A generalized approach to belief learning in repeated games

Abstract

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 rules of behavior predicted.JEL Classification: C51, C92, C72, D03

The Evolution of Cooperation: The Role of Costly Strategy Adjustments

“…The first is to develop new models of within-game learning. These models would likely need to incorporate learning over strategies rather than learning over actions, as standard action-learning models have a difficult time explaining the high incidence of cooperation in repeated prisoner's dilemma games (e.g., Hanaki et al 2005, Ioannou andRomero 2014). The second avenue involves developing new techniques of identifying strategies in the presence of within-game learning.…”

“…Models in which players learn over actions often have a difficult time explaining experimentally observed levels of cooperation. However, models in which players learn over strategies have had more success (Hanaki et al 2005, Ioannou andRomero 2014). When modeling strategy learning, it is not clear whether subjects play a fixed strategy for the entire supergame and then adjust their strategy between supergames, or whether subjects adjust their strategies within the supergame.…”

Learning with repeated-game strategies

“…In Ioannou and Romero (2014), we applied this program of research by building on three leading action-learning models to facilitate their operability with repeated-game strategies. The three modified models approximated subjects' behavior substantially better than their respective models with action learning.…”

Meaningful learning in weighted voting games: an experiment

“…Researchers continue to propose new models that better capture observed human behavior. For example, Marchiori and Warglien (2008) incorporate "regret" in their neural network-based learning model, and show that it better replicates observed human behavior than either the EWA or neural network-based learning models without regret; Hanaki et al (2005) and Ioannou and Romero (2014) extend the reinforcement and the EWA learning models, respectively, to allow players to learn which repeated-game strategies to use in repeated games; Arifovic and Ledyard (2012) report that their "individual evolutionary learning" model captures most of the stylized results in Public Goods game experiments; and Spiliopoulos (2012Spiliopoulos ( , 2013 embed abilities to recognize an opponent's behavioral patterns in a belief-learning model in order to better capture the elicited subjective beliefs about the opponent's strategy in games with a unique mixed strategy Nash equilibrium.…”