Agent-Based Computational Economics Research Area:
Experiments with Real and Computational Agents

Last Updated: 8 July 2009

Site Maintained By: Leigh Tesfatsion Department of Economics Iowa State University Ames, Iowa 50011-1070 http://www.econ.iastate.edu/tesfatsi/ tesfatsi AT iastate.edu ACE Course (Self-Study eBook) ACE Website

Table of Contents: Basic Issues Introductory Materials Other Readings Software, Toolkits, and Demos Resource Sites and Groups Individual Researchers

Basic Issues

• What are the strengths and weaknesses of conducting agent-based computational experiments and human-subject experiments as an aid to understanding real-world economic processes?
• Should computational experiments be viewed as a potential substitute for human-subject experiments? A potential complement?
• Is is feasible and desirable to develop platforms that permit integrated experiments ranging from 100% computatational agents to 100% human participants (and everything in between)?
• In what situations is it both feasible and useful to calibrate the behavior of computational agents to the behavior of humans?
• Can computational agents be used to help infer the decision rules used by humans in parallel situations?
• Do computational agents tend to "do better" than humans in some experimental contexts? Do humans have something to learn from computational agents?
• Should human participants in real-world markets delegate decision-making to computational agents (shop-bots, Web-bots, crawlers, etc.)?

Introductory Materials

• John Duffy, "Agent-Based Models and Human-Subject Experiments", in Leigh Tesfatsion and Kenneth L. Judd (editors), Handbook of Computational Economics, Vol. 2: Agent-Based Computational Economics, Handbooks in Economics Series, North-Holland/Elsevier, Amsterdam, Spring 2006.
  Abstract: This chapter examines the relationship between agent-based modeling and economic decision-making experiments with human subjects. Both approaches exploit controlled "laboratory" conditions as a means of isolating the sources of aggregate phenomena. Research findings from laboratory studies of human subject behavior have inspired studies using artificial agents in "computational laboratories" and vice versa. In certain cases, both methods have been used to examine the same phenomenon. The focus of this chapter is on the empirical validity of agent-based modeling approaches in terms of explaining data from human subject experiments. We also point out synergies between the two methodologies that have been exploited as well as promising new possibilities.

  Francesco Guala, History of Experimental Economics (pdf,121K), in Steven Durlauf and Lawrence Blume (Eds.), New Palgrave Dictionary of Economics, Second Edition (Palgrave-MacMillan).

  John H. Kagel and Alvin E. Roth (eds.), Handbook of Experimental Economics, Princeton University Press, Princeton, NJ, 1995.

  Abstract: This book presents a comprehensive critical survey of the results and methods of human-subject laboratory experiments in economics. The first chapter provides an introduction to human-subject experimental economics as a whole, while the remaining chapters provide surveys by leading practitioners in areas of economics that have a concentration of human-subject experimental research: public goods; coordination problems; bargaining; industrial organization; asset markets; auctions; and individual decision making. For more information about this book and related topics, visit Al Roth's Experimental Economics Handbook Site and his website on Game Theory and Experimental Economics.

  Leigh Tesfatsion, "From Human-Subject Experiments to Computational-Agent Experiments (and Everything In Between)" (pdf,1.4M), Plenary Address, International Economic Science Association Meeting, Arlington, VA, June 27, 2009.

  Abstract:Human-subject experimentation is now a mainstream tool within economics. Computational-agent research is more recent, having emerged in parallel with the growing capability and availability of computers in the late1980s. This talk will focus on the largely untouched span between the two methodologies. What advantages might there be to conducting economic experiments within integrated test beds involving the combined use of humans and computational agents? More generally, what advantages might there be to designing economic systems in which humans routinely interact with cognitive computational agents? Indeed, as will be elaborated in the talk, this is not science fiction; it is already on the drawing boards, exemplified by proposed new smart grid technology for undergirding the operation of electric power markets. Smart grid technology envisions artificially intelligent grid-device and trader agents acting in concert with human market operators and traders, who in turn rely in part on models of such combined systems for operational guidance.

