Strategic Learning, Belief Updating, and Equilibrium Selection in Repeated Coordination Games: Experimental and Theoretical Evidence

Abstract

Coordination games with multiple Nash equilibria pose a fundamental challenge for equilibrium selection in strategic interaction. This review integrates theoretical frameworks and experimental evidence to examine how players resolve strategic uncertainty in repeated coordination settings through learning dynamics, belief updating, and equilibrium selection criteria. Central concepts include payoff dominance versus risk dominance , basins of attraction, stochastic stability, and the role of risk attitudes in tipping dynamics. Experimental findings consistently show that payoff-dominant equilibria are frequently abandoned when coordination risk is high, while risk-dominant outcomes prevail under uncertainty or limited information. Learning models reinforcement learning, fictitious play, Bayesian updating, and level-k reasoning reveal path-dependent convergence patterns, with history, framing, communication, and payoff asymmetry influencing long-run selection. Recent studies highlight the importance of local interaction effects, network structure, and cognitive heuristics in accelerating or hindering convergence to efficient equilibria. The synthesis underscores that equilibrium selection is rarely purely rational but emerges from adaptive processes shaped by psychological, social, and environmental factors, offering insights for mechanism design, policy coordination, and multi-agent AI systems.