Applied Micro Seminar - Lorenzo Magnolfi (Wisconsin)

Abstract: In complex dynamic environments, such as price setting in online marketplaces or bidding in sponsored search auctions, agents may not know or understand key features of their interaction. Therefore, they often delegate decision-making to data-driven algorithms. We propose a method to recover economic primitives from data generated in such environments. Instead of equilibrium, we impose a regret-minimization assumption on behavior. According to regret minimization, the time average increase in past payoffs, had each agent played the best fixed action in hindsight, vanishes in the long run. We first show that each agent's regrets vanish if and only if the time average of play converges to the set of Bayes coarse correlated equilibrium (BCCE) predictions of the stage game. Next, we use the static limiting model of BCCE to construct set estimators for the primitives of interest. The estimators' coverage properties in large and in finite samples directly arise from the theoretical convergence results and from properties of regret-minimizing algorithms. We apply the method to pricing data in an online marketplace. We recover bounds on the distribution of sellers' marginal costs that are useful to inform market design experiments.

Sign up link