Large-scale volatility models: theoretical properties of professionals' practice

This article examines the way in which GARCH models are estimated and used for forecasting by practitioners in particular using the highly popular Riskmetrics-super-TM approach. Although it permits sizable computational gains and provide a simple way to impose positive semi-definitiveness of multivariate version of the model, we show that this approach delivers non-consistent parameter' estimates. The novel theoretical result is corroborated by a set of Monte Carlo exercises. A set of empirical applications suggest that this could cause, in general, unreliable forecasts of conditional volatilities and correlations. Copyright 2008 The Author