This paper compares the performance of three different time-varying betas that have never previously been compared: the rolling OLS estimator, a nonparametric estimator and an estimator based on GARCH models. The study is conducted using returns from the Mexican stock market grouped into six portfolios for the period 2003-2009. The comparison, based on asset pricing perspective and mean-variance space returns, concludes that GARCH based beta estimators outperform the others when the comparison is in terms of time series while the nonparametric estimator is more appropriate in the cross-sectional context.

The authors acknowledge financial support from Ministerio de Ciencia e Innovación under research grants ECO2009-09120, ECO2008-00777/ECON, ECO2008-02599 and ECO2011- 29751, and from Dpto. de Educación, Universidades e Investigación del Gobierno Vasco under research grants IT-313-07 and IT-241-07.