On the Efficiency of Least Squares Regression with Security Abnormal Returns as the Dependent Variable

Monte Carlo procedures are used to compare the finite sample performance of several estimators that may be used in cross-sectional regressions with security abnormal returns as the dependent variable. Alternative models of event-induced increases in stock return variance are examined for the “event-clustering” scenario. Event clustering implies crosssectional correlation and heteroskedasticity in market model prediction errors, violating one of the fundamental ordinary least squares (OLS) assumptions (i.i.d. disturbances). Nonetheless, provided that the conditions for asymptotic validity derived by Greenwald (1983) are met, the OLS estimator is well specified in finite samples. Further, for sufficiently large cross sections there is no advantage to several other more complex estimators.