[If returns are not predictable, dividend growth must be predictable, to generate the observed variation in divided yields. I find that the absence of dividend growth predictability gives stronger evidence than does the presence of return predictability. Long-horizon return forecasts give the same strong evidence. These tests exploit the negative correlation of return forecasts with dividend-yield autocorrelation across samples, together with sensible upper bounds on dividend-yield autocorrelation, to deliver more powerful statistics. I reconcile my findings with the literature that finds poor power in long-horizon return forecasts, and with the literature that notes the poor out-of-sample R² of return-forecasting regressions.]

[I report a bound on the variance of price-dividend ratios and a decomposition of their variance into terms that reflect changes in dividend growth and discount rates. The specification is not restrictive. The test statistics do not require construction of ex post present values; instead, they are restrictions on means, variances, and covariances of price-dividend ratios, dividend growth, and discount rates. I consider implications for the mean price-dividend ratio, and I evaluate whether a low mean discount rate can rationalize the mean and variance of price-dividend ratios. The results do not indicate any striking rejections of present-value models. However, the bulk of the variance of price-dividend ratios must be accounted for by changing forecasts of discount rates, and discount rates must possess some unusual characteristics.]

To question the statistical significance of return predictability, we cannot specify a null that simply turns off that predictability, leaving dividend growth predictability at its essentially zero sample value. If neither returns nor dividend growth are predictable, then the dividend-price ratio is a constant. If the null turns off return predictability, it must turn on the predictability of dividend growth, and then confront the evidence against such predictability in the data. I find that the absence of dividend growth predictability gives much stronger statistical evidence against the null, with roughly 1-2% probability values, than does the presence of return predictability, which only gives about 20% probability values. I argue that tests based on long-run return and dividend growth regressions provide the cleanest and most interpretable evidence on return predictability, again delivering about 1-2% probability values against the hypothesis that returns are unpredictable. I show that Goyal and Welch's (2005) finding of poor out-of-sample R does not reject return forecastability.

The author reports a bound on the variance of price-dividend ratios and a decomposition of their variance into terms that reflect changes in dividend growth and discount rates. The specification is not restrictive. The test statistics do not require construction of ex post present values; instead, they are restrictions on means, variances, and covariances of price-dividend ratios, dividend growth, and discount rates. He considers implications for the mean price-dividend ratio, and he evaluates whether a low mean discount rate can rationalize the mean and variance of price-dividend ratios. The results do not indicate any striking rejections of present-value models. However, the bulk of the variance of price-dividend ratios must be accounted for by changing forecasts of discount rates, and discount rates must possess some unusual characteristics. Article published by Oxford University Press on behalf of the Society for Financial Studies in its journal, The Review of Financial Studies.

[We solve a model with two i.i.d. Lucas trees. Although the corresponding one-tree model produces a constant price-dividend ratio and i.i.d. returns, the two-tree model produces interesting asset-pricing dynamics. Investors want to rebalance their portfolios after any change in value. Because the size of the trees is fixed, prices must adjust to offset this desire. As a result, expected returns, excess returns, and return volatility all vary through time. Returns display serial correlation and are predictable from price-dividend ratios. Return volatility differs from cash-flow volatility, and return shocks can occur without news about cash flows.]

We solve a model with two i.i.d. Lucas trees. Although the corresponding one-tree model produces a constant price-dividend ratio and i.i.d. returns, the two-tree model produces interesting asset-pricing dynamics. Investors want to rebalance their portfolios after any change in value. Because the size of the trees is fixed, prices must adjust to offset this desire. As a result, expected returns, excess returns, and return volatility all vary through time. Returns display serial correlation and are predictable from price-dividend ratios. Return volatility differs from cash-flow volatility, and return shocks can occur without news about cash flows.