Risk-Averse Optimization in Multicriteria and Multistage Decision Making

Abstract

Risk-averse stochastic programming provides means to incorporate a wide range of risk attitudes into decision making. Pioneered by the advances in financial optimization, several risk measures such as Value-at-Risk (VaR) and Conditional-Value-at-Risk (CVaR) are employed in risk-averse stochastic programming for a variety of application areas. In this work, we consider risk-averse modeling approaches for stochastic multicriteria and stochastic sequential decision-making problems. First, we propose a new multivariate definition for CVaR as a set of vectors. We analyze its properties and establish that the new definition remedies some potential drawbacks of the existing definitions for discrete random variables. Motivated by the computational challenges in the optimization of vector-valued multivariate definitions of CVaR, next, we study two-stage stochastic programming problems with multivariate risk constraints utilizing a scalarization scheme. We formulate this problem as a mixed-integer program (MIP) and devise two delayed cut generation algorithms. The effectiveness of the proposed modeling approach and solution methods are demonstrated on a pre-disaster relief network design problem. Finally, we study the Markov Decision Processes (MDPs) under cost and transition probability uncertainty with the objective of optimizing the VaR associated with the expected performance of an MDP model. Based on a sampling approach, we provide an MIP formulation and a branch-and-cut algorithm, and demonstrate our proposed methods on an inventory management problem for long-term humanitarian relief operations.