procedure, by
noting that counterfactual simulation analyses can never replace the generality and the
precision of a theorem. This is certainly true. On the other hand, it must be acknowle-
dged that economic theorems are many times derived on the grounds of heroic
assumptions of little applicability in reality, and/or they are often limited to existence
results about equilibria whose stability (e.g. the problem of whether and how an equi-
librium is reached) is seldom investigated. Once again, it is a matter of making the
right choice in the trade-off between simplicity and economic realism given the issue
at hand. Moreover, one should consider the two approaches (computer simulations
and analytical results) as possible complements rather than substitutes. 16
15. This also goes in the direction of making economics a more empirically disciplined science. See also Stiglitz
(2011) and Dosi (2004) for remarks on this point. It is also worth mentioning the standard models have been
blamed of being unfalsifiable , precisely because a lack of such empirical discipline (see e.g. Fisher, 1989). Indeed,
any observable situation can be interpreted in the terms of the theory that obeys to the first principles (such as
inter-temporal optimization and rational expectations) by giving appropriate specifications to key variables (e.g.
preferences), which are typically not observable.
8 briefing paper no. 3 / 15 march 2012
3. Macroeconomic policies in Agent-Based Models:
an overview of recent results
Several works in the recent years have tried to address macroeconomic policy issues
using the ABM methodology. In what follows we try to provide an overview of these
models, by classifying them into three policy areas: fiscal and monetary policies, bank
regulation, and central bank independence.17
Fiscal and monetary policies
The Great Recession has renewed the interest for fiscal policies as an effective tool to
tackle economic downturns. Dosi, Fagiolo, and Roventini (2010) analyze the role of
fiscal policy in a two sectors ABM, bridging Keynesian theories of demand-generation
and Schumpeterian theories of technology-fueled economic growth (the K+S model).
A novel feature of this model lies in the analysis of both the short-run and long-run
impact of fiscal policies. DSGE models embed a fundamental dichotomy about the
effectiveness of policies. More precisely, the effects of fiscal and monetary policies on
real variables are limited to the short-run, while in the long-run only supply-side poli-
cies can have real effects. This conclusion hinges upon the assumption that the
economy gravitates around an equilibrium characterized by a structural (or “natural”)
level of unemployment where it always returns in the medium-run. However, if the
economy is persistently away from this equilibrium, then the effects of fiscal and mone-
tary policy can span over the short-run. By allowing the possibility of studying the
behavior of economies that are persistently out of equilibrium, ABMs allow one to
study under which circumstances fiscal policy has real effects also in the long-run. This
model shows that the presence of a minimum level of redistributive fiscal policies is a
necessary condition for economic growth. Without fiscal policy the economy gets
stuck into a zero growth trajectory. In addition, the authors show the active fiscal
policie can be sufficiently employed to reduce the volatility of the economy.
Russo, Catalano, Gallegati, Gaffeo, and Napoletano (2007) develop an ABM where a
population of heterogeneous, boundedly rational firms and consumers/workers inte-
ract according to random matching protocols. The model delivers sustained growth
characterized by fluctuations and reproduce micro and macro regularities such as the
Beveridge, Phillips and Okun curves, firm growth-rate distributions, etc. On the policy
side, they find that average output growth rate is non-monotonically linked to the tax
rate levied on corporate profits if revenues are employed to finance R&D investment,
whereas growth is negatively affected if the tax-revenues are employed to provide
unemployment benefits.
DSGE models mostly deal with monetary policy, and searching for the best mone-
tary rule. At the same time the current Great Recession has showed that monetary
policy alone is not sufficient to put economies back on their steady growth path.
Agent-based models can be employed to assess the effects and the limits of monetary
policy, and to compare the ensuing results with policy prescriptions of DSGE models.
16. It is also interesting to note that computer simulations have so far had much warmer welcome in other
scientific disciplines (e.g. physics), where they are often employed as a complement to analytical investigations.
