Research
Published/Forthcoming Papers
1. Impermanent Types and Permanent Reputations (joint with Mehmet Ekmekci and Olivier Gossner), Journal of Economic Theory, Vol 147 (2012), pp 142-178
2. Dynamic Strategic Information Transmission (joint with Mike Golosov, Vasiliki Skreta, Aleh Tsyvinski), Journal of Economic Theory, Vol 151 (2014), pp 304-341
[Online Appendix]
3. Bounded Memory and Biases in Information Processing, Econometrica, Vol 82 No. 6 (2014), pp 2257-2294
4. Comment on "Smith (1995): Perfect Finite Horizon Folk Theorem" (joint with G.H. Demeze-Jouatsa), Econometrica online (2019)
"Memory is the mother of all wisdom." -- AESCHYLUS, Prometheus Bound
Current Projects
1. The Behavioral SIR Model, with Applications to Swine Flu and Covid-19 Pandemics
2. Accept This Paper (joint with Lones Smith)
Slides (working paper in preparation)
3. The Proportionate Likelihood Ratio Property
4. Source Amnesia (joint with Lones Smith)
Working Papers (Rough Drafts):
1. Costly Complexity as a Commitment Device (joint with Mehmet Ekmekci)
Abstract: This paper studies reputation effects in a repeated moral hazard game, with two long-lived players and imperfect monitoring. Player 1 would benefit from committing to an action which is strictly dominated in the stage game; player 2 believes there is a small chance that player 1 is a commitment type who always chooses this action, and observes only a noisy signal about player 1's behavior each period. We depart from the standard literature by assuming that players choose their strategies subject to a complexity cost: each player chooses a finite automaton, both to implement his own strategy and to keep track of his opponent's behavior, but faces a cost on the number of automaton states. Our main result is that if both players face positive complexity costs, then (i) there is a pure strategy equilibrium in which player 1 earns nearly his commitment payoff, and (ii) in all pure strategy equilibria, player 1's payoff tends to his commitment payoff as signal noise vanishes. If complexity is costly for only one player, then we obtain a folk theorem.
2. Optimal Snap Judgements (2013)
Abstract: I consider a symmetric model in which the true state of the world may be high or low (equally likely), and a DM observes the realization of a symmetric binary signal each period until the process terminates, which happens with a fixed chance each period. He must use a finite automaton both to keep track of observed information, and to choose an action each period. The main result is that if signal quality is unknown, then there is a 5-state automaton which is payoff-superior to all other 5-state automata, for all true values of the signal quality; moreover, the expected payoff loss, averaged across possible signal distributions, is less than 1% compared to an unconstrained Bayesian. The 5-state automaton has the feature that in the two states with the most extreme posteriors on the true state of the world, the DM leaves after opposing information with a probability proportional to the square root of the expected number of signals. Thus, in a large-information model, he will appear to reach a "snap judgement" after a short string of successes or failures, then ignoring subsequent information with high probability. In the final section of the paper, I show that if the signal is known to be extremely noisy, then small automata perform badly. However, if the DM can additionally keep track of calendar time, then 2 automaton states is sufficient to earn arbitrarily close to the Bayesian payoff. I relate this to Homeland Security's threat warning system, which recently switched from a 5-color warning scheme to one with just two states.