Open AccessThis paper uses completely new data to study the variations in beta when it deviates from the constancy assumption presumed by the market model. The concentration of the literature on beta is based on post 1926 data. This makes the 19th century Brussels Stock Exchange (BSE) data a very good out-sample dataset to test beta variations. Various models proposed in the literature to capture the variations in beta were studied. Blume’s correlation techniques reveal that beta is not stable at the individual stocks level and that the stability can be fairly improved by portfolio formations. Using root mean square error (RMSE) criterion, it was shown that the market model betas are weak in predicting future betas. The predictability can be improved by adjusting betas with the Blume and Vasicek mean reversion techniques. Further results from this study reveal that few stocks have lead or lag relationship with the market index. Small sized stocks were detected to be more prone to outliers. In effect betas in the 19th century exhibits similar pattern as betas in the post 1926 era.