Statistical Foundations of EITM
James Lo (University of Southern California)
June 24-26, 2020
The basic motivation of EITM
is that it pays to be more conscious and purposeful about how game theoretic
models and statistical tests are connected. Unfortunately, game theory is often
taught in a manner that submerges its relationship to testing, and statistical
training typically begins with the hypothesis for testing having already been
developed. The Foundations Seminar presents statistical methods that lend
themselves well to testing predictions from formal models.
Our course begins with an introduction to probability models, covering random variables, important probability distributions, and Bayes' theorem. We then discuss the use of maximum likelihood as a method to conduct statistical inference, as well as its relationship to the Bayesian framework of inference. Our discussion of Bayesian inference includes coverage of conjugate bayesian models, as well as common techniques for computation such as the Gibbs sampler and Metropolis-Hastings algorithm. We conclude with applications of these ideas to the study of bargaining models, and to models of latent measures.
James Lo is Assistant Professor at the University of Southern California. Previously he was Postdoctoral Research Associate at Princeton University and Research Fellow at the University of Mannheim.