Using short-window regressions, prior studies have shown that the conditional CAPM performs as poorly as the unconditional CAPM and that unconditional alphas are too large to be explained by the covariance between conditional betas and market risk premium. I examine the extent to which these results are driven by estimation biases in short-window regressions. Even when the true time-varying beta has highly transient components, beta estimates from these short-window regressions can be estimated to highly persistent because of yearly portfolio reformulations that are commonly employed in empirical studies. I show that, under these circumstances, large non-zero alpha estimates from short-window regressions are consistent with the conditional CAPM and the covariance estimates between conditional betas and market risk premium are biased downward and can have large estimation errors.