Forecasting multivariate volatility in larger dimensions: some practical issues

The importance of covariance modelling has long been recognised in the field of portfolio management and large dimensional multivariate problems are increasingly becoming the focus of research. This paper provides a straightforward and commonsense approach toward investigating whether simpler moving average based correlation forecasting methods have equal predictive accuracy as their more complex multivariate GARCH counterparts for large dimensional problems. We find simpler forecasting techniques do provide equal (and often superior) predictive accuracy in a minimum variance sense. A portfolio allocation problem is used to compare forecasting methods. The global minimum variance portfolio and Model Confidence Set (Hansen, Lunde, and Nason (2003)) are used to compare methods, whilst portfolio weight stability and computational time are also considered.