An ICA-GARCH approach to computing portfolio VAR with applications to South African financial markets

• Main Author: Mombeyarara, Victor
• Format: Others
• Language: en
• Published: 2017
• Subjects: Finance > Mathematical models; Investments > Mathematical models; Risk management > South Africa
• Online Access: Mombeyarara, Victor (2017) An ICA-GARCH approach to computing portfolio VAR with applications to South African financial markets, University of the Witwatersrand, Johannesburg, <http://hdl.handle.net/10539/23218>; http://hdl.handle.net/10539/23218

Master of Management in Finance & Investment Faculty of Commerce Law and Management Wits Business School University of The Witwatersrand 2016 === The Value-at-Risk (VaR) measurement – which is a single summary, distribution independent statistical measure of losses arising as a result of market movements – has become the market standard for measuring downside risk. There are some diverse ways to computing VaR and with this diversity comes the problem of determining which methods accurately measure and forecast Value-at-Risk. The problem is two-fold. First, what is the distribution of returns for the underlying asset? When dealing with linear financial instruments – where the relationship between the return on the financial asset and the return on the underlying is linear– we can assume normality of returns. This assumption becomes problematic for non-linear financial instruments such as options. Secondly, there are different methods of measuring the volatility of the underlying asset. These range from the univariate GARCH to the multivariate GARCH models. Recent studies have introduced the Independent Component Analysis (ICA) GARCH methodology which is aimed at computational efficiency for the multivariate GARCH methodologies. In our study, we focus on non-linear financial instruments and contribute to the body of knowledge by determining the optimal combination for the measure for volatility of the underlying (univariate-GARCH, EWMA, ICA-GARCH) and the distributional assumption of returns for the financial instrument (assumption of normality, the Johnson translation system). We use back-testing and out-of-sample tests to validate the performance of each of these combinations which give rise to six different methods for value-at-risk computations. === MT2017