Social aspiration reinforcement learning in Cournot games

Abstract

We offer theoretical and experimental evidence showing that social aspiration learning converges to the Walrasian outcome in Cournot games. Reinforcement learning converges to the competitive outcome because the Walrasian quantity is the only action that never yields profits below the average profits in the market. Using data from laboratory experiments, we show that when provided with information about average market profits, they positively (negatively) reinforce actions that yield payoffs above (below) the average payoffs in the market. When provided with both rivals’ individual performance and average market profits, both heuristics (imitation and social learning) are combined by subjects and prices and profits are driven further into competitive levels, closer to the Walrasian quantity. Subjects’ tendency to adjust their choices following the social learning heuristics survives and they adopt it as frequently as imitation when both predictions collide.