Discussion papers in statistics and econometrics 7/10
It is well known that the correlation between financial series varies over time. Here, the forecasting performance of different time-varying correlation models is compared for cross-country correlations of weekly G5 and daily European stock market indices. In contrast to previous studies only the correlation and not the entire covariance matrix is forecasted and multi-step forecasts are considered. The forecast comparison is done by considering statistical and economic criteria. The results suggest that under a statistical criterion time-varying correlation models perform quite well for weekly data, but cannot outperform the constant correlation model for daily data. Considering economic criteria it is hard to beat a constant correlation model.
dynamic conditional correlation regime switching stochastic correlation smooth correlations indirect model comparison portfolio construction