Risk Quadrangle and Robust Optimization Based on $\varphi$-Divergence

Abstract

This paper studies robust and distributionally robust optimization based on the extended $\varphi$-divergence under the Fundamental Risk Quadrangle framework. We present the primal and dual representations of the quadrangle elements: risk, deviation, regret, error, and statistic. The framework provides an interpretation of portfolio optimization, classification and regression as robust optimization. We furnish illustrative examples demonstrating that many common problems are included in this framework. The $\varphi$-divergence risk measure used in distributionally robust optimization is a special case. We conduct a case study to visualize the risk envelope.