Abbreviated Vita: Leigh Tesfatsion

Last Updated: 2 April 2008 Address: Professor of Economics and Mathematics Heady Hall 375 Iowa State University Ames, IA 50011-1070 Tel: (515) 294-7318 FAX: (515) 294-0221 http://www.econ.iastate.edu/tesfatsi/ tesfatsi AT iastate.edu

Educational Background

Current Professional Affiliations

• Professor of Economics and Mathematics, Iowa State University, August 1990 to present.
• Co-Organizer (with Prof. Chen-Ching Liu, ECpE), ISU Electric Energy Economics (E3) Group)
• Faculty Member, ISU Human-Computer Interaction (HCI) Graduate Program, August 2003 to present.
• Affiliate Faculty Member, ISU Center for Computational Intelligence, Learning, and Discovery (CCILD), July 2005 to present.
• Editorial Board, Book Series, New Dimensions in Networks, Edward Elgar Publishers, October 2001 to present.
• Editorial Board, Journal of Economic Interaction and Coordination, Springer, July 2005 to present.
• Associate Editor, Journal of Energy Markets, July 2007 to present.
• Associate Editor, Journal of Economic Dynamics and Control, North Holland, July 2000 to present.
• Co-Organizer, IEEE Power Engineering Society (PES) Task Force on Open-Source Software, June 2007 to present.
• Member, IEEE Computational Finance and Economics Technical Committee, June 2007 to present.
• Member, IEEE PES Multi-Agent Systems Working Group, June 2006 to present.
• Member, IEEE PES Computer and Analytic Methods Subcommittee, June 2007 to present.
• Member, IEEE PES Working Group on Test Systems for Economic Analysis, June 2007 to present.
• Contact Liaison for the Society for Computational Economics (SCE) Special Interest Group on Agent-Based Computational Economics, Fall 1999 to present.

Past Professional Affiliations

• Invited Member, Committee on Organizational Modeling from Individuals to Societies, National Research Council of the National Academies, Washington D.C., March 2005 - March 2008.
• Associate Editor, Applied Mathematics and Computation, Elsevier/North Holland, June 1991 to March 2008.
• Associate Editor, Journal of Public Economic Theory, Basil Blackwell (1997-2006).
• Consultant, Los Alamos National Laboratory (2002-2004).
• Contributing Co-Editor (with D. Ashlock), Complexity-at-Large Section, Complexity, John Wiley (1997-2004).
• Elected Advisory Council Member, Society for Computational Economics (SCE), 1/1997-12/2000, and 1/2002-12/2004.
• Associate Editor, IEEE Transactions on Evolutionary Computation, IEEE Neural Network Council (1996-2002).
• Guest Editor, Journal of Economic Dynamics and Control, Special Double Issue on Agent-Based Computational Economics, Volume 25, Issue Numbers 3-4, March 2001, pages 281-654.
• Guest Editor, Computational Economics, Special Issue on Agent-Based Computational Economics, Volume 18, No. 1, October 2001, pages 1-135.
• Guest Editor, IEEE Transactions on Evolutionary Computation, Special Issue on Agent-Based Modelling of Evolutionary Economic Systems, Volume 5, No. 5, October 2001, pages 437-560.
• Guest Editor, Applied Mathematics and Computation, Special Issue Dedicated to the Work of Robert E. Kalaba, Volume 45, No. 2, Part II, September 1991.
• Professor of Economics, University of Southern California, September 1989 through August 1990.
• Associate Professor of Economics, University of Southern California, September 1981 through August 1989.
• Visiting Researcher, National Bureau of Economic Research, Cambridge, Massachusetts, September 1981 through August 1982.
• Visiting Assistant Professor of Economics, University of Minnesota, Minneapolis, March 1980 through August 1980.
• Assistant Professor of Economics, University of Southern California, September 1975 through August 1981.

Primary Research Areas

• Agent-Based Computational Economics (ACE)

• Agent-Based Research on Restructured Electricity Markets

• Learning and Coordination in Decentralized Market Economies

• Learning via Criterion Filtering (temporal-difference/Q-learning)

• Evolution of Interaction Networks (e.g., endogenization of matching functions in labor markets)

• The Trade Network Game Laboratory (endogenous network formation through choice and refusal of game partners)

• Optimality and Efficiency in Open-Ended Dynamic Economies (e.g., First Welfare Theorem for overlapping generations economies)

• Multicriteria Estimation (e.g., time-varying regression via Flexible Least Squares)

• Nonlinear Estimation Via Multicriteria Associative Memories

• Adaptive Computation Methods for Nonlinear Systems (nonlocal sensitivity analysis, eigenvalue/eigenvector tracking, automatic differentiation, adaptive homotopy continuation)

Current Research Topics

• Agent-based computational economics: A constructive approach to economic theory;
• Testing the efficiency and reliability of restructured wholesale power market designs;
• Open-source software for restructured wholesale power markets;
• Analysis of decentralized market economies as distributed local-interaction systems;
• Concentration and market power in dynamic economies.

General Resource Websites Maintained

• On-Line Guide for Newcomers to Agent-Based Modeling
  in the Social Sciences (R. Axelrod and L. Tesfatsion)
• Agent-Based Computational Economics
• Formation of Economic and Social Networks
• Open-Source Software for Electricity Market Research, Teaching, and Training
• General Resources on Electricity Restructuring
• General Resources for Macro and Financial Economics

Open-Source Software Releases

• AMES Market Package (Java): Agent-Based Modeling of Electricity Systems (homepage)
• DCOPFJ (Java): A DC Optimal Power Flow Solver (homepage)
• TNGLab (C++/VB): Trade Network Game Laboratory (homepage)
• FLS (Fortran): Flexible Least Squares for Time-Varying Linear Regression (homepage)
• GFLS (Fortran): Generalized Flexible Least Squares for Approximately Linear Systems (homepage)
• MAM (Fortran): Multicriteria Associative Memories for Nonlinear Estimation (homepage)
• NASA (Fortran): Nonlocal Automated Sensitivity Analysis (homepage)

Books

• Leigh Tesfatsion and Kenneth L. Judd (eds.), Handbook of Computational Economics: Volume 2, Agent-Based Computational Economics (Contributors and Contents), Handbooks in Economics Series, North-Holland, Amsterdam, 2006.

• Committee on Organizational Modeling from Individuals to Societies, Behavioral Modeling and Simulation: From Individuals to Societies, National Research Council of the National Academies, The National Academies Press, Washington, D.C., 2008.

Other Selected Publications

Topic Listing:

• Agent-Based Computational Economics (ACE)
  • ACE Surveys and Special Issue Introductions
  • ACE Electricity Research
  • ACE Labor Market Research
  • ACE Trade Network Game Research
  • ACE Learning Research (Criterion Filtering)
• Economic Theory (Analytical)
  • Efficiency and Aggregation Issues in Overlapping Generations Models
  • Time Consistency and Boundedly Rational Optimization
  • General Game Theory Research
• Econometrics and Data Analysis
  • Multicriteria Estimation via Cost-Efficient Frontiers
  • Nonlinear Estimation via Associative Memories
• Computational Methods
  • Solution Tracking for General Parameterized Nonlinear Systems
  • Tracking of Eigenvalues and Eigenvectors for General Parameterized Matrices
  • Adaptive Homotopy Continuation
  • Automatic Evaluation of Higher-Order Partial Derivatives

Previewing Information:

Some of the materials available for downloading below are in postscript (ps) or in adobe portable document format (pdf); all others are html documents. Ghostview and Ghostscript for Windows, released as freeware under the GNU General Public License, can be used to preview and print ps files on personal computers. Adobe Reader, also downloadable as freeware, can be used to preview and print pdf files on personal computers.

Agent-Based Computational Economics (ACE)

ACE Surveys and Special Issue Introductions:

• "On-Line Guide for Newcomers to Agent-Based Modeling in the Social Sciences" (html), developed and maintained by Robert Axelrod and Leigh Tesfatsion.

• "Modeling Macroeconomies as Open-Ended Dynamic Systems of Interacting Agents" (pdf,45K) (with Blake LeBaron), American Economic Review Papers and Proceedings, May 2008, to appear.
  Abstract: This review article summarizes the general methodology of Agent-based Computational Economics (ACE) and discusses how ACE researchers are addressing the following three issues: (1) taxonomy - what types of agents for macroeconomic models?; (2) scale robustness - how many agents for macroeconomic models?; and (3) empirical validation - connecting to data.

  "Agent-Based Computational Economics", Scholarpedia, 2(2):1970, acceptance year 2007.

  "Agents Come to Bits: Towards a Constructive Comprehensive Taxonomy of Economic Entities" (pdf,112K), Journal of Economic Behavior and Organization, Vol. 63, Issue 2, June 2007, 333-346.

