Value at Risk Estimation for Non-Gaussian Distributions

This paper presents a methodology for computing Value at Risk for financial assets that does not follow a normal distribution of return. A back-testing approach have been applied in order to select the best theoretical non-Gaussian distributions that can explain the behavior of the empirical data. In this study, Cauchy, Laplace, Logistic and Beta distributions have been considered. As benchmark, historical distribution and Extreme Value Theory (EVT) method have been used.  The experiment suggests differences in estimation of over 5 times between one method and another.