Working Paper
[UTMD-084] Using Big Data and Machine Learning to Uncover How Players Choose Mixed Strategies (by Toshihiko Hirasawa, Michihiro Kandori, Akira Matsushita) (Revised version of UTMD-033)
Author
Toshihiko Hirasawa, Michihiro Kandori, Akira Matsushita
Abstract
We examined how humans learn to choose mixed strategies using our unique big experimental dataset with approximately 75,000 observations. We compared the out-of-sample predictive power of conventional behavioral models and machine learning models and found that a version of the deep learning model (LSTM) substantially outperforms the conventional models. The superiority of the machine learning model is noticeable only when the data size is an order of magnitude larger than that of the typical lab dataset. We tried to open the black box of LSTM and obtained an improved behavioral model with nearly equal predictive power. We provide several key steps one can follow to improve existing behavioral models by means of machine learning and big data.