  Abstract: This essay is an invited comment on Philip Mirowski's essay titled "Markets Come to Bits: Evolution, Computation, and Markomata in Economic Science," also to appear in JEBO. In his usual brilliant and provocative style, Mirowski argues for a constructive approach to economic modeling centered on the computational representation of markets as integrated sets of algorithms that evolve over time. This essay counters that a broader constructive approach to economic modeling can and should be taken. The recent advent of powerful computer technologies supporting Agent-Based Modeling (ABM) renders feasible the computational study of economies modeled as evolving systems of interacting agents. In ABM, an "agent" refers broadly to bundled data and behavioral methods representing an entity constituting part of a computationally constructed world. Examples of possible agent referents include individuals, social groupings, institutions (e.g., markets), biological entities such as crops, and physical entities such as transportation networks and weather. Consequently, ABM provides tremendous opportunities for economists and other social scientists to tailor the breadth and depth of the entities represented in their models to the application at hand. A simple ABM of a two-sector decentralized market economy is used for concrete illustration.

  "Agent-Based Computational Modeling and Macroeconomics" (pdf,148K), Chapter 9 in David Colander (ed.), Post-Walrasian Macroeconomics: Beyond the Dynamic Stochastic General Equilibrium Model, Cambridge University Press, Cambridge, UK, 2006.

  Abstract: Agent-based Computational Economics (ACE) is the computational study of economic processes modeled as dynamic systems of interacting agents. This essay discusses the potential of ACE modeling tools for the study of macroeconomic systems. Points are illustrated using an ACE model of a two-sector decentralized market economy.

  Note: This essay, an abridged version of the following Elsevier/North-Holland handbook chapter, is published with permission from Elsevier/North-Holland.

  "Agent-Based Computational Economics: A Constructive Approach to Economic Theory" (pdf,253K), in Leigh Tesfatsion and Kenneth L. Judd (eds.), Handbook of Computational Economics, Volume 2: Agent-Based Computational Economics (Contributors and Contents), Handbooks in Economics Series, North-Holland, Elsevier, Amsterdam, 2006, pp. 831-880.

  Abstract: This chapter explores the potential advantages and disadvantages of Agent-Based Computational Economics (ACE) for the study of economic systems. General points are concretely illustrated using an ACE model of a two-sector decentralized market economy. Six issues are highlighted: Constructive understanding of production, pricing, and trade processes; the essential primacy of survival; strategic rivalry and market power; behavioral uncertainty and learning; the role of conventions and organizations; and the complex interactions among structural attributes, behaviors, and institutional arrangements.

  "Economic Agents and Markets as Emergent Phenomena" (pdf,167K), Proceedings of the National Academy of Sciences U.S.A., Vol. 99 (supp. 3), 2002, pp. 7191-7192.

  Abstract: A brief overview of recent work in agent-based computational economics is provided, with a stress on the research areas highlighted at the NAS Sackler Colloquium session "Economic Agents and Markets as Emergent Phenomena" held in October 2001 in Irvine, CA.

  "Agent-Based Computational Economics: Growing Economies from the Bottom Up" [ (pdf,212K), (ps,354K)], Artificial Life, Volume 8, Number 1, 2002, pp. 55-82. The published article is available from the MIT Press.

  Abstract: This study is a much-extended version of my brief PNAS article cited above. The main objectives and defining characteristics of agent-based computational economics (ACE) are outlined, and similarities and distinctions between ACE and artificial life research are clarified. Eight ACE research areas are identified, and a number of publications in each area are highlighted for concrete illustration. Open questions and directions for future ACE research are also considered. The study concludes with a discussion of the potential benefits of the ACE approach, as well as some potential difficulties.

  "Agent-Based Computational Economics: Modeling Economies as Complex Systems" (pdf,71K), Information Sciences 149 (2003), 263-269. The published article is available from Science Direct and Computer Science Web.

  Abstract: This much-shortened version of the previous ACE survey was presented at the 2002 Joint Conference on Information Sciences (JCIS).

  "Guest Editorial for the special issue of the IEEE Transactions on Evolutionary Computation on Agent-Based Modeling of Evolutionary Economic Systems" [ (pdf,278K), (ps,111K)], Volume 5, Number 5, October 2001, pp. 437-441. The published article is available from IEEE Xplore.

  Abstract: A brief overview of agent-based computational economics (ACE) is provided, followed by synopses of the articles included in the IEEE-TEC special issue on ACE. Additional readings are also suggested.

  "Introduction to the Computational Economics Special Issue on ACE" [ (pdf,236K), (ps,98K)], Volume 18, Number 1, October 2001, pp. 1-8. The published article is available from SpringerLink.

  Abstract: Following a brief discussion of ACE, this introduction provides a synopses of the articles included in the CE special issue on ACE.

  "Introduction to the Journal of Economic Dynamics and Control Special Issue on ACE" [ (pdf,288K), (ps,100K)], Volume 25, Numbers 3-4, March 2001, pp. 281-293. The published article is available from Science Direct.

  Abstract: Following a brief discussion of ACE, this introduction provides a synopses of the articles included in the JEDC special issue on ACE. A Table of Contents (full text and abstracts) for the JEDC special issue on ACE (Volume 25 Issue 3-4) can be accessed at the Elsevier home page.

  "Agent-Based Computational Economics: A Brief Guide to the Literature" [ (pdf,217K), (ps,82K)], in Jonathan Michie (ed.), Reader's Guide to the Social Sciences, Volume 1, Fitzroy-Dearborn, London, UK, March 2001.

  "How Economists Can Get Alife," [ (ps,226K), (pdf,560K)], pp. 533-564 in W. Brian Arthur, Steven Durlauf, and David Lane (eds.), The Economy as an Evolving Complex System, II, Santa Fe Institute Studies in the Sciences of Complexity, Volume XXVII, Addison-Wesley, 1997.

  Abstract: This study presents a summary overview of the basic artificial life (alife) paradigm, stressing aspects especially relevant for the study of decentralized market economies. In particular, recent work on a Trade Network Game (TNG) framework combining evolutionary game play with endogenous partner selection is used to illustrate how the alife paradigm might be specialized to economics. Analytical and simulation work is reported to show how the TNG is currently being used to study the evolutionary implications of alternative market structures at three different levels: individual trade behavior; trade network formation; and social welfare.

  "How Economists Can Get Alife: An Abbreviated Version" (html).

  Abstract: This is a short summary of the previously cited article that appeared in Arthur et al. (1997). It includes updated references and updated website URLs.

  "Review of J. M. Epstein and R. Axtell, Growing Artificial Societies: Social Science from the Bottom Up" (ps,28K), Journal of Economic Literature XXXVI (March 1998), 233-234

---

ACE Electricity Research:

• Open-Source Software Release: The AMES Market Package (Java) - A Free Open-Source Test Bed for the Agent-Based Modeling of Electricity Systems (homepage). Copyright holders (GNU GPL): J. Sun, H. Li, and L. Tesfatsion.

• Open-Source Software Release: DCOPFJ (Java) - A Free Open-Source Solver for DC Optimal Power Flow Problems (homepage). Copyright holders (GNU GPL): J. Sun and L. Tesfatsion

• "Dynamic LMP Response Under Alternative Price-Cap and Price-Sensitive Demand Scenarios" (with Hongyan Li and Junjie Sun) (pdf,465K), Proceedings, IEEE Power Engineering Society General Meetings, Carnegie-Mellon University, Pittsburgh, July 20-24, 2008.
  Abstract: This study investigates the complicated nonlinear effects of demand-bid price sensitivity and supply-offer price caps on Locational Marginal Prices (LMPs) for bulk electric power when profit-seeking generators can learn over time how to strategize their supply offers. Systematic computational experiments are conducted using AMES, an open-source agent-based test bed developed by the authors. AMES models a restructured wholesale power market operating through time over an AC transmission grid subject to line constraints, generation capacity constraints, and strategic trader behaviors.

  "Modeling of Suppliers’ Learning Behaviors in an Electricity Market Environment" (pdf,277K) (with Nanpeng Yu and Chen-Ching Liu), PSERC working paper 07-11, revised July 2007, to appear in International Journal of Engineering Intelligent Systems.

  Abstract: The Day-Ahead electricity market is modeled as a multi-agent system with interacting agents including supplier agents, Load Serving Entities and a Market Operator. Simulation of the market clearing results under the scenario in which agents have learning capabilities is compared with the scenario where agents report true marginal costs. It is shown that, with Q-Learning, electricity suppliers are making more profits compared to the scenario without learning due to strategic gaming. As a result, the LMP at each bus is substantially higher.

  "Dynamic Testing of Wholesale Power Market Designs: An Open-Source Agent-Based Framework" (pdf,2.2MB) (with Junjie Sun), Computational Economics, Volume 30, Number 3, 2007, pp. 291-327. The published article is available at SpringerLink.

