Abstract

Pastor and Stambaugh’s (PS 2003) aggregate liquidity innovations can be closely replicated, as can their traded factor based on historically estimated liquidity betas, which performs even stronger out of sample. This factor’s performance is highly sensitive to construction details, however, and exhibits significantly weaker performance when rebalanced at its natural monthly frequency, or when constructed using either more or less extreme sorts. Their predicted liquidity risk factor is more difficult to replicate, and difficult to interpret because characteristics chosen to predict liquidity risk introduce mechanical relations to other known anomalies. Contrary to the claims of PS, liquidity risk appears essentially unrelated to momentum.