Thin-Trading Effects in Beta: Bias "v." Estimation Error

Two regression coefficients often used in Finance, the <link rid="b23">Scholes-Williams (1977)</link> quasi-multiperiod 'thin-trading' beta and the <link rid="b19">Hansen-Hodrick (1980)</link> overlapping-periods regression coefficient, can both be written as instrumental-variables estimators. Competitors are Dimson's beta and the Hansen-Hodrick original OLS beta. We check the performance of all these estimators and the validity of the "t"-tests in small and medium samples, in and outside their stated assumptions, and we report their performances in a hedge-fund style portfolio-management application. In all experiments as well as in the real-data estimates, less bias comes at the cost of a higher standard error. Our hedge-portfolio experiment shows that the safest procedure even is to simply match by size and industry; any estimation just adds noise. There is a clear relation between portfolio variance and the variance of the beta estimator used in market-neutralizing the portfolio, dwarfing the beneficial effect of bias. Copyright (c) 2008 The Authors Journal compilation (c) 2008 Blackwell Publishing Ltd.