  Abstract: In April 2003 the U.S. Federal Energy Regulatory Commission proposed a complicated market design -- the Wholesale Power Market Platform (WPMP) -- for common adoption by all U.S. wholesale power markets. Versions of the WPMP have been implemented in New England, New York, the mid-Atlantic states, the Midwest, the Southwest, and California. Strong opposition to the WPMP persists among some industry stakeholders, however, due largely to a perceived lack of adequate performance testing. This study reports on the model development and open-source implementation (in Java) of a computational wholesale power market organized in accordance with core WPMP features and operating over a realistically rendered transmission grid subject to congestion effects. The traders within this market model are strategic profit-seeking agents whose learning behaviors are based on data from human-subject experiments. Our key experimental focus is the complex interplay among structural conditions, market protocols, and learning behaviors in relation to short-term and longer-term market performance. Findings for a dynamic 5-node transmission grid test case are presented for concrete illustration. It is shown for this example that generators easily learn to implicitly collude on higher-than-true marginal costs when the demand bids of load-serving entities take the form of fixed loads.

  "An Agent-Based Computational Laboratory for Wholesale Power Market Design" (pdf,724K) (with J. Sun), Proceedings, IEEE Power Engineering Society General Meeting, Tampa, Florida, June 2007.

  Abstract: This proceedings paper is an abridged version of ISU Economics Working Paper No. 06025 (see above). It reports on the model development and open-source implementation (in Java) of an agent-based computational wholesale power market organized in accordance with core FERC-recommended design features and operating over a realistically rendered transmission grid subject to congestion effects.

  "DC Optimal Power Flow Formulation and Testing Using QuadProgJ" (pdf,521K) (with J. Sun), ISU Economics Working Paper No. 06014, Department of Economics, Iowa State University, Revised October 2007.

  Abstract: Nonlinear AC Optimal Power Flow (OPF) problems are commonly approximated by linearized DC OPF problems to obtain real power solutions for restructured wholesale power markets. We first present a standard DC OPF problem, which has the numerically desirable form of a strictly convex quadratic programming (SCQP) problem when voltage angles are eliminated by substitution. We next augment this standard DC OPF problem in a physically meaningful way, still retaining an SCQP form, so that solution values for voltage angles and locational marginal prices are directly obtained along with real power injections and branch flows. We then show how this augmented DC OPF problem can be solved using QuadProgJ, an open-source Java SCQP solver newly developed by the authors that implements the well-known dual active-set SCQP algorithm by Goldfarb and Idnani (1983). To demonstrate the accuracy of QuadProgJ, comparative results are reported for a well-known suite of numerical QP test cases with up to 1500 decision variables plus constraints. Detailed QuadProgJ results are also reported for 3-node and 5-node DC OPF test cases taken from power systems texts and ISO-NE/MISO/PJM training manuals.

  "Open-Source Software for Power Industry Research, Teaching, and Training: A DC-OPF Illustration" (pdf,115K) (with J. Sun), Proceedings, IEEE Power Engineering Society General Meeting, Tampa, Florida, June 2007.

  Abstract: This proceedings paper is an abridged version of ISU Economics Working Paper No. 06014 (see above). It reports on the development and implementation of a stand-alone open-source Java solver for DC optimal power flow (DC-OPF) problems suitable for research, teaching, and training purposes. The DC-OPF solver is shown to match or exceed the accuracy of BPMPD (a proprietary third-party QP solver highly recommended by MatPower) when tested on a public repository of small to medium-sized QP problems. The capabilities of the DC-OPF solver are illustrated for a 5-node DC-OPF test case commonly used for training purposes.

  "Market Design Test Environments" (pdf,136K) (with J. Sun and S. Widergren), Proceedings, IEEE Power Engineering Society General Meeting, Montreal, June 2006.

  Abstract: Power industry restructuring continues to evolve at multiple levels of system operations. At the bulk electricity level, several organizations charged with regional system operation are implementing versions of a Wholesale Power Market Platform (WPMP) in response to U.S. Federal Energy Regulatory Commission initiatives. Recently the Energy Policy Act of 2005 and several regional initiatives have been pressing the integration of demand response as a resource for system operations. These policy and regulatory pressures are driving the exploration of new market designs at the wholesale and retail levels. The complex interplay among structural conditions, market protocols, and learning behaviors in relation to short-term and longer-term market performance demand a flexible computational environment where designs can be tested and sensitivities to power system and market rule changes can be explored. This paper discusses the use of agent-based computational methods for the study of electricity markets at the wholesale and retail levels, and explores distinctions in problem formulation between these levels.

  "An Agent-Based Computational Laboratory for Testing the Economic Reliability of Wholesale Power Market Designs" (pdf,112K) (with D. Koesrindartoto and J. Sun), Proceedings, Vol. 1, IEEE Power Engineering Society General Meeting, San Francisco, California, June 2005, pp. 931-936.

  Abstract: Previous work by Koesrindartoto and Tesfatsion (2004) reports on the development of an agent-based model for exploring the economic reliability of the Wholesale Power Market Platform proposed by FERC in April 2003. This paper reports on the Repast/Java implementation of this model as an agent-based computational laboratory. Initial experiments focusing on optimal power flow solution methods for the day-ahead and real-time markets are discussed.

  "Testing the Reliability of FERC's Wholesale Power Market Platform: An Agent-Based Computational Economics Approach" (pdf,45K) (with D. Koesrindartoto), Energy, Environment, and Economics in a New Era, Proceedings of the 24th USAEE/IAEE North American Conference, Washington, D.C., July 8-10, 2004.

  Abstract: In April 2003 the U.S. Federal Energy Regulatory Commission (FERC) proposed the Wholesale Power Market Platform (WPMP) for common adoption by U.S. wholesale power markets. The WPMP is a complicated market design that has been adopted in some regions of the U.S. but resisted in others on the grounds that its reliability has not yet been sufficiently tested. This article reports on the development of an agent-based computational framework for exploring the economic reliability of the WPMP. The key issue under study is the extent to which the WPMP is capable of sustaining efficient, orderly, and fair market outcomes over time despite attempts by market participants to gain advantage through strategic pricing, capacity withholding, and/or induced transmission congestion.

  "Market Power and Efficiency in a Computational Electricity Market with Discriminatory Double-Auction Pricing" [ (pdf,162K), (ppt,116K)], (with J. Nicolaisen and V. Petrov), IEEE Transactions on Evolutionary Computation, Vol. 5, No. 5, October 2001, pp. 504-523. The published article is available at IEEEXplore (pdf,367K).

  Abstract: This study reports experimental market power and efficiency outcomes for a computational wholesale electricity market operating in the short run under systematically varied concentration and capacity conditions. The pricing of electricity is determined by means of a clearinghouse double auction with discriminatory midpoint pricing. Buyers and sellers use a modifed Roth-Erev individual reinforcement learning algorithm to determine their price and quantity offers in each auction round. It is shown that high market efficiency is generally attained, and that market microstructure is strongly predictive for the relative market power of buyers and sellers independently of the values set for the reinforcement learning parameters. Results are briefly compared against results from an earlier electricity study in which buyers and sellers instead engage in social mimicry learning via genetic algorithms.

---

ACE Labor Market Research:

• "Evolution of Worker-Employer Networks and Behaviors Under Alternative Non-Employment Benefits: An Agent-Based Computational Study" [ (pdf,269K), (ppt,88K)], (with M. Pingle), pp. 256-285 in Anna Nagurney (ed.), Innovations in Financial and Economic Networks, New Dimensions in Networks Book Series, Edward Elgar Publishers, 2003.
  Abstract: This study replaces the standard exogenously-given worker-employer matching function with endogenous preferential worker-employer matching based on past work-site experiences. Workers and employers participate in a sequential employment game with adaptive job search and incomplete contracts. Matched workers and employers participate in a work-site game. The effects of a non-employment payoff (NEP) on network formation and work-site behaviors are systematically examined through computational experiments. Taking both utility benefits and program costs into account, the highest efficiency is achieved with a moderate NEP level. A zero NEP encourages too much shirking on the work-site, while a high NEP results in too high a risk of lost earnings due to coordination failure.

  "Hysteresis in an Evolutionary Labor Market with Adaptive Search," [ (ps,216K), (pdf,534K)], pp. 189-210 in S.-H. Chen (ed.), Evolutionary Computation in Economics and Finance, Physica-Verlag Heidelberg, New York, 2002.

  Abstract: This study undertakes a systematic experimental investigation of hysteresis (path dependency) when workers and employers preferentially match based on past worksite experiences. It is shown that two distinct hysteresis effects can arise, network and behavioral, when workers and employers interact strategically and evolve their worksite behaviors over time. These hysteresis effects result in persistent heterogeneity in earnings and employment histories across agents who have no observable structural differences. At a more global level, these hysteresis effects are shown to result in a one-to-many mapping between treatment factors and experimental outcomes. These hysteresis effects may help to explain why excess earnings heterogeneity is commonly observed in real-world labor markets.

