Performance of Bayesian Latent Factor Models in Measuring Pricing Errors

This study offers a Bayesian factor modelling framework to obtain the exact distribution of pricing errors in the bounds of Arbitrage Pricing Theory with the aim of observing, first if the usage of a dynamic model, second increasing the number of factors beyond one contributes to a significant reduction of the pricing errors obtained. In doing so, we compare the pricing errors we get from a static and dynamic latent factor model, while adopting the Fama-French data for US monthly industry returns. We observe that the pricing errors increase slightly using a dynamic factor model, when compared with the static factor model. Besides, inclusion of factors beyond the first one pose an improvement with respect to the pricing errors both for the static and the dynamic factor model. When we introduce time-varying betas to the dynamic factor model we get the lowest pricing errors at t=1 compared to static and dynamic model with fixed betas, where the mean pricing errors decreased by 33 percent compared to the static model. Yet pricing errors also become time varying as their dynamics now depend on the dynamics of beta.