Mitigate Overconservatism for Robust Optimization by Adapting to Opportunities

Abstract

Overconservatism has long been recognized as a major issue with robust optimization, despite its key advantages of tractability, performance guarantee, and limited information. To address this issue, a new criterion is proposed that can adapt its level of conservatism continuously to the opportunities out there, while maintaining all the key advantages just mentioned. With this criterion, a general framework of conservatism control based on optimal performance guarantee is developed and characterized, and a new approach to competitive ratio analysis is established. The criterion is then applied to the robust one-way trading problem, where analytical solution is obtained, and the competitive ratio is derived directly via the new approach. Numerical experiments are conducted to demonstrate the effectiveness of conservatism control based on the new criterion, with the average reward improvable by 4% - 17% over the other commonly used criteria.