  "Structure, Behavior, and Market Power in an Evolutionary Labor Market with Adaptive Search" (pdf,295K), Journal of Economic Dynamics and Control 25 (2001), pp. 419-457. The published article is available from Science Direct.

  Abstract: This study undertakes a systematic experimental investigation of the relationship between market power and labor market structure (concentration and capacity conditions) when workers and employers preferentially match based on past worksite experiences. For each tested market structure, workers and employers repeatedly seek preferred worksite partners based on continually updated expected utility, engage in efficiency-wage worksite interactions modeled as prisoner's dilemma games, and evolve their worksite behaviors over time. A key finding is the presence of strong learning and network effects. Each tested market structure maps into a "spectral" distribution of observed interaction networks exhibiting one dominant attractor (frequent network pattern) with one or two weaker attractors (less frequent network patterns). Market structure is strongly predictive for the relative market power of workers and employers across all network attractors, but the magnitudes of the market power levels attained by workers and employers vary widely across the network attractors.

  "Concentration, Capacity, and Market Power in an Evolutionary Labor Market" [ (pdf,344K), (ps,161K)], pp. 1033-1040 in Evolution at Work for the New Millenium, Proceedings of the 2000 Congress on Evolutionary Computation, Volume II, IEEE, Inc., N.J., 2000, pp. 1033-1040. The published article is available at IEEE Xplore.

  This conference paper provides a summary overview of the detailed experimental findings reported in my previously cited JEDC (2001) study.

  "Preferential Partner Selection in Evolutionary Labor Markets: A Study in Agent-Based Computational Economics" [ (pdf,174K), (ps,130K)], pp. 15-24 in V. W. Porto, N. Saravanan, D. Waagen, and A. E. Eiben (eds.), Evolutionary Programming VII, Proceedings of the Seventh Annual Conference on Evolutionary Programming, Springer-Verlag, Berlin, 1998.

  Abstract: This study develops a computational labor market framework for studying the formation and evolution of contractual networks between workers and employers. Resource-constrained workers and employers choose and refuse contractual partners on the basis of continually updated expected utility, engage in risky worksite interactions modelled as two-person "efficiency-wage" prisoner's dilemma games, and evolve their work-site strategies over time on the basis of past worksite earnings. Illustrative computational experiments are reported and interpreted.

---

ACE Trade Network Game Research:

• Open-Source Software Release: The Trade Network Game (TNG) Laboratory (homepage). Copyright holders (Artistic License): David McFadzean, Deron Stewart, and Leigh Tesfatsion.

• "A Computational Laboratory for Evolutionary Trade Networks" (pdf,508K) (with D. McFadzean and D. Stewart), IEEE Transactions on Evolutionary Computation, Vol. 5, No. 5, October 2001, pp. 546-560. The published article is available at IEEE Xplore.
  Abstract: This report presents, motivates, and illustrates the use of a computational laboratory for the investigation of evolutionary trade network formation among strategically interacting buyers, sellers, and dealers. The computational laboratory, referred to as the Trade Network Game Laboratory (TNG Lab), is targeted for the Microsoft Windows desktop. The TNG Lab is both modular and extensible and has a clear, easily operated graphical user interface. It permits visualization of the formation and evolution of trade networks by means of run-time animations. Data tables and charts reporting descriptive performance statistics are also provided in real time. The capabilities of the TNG Lab are demonstrated by means of labor market experiments.
  An automatic installation program for the TNG Lab is available online, as well as TNG Lab tutorials. See the TNG Home Page for further information.

  "A C++ Platform for the Evolution of Trade Networks" [ (pdf,449K), [ (ps,244K)] (with D. McFadzean), Computational Economics 14 (1999), pp. 109-134. The published article is available from SpringerLink.

  Abstract: This study provides a detailed discussion of the C++ implementation of the Trade Network Game (TNG), a computational framework for studying the formation and evolution of trade networks in buyer-seller markets modelled as decentralized systems of autonomous interacting agents. The C++ source code is available online as freeware. For further information, visit the TNG Home Page.

  "An Agent-Based Computational Model for the Evolution of Trade Networks" (ps,132K) (with D. McFadzean), pp. 73-83 in P. Angeline, R. Reynolds, J. McDonnell, and R. Eberhart (eds.), Evolutionary Programming VI, Proceedings of the Sixth International Conference on Evolutionary Programming, Springer-Verlag, Berlin, 1997

  Abstract: This proceedings paper is a summary version of the previously cited CE (1999) article. It briefly discusses the computational (C++) implementation of the Trade Network Game (TNG).

  "A Trade Network Game with Endogenous Partner Selection" [ (pdf,401K), (ps,151K)], pp. 249-269 in H. M. Amman, B. Rustem, and A. B. Whinston (eds.), Computational Approaches to Economic Problems, Kluwer Academic Publishers, 1997.

  Abstract: This study develops a Trade Network Game (TNG) framework for studying the interplay between evolutionary game dynamics and preferential partner selection in buyer-seller markets. The TNG consists of successive generations of resource-constrained traders who choose and refuse trade partners on the basis of continually updated expected payoffs, engage in risky trades modelled as two-person games, and evolve their trade strategies over time. Preliminary computer experiments are reported which suggest that the standard optimality properties used to judge the desirability of matching mechanisms in static market contexts may be inadequate measures of optimality from an evolutionary perspective.

  For additional information regarding TNG research articles and software, see the TNG Home Page.

  "Preferential Partner Selection in an Evolutionary Study of Prisoner's Dilemma" (pdf,2.6M) (with D. Ashlock, M. D. Smucker, and E. A. Stanley), BioSystems 37, No. 1-2 (1996), pp. 99-125. The published article is also available from Science Direct.

  Abstract: This study reports on extensive computer experiments for the "Iterated Prisoner's Dilemma with Choice and Refusal of Partners (IPD/CR)" framework introduced in the next article. For example, it is shown that the introduction of choice and refusal tends to enhance the emergence of cooperation in IPD games in comparison to either random or round-robin partner selection. In addition, with choice and refusal of partners, the fitness values of the players tend to cluster fairly tightly into a small number of narrow regions. Within any one fitness region, however, the players consist of genetically diverse players with interaction patterns peculiarly adapted to the choice and refusal mechanism. Detailed sensitivity studies are also reported with respect to key parameters such as the learning algorithm "memory weight," the "minimum tolerance level" below which game offers are refused, and the "refusal payoff" incurred by a player whenever one of his game offers is refused.

  "Iterated Prisoner's Dilemma with Choice and Refusal of Partners" (with D. Ashlock and E. A. Stanley), pp. 131-175, in C. Langton (ed.), Artificial Life III, Volume XVII, Santa Fe Institute Studies in the Sciences of Complexity, Addison-Wesley, 1994.

  Abstract: This article extends the traditional iterated prisoner's dilemma with round-robin partner matching by permitting players to choose and refuse partners in each iteration on the basis of continually updated expected payoffs. Comparative computer experiments are reported that indicate the introduction of partner choice and refusal accelerates the emergence of mutual cooperation in the IPD relative to round-robin partner matching. Moreover, in contrast to findings for round-robin partner matching (in which the average payoffs of the players tend to be either clustered around the mutual cooperation payoff or widely scattered), the average payoff scores of the players with choice and refusal of partners tend to cluster into two or more distinct narrow bands. Preliminary analytical and computational sensitivity studies are also reported for several key parameters.

---

ACE Learning Research (Criterion Filtering):

• "Games, Goals, and Bounded Rationality", Theory and Decision, Vol. 17 (1984), pp. 149-175. The published article is available from SpringerLink.
  Abstract: This article presents a generalization of the standard N-person game with flexible information requirements suitable for players constrained by certain types of bounded rationality. In particular, strategies (complete contingency plans) are replaced with partial contingency plans augmented by goals. Both utility and probability are conditioned on selected goals and actions (g,a), and both are defined over the same set of possible (g,a)-conditioned events. Well-known existence theorems for Nash equilibria and Nash bargaining solutions are extended to this context. For adaptive sequential games, the symmetrical treatment of payoffs and probability assessments permits players to learn their successive moves via "criterion filtering." That is, the expected utility criterion function of each player can be directly updated in each decision period via transitional utility assessments in a manner analogous to Bayes' rule for updating probability distributions via transitional probability assessments.

  "A Dual Approach to Bayesian Inference and Adaptive Control" (pdf,774K), Theory and Decision, Vol. 14 (1982), pp. 177-194. The published article is available from SpringerLink.

  Abstract: This article surveys results established to date for the "criterion filtering" approach to adaptive control developed in Tesfatsion (JOTA,1978). Criterion filtering bypasses the usual preliminary updating of probability distributions via transitional probability assessments (Bayes' rule) and focuses instead on the direct updating of the criterion function via transitional return assessments.

