On the statistical and economic performance of stock return predictive regression models: an international perspective

The predictability of stock returns is assessed in 10 countries using the linear predictive regression framework. We use recently developed out-of-sample statistical tests and include both valuation ratios and interest rates as predictive variables. Contrary to previous studies, we explicitly address the issue of the small-sample bias, deal with trading profitability, and employ several risk-adjusted metrics. When statistical forecastability is found, it cannot be exploited to consistently deliver abnormal returns across countries and investment horizons. We hold the view that returns are predictable to some extent, but show that such forecasts are not useful for portfolio advice.