2002 Graduate Workshop in Computational Economics
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
Margo Bergman, Economics, U. of Houston (firstname.lastname@example.org).
Margo is exploring the evolution of risk preferences. Agents, with
degrees of risk aversion ranging from risk loving to risk averse,
must make a series of choices over various lotteries. Agents who receive
low payoffs during the lotteries, imitate the risk preferences of more
successful agents. If agents are allowed to accumulate wealth over time,
then the system evolves toward risk loving behavior; if wealth is not
allowed to accumulate, then risk averse behavior arises. When agent evolution
is more tightly coupled to neighborhoods, important skews in the distribution
of final risk aversion appear to arise.
Matthew Cronin, Organizations, Carnegie Mellon U. (email@example.com).
Matt is investigating how groups of individuals can form productive, problem-solving
He assumes that his agents face various bounds to rationality and he also
incorporates ideas from psychology such as insight and incremental
The model has tasks represented by bit strings, with different configurations
of the bits being mapped into outcomes. Agents are able to
"know" only a few of the bits initially; with time, the agents can begin
to investigate other bits, as well as share their knowledge with others through
processes that capture notions of trust, respect, and liking.
He finds that trust improves information discovery but slows down progress,
while the respect and liking operators both increase communication and insight.
Ahlam Fakhar, Economics, Claremont Graduate University (firstname.lastname@example.org).
Ahlam is studying how norms and laws interact in social taboos, such
as drug use or prostitution. Agents have different beliefs about the
appropriateness of the taboo and perceptions about the current degree of
enforcement via the law and social norms. The beliefs of the agents get
modified based on their experiences with the law and the influence of
friendship networks. She finds that there are some important asymmetries in the
belief dynamics, and that effective policy options need to be tailored to
these various dynamics.
Nicolas Garrido, Economics, U. de Alcala de Henares (Nicolas_Garrido@brown.edu).
Nico linked an equation-based market model to an agent-based one. In this
work, he first derived an equation-based approach and solved it using
some new methodological techniques more typically used in the analysis
of solid state physical systems. He then used a closely-tied agent-based
model to test the efficacy of the new methodological techniques.
Next, he expanded the agent-based model by
allowing more elaborate forms of behavior. This research strategy
allowed him to compare the various approaches and identify key similarities
and differences among them. One result is that while the ultimate demand patterns did not differ
across the ensemble of models, adaptive firms appeared resulted in higher prices.
Robert Gazzale, Economics, U. of Michigan (email@example.com).
Bob is analyzing product differentiation with "experience" goods, that
is, with goods that must first be consumed by an agent before the good's
value can be known fully. Goods have various attributes, and these attributes
may interact with one another in a nonlinear way to produce value.
The transparency of a good is the number of attributes that an agent can
know a priori. Firms attempt to adaptively increase their profits by
developing different bundles of attributes in response to consumer demand.
He finds that as firms use more sophisticated adaptive algorithms
aggregate consumer utility increases as the firms are able to better
differentiate their products from one another in meeting the demand. Moreover,
there appears to be an important nonlinearity at lower levels of transparency.
A. Joseph Guse, Economics, U. of Wisconsin (firstname.lastname@example.org).
Joseph is studying group formation in the absence of a state. Agents,
with the possibility of forming into groups, need access to a common, and
scarce, resource to produce an output. Groups meeting at the common
resource play a hawk/dove game biased by the total fitness of each group.
Individuals enter into new groups by bargaining
over the ultimate division of total group output, while attempting to maximize
their own fitness. Their group formation strategies are based on the agent's
access to the common resource last period and the relative size of the other group.
It appears as though the model can produce a variety of group formation and
dissolution dynamics, including a driving force for fairness in the bargaining
outcomes driven by maintaining overall group fitness.
Nobuyuki Hanaki, Economics, Columbia U. (email@example.com).
Nobi is using adaptive models to understand the emergence of efficient and
fair outcomes in laboratory experiments. In games like the battle of
the sexes, human subjects appear to easily find the efficient and fair
outcome. Typical learning models, however, have a difficult time replicating this
particular result. In his model, he uses individual learning (over a
constrained set of strategies) and evolutionary selection (including the
possibility of a rematch if they are unhappy with
their current partner). Using such an algorithm, he finds that an efficient and fair outcome
emerges relatively quickly. Moreover, this result appears to be quite robust
to variations in the algorithm.
Marc Reimann, Center for Business Studies, U. of Vienna (firstname.lastname@example.org).
Marc is modeling disruptive technologies and the evolution of market shares.
The fundamental question he focuses on is the innovators dilemma discussed by
Christensen: namely why do well managed companies fail to heed
the emergence of potentially disruptive technologies. The model
looks at entry dynamics over a two-dimensional product space, where
new entrants are capable of improving the product on one dimension but suffering on
the other. In the simple model, he can link characteristics of
the entering technology to key aspects of the resulting market dynamics.
Michael S. Rimler, Economics, U. of Michigan (email@example.com).
Michael is looking at evolving strategic behavior in a game among a group
of agents where information is slowly revealed over time, such as stud poker.
Such a game has a variety of economic applications, ranging from patent races
to labor negotiations. A genetic algorithm is used to model the strategic
behavior of the agents. He finds that interdependencies across the elements of
a given strategy can cause interesting internal dynamics whereby a given
agent refines and trade offs various aspects of the strategic decision.
Sibel Sirakaya, Economics, U. of Wisconsin (firstname.lastname@example.org).
Sibel, incorporating ideas from non-cooperative game theory, created a new
way to train neural networks using genetic algorithms.
In her work, "competing" genetic algorithms are assigned
to different parts of the neural network, and then adapt based on the current
best efforts of the other algorithms. She finds that this process results
in both improved speed and fit over an ensemble of problems. Furthermore,
she plans to use the above system to gain insight into broader themes
surrounding organizational problem solving.
Miller , email@example.com.