Comparison of Evolutionary Techniques for Value-at-Risk Calculation

The Value-at-Risk (VaR)approach has been used for measuring and controlling the market risks in financial institutions. Studies show that the t-distribution is more suited to representing the financial asset returns in VaR calculations than the commonly used normal distribution. The frequency of extremely positive or extremely negative financial asset returns is higher than that is suggested by normal distribution. Such a leptokurtic distribution can better be approximated by a t-distribution. The aim of this study is to asses the performance of a real coded Genetic Algorithm (GA) with Evolutionary Strategies (ES) approach for Maximum Likelihood (ML) parameter estimation. Using Monte Carlo (MC) simulations, we compare the test results of VaR simulations using the t-distribution, whose optimal parameters are generated by the Evolutionary Algorithms (EAs), to that of the normal distribution. It turns out that the VaR figures calculated with the assumption of normal distribution significantly understate the VaR figures computed from the actual historical distribution at high confidence levels. On the other hand, for the same confidence levels, the VaR figures calculated with the assumption of t-distribution are very close to the results found using the actual historical distribution. Finally, in order to speed up the MC simulation technique, which is not commonly preferred in financial applications due to its time consuming algorithm, we implement a parallel version of it.