Strategic Learning and its Limits
In this concise book based on his Arne Ryde Lectures in 2002, Young suggests a conceptual framework for studying strategic learning and highlights theoretical developments in the area. He discusses the interactive learning problem; reinforcement and regret; equilibrium; conditional no-regret learning; prediction, postdiction, and calibration; fictitious play and its variants; Bayesian learning; and hypothesis testing. Young's framework emphasizes the amount of information required to implement different types of learning rules, criteria for evaluating their performance, and alternative notions of equilibrium to which they converge. He also stresses the limits of what can be achieved: for a given type of game and a given amount of information, there may exist no learning procedure that satisfies certain reasonable criteria of performance and convergence. In short, Young has provided a valuable primer that delineates what we know, what we would like to know, and the limits of what we can know, when we try to learn about a system that is composed of other learners.
What people are saying - Write a review
We haven't found any reviews in the usual places.
Other editions - View all
absolute continuity action space approach assume asymptotically average payoff Bayesian learning behavioral strategy beliefs best reply Blackwell’s theorem calibrated Chapter choose close to equilibrium coarse correlated equilibria conditional regret convex set coordination game cumulative payoff define denote empirical distribution equilib equilibrium behavior expected payoff fictitious play finite game follows forecasting rule Foster and Vohra Foster and Young Fudenberg and Levine game G given Hart and Mas-Colell Hence history jt incremental learning process learning rules matching pennies mixed strategy n-tuple Nash equilibrium opponent’s opponents optimal orthant outcome pÀi parameters payoff function period t þ Peyton Young play paths player i’s posterior prior probability distribution probability vector qtþ1 random variable rational realized payoffs regret matching regret vector reinforcement reinforcement learning repeated game response function robust row player rt þ rtþ sequence set of coarse set of correlated smoothed fictitious play stage game stochastic Suppose updating xÀi zero