  "A Conditional Expected Utility Model for Myopic Decision Makers", Theory and Decision, Vol. 12 (1980), pp. 185-206. The published article is available from SpringerLink.

  Abstract: This article formulates and axiomatizes a conditional expected utility model that allows a decision maker to specify his actions in the form of partial rather than complete contingency plans and to simultaneously choose goals and actions in end-mean pairs. Both utility and probability are conditioned on selected goals and actions (g,a), and both are defined over the same set of possible (g,a)-conditioned events. For adaptive sequential decision problems, this symmetrical treatment of utility and probability permits agents to learn via "criterion filtering." That is, the expected utility criterion function can be directly updated in each decision period via transitional utility assessments in a manner analogous to Bayes' rule for updating probability distributions via transitional probability assessments.

  "C³ Modeling with Symmetrical Rationality", Applied Mathematics and Computation, Vol. 6 (1980), pp. 51-61. The published article is available from Science Direct.

  Abstract: In the absence of contrary information, it would seem prudent for competitors to attribute to their opposition the same level of rationality they attribute to themselves. Using a simple but interesting C³ (command, control, and communication) problem for illustration, a method is proposed for incorporating symmetrical rationality without resorting to the general multistage game framework which has proved difficult to apply in practice. A "criterion filtering" technique is then proposed for the approximate solution of the resulting model which does not require integration operations and which appears to be especially well suited for C³ problems with finite admissible control sets.

  "Direct Updating of Intertemporal Criterion Functions for a Class of Adaptive Control Problems", IEEE Transactions on Systems, Man, and Cybernetics, Vol. SMC-9 (1979), pp. 143-151. The published article is available from IEEE Xplore.

  Abstract: This article extends the criterion filtering approach originally developed in Tesfatsion (JOTA,1978) to intertemporal stochastic optimization problems. It demonstrates the feasibility of directly updating dynamic programming value functions on the basis of sequentially obtained return assessments, bypassing the need for explicit probability updating via Bayes' rule. The performance of the method for a class of adaptive control problems is systematically explored using computational experiments.

  "Two Solution Techniques for Adaptive Reinvestment: A Small-Sample Comparison" (with R. Kalaba), Journal of Cybernetics, Vol. 8 (1978), pp. 101-111.

  Abstract: Two adaptive control techniques are compared in the context of an adaptive reinvestment two-armed bandit problem. The first solution is the Bayesian-dynamic programming approach. The distinctive feature of the second solution technique -- the "criterion filtering" technique -- is the direct estimation and updating of the criterion function without recourse to explicit probability updating via Bayes' rule. The control selections generated by the two solution techniques are shown to closely approximate each other. An explanation for this close approximation is provided by means of an equivalence theorem for control objective specification.

  "A New Approach to Filtering and Adaptive Control", Journal of Optimization Theory and Applications (JOTA), Vol. 25 (1978), pp. 247-261. The published article is available from SpringerLink.

  Abstract: A new approach to adaptive control is proposed, referred to as criterion filtering. The principle distinguishing feature of criterion filtering is the direct updating of the current expected return function by means of a filtering operation on a vector of past return functions. The data storage and computational problems often associated with explicit probability updating via Bayes' rule are thus avoided.

  "A New Approach to Filtering and Adaptive Control: Stability Results", Applied Mathematics and Computation, Vol. 4, No. 1 (1978), pp. 27-44. The published article is available from Science Direct.

  Abstract: In a companion study [Tesfatsion (JOTA, 1978)] a new approach to adaptive control is proposed referred to as "criterion filtering." The principle distinguishing feature of criterion filtering is the direct updating of the current expected return function by means of a filtering operation on a vector of past return functions. In this paper convergence properties are established for a simple linear criterion function filter designed for a class of adaptive control problems typified by a well-known two-armed bandit problem.

  "A New Approach to Filtering and Adaptive Control: Optimality Results", Journal of Cybernetics, Vol. 7 (1977), pp. 133-146.

  Abstract: In a companion study [Tesfatsion (JOTA, 1978)], a new approach to adaptive control is proposed referred to as "criterion filtering." The principle distinguishing feature of criterion filtering is the direct updating of the current expected return function by means of a filtering operation on a vector of past return functions. In this paper sufficient conditions are established for control variables selected in accordance with a simple linear criterion filter to converge to a global maximum of the true criterion function. When states depend nontrivially on control variable selection, the decision maker determines the trade-off between rate of convergence and global optimality by the choice of the greatest lower bound for the prior (initial period) criterion function. When states are independent of control variable selection, the asymptotic global optimality of control variable selections holds under weak restrictions.

Economic Theory (Analytical)

Efficiency and Aggregation Issues in Overlapping Generations Models

• "Active Intermediation in a Monetary Overlapping Generations Economy" (pdf,225K) (with M. Pingle), Journal of Economic Dynamics and Control 22 (1998), pp. 1533-1574. The published article is also available from Science Direct.
  Abstract: Why does the First Welfare Theorem fail in standard monetary overlapping generations economies such as Grandmont (Econometrica,1985), and how is this failure related to business cycle fluctuations? This study argues that the failure of the First Welfare Theorem, and the potentially exotic nature of the business cycle fluctuations that arise, can both be attributed to the passive intermediation role played by the Walrasian Auctioneer in these models. Specifically, the study demonstrates that when private intermediaries pursuing normal profit-maximizing activities are introduced into such models, a Pareto efficient outcome is ensured and only simple periodic types of business cycle fluctuations can occur in equilibrium.

  "Active Intermediation in Overlapping Generations Economies with Production and Unsecured Debt" (pdf,187K) (with M. Pingle), Macroeconomic Dynamics 2 (1998), pp. 183-212.

  Abstract: Why does the First Welfare Theorem fail in standard overlapping generations economies with production, such as Diamond (AER, 1965) and Tirole (Econometrica, 1985)? This study argues that the reason for this failure can be attributed to the passive intermediation role played by the Walrasian Auctioneer in this models. Specifically, the study demonstrates that when intermediation is instead modeled as a contestable activity carried out by corporate intermediaries owned by consumer-shareholders and operated in their interest, every equilibrium is Pareto efficient.

  "Overlapping Generations, Intermediation, and the First Welfare Theorem" (with M. Pingle), Journal of Economic Behavior and Organization 15 (1991), pp. 325-345. The published article is available from Science Direct.

  Abstract: Why does the First Welfare Theorem fail in standard pure-exchange overlapping generations economies such as Samuelson (Journal of Political Economy, 1958)? This study argues that the failure can be attributed to the passive intermediation role played by the Walrasian Auctioneer in such models. Specifically, no agent actually residing within the model is permitted to exploit the earnings opportunities that can arise from mediating intertemporal trade. The study shows that the introduction of active profit-seeking private intermediaries in such models restores the First Welfare Theorem.

  "Walras' Law in Overlapping Generations Economies" [ (pdf,68K), (ps,133K)], (with M. Pingle), Economic Report No. 34, Iowa State University, revised October 1997.

  Abstract: It is widely known that Walras' Law can fail for an overlapping generations economy. This report demonstrates that: (1) when this failure occurs, it represents an economic opportunity that can be exploited by the issuance of unsecured debt by private agents; and (2) when unsecured debt is issued, Walras' Law does not fail.

  "Macrodynamic Implications of Income Transfer Policies for Human Capital Investment and School Effort" (with P. Orazem), Journal of Economic Growth 2 (November 1997), pp. 305-329. The published article is available from SpringerLink. This article is a revised abridged version of "Human Capital Investment and the Locally Rational Child" [ (pdf,534K), (ps,255K)] (with P. Orazem), Economic Report No. 31, Iowa State University, March 1997.

  Abstract: What happens when children are able to affect the productivity of human capital investment by endogenously varying the degree of effort they exert in school? This study explores the effectiveness of tax-financed human capital investment for the education of children when the intensity of effort which children devote to schooling depends on their expected rate of return as proxied by the rate of return experienced by their parents. The context of the investigation is an overlapping generations model consisting of multiple dynastic families with randomly determined ability levels. The main finding is that endogenous schooling effort choices by children can distort the intended effects of government income transfers meant to counter the imperfect access of poor families to credit markets.

  "Macro Implications of Government Redistributive Tax-Transfer Policies", Journal of Public Economics 19(2) (November 1982), pp. 139-169. The published article is available from Science Direct.

  Abstract: Aggregation conditions permitting macro variables to be expressed as functions of other macro variables have long been sought in macroeconomics. For example, conditions have been sought permitting aggregate consumption to be expressed as a function of aggregate income, independently of the distribution of income across agents. This article uses a 3-period lived overlapping generations economy to examine the extent to which macro variables such as aggregate consumption are invariant to government attempts to redistribute income across agents through selective tax-transfer policies. It shows that conditions for invariance in this context are much stronger than found for the standard Walrasian general equilibrium economy.

