Date: 14:15 on Fri, 17 Feb 2017
Seminar Room A, Manor Road Building,Manor Road, Oxford, OX1 3UQ
Keeping Your Story Straight: Truthtelling and Liespotting
Speaker(s): Johannes Horner - Toulouse School of Economics
An agent privately observes a Markov chain online, and reports it to a designer. What patterns in the reported data should the designer pay attention to? We show that, in general, keeping track of the empirical frequency of transition counts in the agent’s reports is not enough, despite the true state being Markovian. Nonetheless, we derive conditions under which any deviation that can be told apart from truth-telling by checking the frequency of strings of an arbitrary (finite) size can be detected by “checking pairs.” Further, weaker conditions suffice given that, as it turns out, some undetectable deviations are unprofitable for the agent, independent of his preferences. We explore the implications of these results for the literature on (i) linking incentives, (ii) dynamic implementation, and (iii) repeated games and agency models.