Econ Papers Daily / 期刊论文
递归期望不确定效用
Recursive Expected Uncertain Utility
摘要
Gul and Pesendorfer’s expected uncertain utility (EUU) theory describes a decision-maker who assigns probabilities to some events, regards others as unquantifiable, and values a prospect by a probability-weighted average of a utility of the best and worst outcomes it may yield on each quantifiable event. I extend EUU to dynamic settings where information arrives before an act is chosen. Outside subjective expected utility, three natural requirements cannot generally coexist: an act’s value should not depend on how the resolution of uncertainty is described; choices after a signal should depend only on what remains possible; and optimizing at each realization should yield a plan that is optimal in advance. The recursive extension characterized here secures the last two by associating with every signal, ambiguous or not, a unique system of conditional beliefs, while retaining EUU’s treatment of risk and ambiguity. For unambiguous signals it reduces to static EUU; otherwise an act’s value depends on how information is revealed. A tractable subclass accommodates choice patterns, such as Machina’s reflection example, that many ambiguity models cannot.
中文摘要
Gul和Pesendorfer的期望不确定效用（EUU）理论描述了一个决策者：他对某些事件赋予概率，将另一些事件视为不可量化，并通过对每个可量化事件上可能产生的最佳与最差结果的效用进行概率加权平均来评估一个前景。我将EUU拓展至信息在行动选择之前到达的动态情形。在主观期望效用之外，三项自然要求通常无法同时成立：一个行动的价值不应取决于不确定性的解决方式如何被描述；信号之后的选择应仅取决于仍然可能的情况；在每个实现处进行优化应得到一个事前最优的计划。本文所刻画的递归拓展通过为每一个信号——无论模糊与否——关联一个唯一的条件信念系统，确保了后两项要求，同时保留了EUU对风险与模糊的处理方式。对于非模糊信号，它退化为静态EUU；否则，一个行动的价值取决于信息如何被揭示。一个易于处理的子类能够容纳许多模糊模型无法容纳的选择模式，例如Machina的反射例子。