Merits and Drawbacks of Variance Targeting in GARCH Models

Variance targeting estimation (VTE) is a technique used to alleviate the numerical difficultiesencountered in the quasi-maximum likelihood (QML) estimation of GARCH models. It relieson a reparameterization of the model and a first-step estimation of the unconditional variance.The remaining parameters are estimated by QML in a second step. This paper establishesthe asymptotic distribution of the estimators obtained by this method. Comparisons with thestandard QML are provided. In particular, when the model is misspecified the VTE can besuperior to the QMLE for long-term prediction or Value-at-Risk calculation. An empiricalapplication based on stock market indices is proposed.