…

                                                                                        8 July 2005
PREFACE TO THE HANDBOOK1
Purpose
    The explosive growth in computer power over the past several decades offers new tools
and opportunities for economists. Volume 1 of the Handbook of Computational Economics
(Amman et al. 1996) surveyed the growing literature on computational methods for solving
standard economic models such as Arrow-Debreu-McKenzie general equilibrium models and
rational expectations models. This second volume focuses on Agent-based Computational
Economics (ACE), a computationally intensive method for developing and exploring new
kinds of economic models.
    ACE is the computational study of economic processes modeled as dynamic systems of in-
teracting agents who do not necessarily possess perfect rationality and information. Whereas
standard economic models tend to stress equilibria, ACE models stress economic processes,
local interactions among traders and other economic agents, and out-of-equilibrium dynamics
that may or may not lead to equilibria in the long run. Whereas standard economic models
require a careful consideration of equilibrium properties, ACE models require detailed spec-
ifications of structural conditions, institutional arrangements, and behavioral dispositions.
    Although the tools and language may differ, the agendas of standard economics and ACE
are thus complementary. For example, many ACE modelers study the processes by which
prices are set in decentralized market economies, a problem not considered in standard
equilibrium modeling. Morever, the two modeling approaches share the long-run goal of
understanding more fully the dynamic properties of realistically rendered economic systems,
an understanding that requires knowledge of potential equilibria together with their basins
of attraction.
    As noted in the preface of Volume 1, there is no clearly defined field that we call Com-
putational Economics. However, the body of ACE research focusing on core topics is now
substantial, and it is a good time to take stock of where we are and to communicate this
summary to a wider audience of economists.
    Moreover, having an ACE handbook at this time also serves an important pedagogical
purpose. The ACE approach to economic problems is novel. ACE research requires training
in computational modeling skills that few graduate economic programs currently provide,
and that relatively few professional economists currently possess. Individuals desiring to
take this path will therefore need to have a certain amount of boldness, a willingness to take
risks, a willingness to operate outside the boxes outlined by those who have gone before.
This ACE Handbook is dedicated to the support and encouragement of these individuals.

Organization
   The ACE Handbook is divided into sixteen chapters, six shorter perspective essays, and
an Appendix. These materials cover the following topic areas.
  1
   Forthcoming in L. Tesfatsion and K. L. Judd (editors), Handbook of Computational Economics, Volume 2:
Agent-Based Computational Economics, Handbooks in Economics Series, North-Holland, to appear.

                                                   1
    In the first two chapters, the Editors present overviews of the substantive aims of the ACE
literature and the relationship of the ACE methodology to more standard economic modeling.
Chapter 1, by L. Tesfatsion, discusses the ACE approach to the study of economic systems
and contrasts this approach with more standard equilibrium approaches using a relatively
simple two-sector decentralized market economy for concrete illustration. In Chapter 2,
K. Judd focuses on the problems of determining and communicating the economic content
of the results of computationally intensive research, and the trade-offs between standard
approaches and computational methods. These two introductory pieces are intended as
gateways into the handbook for economists new to ACE modeling.
    Chapter 3, by T. Brenner, discusses the key role played in ACE models by learning
agents and critically surveys a wide variety of possible agent learning representations. In
Chapter 4, J. Duffy examines the potential synergies between experiments conducted with
human subjects and experiments conducted with computational agents, with a stress on
empirical validation issues.
    The determination of agent interaction patterns is a basic foundation for all ACE models.
In Chapter 5, A. Wilhite undertakes a series of experiments to explore how bilateral trading
and other forms of economic interactions are influenced when conducted within alternative
types of networks (e.g., a small-world network). N. Vriend extends this focus in Chapter 6
by considering how ACE researchers have modeled the endogenous formation of interaction
networks. In the latter models, agents have some degree of choice regarding not only how to
behave in any given interaction but also with whom to interact and with what regularity. In
Chapter 7, H. P. Young presents and concretely illustrates a rigorous method for analyzing
the long-run behavior of systems constituting large numbers of interacting agents with widely
differing characteristics.
    Financial economics is one of the more active ACE research areas. Chapters 8 and 9
provide extensive surveys of financial market research in which the endogeneous heterogeneity
of dynamic investment behavior appears to be critically important for the explanation of
observed regularities in financial time series. In Chapter 8, C. Hommes focuses on relatively
simple financial market models that are at least partly tractable by analytic methods and
that are being used as benchmarks in support of more complex ACE modeling efforts. In
contrast, B. LeBaron in Chapter 9 focuses on ACE financial market studies for which the
complexity of the models requires the intensive use of computational tools.
    Technological change and innovation concern the generation and diffusion of new knowl-
edge, technologies, and products. In Chapter 10, H. Dawid discusses the current and poten-
tial contributions of the ACE modeling approach to this difficult topic area. For example,
he demonstrates how several empirically established stylized facts regarding technological
change and innovation, viewed as puzzles within standard equilibrium modeling, emerge
quite naturally in agent-based models.
    Organizations are collections of agents who interact with each other within the confines
of some formally or informally structured set of rules, and whose activities are guided in part
by personal preferences and in part by collective objectives. In Chapter 11, M. Chang and
J. Harrington survey a wide variety of organization models, including models of multi-agent


