Experience-weighted attraction learning in network coordination games
Abstract
This paper studies the action dynamics of network coordination games with bounded-rational agents. I apply the experience-weighted attraction (EWA) model to the analysis as the EWA model has several free parameters that can capture different aspects of agents' behavioural features. I show that the set of possible long-term action patterns can be largely different when the behavioural parameters vary, ranging from a unique possibility in which all agents favour the risk-dominant option to some set of outcomes richer than the collection of Nash equilibria. Monotonicity and non-monotonicity in the relationship between the number of possible long-term action profiles and the behavioural parameters are explored. I also study the question of influential agents in terms of whose initial predispositions are important to the actions of the whole network. The importance of agents can be represented by a left eigenvector of a Jacobian matrix provided that agents' initial attractions are close to some neutral level. Numerical calculations examine the predictive power of the eigenvector for the long-run action profile and how agents' influences are impacted by their behavioural features and network positions.
Cite
@article{arxiv.2310.18835,
title = {Experience-weighted attraction learning in network coordination games},
author = {Fulin Guo},
journal= {arXiv preprint arXiv:2310.18835},
year = {2023}
}