In Search of Sparsity: Bayesian Sparse Factor Models and the Factor Zoo (JMP)

with Fahiz Baba-Yara (Indiana University)

slides

A large number of potential observable risk factors that can explain the cross-section of stock returns have been proposed in asset pricing literature. Many of these factors are closely related, and there is often no clear group of factors that substantially better explain expected returns than others. This fact suggests a handful of latent factors driving variations in most observable risk factors. Although classical dimension reduction techniques can be used to uncover these latent factors, they mostly remain uninterpretable.

We apply a Bayesian Sparse Factor Model to the time series of observable risk factors to reduce the dimensionality and uncover latent factors. We combine the latent factors with a recently developed Bayesian Fama Macbeth regression (BFM) to assess how well they price the cross-section of returns. Unlike standard dimension reduction techniques, the sparsity prior allows us to efficiently describe the space of commonly proposed observable risks with latent factors that retain an economic interpretation. We show that these sparse latent factors better fit the cross-section of expected returns than models of the same size that define pricing factors as individual characteristic sorted portfolios.

Panel Thresholds Predictive Regressions: An Application to Emerging Market Exchange Rates

with Paulo Rodrigues (Bank of Portugal)

slides

A methodology is proposed for identifying threshold effects in panel predictive regressions in which the predictor variables are strongly persistent. Our statistic of interest is the average of individual supremum type statistics across the units of the panel. In certain cases, this statistic converges to a standard distribution while in other cases we show that a block bootstrap methodology can be applied and offers good finite sample characteristics when cross sectional dependence and serial correlation are present. This methodology is applied to assess the existence of threshold effects in an exchange rate predictability setting for emerging Sub-Saharan African economies during the post-2008 crisis period.