| Abstract: | In this paper, we show that estimating the correlation structure of domestic share prices via the Overall Mean method cannot be considered universally superior to estimation at the Full Historical level for all countries. Specifically, the Japanese data show that the Full Historical Model outperforms the Overall Mean Model in forecasting accuracy, while the opposite is true with the U.S. data. We derive a Composite Model that analytically explains this contrasting result. The Industry Mean Model, which allows for efficient ex ante portfolio selection via a simple algorithm, is likely to be the best forecasting model applicable to both the U.S. and Japanese stock markets. |
