Asset pricing models represent one of the most discussed and researched areas in finance. They are widely used in a theoretical and practical manner to model and predict risk and return to price securities and portfolios as well as in corporate finance to estimate the cost of capital and rank investment projects. They provide a usable measure of risk that helps managers and investors determine what return they deserve for putting their money at risk. The objective of this paper is to analyze the performance of the learning-augmented conditional CAPM model of Adrian and Franzoni (2009) when applied to the returns of the most liquid stocks transactioned in the Brazilian stock market from 1987 to 2010. Adrian and Franzoni (2009), in their paper, complemented the conditional CAPM literature by modeling a new type of time-variation in conditional betas. In this environment, investors form expectations about the long run level of factor loadings from the observation of realized returns of exogenous variables. As a direct consequence of this assumption, conditional betas are modeled using the Kalman filter. Using data of 25 portfolios sorted by size and book-to-market ratio, the authors concluded that the learning-augmented conditional CAPM is able to substantially reduce the pricing errors when compared to the original version of CAPM. Thus, we contribute to the pricing asset literature, as we evaluate whether this model is able to reduce pricing errors in relation to its original version when applied to Brazilian individual asset data. The results of this article showed a decreasing in the pricing errors of learning-augmented conditional CAPM in relation to CAPM in its original version. Our empirical results suggests that the learning about betas should be taken into account when estimating both conditional and unconditional CAPM.
Conditional CAPM; Kalman filter; Forecasting; Beta coefficient; Pricing errors
