Least Squares Predictions and Mean-Variance Analysis

We compare the Sharpe ratios of traders who combine one riskless and one risky asset following (i) buy and hold strategies; (ii) timing strategies with forecasts from simple; or (iii) multiple regressions; and (iv) passive allocations of (i) and (ii) with mean-variance optimizers. We show that (iv) implicitly uses the linear forecasting rule that maximizes the Sharpe ratio of managed portfolios, but the remaining rankings are unclear. We also suggest generalized method of moments (GMM) estimators to make (iv) operational and evaluate their significance with spanning tests. Finally, we characterize the equivalence between (iii) and (iv), and propose moment tests to assess it. Copyright 2005, Oxford University Press.