Quality of PIN Estimates and the PIN-Return Relationship

We estimate the probability of informed trading (PIN) for nearly 80,000 stock-quarters between 1993 and 2004 by using two methods to classify buyer-initiated and seller-initiated trades. The first method is the Lee and Ready (1991) algorithm with adjustment for 5-second delay in trade time. The second method is the Ellis, Michaely, and O’Hara (2000) algorithm without time adjustment. The second set of PIN estimates appears to be better than the first set, especially in the years since 2001. More importantly, the Fama-MacBeth cross-sectional regressions show that, for the period between April 1993 and March 2005 inclusively, after excluding the January returns and controlling for beta, size, and book-to-market-equity ratio, monthly stock returns are significantly positively related to the second set of PIN estimates, whereas this is not true for the first set of estimates. Our results suggest that the positive PIN-return relationship discovered by Easley, Hvidkjaer, and O’Hara (2002) is stronger when high-quality PIN estimates are used