A0974
Title: Data and model uncertainty in the cross-section of equity returns
Authors: Serhiy Kozak - University of Maryland (United States)
Jiantao Huang - University of Hong Kong (Hong Kong) [presenting]
Abstract: A two-layer hierarchical Bayesian framework is developed to study the effects of data and model uncertainty on the pricing of the cross-section of equity returns. In the first layer, which handles data uncertainty, a Bayesian Tensor Model is proposed to generate a probability distribution of missing, infrequently, or imprecisely observed characteristic data. It is then drawn from this distribution to estimate a Bayesian factor pricing model of stock returns, which can be seen as a probabilistic generalization of the IPCA model. Bayesian averaging across the space of factor pricing models provides regularization similar to a prior study. Further averaging across posterior draws of characteristics data provides additional robustness with respect to uncertainty in the characteristics data. Jointly, accounting for both sources of uncertainty more than doubles the Sharpe ratios of alpha portfolios.
