Modeling Value-at-Risk for Commodities
| Author | : Lionel Gerboth |
| Publisher | : |
| Total Pages | : |
| Release | : 2013 |
| Genre | : |
| ISBN | : |
Download Modeling Value-at-Risk for Commodities Book in PDF, ePub and Kindle
Assessing the significance of the market risk of a portfolio of financial securities has long been acknowledged by academics and practitioners. In measuring market risk, one of the most advanced techniques in the literature is the value-at-risk (VaR) measurement. Although widely used for traditional markets, in commodity markets which usually exhibit peculiar features like volatility jumps and price spikes, the use of nonparametric and semi-parametric VaR modeling techniques has not yet been analyzed exhaustively. This master thesis studies in a comprehensive out-of-sample backtesting procedure traditional models like historical simulation, filtered historical simulation, univariate GARCH-type models with and without leverage and compares them to state of the art modeling techniques like EVT-EGARCH models using the peak-over-threshold procedure as well as EVT-EGARCH-Copula models with Gaussian and Student-t-copulas. Goal of the analysis is to provide a comprehensive comparison of the forecast ability of each model with respect to each out of the four main commodity classes (agricultural, energy, livestock and metals). Among all models presented, empirical results for a number of adequacy and accuracy tests suggest that the conditional t-copula approach with EVT modeled tails and EGARCH standardized residuals performs best for high confidence levels and across all commodity classes analyzed, shortly followed by the Gaussian copula approach. For lower significance levels however, the GARCH and EGARCH model seem to outperform. As anticipated, the naïve historical simulation approach performs worst over all confidence levels and commodity portfolios. The filtered historical simulation and the EVT model demonstrate a mixed performance.
