Differences in inputs, particularly estimated correlation coefficients, can have a significant impact on the weighting of asset classes in an asset allocation scheme and on the risk and return characteristics of that combination. Most experts in the field of asset allocation have relied heavily on extrapolation of historical data to develop both standard deviation and correlation inputs. But historical data show a great deal of variability and can be misleading indicators of future behavior.

More accurate inputs for asset allocation can be generated by the following four-step process: (1) assess the basic financial characteristics of each asset class; (2) evaluate how, given these characteristics, the asset class is likely to react to changing economic conditions; (3) examine how asset classes have historically reacted to economic changes; and (4) develop scenarios and assign probabilities to them to generate forecasts of return, variability and correlation.

Such a four-step analysis suggests that the pattern of correlation between stocks and bonds over the next several years is likely to be significantly different from the strong positive correlation of the recent past. The consensus forecast, however, seems to be for the persistence of positive correlation, and major asset class pricing may well reflect that. To the extent a new pattern of correlation has begun, asset prices will have to adjust, creating the opportunity for superior performance through a fundamental forecast approach to asset allocation.