Bernardo Pagnoncelli: Contextual chance-constrained programming

Abstract

Uncertainty in classical stochastic programming models is often described solely by independent random parameters, ignoring their dependence on multidimensional features. We propose a novel contextual chance-constrained programming formulation that incorporates features, and argue that a solution that ignores them may not be implementable. Our formulation relies on regression trees and is completely data-driven. Borrowing results from quenched large deviation theory we show the exponential convergence of our scheme as the number of data points increases. We illustrate our findings with a prescriptive model for transfer market decisions for soccer teams, using real data from the Premier League. This is joint work with H. Rahimian (Clemson University).

Bernardo Pagnoncelli
Associate Professor
School of Business, Universidad Adolfo Ibañez