Strategic Experimentation with Heterogeneous agents and Payoff externality

Abstract:

This paper analyses a two-player game of strategic experimentation with two-armed bandits. At least one of the arms is risky in the sense that it may not yield a lump sum payoff. There is payoff externality between the players and they differ in their ability to learn across the risky arm. Either player has to decide in a continuous time regarding which arm to use. Two alternative settings are analysed. The first setting has two risky arms which are perfectly negatively correlated. The other one has one safe arm and one risky arm. I show that in equilibrium (Markovian) there is always too much of duplication which implies that with respect to a social planner's solution, risky arms are explored excessively.

Full Text