9711885 Smith This research makes methodological and substantive contributions to dynamic game theory and learning theory. It seeks to attain unmet goals of explaining observed features of bargaining - offers are concessions, timing of offers is unrestricted, the process itself is a war of attrition - in a perfect information framework. It achieves a characterization of the "recursive" sequential equilibrium payoff set, and later a folk theorem for games with hidden actions and no summary public statistic. It pushes as far as possible the methodology introduced by Abreu, Pearce, and Stacchetti of reducing a dynamic game to a one-shot "auxiliary game." It explores a new conjecture about optimal experimentation - that information demand is quasiconcave in the posterior beliefs. The model offers a simple framework for exploring the stochastic evolution of research expenditures on a given project. It is conceptually similar to both game theory projects in the attempt to understand intertemporal dynamics by carefully analyzing the cross-section of possible continuation "games." Understanding the dynamics of private information is one of the single most important outstanding problems facing game theory and, to a lesser extent, contract theory. This research will open up a very large class of models that economists currently simply cannot solve. It also offers a simple and substantive new pure theory of dynamic R&D founded on dynamic experimentation.