Actions

1999

From Santa Fe Institute Events Wiki

Workshop Navigation


Program Announcement

Program Announcement 1999

Students

  • Alessandra Cassar, UCSC [email]
  • Sean Gailmard, Cal Tech [email]
  • Mike Gibney, University of London [email]
  • Brit Grosskopf, Universitat Pompeu Fabra [email]
  • Serena Guarnaschelli, Cal Tech [email]
  • Kelly Lautt, UCLA [email]
  • Jim Leady, University of Michigan [email]
  • Artur Minkin, University of Wisconsin [email]
  • David Robalino, Rand [email]
  • Sonia Schulenburg, University of Edinburgh [email]

Schedule

Schedule 1999

Faculty

  • Lars Erik Cederman, Political Science, UCLA.
  • Jim Crutchfield, Physics, SFI.
  • Scott deMarchi, Political Science, Duke.
  • Doyne Farmer, Prediction Co.
  • Alex Lancaster, Swarm, Santa Fe Institute.
  • John H. Miller, Economics, Carnegie Mellon University (co-director).
  • Mark Newman, Physics, SFI.
  • Scott E. Page, Economics, Iowa (co-director).

Homework

Homework Problem 1999

Student Projects

Each student began a research project during the two-week workshop. Below are brief descriptions of these various projects. These projects will form the basis for dissertation chapters and/or journal articles.

Alessandra Cassar, UCSC (cassar@cats.ucsc.edu).
Alessandra is investigating how local interactions among banks can precipitate financial crises. In her model, banks receive deposits and underwrite loans to customers, while simultaneously using transfers among "neighboring" banks to balance funds. In such a system, insolvent banks can begin a contagion cycle in which other banks are forced to default. She finds that different connection topologies imply different tradeoffs between maintaining liquidity and avoiding insolvency. The work suggests that policies designed to enhance the right kind of connectivity may greatly improve the robustness of financial systems.

Sean Gailmard, Cal Tech (gailmard@hss.caltech.edu).
Sean is looking at adaptive mechanism design for effectively controlling a common property resource, such as a fishery. The model has fishery managers attempting to better control the fishery by adapting a control mechanism, here consisting of target fish populations, net size restrictions, and violation penalties. He finds that adaptive agents can discover productive mechanisms to control this nonlinear system, though depending on the objective function, complete degradation can still ensue. The computational approach should provide a productive laboratory for mechanism design research.

Mike Gibney, University of London (M.A.Gibney@qmw.ac.uk).
Mike is using a market-based control mechanism to design a distributed control system, such as a telecommunication network. The initial work explores how such mechanisms can effectively allocate complementary goods. Such goods naturally arise in telecommunication, for example, the value of controlling a link between two locations often depends on what other links in the system the agent currently controls. The early evidence indicates that a market mechanism can result in resource allocations that are quite effective at allocating existing resources. Moreover, this high performance is achieved using either a homogeneous collection of the best-known trading agent or a more heterogeneous collection of all types of agents.

Brit Grosskopf, Universitat Pompeu Fabra (grosskop@upf.es).
Brit is exploring models of incomplete information using artificial adaptive agents. The agents are placed in a simple coordination game with a Pareto inferior, but risk dominant, outcome, and they are given different levels of information about the play of the opponent. This environment matches the one she used in some previously conducted human experiments. Finite automata are coevolved in this world using a genetic algorithm. She finds that these agents tend towards the risk-dominant equilibria---perhaps due to an inherent preference of evolutionary systems to promote "survival of the adequate" in such systems.

Serena Guarnaschelli, Cal Tech (serena@hss.caltech.edu).
Serena is analyzing information aggregation in a double auction market using a common value good. She has identified a variety of candidate strategies from the literature, ranging from highly rational to somewhat irrational. By combining these strategies into various ecologies, she hopes to better understand the implications of heterogeneous collections of agents both at the individual and aggregate level in such markets.

Kelly Lautt, UCLA (Kellyl@ucla.edu).
Kelly is interested in the gifts-for-goods corruption that has recently arisen in China. In this system, citizens may proffer gifts to officials in hopes of receiving key goods. Kelly models this system as a replicator dynamic, and early results indicate that the system may evolve in and out of corruption depending on the underlying norms of gifting and the penalties imposed on corrupt officials.

Jim Leady, University of Michigan (leady@umich.edu).
Jim is studying how bounded processing ability impacts the play of adaptive agents facing multiple games. In the model, agents adapt finite automata using a genetic algorithm that must play a set of standard two-by-two games: the Prisoner's Dilemma, Chicken, Battle of the Sexes, and a simple coordination game. Agent's strategies must be implemented using a limited amount of processing ability, and thus strategies for the different games may need to share various scarce resources.

Artur Minkin, University of Wisconsin (aminkin@students.wisc.edu).
Artur is developing new econometric techniques for estimating parameters in spatially connected systems. His initial focus is on the estimation of growth models with heterogeneous countries. In his work, he uses Bayesian inference to identify appropriate groups of countries. To improve the quality of the estimates, he is using artificial worlds to refine the econometric techniques before applying them to the real data.

David Robalino, Rand (David_Robalino@rand.org).
David is applying computational techniques to evaluate the implications of including social interactions in current integrated assessment models. In a preliminary application, David shows that optimal savings rates derived from the standard stochastic neoclassical growth model may tend to underestimate socially optimal savings rates. His results suggest that policy analysis ought to consider expanding current representative agent models to incorporate social networks. His approach may also prove useful to formalize the well known empirical linkage between social capital and economic growth.

Sonia Schulenburg, University of Edinburgh (sonias@dai.ed.ac.uk).
Sonia is using classifier systems to model the behavior of asset traders. Initial work has focused on creating adaptive traders that can effectively identify and trade on price series of stocks generated by actual markets. The insights gained from this work are being used to create artificial market systems composed of adaptive agents that endogenously determine prices.