Predicting and Understanding Initial Play (joint with Drew Fudenberg)

Last updated: August 21, 2017

 

Abstract. We take a machine learning approach to the problem of predicting initial play in strategic-form games. We predict game play data from previous laboratory experiments, and also a new data set of 200 games with randomly distributed payoffs that were played on Mechanical Turk. Prediction rules built on game features outperform the Poisson Cognitive Hierarchy model (PCHM) of Camerer, Ho, and Chong (2004), and suggest a one-parameter extension of the model with substantially better fit. We also show how to improve the PCHM by using game features to predict a best value of the parameter τ for each game. Finally, we discuss the use of human predictions of play (also collected on Mechanical Turk) as a way to predict game play, and show that even naive use of this crowd data performs quite well.

DOWNLOAD PDF

 

BACK