A generalized approach to belief learning in repeated games

• Published in: Games and economic behavior Vol. 87; pp. 178 - 203
• Main Authors: Ioannou, Christos A., Romero, Julian
• Format: Journal Article
• Language: English
• Published: Duluth Elsevier Inc 01.09.2014; Academic Press
• Subjects: Action learning; Adaptive models; Approximation; Battle of the Sexes; Behavior; Belief learning; Chicken; Cognitive models; Cognitive psychology; Data analysis; Economic behaviour; Experimental methods; Finite automata; Game theory; Games; Mapping; Methodology; Prisoner's Dilemma; Repeated-game strategies; Research methodology; Research subjects; Stag-Hunt; Studies; Finite automata; D03; Repeated-game strategies; Belief learning; Prisoner's Dilemma; Chicken; C92; Battle of the Sexes; Stag-Hunt; C51; C72; Adaptive models
• ISSN: 0899-8256; 1090-2473
• DOI: 10.1016/j.geb.2014.05.007

We propose a methodology that is generalizable to a broad class of repeated games in order to facilitate operability of belief-learning models with repeated-game strategies. The methodology consists of (1) a generalized repeated-game strategy space, (2) a mapping between histories and repeated-game beliefs, and (3) asynchronous updating of repeated-game strategies. We implement the proposed methodology by building on three proven action-learning models. Their predictions with repeated-game strategies are then validated with data from experiments with human subjects in four, symmetric 2×2 games: Prisoner's Dilemma, Battle of the Sexes, Stag-Hunt, and Chicken. The models with repeated-game strategies approximate subjects' behavior substantially better than their respective models with action learning. Additionally, inferred rules of behavior in the experimental data overlap with the predicted rules of behavior. •We propose a methodology that generalizes belief learning to repeated games.•This methodology is applied to three leading action learning models.•Simulations are run for action learning and belief learning models in four games.•Proposed modifications fit substantially better when compared to human subject data.