17. Due to space constraints we discuss only a selection of works in Agent Based Macroeconomics. For more
extensive surveys of this literature see Tesfatsion and Judd (2006), Fagiolo and Roventini (2012), and Dawid,
Fagiolo and Roventini (2012). The survey of this section builds on the last two.
briefing paper no. 3 / 15 march 2012 9
Haber (2008) studies the impact of different fiscal and monetary policies under diffe-
rent expectation formation mechanism in an ABM characterized by the presence of
households, firm, banks, a government and a central bank. The model is calibrated in
order to produce “reasonable” time series for GDP, consumption, unemployment and
the inflation rate. Next, positive fiscal (lower tax rate) and monetary shocks (higher
money target) are introduced in the analysis. A central result of the paper is that both
policies increase GDP growth, and inflation and reduce unemployment. The introduc-
tion of more sophisticated expectations of agents reduce the effects of fiscal policy,
whereas it increases the impact of monetary policy. Dosi, Fagiolo, Napoletano, and
Roventini (2011) extend the K+S model by adding a bank, which collects the deposits
of firms, and provide costly loans to financially constrained firms on a pecking-order
basis. The model is then employed to assess the effects of monetary policy in different
income distribution scenarios. Simulation results show that higher income inequality
increases the volatility of output, the unemployment rate and the likelihood of severe
crises. The characteristics of income distribution affects the effectiveness of monetary
policy: higher interest rates do not affect the dynamics of the economy up to an endo-
genous threshold – increasing in the level of economic inequality – above which the
average growth rate of the economy falls and the amplitude of fluctuations, the unem-
ployment rate and the probability of crises rise.
Finally, many central banks have largely adopted inflation targeting policies in the
recent past, based on the belief that focusing on low inflation was enough to
guarantee a good economic performance18. However, very few theoretical studies
have tried to investigate the effects of higher or lower inflation on the dynamics of real
aggregate variables. ABMs are well suited for the task because they allow one to study
how different inflation rates impact on the functioning of the mechanisms of exchange
into an economy. Ahsraf and Howitt (2008) study the effects of inflation on real activity
by extending the ABM model developed in Howitt and Clower (1998). In the model
boundedly rational agents trade several goods, and specialized traders mediate
exchange. In this framework, high inflation rates lead to lower mean GDP growth and
to higher GDP volatility. The reason is that higher mean inflation is also associated with
higher dispersion of prices. In turn, higher price dispersion induces more volatility in
the demand faced by single traders, with the consequence that the rate of bankruptcy
of traders is increased. Bankruptcy can have large effects in this model. This is because
the failure of few traders affects also the exchange patterns of other traders in the
system and let them to fail as well. The consequence is that the functioning of the
whole mechanism of exchange in the economy is altered when the inflation rate is
high.
Bank regulation
The flexibility of ABMs is extremely useful to test the impact of different regulation
frameworks on banks’ behavior. For instance, one can assess how different regulations
affect the liquidity of the interbank payment system or how alternative micro-pruden-
tial rules impact on macroeconomic stability. The latter issue is investigated in Ashraf,
Gershman, and Howitt (2011). They build an ABM wherein heterogeneous firms inte-
18. Howitt (2011) contains for a very recent evaluation of the interplay between central bank practices and
theory, and a discussion of the usefulness of ABMs for the conduct of monetary policy.
10 briefing paper no. 3 / 15 march 2012
ract with banks providing them credit. Banks are subject to various regulations, such as
capital adequacy ratio and limits to loan-to-value ratios. Simulations of the model show
that: i) the economy can be hit by “rare disasters”, where the behavior of banks
strongly affect macroeconomic performance; ii) during a crisis, higher loan-to-value
ratios and lower capital-adequacy ratios allow the economy to recover faster; iii) banks
can be an important “financial stabilizer” of the economy, easing the entry of new
firms and avoiding the incumbents to go bankrupt. Raberto, Teglio, and Cincotti
(2010) find that lower capital adequacy ratios can spur growth in the short-run, but
the higher stock of private debt can lead to higher firm bankruptcies, credit rationing
and serious economic downturns.
Also the network structure between firms and banks can play a significant role in
determining the emergence, the depth and the diffusion of a crisis. Delli Gatti, Galle-
gati, Greenwald, Russo, and Stiglitz (2010) develop an ABM populated by banks and
financially constrained firms to study the properties of a “network-based” financial
accelerator. The topology of the network is continuously evolving because firms can
switch their partner, in order to find better credit conditions (i.e. lower interest rates).
