Introduction. In this paper we interlink a dynamic programming, a game theory and a behavioral simulation approach to the same problem of economic exchange.
The size and complexity of the strategy sets for even a simple infinite horizon exchange economy are so overwhelmingly large that it is reasonably clear that individuals do not indulge in exhaustive search over even a large subset of the potential strategies. Furthermore unless one restricts the unadorned definition of a noncooperative equilibrium to a special form such as a perfect noncooperative equilibrium, almost any outcome can be enforced as an equilibrium by a sufficiently ingenious selection of strategies. In essence, almost anything goes, unless the concept of what constitutes a satisfactory solution to the game places limits on permitted or expected behavior. The latter presumes that the players follow the same introspective process as the game-theorist. As these refinements may be hard to justify, it is interesting to complement this introspective approach with a study of whether interactive market processes provide enough structure to tie down the set of strategies played.
Karatzas, Shubik and Sudderth [1992] formulated a simple infinite horizon economic exchange model involving a continuum of agents as a set of parallel dynamic programs, and were able to establish the existence of a stationary noncooperative equilibrium. In order to obtain an explicit closed form solution for the optimal policy and equilibrium wealth distribution, it relies on a particular utility function. In order to match these analytical results with a behavioral approach, we first develop simulation models of market processes with agents learning through reinforcement. Second, we consider more general classes of utility functions.

J.E.L. classification codes. C61, C63, C72, C73, D83, D91

Keywords. Market game, dynamic programming, Classifier System, adaptive behavior