                                              2
firms, multi-plant manufacturers, and retail chains. They develop their chapter around a
set of research questions common to the organization literature, comparing and contrasting
traditional and agent-based modeling approaches and highlighting new insights afforded by
the latter approach.
    Over the past thirty years a whole new field of study has blossomed within economics,
called market design. The normative focus of this field is how institutional rules governing
trade can be treated as variables subject to optimization. To date, however, tractability
concerns have forced many researchers to restrict attention to equilibrium models in which
the strategic options open to traders are severely constrained ex ante.
    In Chapter 12, R. Marks first reviews in general terms the manner in which ACE models
with strategic learning agents have been used to evaluate market designs from a dynamic
perspective. He then highlights ten papers that exemplify recent progress in this topic area,
with a particular emphasis on the evaluation of electricity market designs. Chapter 13,
by J. Mackie-Mason and M. Wellman, also addresses market design issues. In contrast
to Marks, however, the authors focus their attention on automated markets with software
trading agents. Their primary concern is the direct use of agent-based tools to achieve a
complete effective automation of the various components of market transactions.
    A particularly exciting aspect of the ACE methodology is the encouragement and facility
it provides for integrative modeling. In keeping with the reasonable Einstein dictum "a
scientific theory should be as simple as possible but no simpler," researchers generally tailor
their models to the type of issue under study, stressing some features while downplaying
or omitting others. But critical model features do not always fall tidily along conventional
disciplinary lines.
    Chapters 14 and 15 focus on issues of importance to economists for which political con-
cerns are paramount. In Chapter 14, K. Kollman and S. Page critically survey a range of
agent-based models developed by economists and political scientists to address collective
action problems, pie-splitting problems, electoral competitions, and security and communal
stability issues at both the national and sub-national levels. In Chapter 15, M. Janssen and
E. Ostrom survey ACE research addressing the governance of systems comprising social and
biophysical agents. A key aim of the latter research has been increased understanding of
institutional arrangements conducive to the cooperative use of collective ecological resources
(e.g., fisheries) in the face of extensive behavioral uncertainty.
    In Chapter 16, C. Dibble discusses the potential of computational laboratories for facili-
tating the design and exploratory analysis of agent-based models with spatial aspects. Illus-
trative examples include spatial small-world network models, social norm diffusion models,
and epidemiology models for the control of infectious diseases.
    The next section of the handbook consists of six essays offering shorter perspectives
on agent-based modeling. Alphabetically ordered by author, these essays elaborate on the
following themes.
    W. B. Arthur explains why the movement now under way towards agent-based model-
ing is not simply an adjunct to neoclassical economics but a major shift to a more general
out-of-equilibrium economics. R. Axelrod uses some of his own personal experiences to ex-


                                              3
emplify how agent-based modeling can help overcome the somewhat arbitrary boundaries
between disciplines. J. Epstein argues that the central contribution of agent-based modeling
to the scientific enterprise is the facilitation of generative explanation: How can an ob-
served regularity be generated through the autonomous local interactions of heterogeneous
boundedly-rational agents?
    P. Howitt contends that current economic growth research focuses too exclusively on in-
dividual incentives and choice, ignoring critical coordination issues. He advocates the use of
agent-based modeling tools as a way of seeing beyond the "individual dots" of an economic
system to the overall patterns that emerge from simple interactions among a large number
of interacting agents. Following a critique of modern macro theory, A. Leijonhufvud argues
that agent-based methods should be used to revive the traditional core of macroeconomics:
namely, supply and demand interactions in markets with adaptive boundedly-rational par-
ticipants. He concludes, in particular, that agent-based methods provide the only means for
exploring the self-regulating capabilities of complex dynamic economies, and for advancing
our understanding of the adaptive dynamics of actual economies.
    In the final perspectives essay, T. Schelling takes the reader back to an airplane trip
in the 1960s during which, for amusement, he began experimenting with x's and o's on a
penciled-in checkerboard on a piece of paper. His purpose was to see what might result
from the repeated location choices of the x and o agents under variously assumed intensities
of preference for residing among neighbors of their own type. Out of such musings, the
now-famous Schelling City Segregation Model was born.
    The handbook concludes with an Appendix by R. Axelrod and L. Tesfatsion offering a
general guideline for newcomers to agent-based modeling in the social sciences. The guideline
provides a short annotated list of suggested introductory readings. It also provides pointers
to additional readings and software materials to help interested readers get started on their
own agent-based research.

Acknowledgements
    We are grateful to the General Editors of the Handbooks in Economics Series, Mike
Intriligator and Ken Arrow, as well as the Publishing Editor Valerie Teng and Publishing
Assistant Pauline Riebeek, for their encouragement and support of this handbook project.
We are tremendously in debt to the contributors whose hard work and enthusiasm helped us
to bring this project to a successful and just-about-on-time conclusion. Our heartfelt thanks,
also, to the many referees who aided these contributors by providing detailed constructive
comments at various draft stages. Finally, special thanks to Scotte Page and Howard Oishi
for hosting and arranging the ACE Workshop for handbook contributors at the University
of Michigan in May 2004.

References
Amman, H. M., D. A. Kendrick, and J. Rust, eds. (1996), Handbook of Computational
Economics, Volume 1 (Handbooks in Economics Series, North-Holland, Amsterdam, the
Netherlands).


                                              4
    LEIGH TESFATSION
     Iowa State University

    KENNETH L. JUDD
      Hoover Institution




5