  "Welfare Implications of Net Social Security Wealth", Journal of Public Economics 24(1) (1984), pp. 1-27. The published article is available from Science Direct.

  Abstract: Real net social security wealth (NSSW), the real present value of social security benefits received minus social security taxes paid, is frequently used as a direct proxy measure for the impact of a social security system on generation welfare. This article examines the relationship between real NSSW and generation welfare in a 3-period lived overlapping generations economy. It demonstrates that NSSW can be a very poor proxy for the effects of a social security system on generation welfare. For example, NSSW can actually be negatively correlated with welfare for every generation due to price distortion effects.

---

Time Consistency and Boundedly Rational Optimization

• "Time Inconsistency of Benevolent Government Economies" (pdf,1.7M), Journal of Public Economics 31 (1986), pp. 25-52. The published article is available from Science Direct.
  Abstract: Why do government policymakers in open-ended dynamic economic models with period-by-period re-optimization tend to exhibit time inconsistency, in the sense that they systematically deviate in later periods from earlier planned policy paths? This article develops necessary and sufficient conditions for time consistency for a general class of dynamic Walrasian economies that includes many previous economic models (Brock, Calvo, Kydland-Prescott, Fischer, etc.) as special cases. It is shown that the time inconsistency of government policymakers can be explained in this class of models as the consequence of successive structural changes in the constraints faced by the government policymakers as private agents carry out decisions in each successive time period conditional on anticipated future government policy settings.

  "Fair Division with Uncertain Needs and Tastes" (pdf,738K), Social Choice and Welfare 2 (1985), pp. 295-309. The published article is available from SpringerLink.

  Abstract: This article examines the extent to which utilitarian social welfare maximization results in an egalitarian outcome for the classic n-agent cake-cutting problem when the agents are characterized by heterogenous subsistence needs as well as by heterogeneous tastes. It is shown that utilitarian social welfare maximization results in surplus-egalitarianism---i.e., first meeting needs, then equally dividing the remaining cake---when tastes are uncertain but needs are known, can be met, and are required to be met by prior restriction. If the meeting of needs is not imposed as a lexicographically prior principal of fairness, however, then---due to the fundamental non-concavity of individual agent utility functions resulting from the introduction of subsistence needs---utilitarian social welfare maximization requires that agents with high subsistence needs be allowed to die with zero shares as a matter of general policy. Moreover, when needs and tastes are both uncertain, an egalitarian allocation only results if the meaning of subsistence needs is suitably weakened to a poverty line definition.

  "Dynamic Investment, Risk Aversion, and Foresight Sensitivity" (pdf,1.5M), Journal of Economic Dynamics and Control 3 (1981), pp. 65-96. The published article is also available from Science Direct.

  Abstract: Since optimal investment strategies generally cannot be obtained in closed form when consumers exhibit non-constant risk aversion, many dynamic investment studies have focused on the constant risk aversion case. This study considers a general class of dynamic investment models in which agents are not restricted to have constant risk aversion. Existence, monotonicity, concavity, differentiability, and absolute risk aversion properties are established analytically for the optimal feedback investment strategies and dynamic progamming value functions. A bound is also obtained on the loss in expected utility resulting from the use of limited foresight horizons. Finally, results are reported for computer experiments conducted to explore the sensitivity of expected utility outcomes to changes in the length of the foresight horizon. For example, the results indicate the optimality of a rolling 2-period foresight horizon for a particular class of exponential utility functions exhibiting decreasing risk aversion.

  "Global and Approximate Global Optimality of Myopic Economic Decisions" (pdf,1.4M), Journal of Economic Dynamics and Control 2 (1980), pp. 135-160. The published article is available from Science Direct.

  Abstract: In actual problem contexts, the time horizon over which plans are formulated must generally be short in relation to the history of the process as a whole. What loss of return is entailed by the use of these relatively short planning horizons? This article develops a general discrete-time dynamic stochastic control model that encompasses many well-known economic models. It derives sufficient conditions in this context for the equivalence of myopic (single period) and global (simultaneous multiple period) expected return maximization, and it provides a bound for the loss in global return when these conditions are not met. It also identifies properties of proxy short-horizon return functions which can be used to partially order them in terms of overall expected return performance.

  "A Note on the Dependence Between a Venture and a Current Prospect" (with Clifford Hildreth), Journal of Economic Theory 15(2) (August 1977), pp. 381-391. The published article is available from Science Direct.

  Abstract: Some sufficient conditions that a random variable be positively correlated with every strictly decreasing function of a second random variable are developed and applied to the problem of choosing the optimal amount of an uncertain venture. Two of the conditions generalize conditions previously employed by Samuelson and Scheffman.

  "Stochastic Dominance and the Maximization of Expected Utility" Review of Economic Studies 43 (1976), pp. 301-315. The published article is available from JSTOR (pdf,1.2M).

  Abstract: This article clarifies and extends various basic relationships between stochastic dominance and the maximization of expected utility.

---

General Game Theory Research

• "Pure Strategy Nash Equilibrium Points and the Lefschetz Fixed Point Theorem", International Journal of Game Theory 12 (1983), pp. 1810-191.
  Abstract: Using the Lefschetz fixed point theorem, a pure strategy Nash equilibrium existence theorem is established for a class of n-person games with possibly nonacyclic strategy sets. It is argued that the Lefschetz approach to fixed point theorems may ultimately prove to be particularly important in economic and game theory due to the generality of spaces that can be considered and the interesting related questions that can be investigated. For example, the Lefschetz approach to fixed point theorems leads naturally to the concept of a "Nielsen Number" of a map f:Y->Y, a homotopy-invariant lower bound for the number of fixed points of f. The Nielsen number provides a lower bound for the number of Nash equilibria in certain n-person games.

  "A Stability Theorem for Symmetrically Rational Counterplanning" (with R. E. Kalaba and K. Spingarn), Journal of Optimization Theory and Applications 37(3) (July 1982), pp. 379-385. The published article is available from SpringerLink.

  Abstract: In the absence of contrary information, it would seem prudent for a competitor to attribute to his opponents the same level of rationality that he himself employs. In the context of a general linear-quadratic nonzero-sum two-person game, it is shown that a counterplanning procedure consistent with this principle of symmetrical rationality always converges to the unique Nash equilibrium for the game.

Econometrics and Data Analysis

Multicriteria Estimation via Cost-Efficient Frontiers

• Open-Source Software Release: Flexible Least Squares (FLS) and Generalized Flexible Least Squares (GFLS) for Multicriteria Model Estimation (homepage). Copyright holders (Artistic License): Robert Kalaba and Leigh Tesfatsion.

• "A Multicriteria Approach to Model Specification and Estimation" (pdf,1.6MB) (with R. Kalaba), Computational Statistics and Data Analysis 21 (1996), pp. 193-214. The published article is also available from Science Direct.
  Abstract: This study considers why multicriteria techniques have not been widely adopted in econometrics to date. It then presents a multicriteria approach to estimation problems for which the basic objective is to learn about the sequence of states through which a process has passed. The multicriteria approach involves the construction of a "cost efficient frontier" which determines the set of state trajectory estimates that are minimally incompatible with a specified set of model criteria. This approach includes flexible least squares (FLS) and generalized flexible least squares (GFLS) as special cases; see the articles on FLS and GFLS cited below. The study also surveys recent theoretical and empirical work that makes use of FLS and GFLS.

  A Unified Approach to Dynamic Estimation" (with Robert Kalaba), Information Sciences 57-58 (September-December 1991), pp. 159-169. The published article is available from Science Direct.

  Abstract: Discrepancies between assumed dynamical models and observations are often handled by making further probabilistic assumptions, a tactic which has both strengths and weaknesses. A re-examination of filtering and smoothing is conducted, and an alternative multicriteria approach, which is probability free, is advanced. This approach involves vector minimization as a key ingredient, and it specializes to the well-known Kalman, Viterbi, Larson-Peschon, and Swerling filters.

  Flexible Least Squares for Approximately Linear Systems (pdf,1.2MB) (with R. Kalaba), IEEE Transactions on Systems, Man, and Cybernetics 20, No. 5 (1990), pp. 978-989.

  Abstract: This article proposes a multicriteria "generalized flexible least squares (GFLS)" method for the smoothing and filtering of state-space systems described by approximately linear dynamic and measurement relations. The basic GFLS objective is to determine the "cost-efficient frontier," that is, the set of state trajectory estimates that are efficient in the sense that they are minimally incompatible with the specified dynamic and measurement relations.

  NOTE: A program implementation for GFLS has been incorporated into the statistical package GAUSS. For more detailed information about GFLS software availability, visit the FLS Home Page

  "U.S. Money Demand Instability: A Flexible Least Squares Approach" (pdf,1.4MB) (with J. Veitch), Journal of Economic Dynamics and Control 14, No. 1 (February 1990), pp. 151-173. The published article is available from Science Direct.

