The recent 2007 – 2009 financial crisis demonstrated, if nothing else, that the financial institutions’ risk management systems were not as adept as previously thought in tracking and anticipating the extreme price movements witnessed during that highly volatile period. Nearly all financial institutions recorded multiple consecutive exceptions, i.e. days in which the trading book losses exceeded the prescribed Value-at-Risk (VaR). In several instances, the total number of exceptions during the previous trading year exceeded the threshold of ten violations which is the set regulatory maximum (Campel and Chen, 2008). Consequently, much doubt was cast and many questions were raised about the reliability and accuracy of the implemented VaR models, systems and procedures.
However, the criticisms faced by the risk management departments can hardly be attributed to a lack of allocated resources or research efforts. VaR measurement and forecasting has been one of the most vigorously researched areas in quantitative risk management and financial econometrics. It has also enjoyed significant investments both in terms of capex and in human capital within banks and financial institutions. In this context, the evaluation of some recently proposed volatility models which make use of the informational content in high frequency data could reveal some attractive alternative VaR modelling specifications.
The foundations of modern risk management were laid with the seminal work of Engle (1982) who introduced the AutoRegressive Conditional Heteroscedasticity (ARCH) model for modeling the conditional heteroscedasticity in financial assets returns. Since then, a plethora of ARCH–type models have been proposed in the open literature (see Bollerslev, 2010 for a short description for almost all ARCH–type models) and most of them have been included in VaR studies. Giot and Laurent (2003a and 2003b) for example, showed that flexible ARCH specifications combined with fat tailed distributions can provide accurate VaR forecasts for a wide range of assets.
More recently, Andersen and Bollerslev (1998), Andersen et al. (2001a), Andersen et al.(2001b) and Barndorff-Nielsen and Shephard (2002) introduced and promoted the realized volatility as a non-parametric approach for measuring the unobserved volatility. In Andersen et al. (2003), the authors also suggested that standard time series techniques can be used in order to model the “observable” realized volatility. These concrete theoretical foundations coupled with the increased availability of high quality intraday data for a wider range of assets, fuelled the research interest on the use of high frequency data for measuring and forecasting the volatility of financial assets. Several authors demonstrated the superiority of realized volatility models over ARCH models for volatility forecasting (see Koopman et al., 2005; Martens et al., 2009; Martens, 2002 among others), while Giot and Laurent (2004) first utilized high frequency intraday data in a VaR forecasting context.
Download
Are realized volatility models good candidates for alternative Value at Risk prediction strategies?