Simulation results show that the interactions of financially constrained agents, occur-
ring through the evolving credit network, give rise to business cycles and to financial
crises. Hence, policy makers can try to reduce the magnifying effect of the network-
based financial accelerator by changing the network structure. On a related ground,
Battiston, Delli Gatti, Gallegati, Stiglitz and Greenwald (2009) show the possibility of a
non-monotonic relation between the degree of connectivity of a financial network
(measured by the average number of connections of each agent in the network) and
the likelihood of systemic crises. When the network is poorly connected, increasing the
number of connections lowers systemic risk, because it widens the possibility of diver-
sifying idiosyncratic shocks. In contrast, when the network is already dense (i.e.
average degree is high), increasing the number of connections has adverse effects. This
is because the risk diversification effect is overwhelmed by the “contagion effect”, i.e.
the fact that everybody in the system is increasingly exposed to everybody else’s
shocks. As a consequence of that, the dynamics of agents’ balance sheets become
strongly correlated, and this leaves open the possibility that small shocks may trigger
systemic crises.
Galbiati and Soramaki (2011) use an ABM in order to assess the efficiency of alterna-
tive interbank payment systems. Heterogeneous banks choose their (costly) liquidity
stock at the beginning of the day in order to satisfy an exogenous random stream of
payment orders. If some banks exhaust their liquidity stock, they must delay their
payments until new liquidity arrives from other banks possibly leading to the gridlock
of the system. The model shows that the efficiency of the payment system increases if
the number of banks is small and if they are encouraged to provide more liquidity.
Moreover, there are strong economies of scale in payment activity (higher volumes
reduce total payment costs).
Central bank independence
ABMs have also been employed to study political economy issues related to the
institutional role of central banks and to the way monetary policy is announced to the
public. Rapaport, Levi-Faur, and Miodownik (2009) study why during the nineties
many governments decided to delegate authority to their central banks, by employing
briefing paper no. 3 / 15 march 2012 11
an ABM where heterogeneous countries decide whether to introduce central bank
independence taking into account the behavior of their neighbors. Simulation results,
conducted under a Monte Carlo exploration of the parameter space, show that the
emergence and the rate of adoption of central bank independence is positively related
to the size of the zone of influence of neighboring countries.
The time-inconsistency problem faced by central banks is analyzed in a more
general framework by Arifovic, Dawid, Deissenberg, and Kostyshyna (2010) using an
ABM where the interaction between a boundedly-rational central bank and a popula-
tion of heterogeneous agents determines the actual inflation rate. The agents can
either believe the inflation rate announced by the central bank or employ an adaptive
learning scheme to forecast future inflation. Computer simulations of this model show
that the central bank learns to sustain an equilibrium with a positive, but fluctuating
fraction of “believers”, and that this outcome is Pareto superior to the equilibrium
determined by standard models.
4. Conclusions
In this note we have provided a brief account of the status of macroeconomic
theory and policy in light of the recent economic crisis. The Great Recession has
revealed the inadequacy of the current standard models in macroeconomics to analyze
economic crises and to provide useful policy prescriptions to restore growth and
employment. This inadequacy of standard models stems from the type of approach
followed to describe the behavior of agents at the microeconomic level. In particular,
three cornerstones of this approach have come under question: i) the use of the repre-
sentative agent to approximate the behavior of the individuals populating the
economy; ii) the assumption that agents form rational expectations when forecasting
economic variables; iii) the assumption that all markets clear. In contrast the recent
economic events seem to be better described by models featuring boundedly rational
heterogeneous agents and wherein markets do not necessarily clear at all times. Agent
Based Models (ABMs) are a new class of models that embed all the above features, and
therefore qualify as a promising alternative to standard models. In this note, we have
described the basic structure of these models, their pros viz. standard models. We have
also, discussed the problems one finds in the construction and analysis of ABMs and
the solutions that so far have been proposed to deal with them. Besides, the possibility
of getting rid of the drawbacks of standard models, another interesting feature of
ABMs is the possibility of using them as computational laboratories to analyze policies
under the different institutional contexts that are observed in reality. This is typically
precluded in simple analytical models. Although the latter models can be useful in a
number of situations (provided that one introduces heterogeneity, bounded rationality
and the possibility of disequilibrium in the picture), the possibility of conducting finer-
grained analysis of policies under more realistic conditions should be considered a
strong added value of Agent Based Models.
12 briefing paper no. 3 / 15 march 2012
References
Akerlof, G. (2002), 'Behavioral macroeconomics and macroeconomic behavior',
American Economic Review, 411-433.
Amendola, M. & Gaffard, J. (1998), Out of equilibrium, Oxford University Press, USA.
