2005 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.
These projects will form the basis for dissertation chapters and/or
journal articles.
Kirill Chernomaz, Economics, Ohio State (chernomaz.1@osu.edu).
Kirill is interested in understanding the difficulty of learning in economics. He relies
on a model of collusion in independent, private-value first-price auctions as an environment
to explore this question, and has agents learning polynomial bidding functions via a genetic algorithm.
Agents have the ability to bound the learning task at hand by reducing the complexity of the
function they are trying to learn.
He finds that agents learn to make tradeoffs between the difficulty of the learning task they
give themselves and the environment. In particular, agents tend to simplify the learning task at hand, at
a cost of deviating from optimal behavior, as long as the environment is not too complex.
Yen-Sheng Chiang, Sociology, Washington (yen506@u.washington.edu).
Yen is exploring a model of corruption in social and economic
systems. Studies have shown that economic development and corruption
tend to be negatively correlated.
In the first model, he considers agents queuing up for
a service. Corruption is introduced by allowing agents to bribe either
one another or a central authority to alter their positions in the queue.
In the second model he uses a three level model that entails agents bribing
an official who is under a supervisor.
Scott Christley, Computer Science, Notre Dame (schristl@nd.edu).
Scott wants to understand how different organizational forms influence
software development. He represents software as being composed of interacting modules
and uses a NK(C) model to capture this notion. Agents can manipulate the software
in various ways, ranging from random modification to more specialized efforts on
particular parts of the software, and also vary in their ability to manipulate
the software ("expertise").
When the software has few interactions, all agent types appear to quickly develop
the optimal software. As software becomes more complex, expertise begins to
dominate, though this advantage tends to diminish as complexity increases.
The underlying model will serve as a productive testbed for exploring
a variety of key questions about software development, both in terms of how to organize
programmers and how to structure software.
Sheila Conway, Systems Engineering, NASA and Old Dominion (Sheila.R.Conway@nasa.gov).
Sheila is investigating ways to improve air traffic control systems. Her model
considers banks of flights arriving and departing from an air traffic control
facility that is operationally constrained. With this model she can identify
the key driving forces of the system, and develop new policy prescriptions that
can improve performance both under current and anticipated future air traffic
conditions.
Adrian de Froment, Ecology and Evolutionary Biology, Princeton (adriande@princeton.edu).
Adrian is modeling niche construction and firms. In the work,
firms need to adapt to a landscape of consumer preferences while
consumers simultaneously adapt their preferences to firms.
Across a variety of model specifications, he found that
the system tended to create a similar number, type, and
distribution of firms. However, there were important
differences, for example in first mover advantage, depending
on the degree of niche construction allowed.
Joel Grus, Social Science, Cal Tech (grus@hss.caltech.edu).
Joel considers the problem of innovation, complexity, and patents.
Technology is modeled as a landscape that can embrace different
levels of nonlinearity. Agents search across this landscape using
various heuristics, and are prevented via "patents" from searching
neighborhoods of previous identified points. The model allows
the analysis of the interaction among search heuristics, underlying
complexity of the technology, and the patent system. He finds that
along with exploring issues of innovation free riding, the model provides
a tractable way to explore how patents alter the dynamics of
search.
Kyle Joyce, Political Science, Pennsylvania State (kjoyce@psu.edu).
Kyle's work focuses on why some wars expand to include third parties
in the conflict. Using an agent-based model, states must decide
whether or not to join an ongoing conflict based on their expectations
of the outcome of the war and their connections to the embattled states.
He finds that the inclusion of third parties can have a big impact
on the dynamics of war, and that as the threshold for joining the
conflict increases the system appears to embrace a very different
dynamics.
Yasmina Khoury, Economics, Columbia (yek2001@columbia.edu).
Yasmina is analyzing how consumers can trust messages when the source
may either be from credible consumers or self-interested merchants. This model
reflects some new marketing trends, like those used by the company Bzzagents. A key question
is when does such word-of-mouth advertising break down. She considers an evolutionary
system driven by a replicator dynamic. She finds that when only firms can evolve, they
learn to lie about the quality of their product; when only consumer evolve, they learn
to distrust signals from firms. Note that both of these behaviors result in an inferior
outcome for the system as a whole. Finally, when both sides of the market can evolve, a
similar outcome arises, but consumers learn their strategy much faster.
Andreas Pape, Economics, Michigan (apape@umich.edu).
Andreas is interested in agents and causal inference. In particular,
how do agents develop useful causal models of the world. The basis
of his work is using recent developments in causal inference techniques
developed by statisticians as a way to model agent behavior.
Agents need to differentiate among potential causal graphs given
a set of observations. The basic framework will be applied to
a variety of economic scenarios.
Markus Schneider, Economics, New School (SchnM869@newschool.edu).
Markus employs an agent-based model to analyze urban dynamics, in particular,
the impact of home ownership versus renting on neighborhood formation. In
the model, agents must decide where to live and whether to buy or rent, based
on neighborhood quality, cost, and availability of credit. He finds that
high quality neighborhoods form anchored by owners. The model will ultimately
serve as a means by which to better understand gentrification processes in
urban areas, changes in racial composition, and how social networks and learning
dynamics influence neighborhoods.
John H.
Miller , miller@santafe.edu.