Convergence of least squares learning in self-referential discontinuous stochastic models

We examine the stability of rational expectations equilibria in the class of models in which the decision of the individual agent is discontinuous with respect to the state variables. Instead of rational expectations, each agent learns the unknown parameters through a recursive stochastic algorithm. If the agents update the estimated value function rapidly enough, then each agent learns the true value function associated with the optimal action with probability 1 and almost always takes the optimal action asymptotically. Journal of Economic Literature Classification Number: C62, D83.