The design and analysis of electronic markets is
a key area of research. A key attribute of such markets is that they
involve both people and intelligent computer agents. I am
particularly interested in applications such as supply chain
management that require solving distributed scheduling and
optimization problems involving many people and organizations with
conflicting interests.
A key goal is to design markets which minimize the cognitive load of
the participants while maintaining economic and computational
efficiency. The work is both theoretical, drawing on a rich body
of theory in economics, artificial intelligence, game theory and
optimization, and empirical, examining real world markets and testing
new results in simulation. Primal-dual theory allows proof of
convergence of certain market mechanisms, while economics and game
theory provide tools for assuring the truth revelation required for
optimization. Unlike classical economics, we do not assume the market
participants know their own utilities; in many situations, exact
determination of utility requires extensive communication with vendors
and solution of large optimization problems. Appropriate market
structures can vastly reduce these costs.
Results include: iBundle, the first provably
allocatively-efficient iterative combinatorial auction for problems
where agents demand bundles of resources, approximation techniques for
optimizations arising in combinatorial auctions, methods for single
item pricing in combinatorial auctions an augmented tatonnement method
for chemical plant control and coordination, and methods for using
combinatorial auctions for supply chain management.
We are also studying the economics of consumer search
for and vendor dissemination of product information, and
the role of intermediaries and shop-bots. We have found that an
"information gap" exists in electronic markets, because sellers often
have no incentive to publish horizontal information, even
though it would increase buyers' utility. A formal model of
buyer search costs provides a clear explanation of when different
forms of product information will, or will not, be supplied by
vendors. Empirical studies of actual market behavior which we are
currently conducting appear to confirm the accuracy of this model.