This paper uses the cross-sectional variance of the betas from the CAPM model to study herd behavior towards market index in Romania. For time-varying beta determination, three different modeling techniques are employed: two bivariate GARCH models (DCC and FIDCC GARCH), two Kalman filter based approaches and two bivariate stochastic volatility models. A comparison of the different models’ in-sample performance indicates that the mean reverting process in connection with the Kalman filter and the stochastic volatility model with a t distribution for the excess return shocks are the preferred models to describe the time-varying behavior of stocks betas. Through the estimated values, the evolution of the herding measure, especially the pattern around the beginning of the subprime crisis is examined. Herding towards the market shows significant movements and persistence independently from and given market conditions and macro factors. Contrary to the common belief, the subprime crisis reduces herding and is clearly identified as a turning point in herding behavior.