IEOR SEMINAR: Erick Delage (HEC Montreal)
IEOR SEMINAR: Erick Delage (HEC Montreal)Mudd Hall 303
Abstract: Decisions often need to be made in situations where parameters of the problem that is addressed are considered uncertain. While there are a number of well-established paradigms that can be used to design an optimization model that accounts for risk aversion in such a context (e.g. using expected utility or convex risk measures), such paradigms can often be impracticable since they require a detailed characterization of the decision maker’s perception of risk. Indeed, it is often the case that the available information about the DM’s preferences is both incomplete, because preference elicitation is time consuming, and imprecise, because subjective evaluations are prone to a number of well-known cognitive biases. In this talk, we introduce preference robust optimization as a way of accounting for ambiguity about the DM’s preferences. In a financial environment, an optimal preference robust investment will have the guarantee of being preferred to the largest risk-free return that could be made available. We show how preference robust optimization models are quasiconvex optimization problems of reasonable dimension when parametric uncertainty is described using scenarios and preference information takes the form of pairwise comparisons of discrete lotteries. Finally, we illustrate numerically our findings with a portfolio allocation problem and discuss possible extensions.
Bio: Erick Delage completed his Ph.D. at Stanford University in 2009, is currently associate professor in the Department of Decision Sciences at HEC Montreal, and was recently appointed as chairholder of Canada Research Chair in Decision Making under Uncertainty. He serves on the editorial board of Management Science, Pacific Journal of Optimization, and Computational Management Science where he recently co-edited a special issue on recent advances in the field of robust optimization. His research interests span the areas of robust optimization, stochastic programming, decision theory, artificial intelligence and applied statistics.