Value at Risk Estimation using GARCH Family Models: A Comparison of Different Specifications and Distributions.

Abstract

The objective of this study is to compare the performance of different GARCH models, under various conditional distribution assumptions, to predict one-day-ahead Value-at-Risk (VaR) for three stocks: Swedbank, Handelsbanken, and SEB over the Covid-19 period. The performance is evaluated using Kupiec, Christoffersen tests and the Quadratic Loss. The results show that the assumed distribution, model specification, and confidence level together play an important role in VaR estimation, as these factors can substantially affect the accuracy of the estimate. Models that assume the skew t-distribution and the t-distribution generally perform well, while the performance of models that assume the normal distribution changes dramatically as a function of the confidence level. Regarding the models, the study shows that the EGARCH specification resulted in the lowest losses and that the worst performing model is ARCH, especially when the assumed distribution is normal.