You are here

“Bayesian Variable Selection and Factor Models"

“Bayesian Variable Selection and Factor Models"

Coauthor(s): Bronson Argyle
Download (PDF)

Bayesian variable selection (BVS) algorithms offer robust, intuitive methods for determining the inclusion of specific regressors. After constructing a BVS framework, I apply this methodology to test a factor asset-pricing model, using 270 different portfolios, spanning 8 different sorting characteristics, onto 5 popular factors - the Fama-French 3 (SMB, HML, MKT), a measure of aggregate liquidity (PS-LIQ), and a measure of momentum (UMD). Results show that the Fama-French 3 are included and have inclusion probabilities .93, .69, and .66, respectively. There is marginal evidence that the momentum risk is priced (.22 probability of inclusion). There is little evidence that the measure of aggregate liquidity is priced (.11 probability of inclusion). The apparent nonpricing of aggregate liquidity may be due to even liquidity-spreading across portfolios. Applying BVS to ten large cap securities, I find a notable reduction in the MKT factor, an increase in the intercept (the stock alpha), but no relevant change to the PS-LIQ inclusion.

Exact Citation:
Bronson Argyle "“Bayesian Variable Selection and Factor Models"." , Columbia Business School, (2011).
Date: 2011