With asset allocation riding a wave of investor popularity, investors have begun to treat asset allocation with the same diligence afforded manager and security selection. The common requirements of essentially every approach to asset allocation are estimates of the returns, risks and correlations of the assets. Of these three estimates, correlations have appeared the most stable and have therefore received the least attention. But the correlation between stocks and bonds, for example, is stable only in relation to the variability of the assets’ risks and returns. In an absolute sense, correlations between assets can be very unstable, hence unpredictable.

The traditional approach to predicting correlations—extrapolating past correlations—leads to unsatisfactory results. The historical correlation between stocks and bonds, for example, provides a poor prediction of future correlations. Consideration of the difference between the current income yields of stocks and bonds can improve upon the correlation predictions derived from a purely historical approach.