This paper compares the performance of nine time-varying beta estimates taken from three different methodologies never previously compared: least-square estimators including nonparametric weights, GARCH-based estimators and Kalman filter estimators. The analysis is applied to the Mexican stock market (2003-2009) because of the high dispersion in betas. The comparison be- tween estimators relies on their financial applications: asset pricing and portfolio management. Results show that Kalman filter estimators with random coefficients outperform the others in capturing both the time series of market risk and their cross-sectional relation with mean returns, while more volatile estimators are better for diversification purposes.