MACHINE LEARNING TECHNIQUES APPLIED TO ROBUST OPTIMAL CONTROL PROBLEMS

Abstract

This project aims to solve the discrete time stochastic optimal control problem of evaluation of Average Value-at-Risk (AVaR) function. AVaR is an important tool in market risk management used to measure the risk. In the paper it was designed as a sequential decision model and solved by formulating an optimal control problem of minimizing the value. Brute force and Approximate Dynamic Programming (ADP) techniques were used for exact and approximate solutions respectively. Golden section search was used to solve the problem completely. The numerical experiments conducted at the end showed the effectiveness of the algorithm in evaluating the AVaR.