  Abstract: This article uses the flexible least squares (FLS) method to investigate coefficient stability for the Goldfeld U.S. money demand model over the volatile period 1959:Q2 to 1985:Q3. The only constraint imposed on coefficient variation over time is a smoothness prior. Nevertheless, the time paths traced out by the FLS coefficient estimates exhibit systematic idiosyncratic time variations as well as simultaneous shift movements in 1974 during the time of the first oil price shock. Moreover, the FLS estimates also indicate that the "unit root" nonstationarity problem reported by OLS money demand studies disappears if the coefficients are allowed to exhibit even small amounts of time variation.

  "An Organizing Principle for Dynamic Estimation" (with R. Kalaba), Journal of Optimization Theory and Applications, Vol. 64, No. 3 (March 1990), 445-470. The published article is available from SpringerLink.

  Abstract: This paper develops a general multicriteria framework for the sequential estimation of process states. Three well-known state estimation algorithms (the Viterbi, Larson-Peschon, and Kalman filters) are derived as monocriterion specializations. The multicriteria estimation framework is used to clarify both Bayesian and classical statistical procedures for treating potential model specification errors. A recently developed bicriteria specialization (flexible least cost), explicitly designed to take specification errors into account, is also reviewed. The latter application suggests how the multicriteria framework might be used to construct estimation algorithms capable of handling disparate sources of information coherently and systematically, without forced scalarization.

  "Time-Varying Linear Regression Via Flexible Least Squares" (pdf,2.4M) (with R. Kalaba), Computers and Mathematics with Applications 17 (1989), pp. 1215-1245. The published article is available from Science Direct.

  Abstract: This article develops a multicriteria "flexible least squares (FLS)" method for time-varying linear regression. The basic FLS objective is to determine the "residual efficiency frontier," that is, the set of all coefficient trajectory estimates that yield vector-minimal sums of squared residual measurement and dynamic errors conditional on a given set of observations.

  NOTE: A program implementation for FLS has been incorporated into the statistical packages GAUSS and SHAZAM. For more detailed information about FLS software availability, visit the FLS Home Page

  "Sequential Nonlinear Estimation with Nonaugmented Priors" (with R. Kalaba), Journal of Optimization Theory and Applications, Vol. 60, No. 3 (March 1989), pp. 421-438. The published article is available from SpringerLink.

  Abstract: A "flexible least cost method" is proposed for investigating the basic compatibility of theory and observations in the absence of valid or known stochastic characterizations for residual error terms.

  "The Flexible Least Squares Approach to Time-Varying Linear Regression" (pdf,360K) (with Robert Kalaba), Journal of Economic Dynamics and Control 12(1) (March 1988), 43-48. The published article is available from Science Direct.

  Abstract: This article proposes a "flexible least squares" (FLS) method for state estimation when the dynamic equations are unknown but the process state evolves only slowly over time. A smoothness prior is introduced in place of an explicit specification for the unknown dynamic equations governing the evolution of the process state. Simulation experiments illustrating the method are presented.

  "Exact Sequential Filtering, Smoothing, and Prediction for Nonlinear Systems" (with R. Kalaba), Nonlinear Analysis, Vol. 12, Issue 6, June 1988, 599-615. The published article is available from Science Direct.

  Abstract: This article develops two algorithms for the exact sequential updating of the optimal solution for a general discrete-time nonlinear least squares estimation problem as the process length increases and new observations are obtained. One algorithm proceeds via an imbedding on the process length and the final state vector. The second algorithm proceeds via an imbedding on the process length and the final observation vector. Each algorithm generates optimal (least cost) filtered and smoothed state estimates, together with optimal one-step-ahead state predictions.

  "A Sequential Method for Nonlinear Filtering: Numerical Implementation and Comparisons" (with R. Kalaba and K. Spingarn), Journal of Optimization Theory and Applications, Vol. 34, No. 4 (August 1981), pp. 1144-1149. The published article is available from SpringerLink.

  Abstract: Exact equations are presented for sequentially updating the optimal solution for a discrete-time analog of the basic Sridhar nonlinear filtering problem as the process length increases and new observations are obtained. A tabular method is described for implementing numerically the sequential filtering equations. The accuracy and efficiency of the tabular method are illustrated by means of several numerical examples.

  "An Exact Sequential solution Procedure for a Class of Discrete-Time Nonlinear Estimation Problems" (pdf,595K) (with R. Kalaba), IEEE Transactions on Automatic control 26 (1981), pp. 1144-1149. The published article is available at IEEE Xplore.

  Abstract: This article develops an exact procedure for sequentially updating the optimal solution for a general discrete-time nonlinear least squares estimation problem as the process length increases and new observations are obtained.

  "A Least-Squares Model Specification Test for a Class of Dynamic Nonlinear Economic Models with Systematically Varying Parameters" (with R. Kalaba), Journal of Optimization Theory and Applications, Vol. 32, No. 4 (December 1980), pp. 537-567. The published article is available from SpringerLink.

  Abstract: This study develops a least-squares measure for simultaneously testing the basic compatibility of prior dynamical, observational, and distributional model specifications against actual data for a class of dynamic nonlinear economic models with parameters explicitly modeled as nonlinear functions of endogenous and exogenous variables. Using invariant imbedding techniques, an algorithm is derived for sequentially updating the optimal least-squares estimates for parameters, endogenous variables, and squared residual modeling error sums as the duration of the process increases and new observations are obtained.

---

Nonlinear Estimation via Associative Memories

• Open-Source Software Release: Multicriteria Associative Memories (MAM) for Nonlinear Estimation (homepage). Copyright holders (Artistic License): Robert Kalaba and Leigh Tesfatsion

• "Linear and Nonlinear Associative Memories for Parameter Estimation" (with R. Kalaba, Z. Lichtenstein, and T. Simchony), Information Sciences 61, Issues 102 (April 1992), pp. 45-66. The published article is available from Science Direct.
  Abstract:This article discusses the use of associative memories for obtaining preliminary parameter estimates for nonlinear systems.

  "Obtaining Initial Parameter Estimates for Nonlinear Systems Using Multicriteria Associative Memories" (pdf,1.3M) (with R. Kalaba), Computer Science in Economics and Management (now Computational Economics) 4 (1991), pp. 237-259.

  Abstract: Parameter estimation problems for nonlinear systems are typically formulated as nonlinear optimization problems. For such problems, one has the usual difficulty that standard successive approximation schemes require good initial estimates for the parameter vector. This article develops a simple multicriteria associative memory (MAM) method for obtaining good initial estimates for nonlinear parameter estimation problems. The method is implemented in Fortran by the MAM program

Computational Methods

Solution Tracking for General Parameterized Nonlinear systems

• Open-Source Software Release: Nonlocal Automated Sensitivity Analysis (NASA) (homepage). Copyright holders (Artistic License): Robert Kalaba and Leigh Tesfatsion.

• "Nonlocal Sensitivity Analysis with Automatic Differentiation" [ (ps,249K), (pdf,114K)], C. A. Floudas and P. M. Pardalos (eds.), Encyclopedia of Optimization, Kluwer Academic Publishers, Volume 4, 2001, pp. 92-97
  Abstract: This encyclopedia entry is a summary report on the design and use of the NASA program for the Nonlocal Automated Sensitivity Analysis of parameterized nonlinear systems H(x,b)=0.

  "Nonlocal Automated Comparative Static Analysis", Computational Economics 5(4) (November 1992), pp. 313-331. The published article is available from SpringerLink.

  Abstract: This paper reviews work on the development of the NASA program for the automated comparative static analysis of parameterized nonlinear systems over parameter intervals. NASA incorporates a fast and efficient algorithm Feed for the automatic evaluation of higher-order partial derivatives, as well as an adaptive homotopy continuation algorithm for obtaining all required initial conditions. Applications are envisioned for fields such as economics where models tend to be complex and closed-form solutions are difficult to obtain.

  "Nonlocal Automated Sensitivity Analysis" (pdf,1M), (with Robert Kalaba), Computers and Mathematics with Applications Volume 20, Issue 2, 1990, 53-65. The published article is available from Science Direct.

  Abstract: This article presents and illustrates the applicability of the NASA program for the Nonlocal Automated Sensitivity Analysis of parameterized nonlinear systems H(x,b)=0. The NASA program incorporates automated procedures for initialization, derivative evaluation, and the tracking of solution branches x(b) along user-designated paths for the parameter vector b.

  "Nonlocal Sensitivity Analysis, Automatic Derivative Evaluation, and Sequential Nonlinear Estimation" (with Robert Kalaba), Computational Statistics and Data Analysis 4(2) (July 1986), pp. 79-81. The published article is available from Science Direct.

