Essays on stock return predictability and portfolio allocation

Chapter One presents evidence of instability in models of ex-post predictable components in stock returns related to structural breaks in the coefficients of state variables such as the lagged dividend yield, Treasury Bill rate, term spread and default premium. The empirical analysis considers excess returns on both US and international stock indices. The results indicate systematic evidence of breaks in the vast majority of regression models for returns. The breakpoints most frequently identified are 1962 and 1974. The evidence suggests that the predictable component in US stock returns has diminished following the most recent break and is largely non-existent in the 1990s.;Chapter Two examines whether portfolio allocation strategies based upon predictable components in stock returns outperform unconditional strategies in an out-of-sample, as opposed to an in-sample, setting. Optimal portfolio decision rules are estimated for a buy and hold portfolio allocation problem using a flexible class of parametric models. In a finite sample setting, it is possible that a misspecified parametric estimator achieves a higher expected utility than a consistent nonparametric estimator. There is an implicit trade-off between bias and estimation uncertainty in selecting models for portfolio decision rules entirely analogous to the same trade-off in selecting econometric forecasting models.;Chapter Three considers finite sample inference problems that arise in regressions of "realized volatility" measures on persistent macroeconomic predictors such as the dividend yield and short interest rates. A spurious regression bias arises from persistent components in both realized volatility and macroeconomic predictors. A Monte Carlo simulation analysis reveals that the common practice of including one or two lags of the dependent variable in the regression specification may not be sufficient to control spurious regression bias when realized volatility exhibits long range dependence. The true extent of the "Sharpe ratio puzzle" suggested by Lettau and Ludvigson (2003) is difficult to ascertain since finite sample inference problems plague regression models for both expected returns and realized volatility.