In this paper we analyze the theoretical implications of sorting data into groups and then running asset pricing tests within each group. We show that the way this procedure is implemented introduces a bias in favor of rejecting the model under consideration. By simply picking enough groups to sort into, the true asset pricing model can be shown to have no explanatory power within each group.

Exchange seat prices are widely reported and followed as measures of market sentiment. This paper analyzes the information content of NYSE seat prices using: (1) annual seat prices from 1869 to 1998, and (2) the complete record of trades, bids and offers for the seat market from 1973 to 1994. Seat market volumes have predictive power regarding future stock market returns, consistent with a model where seat market activity is a proxy for unobserved factors affecting expected returns. We find abnormally large price movements in seats prior to October 1987, consistent with the hypothesis that seat prices capture market sentiment.