ABSTRACT
Decisions often need to be made in situations where one has incomplete knowledge about some of the parameters of the problem that he is addressing. While expected utility theory can provide essential guidance in managing such decision problems, it relies on two key assumptions:
1) that the decision maker can dedicate enough resources to identify a stochastic model that accurately embodies the potential realizations of these variables
2) that he can, after a tolerable amount of introspective questioning, clearly identify a utility function that characterizes his attitude toward risk. In practice, parsimony (or perhaps a lack of consensus among the stakeholders) often requires that modeling be interrupted at a moment when these elements of the model remain ambiguous. In particular, one might only have established some statistics that are satisfied by the distribution of parameters (e.g., mean, variance, correlation, some probabilities) or only know that the decision maker is risk averse and prefers a list of lotteries over others. While common practice will suggest ways of selecting the most plausible distribution model and utility function, this talk will describe how to modify the expected utility model to account for either form of ambiguity while preserving tractability of the solution process. We use a portfolio allocation problem to illustrate our findings.
The talk is based on three papers that are joint work with Yinyu Ye (Stanford University), Sharon Arroyo (The Boeing Company) and Benjamin Armbruster (Northwestern University)
