Heteroscedasticity and Interval Effects in Estimating Beta : UK Evidence

The paper compares beta estimates obtained from OLS regression with estimates corrected for heteroscedasticity of the error term using ARCH models, for 145 UK shares. The differences are mainly less than 0.10, for betas calculated using daily returns, but even such small differences can matter in practice. OLS tends to overestimate the beta coefficients compared with ARCH models, and selecting an ARCH-type estimate makes most difference for large-cap shares. Regarding the measurement interval, the downward bias in betas from daily returns is associated not only with thin trading but also with the volatility of the share's daily returns. We infer that the idiosyncratic component in daily returns, as well as lack of trading, is responsible for low daily betas