Abstract: | Recently, the case for international portfolio diversification has been convincingly argued in the framework of mean-variance portfolio analysis by a number of researchers. However, virtually no empirical documentation exists concerning the best method for estimating the correlation structure of international share prices. In this paper, 12 models for estimating the international correlation matrix are presented and empirically tested relative to full historical extrapolation. The major evaluation criteria are the mean squared error and stochastic dominance based on the frequency distribution of the squared forecast errors. The results indicate that the National Mean Model strictly dominates all the others in terms of forecasting accuracy. |