  Experimental Economics is the official journal of the Economic Science Association. The journal serves the growing group of economists around the world who use laboratory methods to study phenomena that are difficult to observe directly in naturally occurring economic contexts. The journal publishes high quality papers in any area of experimental research in economics and in related fields such as accounting, finance, political science, and the psychology of decision-making. Readers will also find state-of-the-art theoretical work and econometric work motivated by experimental data. Lastly, the journal publishes articles with a primary focus on methodology or replication of controversial findings.

Other Readings

• James Andreoni and John H. Miller, "Auctions with Artificial Adaptive Agents," Games and Economic Behavior 58 (1995), 211-221.

• Jasmina Arifovic, "The Behavior of the Exchange Rate in the Genetic Algorithm and Experimental Economics," Journal of Political Economy 104(3), 1993, 510-541.

• W. Brian Arthur, "Designing Economic Agents that Act Like Human Agents: A Behavioral Approach to Bounded Rationality," American Economic Review Papers and Proceedings 81(2), 1991, 353-359.

• P. J. Brewer, M. Huang, B. Nelson, and C. R. Plott, "On the Behavioral Foundations of the Law of Supply and Demand: Human Convergence and Robot Randomness," Experimental Economics 5(3), 2002, 179-208.

• Nicholas T. Chan, Blake LeBaron, Andrew W. Lo, and Tomaso Poggio, "Agent-Based Models of Financial Markets: A Comparison with Experimental Markets," Working Paper, September 5, 1999.

• Bruno Contini, Roberto Leombruni, and Matteo G. Richiardi, "Exploring a New ExpAce: The Complementarities between Experimental Economics and Agent-Based Computational Economics (pdf,129K), Journal of Social Complexity, Vol. 3, No. 1, 2006.
  Abstact: This paper addresses whether experimental economics and agent-based computational economics can complement and integrate with each other. The authors argue that the answer is yes, that there are many benefits to be gained by both communities of researchers from increased interactions.

  John Duffy, "Learning to Speculate: Experiments with Artificial and Real Agents," Journal of Economic Dynamics and Control 25(3/4), March 2001, pages 295-319.

  Abstract: This paper employs parallel experiments with real and computational agents to explore issues originally raised by Kiyotaki and Wright in their well-known search model of money (JPE, 1989). The primary issue of interest is how individuals might come to accept or learn to adopt a convention in which the particular commodity functioning as "money" is dominated in rate of return by other assets, in the sense that it has a higher storage cost. The key offsetting factor is anticipations ("speculation") concerning the ease with which the "money" good can be turned over in trade for other goods that agents have a higher desire to consume. The author shows how each type of experiment can contribute to the experimental design and interpretation of results for the other.

  Herbert Gintis, A Framework for the Unification of the Behavioral Sciences (pdf,1M), Behavioral and Brain Sciences, Vol. 30, 1-61.

  Abstract: "The various behavioral disciplines model human behavior in distinct and incompatible ways. Yet, recent theoretical and empirical developments have created the conditions for rendering coherent the areas of overlap of the various behavioral disciplines. The analytical tools deployed in this task incorporate core principles from several behavioral disciplines. The proposed framework recognizes evolutionary theory, covering both genetic and cultural evolution, as the integrating principle of behavioral science. Moreover, if decision theory and game theory are broadened to encompass other-regarding preferences, they become capable of modeling all aspects of decision making, including those normally considered “psychological,” “sociological,” or “anthropological.” The mind as a decision-making organ then becomes the organizing principle of psychology."

  D. K. Gode and S. Sunder, "Allocative Efficiency of Markets with Zero Intelligence Traders: Market as a partial substitute for individual rationality" (pdf,1.4M), Journal of Political Economy, Vol. 101, Number 1, 1993, pp. 119-137.

