Design of market mechanisms using computer testbeds
and experiments with human
This paper is part of a wider research project with the
objective of creating computational testbeds for designing and testing new
mechanisms – new economic and political institutions. In Arifovic and Ledyard
(2001), we took this approach for a collection of Groves-Ledyard mechanisms
in an environment with a public good. In this paper, we describe the implementation
of our methodology for two forms of call markets in a private goods environment
(the simple, one good, demand-supply world). The environments we look at
have a fixed number of buyers and sellers. Sellers each own 1 unit of a commodity
and buyers each want to consume 1 unit of the commodity.
Sellers must pay a cost if they sell, buyers receive a
value if they buy. In this world we test a call market, a sealed-bid auction
in which buyers submit bids, willingness-to-pay, and sellers submit offers
willingness-to-accept, to a “market”. When all bids and offers are collected,
the market is “called”. That is, the market computes a demand-supply
equilibrium price. Every buyer whose bid is above that price receives a unit
at that price and every seller whose offer is below that price sells a unit
at that price. There are many micro-forms of call markets depending on the
number of calls before a trade, the information about bids and offers submitted
to the call, etc. In this paper we focus on the simplest version. There is
only one call before trade occurs and there is no information revealed about
submitted bids or offers before the call is made.
The testbed is created by implementing a particular agent-based model. Our
goal is to have the testbed be independent of the mechanisms we want to test.
Agents, buyers and sellers are endowed with sets of strategies. In each round,
agents will send messages to the mechanism based on random selection from
a set; that is, they use a mixed strategy. The mechanism will pick outcomes
and then inform the agents about them and (as part of the mechanism design)
other information in the form of a signal. The agents
then adjust the set they are selecting from and the probability density that
determines their selection.
We find there to be significant differences in performance depending on the
information provided to the traders between calls. In particular, we find
that both dynamic and static performance is better, less volatility and higher
gains from trade, if traders receive less information between calls. We also
design and conduct experiments with human subjects to test if the same behavioral
features can be observed.
Jasmina Arifovic
Simon Fraser University
Department of Economics
http://wwww.sfu.ca/~arifovic
arifovic@sfu.ca
John Ledyard
California Institute of Technology
Division of Humanitis and Social Sciences
http://www.hss.caltech.edu/~jledyard
jledyard@hss.caltech.edu,