讲座题目：Strategic Experimentation with Privately Observed Payoffs私下观察收益的战略实验

主讲人：Jérôme Renault  教授  图卢兹经济学院

讲座时间：4月20日 周四14:30

讲座地点：腾讯会议542 566 273

讲座内容摘要：

We study a model of strategic experimentation where at every period, each of the two players can pull either a safe arm, or a risky arm at a positive cost. The risky arms of the players are perfectly correlated and may be either bad (with return always being zero), or good with i.i.d. returns (a certain positive return with a positive probability and zero payoff otherwise). Players play in turn and maximize discounted payoffs. The novelty of the model is that each player observes after each period the arm pulled by the other player, but not the return. We find that this leads to more experimentations than in the public observation case, and exhibits a new cutoff for the belief of a player for the arm begin good. At pure sequential equilibria, the players experiment up to this new cutoff, under-experimentation never happens but there are examples where players experiment way too much compared to the social optimum. At a mixed sequential equilibrium, the number of experiments in the bad state is almost surely finite, and at least one final belief is below the new cutoff. This is a joint work with Eilon Solan (Tel-Aviv U.) and Nicolas Vieille (HEC Paris).

我们研究了一种策略性实验模型，在每个时期，两个玩家中的每一个都可以花费一定的成本拉动一个安全臂或冒险臂。玩家的冒险臂是完全相关的，可以是坏的（回报始终为零）或者是好的（具有独立同分布的回报，以正概率获得正回报，否则回报为零）。玩家轮流进行并最大化折现收益。这个模型的新奇之处在于，每个玩家在每个时期观察另一个玩家拉动的手臂，但无法观察到回报。我们发现，相比于公开观察的情况，这会导致更多的实验，并展现出一种玩家对手臂好坏的信念的新截止点。在纯顺序均衡中，玩家进行实验直到达到这个新的截止点，不会出现过少实验的情况，但有些情况下玩家进行的实验远远超过了社会最优。在混合顺序均衡中，差状态下的实验次数几乎肯定是有限的，并且至少存在一种最终信念低于新的截止点。这是与Eilon Solan（特拉维夫大学）和Nicolas Vieille（HEC巴黎）的合作研究成果。

主讲人信息：

Jérôme Renault is full professor of Mathematics at Toulouse School of Economics, and holder of the chair "Game Theory and Artificial Intelligence » at the AI institute ANITI. He obtained his PhD from University Paris 1 Pantheon-Sorbonne and has worked at University Paris Dauphine and Ecole Polytechnique. As a world-wide known expert on repeated games, stochastic games and incomplete-information games, his research works are published in leading journals of related fields likeMathematics of Operations Research,Theoretical Economics,Journal of Economic Theory,Annals of Applied Probability,SIAM Journal on Control and Optimization,Games and Economic Behavior,International Journal of Game Theory, etc. Having served for years on the editorial boards ofMathematics of Operations Research,Dynamic Games and Applications,Journal of Games and Dynamics, he is now the new editor-in-chief of theInternational Journal of Game Theory.

Jérôme Renault是图卢兹经济学院数学教授，也是人工智能研究所ANITI的“博弈论与人工智能”讲席教授。他在巴黎第一大学巴黎索邦大学获得博士学位，并曾工作于法国知名高校巴黎第九大学和巴黎综合理工。作为一位全球知名的重复博弈、随机博弈和不完全信息博弈专家，他的研究成果发表在相关领域的顶级期刊上，如《运筹学数学》、《理论经济学》、《经济理论杂志》、《应用概率学年刊》、《SIAM控制与优化杂志》、《博弈与经济行为》、《国际博弈论杂志》等。他曾在《运筹学数学》、《动态博弈与应用》、《游戏与动态学》等期刊担任编委多年，即将成为《国际博弈论杂志》的新任主编。