  Abstract: Gode and Sunder report on continuous double auction experiments with budget-constrained computational traders. They find that market efficiency levels are close to 100 percent even when buyers submit purely random bids and sellers submit purely random offers. The authors conclude that the high market efficiency typically observed in continuous double auction experiments with human subjects is due to the structure of the auction and not to learning. Their seminal work has highlighted an important issue now being actively pursued by many other researchers: what are the relative roles of learning and institutional arrangements in the determination of economic, social, and political outcomes?

  Marton Ivanyi, Rajmund Bocsi, Laszlo Gulyas, Vilmos Kozma, and Richard Legendi, The Multi-Agent Simulation Suite (pdf,600K), AITIA International, Inc., Budapest Hungary.

  Abstract: "Agent-based modeling is a branch of computer simulation, especially suited for studying complex social systems. It models the individual, together with its imperfections (e.g., limited cognitive or computational abilities), its idiosyncrasies and personal interactions. The approach builds the model from ’the bottom-up’, focusing mostly on micro rules and seeking to understand the emergence of macro behavior. Participatory simulation - a branch of agent-based simulation - is a methodology building on the synergy of human actors and artificial agents, excelling in the training and decision-making support areas. In participatory simulations some agents are controlled by users, while others are software governed. The Multi-Agent Simulation Suite is a software package intended to enable modelers to utilize the tools of agent-based simulation in various fields, without having to develop heavy programming skills."

  Wander Jager Marco A. Janssen, "Using Artificial Agents to Understand Laboratory Experiments of Common-Pool Resources with Real Agents," Chapter 6 in Janssen, M. A. (ed.), Complexity and Ecosystem Management: The Theory and Practice of Multi-Agent Systems, Edward Elgar Publishers, Cheltenham UK/Northampton, MA, 2003.

  Wander Jager and Marco Janssen, "The Need for and Development of Behaviourally Realistic Agents" (pdf,196K), pp. 36-49 in J. S. SIchman, F. Bousquet, and P. Davidsson (Eds.), MABS 2002, LNAI2581, Springer-Verlag Berlin Heidelberg, 2003.

  Abstract: "In this paper we argue that simulating complex systems involving human behaviour requires agent rules based on a theoretically rooted structure that captures basic behavioural processes. Essential components of such a structure involve needs, decision-making processes and learning. Such a structure should be based on state-of-the-art behavioural theories and validated on the micro-level using experimental or field data of individual behaviour. We provide some experiences we had working with such a structure, which involve the possibility to relate the results of simulations on different topics, the ease of building in extra factors for specific research questions and the possibility to use empirical data in calibrating the model. A disadvantage we experienced is the lack of suiting empirical data, which necessitates in our view the combined use of empirical and simulation research."

  Robert Kurzban and Daniel Houser, "Experiments Investigating Cooperative Types in Humans: A Complement to Evolutionary Theory and Simulations" (pdf,307K), Proceedings of the National Academy of Sciences, Vol. 102, No. 5, February 1, 2005, 1803-1807.

  Abstract: "Unlike other species, humans cooperate in large, distantly related groups, a fact that has long presented a puzzle to biologists. The pathway by which adaptations for large-scale cooperation among nonkin evolved in humans remains a subject of vigorous debate. Results from theoretical analyses and agent-based simulations suggest that evolutionary dynamics need not yield homogeneous populations, but can instead generate a polymorphic population that consists of individuals who vary in their degree of cooperativeness. These results resonate with the recent increasing emphasis on the importance of individual differences in understanding and modeling behavior and dynamics in experimental games and decision problems. Here, we report the results of laboratory experiments that complement both theory and simulation results. We find that our subjects fall into three types, an individual’s type is stable, and a group’s cooperative outcomes can be remarkably well predicted if one knows its type composition."

