A comparison of different procedures for combining high-dimensional multivariate volatility forecasts

The paper investigates the effect of model uncertainty on multivariate volatility prediction. Our aim is twofold. First, by means of a Monte Carlo simulation, we assess the accuracy of different techniques in estimating the combination weights assigned to each candidate model. Second, in order to investigate the economic profitability of forecast combination, we present the results of an application to the optimization of a portfolio of US stock returns. Our main finding is that, for both real and simulated data, the results are highly sensitive not only to the choice of the model but also to the specific combination procedure being used.