Out-of-sample forecasting performance of the QGARCH model

Abstract

The population value of the R ² is derived from the Mincer-Zarnowitz volatility forecast regression for a QGARCH(1,1). The study shows that the population R ² exceeds that of the standard GARCH(1,1). This indicates that accounting for asymmetry in the conditional variance process can increase the predictive power of volatility forecasts. As with the standard GARCH(1,1) model, however, the R ² is still bounded by the reciprocal of the innovation kurtosis. As a result, small values of the R ² should be anticipated when using the QGARCH(1,1) in empirical work.

Notes

¹Similar results can be found in more recent papers cited in Granger and Poon's (2003) comprehensive survey.

²See Sentena (1995) for a detailed comparison.