Do Power GARCH models really improve value-at-risk forecasts?

Traditional heteroskedastic models either rely on the specification of the conditional variance as in Bollerslev (1986) or on a direct modeling of the conditional standard deviation as in Taylor (1986). With its endogenous estimation of the optimal power transformation, the Power GARCH (PGARCH) of Ding, Granger, and Engle (1993) represents a flexible alternative that also nests the previous competing families. Building on a “dynamic” estimation and out-of-sample tests, the current paper undertakes a comparison of the three models in a value-at-risk setting. Despite existing fluctuations in the optimal power transformation obtained with the Ding, Granger, and Engle model, our empirical investigations suggest that the parameter is rarely found different from one or two. Although the volatility dynamics may switch from Taylor's to Bollerslev's specification during the life of the future contract, the measures of accuracy and efficiency used to assess the performance of VaR forecasts indicate that the additional flexibility brought by the PGARCH model provides little, if any, improvement for risk management. *** DIRECT SUPPORT *** A00DH023 00004 Copyright Springer 2005