We use predictions of aggregate stock return variances from daily data to estimate time‐varying monthly variances for size‐ranked portfolios. We propose and estimate a single factor model of heteroskedasticity for portfolio returns. This model implies time‐varying betas. Implications of heteroskedasticity and time‐varying betas for tests of the capital asset pricing model (CAPM) are then documented. Accounting for heteroskedasticity increases the evidence that risk‐adjusted returns are related to firm size. We also estimate a constant correlation model. Portfolio volatilities predicted by this model are similar to those predicted by more complex multivariate generalized‐autoregressive‐conditional‐heteroskedasticity (GARCH) procedures.

We use predictions of aggregate stock return variances from daily data to estimate time-varying monthly variances for size-ranked portfolios. We propose and estimate a single factor model of heteroskedasticity for portfolio returns. This model implies time-varying betas. Implications of heteroskedasticity and time-varying betas for tests of the capital asset pricing model (CAPM) are then documented. Accounting for heteroskedasticity increases the evidence that risk-adjusted returns are related to firm size. We also estimate a constant correlation model. Portfolio volatilities predicted by this model are similar to those predicted by more complex multivariate generalized-autoregressive-conditional-heteroskedasticity (GARCH) procedures.

The authors use predictions of aggregate stock return variances from daily data to estimate time-varying monthly variances for size-ranked portfolios. The authors propose and estimate a single factor model of heteroskedasticity for portfolio returns. This model implies time-varying betas. Implications of heteroskedasticity and time-varying betas for tests of the capital asset pricing model are then documented. Accounting for heteroskedasticity increases the evidence that risk-adjusted returns are related to firm size. The authors also estimate a constant correlation model. Portfolio volatilities predicted by this model are similar to those predicated by more complex multivariate generalized autoregressive conditional heteroskedasticity procedures.