APPRENDIMENTO INTERATTIVO «NON RAZIONALE» ED EQUILIBRI CONSEGUENTI

• Published in: Giornale degli economisti e annali di economia 1990-01, Vol.49 (1/2), p.3-28
• Main Author: Rampa, Giorgio
• Format: Article
• Language: Italian
• ISSN: 0017-0097

An abstract model is envisaged in which agents perceive the strategic environment and optimize their expected payoffs under genuinely incomplete information, i.e. absence of common prior. It is assumed in addition that agents have «limited» computability capacities: by applying the concept of «rationalizability», this implies that individuals rationalize their choices up to a finite degree, on the basis of their subjective priors on the characteristics of the remaining agents (including the latters' priors). Given incomplete information, agents perceive that their distributios (like any other agent's distribution) might be possibly disconfirmed by actual realizations. By anticipating the use of Bayes's rule on the part of any agent, a single individual sets up and solves ex ante a stochastic dynamic programming problem, uniquely induced by his prior. After any actual realization, agents form their posteriors and revise their plans accordingly. A concept of (temporary) equilibrium is introduced for any finite time horizon, as a fixed point of the learning dynamics. The existence of equilibria is proved, together with a result regarding the dimensionality of their set. It is shown that a higher level of rationalization increases this dimensionality, instead of refining it.