  Abstract: This article summarizes work by the authors on computational methods for nonlinear systems. Section 2 develops a complete set of ordinary differential equations for generating solutions to parameterized systems of nonlinear equations over parameter intervals of interest. Section 3 presents a simple finite algorithm, FEED, for the exact forward-mode automatic evaluation of higher-order partial derivatives using derivative arrays. Section 4 obtains an exact sequential characterization of the solution to a general nonlinear least-squares estimation problem as the duration of the process increases and additional observations are made.

  "Local and Nonlocal Comparative Static Analysis of Economic Systems" (with Robert Kalaba and Jone-Line Wang), Applied Mathematics and Computation Vol. 9 (1981), 227-234. The published article is available from Science Direct.

  Abstract: The complete system of ordinary differential equations developed by Kalaba and Tesfatsion (1981) for tracking solution branches of parameterized nonlinear systems (see the next article) is tested using several illustrative examples. One example is the standard Ramsey optimal growth model, for which analytical solutions can be obtained. For this example, the complete system is used to generate solutions c(rho) and k(rho) for the steady-state per-capita levels for consumption and capital as the time preference parameter rho varies from 0 to 0.50. Accuracy to four decimal places is obtained. This represents a stringent test of the method, since the derivative of k(rho) near rho=0 is on the order of -10² whereas the derivative of k(rho) near rho=0.50 is on the order of -10⁰.

  "Complete Comparative Static Differential Equations" (with R. Kalaba), Nonlinear Analysis: Theory, Methods, and Applications 5(8) (August 1981), pp. 821-833. The published article is available from Science Direct.

  Abstract: This article develops a complete system of ordinary differential equations for tracking solution branches x(b) of a parameterized system of nonlinear equations H(x,b) = 0 along arbitrary user-designated paths for the parameter vector b. The complete system is obtained by augmenting the usual fundamental equation of comparative static analysis,
  dx(b)/db = -H_x(x(b),b)^-1H_b(x(b),b), with ordinary differential equations for the Jacobian inverse. The feasibility and accuracy of the method are illustrated by several numerical examples.

  Note: This complete differential system constitutes the core of the NASA Program for the Nonlocal Automated Sensitivity Analysis of parameterized nonlinear systems H(x,b)=0.

  "Parameter-Sensitivity Study for a Linear-Quadratic Control Problem with Random State Coefficients", Information Sciences 15(2), (July 1978), pp. 101-125. The published article is available from Science Direct.

  Abstract: This study undertakes a detailed parameter sensitivity study for the optimal feedback-control law and dynamic programming optimality equations associated with a discrete-time finite-horizon linear-quadratic control problem with random state coefficients. Both analytical and computational findings are reported. One interesting characteristic revealed by the analysis is the existence of simple linear relationships between parameter sensitivities for the optimal feedback-control selections and for the corresponding dynamic programming value ("cost-to-go") functions.

---

Tracking of Eigenvalues and Eigenvectors for General Parameterized Matrices

NOTE: All of the eigenvalue/eigenvector tracking algorithms below can be implemented via the NASA Program for the Nonlocal Automated Sensitivity Analysis of parameterized nonlinear systems H(x,b)=0.

• "Variational Equations for the Eigenvalues and Eigenvectors of Nonsymmetric Matrices" (pdf,359K) (with R. Kalaba and K. Spingarn), Journal of Optimization Theory and Applications 33 (1981), pp. 1-8. The published article is available from SpringerLink.
  Abstract: This article develops a complete system of ordinary differential equations for tracking the eigenvalues and the right and left eigenvectors of nonsymmetric parameterized matrices over parameter intervals. A simpler reduced form of the ODE system is then derived for tracking the eigenvalues and eigenvectors of symmetric parameterized matrices. The feasibility and accuracy of the tracking method are illustrated by numerical examples.

  "Individual Tracking of an Eigenvalue and Eigenvector of a Parameterized Matrix" (with R. Kalaba and K. Spingarn), Nonlinear Analysis: Theory, Methods, and Applications 5(4) (1981), pp. 337-340. The published article is available from Science Direct.

  Abstract: This article develops a complete system of ordinary differential equations for tracking a single eigenvalue of a parameterized matrix, together with one of its corresponding right or left eigenvectors, over parameter intervals. The feasibility and accuracy of the method are illustrated by numerical examples.

  "A New Differential Equation Method for Finding the Perron Root of a Positive Matrix" (with R. Kalaba and K. Spingarn), Applied Mathematics and Computation 7(3), (October 1980), pp. 187-193. The published article is available from Science Direct.

  Abstract: This article develops a complete system of ordinary differential equations for tracking the Frobenius-Perron root (largest eigenvalue) of a parameterized matrix, together with a unit-normalized right eigenvector, over parameter intervals. The feasibility and accuracy of the method are illustrated by numerical example.

---

Adaptive Homotopy Continuation

• "Solving Nonlinear Equations By Adaptive Homotopy Continuation" (pdf,1MB) (with R. Kalaba), Applied Mathematics and Computation 41, No. 2:Part II (January 1991), pp. 99-115. The published article is available from Science Direct.
  Abstract: This article introduces and constructively illustrates the concept of an adaptive homotopy for solving systems of nonlinear equations. Standard homotopy methods rely on a passive continuation parameter moving from 0 to 1 along the real line and are stymied if the homotopy Jacobian matrix becomes ill-conditioned along this path. In contrast, an adaptive homotopy replaces the passive continuation parameter by a "smart agent" that adaptively makes its way by trial and error from 0+0i to 1+0i in the complex plane in accordance with certain specified objectives. The homotopy thus adapts to the physical problem at hand rather than requiring the user to reformulate his physical problem to conform to homotopy requirements. The adaptive homotopy algorithm designed and tested in the current study incorporates two objectives: (a) short continuation path; and (b) avoidance of regions where the homotopy Jacobian matrix becomes ill-conditioned.

  Note: This adaptive homotopy method has been incorporated into the NASA Program for the Nonlocal Automated Sensitivity Analysis of parameterized nonlinear systems H(x,b)=0 in order to generate all needed initial conditions automatically.

---

Automatic Evaluation of Higher-Order Partial Derivatives

• "Automatic Evaluation of Higher-Order Partial Derivatives for Nonlocal Sensitivity Analysis", pp. 157-165 in A. Griewank and G. Corliss (eds.), Automatic Differentiation of Algorithms: Theory, Implementation, and Application, SIAM, Philadelphia, 1991.
  Abstract: This proceedings paper surveys work on the FEED (Fast Efficient Evaluation of Derivatives) algorithm originally developed by Kalaba, Tesfatsion, and Wang (1983), with particular reference to the use of FEED for the implementation of nonlocal automated sensitivity techniques; see the articles below. The relationship of FEED to the automatic differentiation algorithms developed by other SIAM Workshop participants is clarified.

  "Automation of Nested Matrix and Derivative Operations" (pdf,863K) (with Robert Kalaba and Thomas Plum), Applied Mathematics and Computation 23(3) (September 1987), pp. 243-268. The published article is available from Science Direct.

  Abstract:In Kalaba, Tesfatsion, and Wang (1983) - see below -- an algorithm was developed for the exact forward-mode automatic evaluation of higher-order partial derivatives of functions of many variables using derivative arrays. This algorithm was supported by a library of "calculus subroutines" for many standard one-variable and two-variable functions. This article extends the FEED library to permit the automatic differentiation of expressions involving nested matrix and derivative operations. The need to differentiate such expressions arose naturally in the course of designing sequential filters for a class of nonlinear tracking problems.

  "Automatic Differentiation of Functions of Derivatives" (with Robert Kalaba), Computers and Mathematics with Applications, Volume 12, Issue 11, Part 1, November 1986, 1091-1103. The published article is available from Science Direct.

  Abstract: In Kalaba, Tesfatsion, and Wang (1983) -- see below -- the FEED algorithm was introduced for the exact forward-mode automatic evaluation of higher-order partial derivatives of functions of many variables using derivative arrays. This study demonstrates that FEED can be applied to a much broader class of functions than originally envisioned: namely, functions defined in terms of the derivatives of other functions. Such functions arise in many applications.

  "A Finite Algorithm for the Exact Evaluation of Higher Order Partial Derivatives of Functions of Many Variables" (with R. Kalaba and J.-L. Wang), Journal of Mathematical Analysis and Applications 92 (1983), pp. 552-563. The published article is available from Science Direct.

  Abstract: This article develops an algorithm for the exact forward-mode automatic evaluation of higher order partial derivatives using derivative arrays, now referred to as the FEED (Fast Efficient Evaluation of Derivatives) algorithm. Building on previous work by R. Wengert, the FEED algorithm proceeds by decomposing the evaluation of complicated functions of many variables into a sequence of simpler evaluations of special functions of one or two variables.

  Note: A library of FEED calculus subroutines has been incorporated into the NASA program for the Nonlocal Automated Sensitivity Analysis of parameterized nonlinear systems H(x,b)=0 in order to generate all needed derivative evaluations automatically.