Forecasting Exchange Rate Value-at-Risk and Expected Shortfall: A GARCH-EVT Approach

Abstract: This thesis aims to investigate the accuracy of Value-at-Risk and Expected Shortfall forecasts of various GARCH-type models based on five currency exchange rate pairs. The GARCH models are employed under different conditional distributional assumptions, and extended using the two-stage Extreme Value Theory (EVT) approach of McNeil and Frey (2000). The forecasts are evaluated through simulation using the backtesting methodologies of Christoffersen (1998) and Acerbi & Szekely (2014). We find that forecasts of models assuming a skewed t-distribution are rejected the least number of times. Furthermore, the usefulness of the EVT approach of McNeil and Frey (2000) appears to be dependent on the distributional assumption as well as the choice of quintile. No conditional volatility model is consistently found to be superior to the others.