Research Seminar: Asymptotic Optimality in Data-Driven Decision Making
A fundamental challenge in data-driven decision making is to construct estimators for the optimal solutions of stochastic optimization problems based on limited training data.
Asymptotic Optimality in Data-Driven Decision Making
On April 13, we had the pleasure to have Tobias Sutter, Associate Professor of Econometrics, University of St. Gallen, presenting his recent works on statistically optimal approaches to construct data-driven decisions for stochastic optimization problems.
A data-driven decision can always be expressed as the minimizer of a surrogate optimization model constructed from the available training data. The quality of a data-driven decision is measured by its out-of-sample risk. An ideal data-driven decision should minimize the out-of-sample risk simultaneously with respect to every conceivable probability measure. As such ideal decisions are generally unavailable, the seminar proposed Pareto dominant data-driven decisions that minimize the in-sample risk subject to an upper bound on the out-of-sample risk.