Rational choice theory and ABMs (was RE: Criticisms and defense of ABM)

[Alexander,J]

> We may dislike rational choice theory but I am afraid that it is the only
> theory that has a coherent and consistent set of assumptions by which we
> can predict prices, and it is the theoretical/behavioral basis of hedonic
> prices models (a fancy term for regressions). As I see it ABMs have to be
> combined at some level with rational choice theory.
 
It's not like there is a great deal of work involved in combining ABMs with 
rational choice
theory: one simply needs to construct the ABM to include a model of the 
deliberative 
process by which agents arrive at a decision.  ABMs, in many senses, are ideal 
for modeling the
deliberative processes of boundedly rational individuals, as such individuals 
are typically
conceived of as arriving at choices by employing inductive generalizations of 
past
experiences or heuristics.  If it fit the problem at hand, you could model 
agents as Bayesians,
initially endowed with a somewhat random prior, who updated their prior in the 
light
of new experience from time to time.
 
In my own work, I've modelled boundedly rational agents faced with simple
distribution problems (a simple version of the Nash bargaining problem and the 
ultimatum
game) as well as coordination games.  Although I typically assume that the 
agents 
usually follow imitative learning rules (it makes the analysis afterwards 
easier), I've also modeled 
them as using fictitious play.  There's no reason why more complex choice 
processes couldn't
be modelled.  If you can describe it, you can model it.
 
From my point of view, the most natural point of contact between rational 
choice theory and 
ABMs is through evolutionary game theory.  Traditional models of rational 
choice, i.e.,
von Neumann-Morgenstern game theory, lack a proper treatment of the dynamic 
aspect of
choice.  Evolutionary game theory (at least the economic interpretations a la 
Binmore, Samuelson,
and others) seeks to provide such a dynamic component, and usually assumes that 
agents
are boundedly rational.  ABMs provide a framework for a detailed study of both 
of
these elements.
 
As an aside: the problem mentioned with markets points out what, to my mind, is 
a more interesting
problem for ABMs.  Namely, how does one smoothly incorporate, into a single 
problem, problems
involving group action with problems involving individual action.  One could 
always force upon the model
some way of having aggregative individual action lead to group action (if, say, 
50%+1 of the group make
an individual choice to do X, then the group does X) but, in many cases, this 
requires making
substantive assumptions as to the causal relations holding between individuals 
and the group.
Ideally, one would like to frame very general rules of interaction which 
allowed group action
to develop out of individual action.  This is a hard problem, though.
 
Cheers,
 
Jason
 
--
J. McKenzie Alexander
Co-ordinator, MSc in Philosophy of the Social Sciences
Department of Philosophy, Logic and Scientific Method
London School of Economics, Houghton Street, London WC2A 2AE

<<winmail.dat>>