Multi-period market risk estimation and performance evaluation : evidence from univariate, multi-variate and options data

There are different risk management approaches available, as different firms have different risk goals. Value at risk (VaR) is the most frequently used risk measure for asset or portfolio risk and certainly, per the Basel framework, is a preferred measure for market risk for banks and financial institutions. VaR is still the most popular method for performing financial risk, although it has been criticized on many grounds by academic researchers. A coherent measure of financial risk referred to as expected shortfall (hereinafter ES) was proposed by Artzner et al. (1999) to overcome problems associated with VaR. In the first part of the thesis we evaluate expected shortfall (ES) with a new 6-parameter heavy tailed distribution by Baker (2014) alongside recent generalizations of the asymmetric Student t by Zhu and Galbraith (2010) and exponential power distributions by Zhu and Zinde-Walsh (2009). This is allowing separate parameters to control skewness and tail thickness for both stocks and indexes. The results suggest that GAT of Baker (2014) outperforms both AST of Zhu and Galbraith (2010) and APED by Zhu and Zinde-Welsh (2009) for both 1-day and multi-day ES forecasts. In the second part of the thesis, we present and discuss the use of copulas and vine copulas for financial risk management, also introduce the term structure of risk for bivariate and multivariate data. To the best of our knowledge, this study is the first to explore multivariate term structure of risk with both static and dynamic conditional correlation. The results suggest that copula models for two-dimensional data and vine copula models for five, seven and fifteen-dimensional data provide a good fit and accurately and efficiently forecast the expected shortfall as compared to DCC-norm and DCC-t. In the third part of the thesis, we compare the performance of the Heston option pricing model, Bates option pricing model, Merton jump diffusion option pricing model, Kou option pricing model and variance gamma option pricing model with a traditional Black-Scholes option pricing model. We also evaluate expected shortfall estimates for European options for 1-day and 10-days at a range of confidence levels with full Monte Carlo and Monte Carlo delta and Monte Carlo delta gamma derived from option pricing models tested in our research. The results indicate that full valuation appears to be one of the top models for both 1-day ahead and multi-days ahead ES. This gives us clear implications for calculation of ES beyond 10-days