A major drawback with the classical implementation of mean-variance analysis is that it completely ignores the effect of measurement error on optimal portfolio allocations. A simple simulation approach can provide insight into the distribution of optimal portfolio weights.
As an example, an ex post optimal portfolio of U.S. and foreign bonds is compared with two benchmarks a world bond index and a U.S. bond index. Taking sampling variability into account, there is no evidence that the optimal portfolio outperformed the world index over the 1978-88 period. The optimal portfolio did, however, perform significantly better than the U.S. index.
These results suggest that, over the time period studied, international diversification into foreign bonds has offered some benefits. These benefits are best measured, however, by comparing the performance of a passive world index with that of a U.S. index. An ex post mean-variance analysis systematically overstates the possible gains from going international.