  Sheri Markose, Jasmina Arifovic, and Shyam Sunder, "Advances in experimental and agent-based modelling: Asset markets, economic networks, computational mechanism design, and evolutionary game dynamics", Journal of Economic Dynamics and Control 31 (2007), pp. 1801-07.

  Sheri Markose (Guest Editor), Editorial Overview: Special issue on Developments in Experimental and Agent-Based Computational Economics (ACE) (pdf,119K), Journal of Economic Interaction and Coordination,Volume 1, Number 2, December 2006.

  Claudia Pahl-Wostl and Eva ebenhoh, Heuristics to Characterise Human Behaviour in Agent-Based Models (pdf,7pp), Working Paper, Institute of Environmental Systems Research, University of Osnabruck, Germany, downloaded 1/29/05.

  Abstract: The authors pursue a pragmatic approach to the representation of human behavior in agent-based models, assuming that agents can be characterized by a set of attributes and their behavior can be described by a set of simple decision heuristics. These assumptions are tested and refined by using data from human-subject experiments describing the behaviors of players in simple resource allocation games.

  Mark Pingle and Leigh Tesfatsion, "Evolution of Worker-Employer Networks and Behaviors Under Alternative Non-Employment Benefits: An Agent-Based Computational Study", pp. 254-283 in Anna Nagurney (ed.), Innovations in Financial and Economic Networks, Edward Elgar Publishers, 2003 (pdf preprint,269K).

  Note: The results of this ACE labor market study are briefly compared against results from a parallel human-subject experiment.

  Juliette Rouchier, "Re-Implementation of a Multi-Agent Model Aimed at Sustaining Experimental Economic Research: The Case of Simulations with Emerging Speculation", Journal of Artificial Societies and Social Simulation, Vol. 6(4), 2003, (html).

  Note: This paper explores replication issues for the article by John Duffy (JEDC,2001) cited above.

  Leonidas Spiliopoulos, "Human versus Computer Algorithms in Repeated Mixed Strategy Games" (pdf,1.8M), Munich Personal RePEc Archive (MPRA) Paper No. 6672, January 2008.

  Absract: "This paper is concerned with the modeling of strategic change in humans’ behavior when facing different types of opponents. In order to implement this efficiently a mixed experimental setup was used where subjects played a game with a unique mixed strategy Nash equilibrium for 100 rounds against 3 preprogrammed computer algorithms (CAs) designed to exploit different modes of play. In this context, substituting human opponents with computer algorithms designed to exploit commonly occurring human behavior increases the experimental control of the researcher allowing for more powerful statistical tests. The results indicate that subjects significantly change their behavior conditional on the type of CA opponent, exhibiting within-subjects heterogeneity, but that there exists comparatively little between-subjects heterogeneity since players seemed to follow very similar strategies against each algorithm. Simple heuristics, such as win-stay/lose-shift, were found to model subjects and make out of sample predictions as well as, if not better than, more complicated models such as individually estimated EWA learning models which suffered from overfitting. Subjects modified their strategies in the direction of better response as calculated from CA simulations of various learning models, albeit not perfectly. Examples include the observation that subjects randomized more effectively as the pattern recognition depth of the CAs increased, and the drastic reduction in the use of the win-stay/lose-shift heuristic when facing a CA designed to exploit this behavior."

Software, Toolkits, and Demos

• Annotated list of pointers to ACE/CAS General Software and Toolkits

• Annotated list of pointers to ACE/CAS Computational Laboratories and Demonstration Software

• Trade Network Game Lab used to run ACE labor market experiments (Pingle/Tesfatsion).

• ComLabGames provides a simple setup that is explicitly designed for running classroom game theory experiments. The site supports both normal-form and extensive-form games. The comlabgame software can be downloaded and installed on personal computers. The ComLabGames support team includes Robert A. Miller, Vesna Prasnikar, and Darko Zupanic.