Arifovic, J.; Dawid, H.; Deissenberg, C. & Kostyshyna, O. (2010), 'Learning benevolent
leadership in a heterogenous agents economy', Journal of Economic Dynamics and
Control, 34(9): 1768-1790.
Ashraf, Q.; Gershman, B. & Howitt, P. (2011), 'Banks, Market Organization, and
Macroeconomic Performance: An Agent-Based Computational Analysis', NBER
Working Paper 1702, National Bureau of Economic Research, Mass., USA.
Barro, R. & Grossman, H. (1976), Money, Employment and Inflation, Cambridge:
Cambridge University Press.
Battiston, S.; Delli Gatti, D.; Gallegati, M.; Greenwald, B. & Stiglitz, J. (2009), 'Liaisons
dangereuses: increasing connectivity, risk sharing, and systemic risk' NBER Working
Paper 15611, National Bureau of Economic Research Cambridge, Mass., USA.
Baumol, W. (2000), 'Review of Out of Equilibrium', Structural Change and Economic
Dynamics, 11: 227-233.
Berka, M. & Zimmermann, C. (2011), 'Basel Accord and financial intermediation: the
impact of policy', mimeo.
Caballero, R. (2010), 'Macroeconomics after the Crisis: Time to Deal with the Pretense-
of-Knowledge Syndrome', Journal of Economic Perspectives, 24(4): 85-102.
Camerer, C. (2004), Advances in behavioral economics, Princeton University Press.
Campbell, J. R. & Hercowitz, Z. (2005), 'The Role of Collateralized Household Debt in
Macroeconomic Stabilization' (11330), Technical report, National Bureau of
Economic Research, Inc.
Canning, D.; Amaral, L.; Lee, Y.; Meyer, M. & Stanley, H. (1998), 'Scaling the volatility
of GDP growth rates', Economics Letters, 60(3), 335-341.
Castaldi, C. & Dosi, G. (2009), 'The patterns of output growth of firms and countries:
Scale invariances and scale specificities', Empirical Economics, 37(3): 475-495.
Cincotti, S.; Raberto, M. & Teglio, A. (2010), 'Credit money and macroeconomic insta-
bility in the agent-based model and simulator Eurace', Economics: The Open-
Access, Open-Assessment E-Journal 4(26): 2010-26.
Colander, D.; Goldberg, M.; Haas, A.; Juselius, K.; Kirman, A.; Lux, T. & Sloth, B.
(2009), 'The financial crisis and the systemic failure of the economics profession',
Critical Review, 21(2-3): 249-267.
Dawid, H.; Gemkow, S.; Harting, P.; van der Hoog, S. & Neugart, M. (2011), 'The
Eurace@ Unibi Model: An Agent-Based Macroeconomic Model for Economic Policy
Analysis'.
Dawid, H. & Neugart, M. (2011), 'Agent-Based Models for Economic Policy Design',
Eastern Economic Journal, 37(1), 44-50.
Dawid, H., Fagiolo, G., & Roventini, A. (2012), ‘Macroeconomic policy in agent-based
and DSGE models’, LEM Working Paper, forthcoming.
Delli Gatti, D.; Gallegati, M.; Greenwald, B.; Russo, A. & Stiglitz, J. (2010), 'The finan-
cial accelerator in an evolving credit network', Journal of Economic Dynamics and
Control, 34(9): 1627-1650.
briefing paper no. 3 / 15 march 2012 13
Delli Gatti, D.; Gallegati, M.; Greenwald, B.; Russo, A. & Stiglitz, J. (2009), 'Business
fluctuations and bankruptcy avalanches in an evolving network economy', Journal of
Economic Interaction and Coordination, 4(2), 195-212.
Dosi, G. (2007), 'Statistical Regularities in the Evolution of Industries. A Guide through
some Evidence and Challenges for the Theory', Perspectives on innovation, 153-186.
Dosi, G. (2004), 'A very reasonable objective still beyond our reach: economics as an
empirically disciplined social science', Models of a Man: Essays in Memory of Herbert
A. Simon, 211-26.
Dosi, G.; Fagiolo, G.; Napoletano, M. & Roventini, A. (2011), 'Income Distribution,
Finance, and Macroeconomic Policies', Technical report, LEM.
