Empirical comparison of conventional methods and extreme value theory approach in value-at-risk assessment

Abstract

Value-at-risk (VaR) has become a standard tool in contemporary risk management practice. However, the latest financial crisis has put in question the adequacy of different methodologies for VaR estimation. This paper investigate the predictive performances of eight VaR models, ranging from well-known historical simulation and exponentially weighted moving average (EWMA) models to more advanced models such as generalized autoregressive conditionally heteroscedastic (GARCH) and extreme value theory (EVT). The special emphasis was paid to the approach that used GARCH model to estimate volatility of returns and EVT model to estimate the tails of GARCH residuals. The research covers the sample of daily returns of FTSE100 index from March 25, 1997 to March 22, 2011. This sample period was chosen since it covers some major crisis and shocks, thus being suitable for testing robustness of these models. All models are statistically backtested and obtained results proved that EVT-based methodology generated the most accurate VaR estimates.