(A)symmetric equilibria and adaptive learning dynamics in small-committee voting

We study equilibrium selection in a common-interest voting model with three alternatives. In the model, symmetric Bayesian Nash Equilibria (BNE) of varying efficiency are known to exist. Employing evolutionary adaptive learning simulations, we find interesting new equilibria. In simulations, we distinguish between individual learning (agents learn from their own experience) and social learning (agents may also imitate each other’s strategies). We also vary whether voters are randomly re-matched. Social learning consistently converges to steady states that match efficiency-maximizing symmetric BNE. Individual learning with fixed matching converges to, on average, more efficient steady states, which we confirm as pure-strategy asymmetric BNE. This class of BNE has received little attention in the literature. We show that these BNE may be more efficient and discoverable through an adaptive learning process.