• The Multi-Agent Simulation Suite, developed by Marton Ivanyi, Rajmund Bocsi, Laszlo Gulyas, Vilmos Kozma, and Richard Legendi, supports Participatory Simulation in which some agents are controlled by users while other agents are software governed.

• Charles A. Holt, Susan K. Laury, and David Lucking-Reiley,Classroom Experiments on the Internet (pdf,16K), Notes, January 2001.
  Abstract: Classroom experiments can be easily set up to run through standard internet browsers, which avoids the need to install special software on the students' ("client") personal computers. The instructions, decisions, and market signals are communicated via interactive web pages, with data stored in a database on the web server for later use in classroom discussions. The advantages of web-based interactions (over in-class experiments or programs that run on a local area network) are 1) scalability to accommodate potentially large numbers of students, 2) flexible hours to save class time for discussion, and flexible locations to allow students to connect from any personal computer with a standard web browser. This paper surveys a number of different sites that provide classroom applications of economics experiments. Some technical issues, of interest to those who want to develop their own web applications, are also addressed."

  Robert Goldstone, Allen Lee, and Andy Jones (Percepts and Concepts Laboratory, Indiana University, Bloomington) are encouraging people to participate in an ongoing on-line group experiment on resource foraging accessible at the Group Experimentation Environment (GEE) Project Site. They describe the experiment as follows: "The experiment is rather educational and engrossing. Your goal in a four-minute experiment is to pick up as many resources as you can by moving your icon with the arrow keys. You compete against other humans when they are available or Artificial Intelligence Bots (programmed by Michael Roberts) when no other humans are currently on-line. We have been beta-testing the environment locally at Indiana University for awhile, and now feel ready to announce it to the broader community, at least the broader ACADEMIC community at this point. If you find any bugs or have suggestions for improvements, please email Andy Jones or Allen Lee , and CC me (rgoldsto@indiana.edu)."

Resource Sites and Groups

• ACE/CAS Research Sites and Groups

• The Institute for Empirical Research in Economics (University of Zurich, Switzerland), headed by Prof.Dr. Ernst Fehr, provides links to a variety of resources related to microeconomics and experimental economics on its home page. Institute researchers combine insights from modern economic theory with results from social psychology and sociology to understand important economic phenomena. Topics stressed include the functioning of labor markets, the organization of the modern corporation, the private and public provision of public goods, and intertemporal choice problems.

• Al Roth (Economics Department, Harvard University) maintains an extensive informative website on Game Theory and Experimental Economics.

Individual Researchers Interested in Parallel Experiments

• Individual Researchers using ACE/CAS Tools

• Jasmina Arifovic (Economics Department, Simon Fraser U, Canada): Learning and Adaptation, Experimental Economics, Macroeconomics, Monetary Economics, Evolutionary Game Theory, Computational Mechanism Design

• John Duffy (Department of Economics, University of Pittsburgh, Pennsylvania): Incorporation of learning in computational economic models; Using genetic algorithms to model how agents learn and adaptively update their forecasts; Parallel experiments with real and computational agents.

• Herbert Gintis (Department of Economics, University of Massachusets, Amherst): Agent-based evolutionary game dynamics; Evolution of strong reciprocity; Moral economy of communities; Evolution of social norms; Experimental games.

• Blake LeBaron (Graduate School of International Economics and Finance, Brandeis University, Waltham, Massachusetts): Quantitative dynamics of interacting systems of adaptive agents, and how these systems replicate real-world phenomena; Behavior of traders in financial markets; Nonlinear behavior of financial and macroeconomic time series; Parallel experiments with real and computational agents.

• M. Utku Ünver (Department of Economics, University of Pittsburg, Pennsylvania): Social learning in market games using genetic algorithms; Experimental economics; Game theory; Two-sided and one-sided matching; Auctions; Parallel experiments with real and computational agents.

Copyright © 2009 Leigh Tesfatsion. All Rights Reserved.