Dosi, G.; Fagiolo, G. & Roventini, A. (2010), 'Schumpeter meeting Keynes: A policy-
friendly model of endogenous growth and business cycles', Journal of Economic
Dynamics and Control, 34(9): 1748-1767.
Epstein, J. & Axtell, R. (1996), Growing artificial societies: social science from the
bottom up, The MIT Press.
Fagiolo, G.; Napoletano, M. & Roventini, A. (2008), 'Are output growth-rate distribu-
tions fat-tailed? some evidence from OECD countries', Journal of Applied
Econometrics, 23(5): 639-669.
Fagiolo, G. & Roventini, A. (2012), 'On the scientific status of economic policy: a tale of
alternative paradigms', Knowledge Engineering Review, forthcoming
Fitoussi, J. & Saraceno, F. (2010), 'Inequality and macroeconomic performance', Docu-
ments de Travail de l'OFCE.
Galbiati, M. & Soramaki, K. (2010), 'An agent-based model of payment systems',
Journal of Economic Dynamics and Control.
Haber, G. (2008), 'Monetary and fiscal policy analysis with an agent-based macroeco-
nomic model', Journal of Economics and Statistics (Jahrbuecher fuer
Nationaloekonomie und Statistik) 228(2+ 3): 276-295.
Haltiwanger, J. (1997), 'Measuring and analyzing aggregate fluctuations: the impor-
tance of building from microeconomic evidence', Federal Reserve Bank of St. Louis
Review, 55-78.
Hommes, C. (2006), 'Heterogeneous agent models in economics and finance', Hand-
book of computational economics, 2: 1109-1186.
Iacoviello, M. (2005), 'House prices, borrowing constraints, and monetary policy in the
business cycle', American Economic Review, 739-764.
Kirman, A. (1992), 'Whom or what does the representative individual represent?', The
Journal of Economic Perspectives, 6(2): 117-136.
Kirman, A. (2010), 'The economic crisis is a crisis for economic theory', CESifo Economic
Studies, 56(4): 498.
Krueger, D. & Perri, F. (2006), 'Does Income Inequality Lead to Consumption Inequa-
lity? Evidence and Theory', Review of Economic Studies, 73(1): 163-193.
Krugman, P. (2009), 'How Did Economists Get It So Wrong?', The New YorkTimes.
Leijonhufvud, A. (2011), 'Nature of an economy', CEPR Policy Insight, (53).
Leijonhufvud, A. (1993), 'Towards a not-too-rational macroeconomics', Southern
Economic Journal, 1-13.
14 briefing paper no. 3 / 15 march 2012
Lucas, R. (1976), Econometric Policy Evaluation: A Critique, in Meltzer A Brunner, K.,
ed., 'The Phillips Curve and Labor Markets', Carnegie-Rochester Series on Public Policy,
pp. 19-46.
Mishkin, F. (2011), 'Monetary policy strategy: lessons from the crisis', Technical report,
National Bureau of Economic Research.
Pyka, A. & Fagiolo, G. (2007), '29 Agent-based modelling: a methodology for neo-
Schumpeterian economics', Elgar companion to neo-schumpeterian economics, 467.
Rapaport, O.; Levi-Faur, D. & Miodownik, D. (2009), 'The Puzzle of the Diffusion of
Central-Bank Independence Reforms: Insights from an Agent-Based Simulation',
Policy Studies Journal, 37(4): 695-716.
Russo, A.; Catalano, M.; Gaffeo, E.; Gallegati, M. & Napoletano, M. (2007), 'Industrial
dynamics, fiscal policy and R&D: Evidence from a computational experiment',
Journal of Economic Behavior & Organization, 64(3-4): 426-447.
Schiller, R. (2001), Irrational exuberance, Wiley Online Library.
Stiglitz, J. (2011), 'Rethinking macroeconomics: what failed and how to repair it’,
Journal of the European Economic Association.
Tesfatsion, L. (2006), Agent-based computational economics: A constructive approach to
economic theory, in L. Tesfatsion & K. Judd, ed., 'Handbook of Computational
Economics', Elsevier, pp. 831-880.
Thaler, R. (2000), 'From homo economicus to homo sapiens', The Journal of Economic
Perspectives, 14(1): 133-141.
Trichet, J. (2010), Reflections on the nature of monetary policy non-standard measures
and finance theory, in 'opening address at the 6th ECB Central Banking Conference,
Frankfurt am Main'.
briefing paper no. 3 / 15 march 2012 15
