Value-at-risk forecasts with conditional volatility for structured products

Fen-Ying Chen

Abstract

ABSTRACT

The existing literature commonly concludes that generalized autoregressive conditional heteroskedasticity (GARCH) models provide better volatility forecasts in financial markets, using mean absolute squared errors or mean squared error criteria based on normality and serially uncorrelated assumptions for forecast errors. In contrast to the majority of the literature, this paper adopts the Diebold and Mariano test to reexamine the performance of GARCH models, allowing for forecast errors that can be non-Gaussian, nonzero mean, serially correlated and contemporaneously correlated for structured products. The results consistently show that the performance of GARCH-type models is not significantly better during the period of low oil prices or the period of high oil prices.