Value-at-Risk and Expected Shortfall estimation using Machine Learning

Managing and estimating risk are one of the most important functions of financial institutions. Markets trends are complicated and often unpredictable so it is crucial for investment decisions to be based on accurate measures of risk. In this thesis, we ask ourselves how machine learning can be used to estimate the following two risk measures: Value-at-Risk (VaR) and Expected Shortfall (ES). We provide an univariate and multivariate analysis on two portfolios of assets and we show that kernel quantile regression and ICA can be both used to compute those risk measures. We observe that these techniques seem to outperform more traditional methods. We then conclude and discuss about how other machine learning algorithms could be used for the same purpose. Overall, it seems these new methods show great potential for more accurate risk estimation.