Annie Liang

Abstract

Players are statistical learners who form beliefs about payoff-relevant parameters based on data. They may update from the same data in different ways, but have common knowledge that all beliefs are formed by updating to this data from a set of "reasonable" learning rules. The strategic predictions that are compatible with this belief restriction depend on the realization of the data, and I define the plausibility of a strategic prediction based on the typicality of the data sets that support it. The main results characterize how the plausibility of a given strategic prediction varies depending on properties of the learning problem, including the quantity of data that players see. The approach generates new predictions, e.g. that speculative trade is more plausible when players see sparse data and when the learning problem is high-dimensional.