1997 Graduate Workshop in Computational Economics
Student Projects
Each student began a research project during the two-week workshop.
Below are brief descriptions of these various projects.
The expectation is that these projects will form the basis for dissertation
chapters and/or journal articles.
Robert Bernard, Rutgers/C&L
Rob has created an adaptive model of an apartment rental market.
Agents, based on limited knowledge of the available housing stock, seek utility
maximizing apartments, while landlords adaptively adjust their pricing policies so as
to best fill their units. Using this model as an "adaptive flight simulator," Rob is
investigating the impact of various rent control/decontrol policies. One phenomena
that has emerged from the model is a natural tendency towards seasonal rental markets.
Diego Comin, Harvard (dcomin@kuznets.fas.harvard.edu)
Diego is investigating learning and the non-neutrality of money in a neutral
economy. Two types of agents coexist in an asset market, one of the types pays
attention to market fundamentals while the other also incorporates some outside
information, for example, projected rates of inflation. Using both replicator
dynamic and least squares learning algorithm, he finds that even a few non-fundamental
traders (2%) can seriously impact the behavior of the system.
Simon Emrich, London School of Economics (S.Emrich@lse.ac.uk)
Simon is exploring the dynamics of asset price formation. In the model agents
update trading strategies by imitating their more successful neighbors. This spatial
neighborhood structure appears to alter the price dynamics in fundamental ways.
Tomas Klos, U. of Groningen (t.b.klos@bdk.rug.nl)
Tomas is concerned with the formation of buyer/supplier relationships.
The model allows endogenous trade networks to form among agents based on
both potential profitability and past history of interactions (instantiated
in a "loyalty" parameter). Tomas is using this model to better understand how
such networks can be maintained over time, as well as how various exogenous factors,
for example, diversity in technical capabilities, alters the formation of such networks.
Brian Krauth, Wisconsin (bvkrauth@students.wisc.edu)
Brian is using computational methods to enhance his understanding of an analytic
model of job networks and social stratification. In the model, social links allow
managers to better observe the innate ability of agents in hiring decisions. Analytically,
Brian has shown that if the quality of social links exceed a critical value, then
stratification will ensue. Computationally, Brian is exploring the behavior of this
critical parameter. He has also found that an analytically questionable approximation of
the model that lends a variety of new insights into the theory, appears to capture the
fundamental behavior of the more complex model.
Sylvain Leduc, U. of Rochester (sl013d@uhura.cc.rochester.edu)
Sylvain is attempting to explain the forward discount bias puzzle arising in
international finance via an adaptive agent model. Grounded in a standard expectations
model, he allows the expectations parameters to enter adaptively. Early results indicate
that the forward discount bias disappears as more learning is allowed in the system.
Michael Lenox, MIT (mlenox@mit.edu)
Michael has created an abstract model of the role of knowledge transfer in the
diffusion of innovations. Each agent has a basic core capability to both understand
new outside innovations as well as communicate these developments to the the colleagues
to whom they are connected. Initial results with standard organizational forms
indicate that there is a tradeoff between having an organization that is quickly
able to exploit innovations once they become known inside of the firm and being
able to discover new innovations as they arise outside of the firm. Michael is
also using a genetic algorithm to understand how adaptive systems might be able
to better design such organizations.
Wei Lin, C&L
Wei has developed a model of decentralized organizational task solving.
In the model, organizations form by incorporating agents with various
skills. The model provides a bases from which to explore the dynamics of
self-organizing problem-solving organizations. Wei has found that as tasks become
more complicated, there is more cooperation among the smaller firms, while medium
sized tasks can engender complicated cycles of non-cooperative behavior.
Paolo Lupi, York (paolo@shiva.york.ac.uk)
Paolo is exploring the impact of aspiration-based learning on spatial models of
economic systems. In the model, duopoly competition occurs at every point on
a lattice. When the aspirations of the firm fall below the average profits of
all the firms in the model, the firm alters its strategy by imitating the behavior
of a randomly local firm. Notwithstanding the seemingly simple learning dynamics,
Paolo has found that strategic ideas are rapidly propagated in this setting, resulting
in the emergence of cooperation throughout the lattice.
Agostino Manduchi, Columbia (am195@columbia.edu)
Agostino is trying to understand how learning affects a spatial model of
monopolistic competition. Agents are arrayed on a circle and compete with their
immediate neighbors for consumers. Learning proceeds by imitating better-performing
firms that find themselves in relatively similar situations vis-a-vis competitors.
The results suggest that
such a learning process leads the producers to quote prices that are, on
average, above the level that would prevail if they faced different
competitors in each round of the game, and in so doing realize higher
average profits.
Salvatore Pitruzzello, Columbia (pitruzz@columbia.edu)
Salvatore wants to understand the impact of globalization on policy convergence.
He has created a dynamic model of international interaction that incorporates
many current theoretical ideas from international relations research. By incorporating
these ideas in a single, dynamic model, he can now begin to unravel the full implications
of each of the various theories.
Francisco Rodriguez, Harvard (frodrigu@kuznets.fas.harvard.edu)
Francisco is using computational methods to better understand the relationship
between inequality and redistribution in a simple voting model. Different income
classes in the electorate are courted by adaptive parties attempting to implement
different taxation policies. Although this is a simple model, standard theoretical
tools yield few insights. The computational approach has begun to generate a variety
of new insights into this problem.
Bill Watkins, UC Santa Barbara (bill@econ.ucsb.edu)
Bill is using adaptive algorithms to create rapid information diffusion networks
within an organization. In the model, agents probabilistically learn new information
from those colleagues with whom they are connected. The computational models have
revealed new organizational forms from which new analytic explorations can begin.
Peter Wurman, U. of Michigan (pete.wurman@umich.edu)
Peter is developing new forms of discrete multi-dimensional auctions which can
be used as decentralized solution mechanisms for the knapsack problem.
In these auctions, simple adaptive agents submit bids over multi-dimensional goods
tempered by information generated by more traditional knapsack algorithms.
Initial results indicate that such auctions can indeed lead to good outcomes in
this difficult domain.
John H. Miller , miller@zia.hss.cmu.edu.