This paper evaluates the predictive ability of financial ratios—dividend yield (DY), book-to-market (B/M), and earnings-price (E/P)—for forecasting stock returns. The study addresses the issue of small-sample biases in predictive regressions, which can significantly underestimate the true predictive power of these ratios. The analysis focuses on monthly returns and uses a conditional distribution approach to account for the autocorrelation in the predictive variables. The results show that DY, B/M, and E/P have strong predictive power for stock returns. DY predicts returns from 1946–2000, as well as in various subperiods. B/M and E/P also predict returns, though the evidence is less reliable than for DY. The study corrects for small-sample biases by incorporating information about the sample autocorrelation of the predictive variables. This adjustment leads to more accurate estimates and stronger statistical significance. The paper finds that DY has a significant predictive effect on both equal- and value-weighted NYSE indices. The conditional tests, which account for the autocorrelation in DY, show that the predictive power of DY is much stronger than previously thought. For example, the bias-adjusted slope for DY on the value-weighted index is 0.66 with a t-statistic of 4.67, significant at the 0.000 level. Similar results are found for B/M and E/P, though their predictive power is weaker than that of DY. The study also examines the impact of recent data (1995–2000) on the results. While the inclusion of this data reduces the OLS slope estimates for DY, B/M, and E/P, the bias-adjusted estimates remain significant. The sharp increase in the sample autocorrelation of DY during this period reduces the bias adjustment, leading to stronger predictive power. Overall, the paper concludes that financial ratios like DY, B/M, and E/P have significant predictive power for stock returns, and that the conditional tests provide a more accurate assessment of this power than traditional unconditional tests. The findings highlight the importance of accounting for autocorrelation in predictive regressions to avoid underestimating the true predictive ability of financial ratios.This paper evaluates the predictive ability of financial ratios—dividend yield (DY), book-to-market (B/M), and earnings-price (E/P)—for forecasting stock returns. The study addresses the issue of small-sample biases in predictive regressions, which can significantly underestimate the true predictive power of these ratios. The analysis focuses on monthly returns and uses a conditional distribution approach to account for the autocorrelation in the predictive variables. The results show that DY, B/M, and E/P have strong predictive power for stock returns. DY predicts returns from 1946–2000, as well as in various subperiods. B/M and E/P also predict returns, though the evidence is less reliable than for DY. The study corrects for small-sample biases by incorporating information about the sample autocorrelation of the predictive variables. This adjustment leads to more accurate estimates and stronger statistical significance. The paper finds that DY has a significant predictive effect on both equal- and value-weighted NYSE indices. The conditional tests, which account for the autocorrelation in DY, show that the predictive power of DY is much stronger than previously thought. For example, the bias-adjusted slope for DY on the value-weighted index is 0.66 with a t-statistic of 4.67, significant at the 0.000 level. Similar results are found for B/M and E/P, though their predictive power is weaker than that of DY. The study also examines the impact of recent data (1995–2000) on the results. While the inclusion of this data reduces the OLS slope estimates for DY, B/M, and E/P, the bias-adjusted estimates remain significant. The sharp increase in the sample autocorrelation of DY during this period reduces the bias adjustment, leading to stronger predictive power. Overall, the paper concludes that financial ratios like DY, B/M, and E/P have significant predictive power for stock returns, and that the conditional tests provide a more accurate assessment of this power than traditional unconditional tests. The findings highlight the importance of accounting for autocorrelation in predictive regressions to avoid underestimating the true predictive ability of financial ratios.