This paper explores two issues in beta estimation, specifically, time variation and thin trading. In a multivariate GARCH approach, the paper conducts an analysis of the importance of assumptions made about the correlation structure in the multivariate GARCH model. The results of Monte Carlo analysis and an empirical application to Australian stock data demonstrate that it is better to allow for time variation in the correlation structure. The paper then develops a selectivity corrected time varying beta estimator. The results of a Monte Carlo experiment show that the new estimator performs well in handling the censoring in the data. Further, when the model is applied to individual stock data for Australia it provides a model that captures the impacts of censoring and thin trading on time varying beta risk.