|
Institutt for informatikk seminar: Thursday 25 May, kl. 14:15
Experiments in economic rationality with genetic programmingTerje LensbergNHH, BergenThe "as if" view of economic rationality defends the profit maximization hypothesis by pointing out that only those firms who act as if they maximize profits can survive in the long run. Recently, the problem of arriving at a logically consistent definition of rational behavior in games has shown that one must sometimes study explicitly the evolutionary processes that form the basis of this view. In this paper, we investigate the usefulness of genetic programming as a tool for generating hypotheses about rational behavior in situations where explicit maximization is not well defined. We first consider an investment decision problem with unknown probability distributions as a borderline test-case, and show that the artificial agents develop behavior rules as if they were expected utility maximizers with Bayesian learning rules. We next consider repeated game-playing situations where the agents never play the same game twice, and where they must learn to generalize from their experience with a finite number of games in order to deal efficiently with games with which they have no previous experience. In this setting also, the evolved behaviour rules act in agreement with economic (game theoretic) rationality for most games, but deviates from it in some cases where the evidence from experiments with human subjects also deviates from the predictions of standard game theory. |