Theory/Experimental Seminar - Drew Fudenberg (MIT)

Title: Adversarial Forecasters, Surprises and Randomization

Abstract: In an adversarial forecaster model, utility over lotteries is the sum of an expected utility function and a “suspense function” measuring how surprising the outcome is given the forecast made by an adversarial forecaster who attempts to find the forecast that minimizes the surprise. We show that an adversarial forecaster model gives rise to preferences that are concave and satisfy a form of differentiability condition, and that any preference relation that has a concave representation that satisfies the differentiability condition arises from an adversarial forecaster model. Because of concavity, the agent typically prefers to randomize. We characterize the support size of optimally chosen lotteries, and how it depends on risk preferences. We also show that the induced preferences in some problems of Bayesian persuasion and resource allocation have an adversarial forecaster representation, so that our results apply there.

Link to paper: http://www.dklevine.com/archive/refs411694000000000130.pdf

To meet the speaker please click here

Location: 321, Drayton House or join via Zoom. https://ucl.zoom.us/j/98871239429?pwd=cVZXMmhsT291K3p2MlNQd0R2UkY3QT09