Emergenct conventions and equilibrium selection : learning dynamic in repeated stag hunt games

Chapman, Archie C. (2004). Emergenct conventions and equilibrium selection : learning dynamic in repeated stag hunt games Honours Thesis, School of Economics, The University of Queensland.

       
Attached Files (Some files may be inaccessible until you login with your UQ eSpace credentials)
Name Description MIMEType Size Downloads
THE18091.pdf THE18091.pdf Click to show the corresponding preview/stream application/pdf 6.04MB 0
Author Chapman, Archie C.
Thesis Title Emergenct conventions and equilibrium selection : learning dynamic in repeated stag hunt games
School, Centre or Institute School of Economics
Institution The University of Queensland
Publication date 2004
Thesis type Honours Thesis
Total pages 111
Language eng
Subjects 14 Economics
Formatted abstract Inductive learning rules are used to predict the behaviour of experimental subjects playing repeated games. Battalio, Samuelson and Van Huyck (2001) provide an experiment that tests for the effects of varying the risk involved with playing a payoff-dominant strategy on the behaviour of subjects playing a repeated Stag Hunt game. The behavioural game theorist can use results from this experiment to evaluate the predictive ability of a learning rule under different risk environments.

Learning rules are often divided into three categories: belief learning, reinforcement learning, and rules that are a hybrid of the two. Battalio, Samuelson and Van Huyck (2001) introduce a new belief learning rule. This thesis introduces the Q-learning rule, novel in experimental economics literature. Q-learning is a member of the class of temporal-difference reinforcement learning algorithms studied extensively in machine learning. A hybrid rule that has been shown to outperform many others across numerous games is Camerer and Ho's (1999) experience-weighted attraction learning.

The predictive abilities of the above three learning rules are evaluated by comparing each rule's out of sample predictions of behaviour of agents in each round to empirical data from Battalio, Samuelson and Van Huyck (2001). Across all three games and according to several criteria experience weighted attraction learning performs best. In addition, this paper explores the use of initialised learning rules to predict the trajectory of aggregate play using Monte Carlo simulations of the repeated game. The results of the Monte Carlo experiment are inconclusive.


Document type: Thesis
Collection: UQ Theses (non-RHD) - UQ staff and students only
 
Citation counts: Google Scholar Search Google Scholar
Access Statistics: 73 Abstract Views, 0 File Downloads  -  Detailed Statistics
Created: Wed, 01 Dec 2010, 18:17:03 EST by Ning Jing on behalf of The University of Queensland Library