An Empirical Analysis of the Influence of Jump Dynamics on Value-at-Risk Estimation

Abstract: Abstract Recent financial crises have demonstrated the importance of accurately measuring financial risks in order to be able to employ mitigating risk policies. This essay investigates the question whether a GARCH model augmented by an autoregressive conditional jump intensity component can improve value-at-risk forecasts in comparison to common GARCH models. Therefore, the VaR forecasting performance of the GARJI model, introduced by Chan and Maheu (2002) and Maheu and McCurdy (2004), is compared to that of a normal-GARCH model and three GARCH models with leptokurtic distributional assumptions. Estimations and calculations are based on single stock return data of ten European banks over the period 27th February 2003 to 31st December 2015. The results show that the GARJI model yields an improved VaR forecasting performance over the normal-GARCH model, in particular for very high VaR levels. However, it is nevertheless outperformed by the GARCH models with leptokurtic distributional assumptions as fat-tails properties appear to matter more for producing accurate VaR forecasts than accounting for jump dynamics in the return data.