1. Philippe Jehiel and Dov Samet Learning To Play Games In Extensive Form By Valuation
Extensive game theoretic models of reinforcement learning assume that players make their decisions based on their experienced valuation of the extensive game strategies. A new and exciting direction of research is suggested. Each player evaluates each move separately and chooses the move with the highest valuation. When applied to win-lose games and when the valuation of a move is taken to be the payoff incurred the last time the action was used, a player with a winning strategy always wins. For general payoffs, when the valuation of a move is the average of the payoffs incurred when it was used, and with a small "exploration" probability the players converge to a subgame perfect equilibrium.
2. Ran Spiegler Testing Threats in Repeated Games
Two players play a 2X2 repeated game. Strategies are implemented by automatae. When a player responds to a state of the other machine with an action that is not the one-shot best response, he is deterred by some threat from the other player's machine. The paper suggests a solution concept which essentially requires that for any player (1) if the other player's machine has a recurrent state, the machine will eventually play the best response against it and (2) the solution path has the property that a player who does not play the one shot best response will be able to point to an event in the past which shows that the deterring threat exists. A partial characterization of the solution for repeated chicken and prisoner's dillema and a folk theorem are provided.
3. Mikhail Golosov, Narayana Kocherlakota and Aleh Tsyvinski Optimal Indirect and Capital Taxation
This paper analyzes the classic Mirrlees problem of designing a taxation scheme to provide insurance, when agents' skills are private information. However, this paper considers a much more general model, where agents' skills may be multidimensional and can follow any stochastic process, and the tax system can be nonlinear and history-dependent. The paper provides an important and general insight: investment should be discouraged relative to the complete-information solution, because future investment income makes it more costly to provide incentives for truthful revelation. Thus, the optimal tax scheme has a positive capital income tax.
4. Gil Kalai Social Choice without Rationality
The power of high-caliber mathematicians is knocking on the doors of Social Choice Theory with some intersting and general results. The paper overviews two involved mathematical results proved by Saharon Shelah and Gil Kalai, which relate to the aggregation of classes of choice functions. In particular the following is result is discussed: Let C be a class of choice functions which does not contain all choice functions, and which is closed to all permutations of the names of the alternatives. Let F be a function which aggregates profiles of functions in C into C and which satisfies that if all individuals agree on the choice from a set so does the aggregator and that the choice from a set depends only on the individuals' choices from that set. Then, F must be a "dictatorship".
5. Jesus Fernandez-Villaverde and Arijit Mukherji Can We Really Observe Hyperbolic Discounting?
Short answer: no. In the presence of uncertainty about future preferences geometric discounting gives rise to exactly the type of "preference reversal" observed in psychology experiments. However, hyperbolic discounting implies a preference for commitment, a preference that is not present in an experiment carefully designed to distinguish the two theories.