The Consistency of Size Effect: Time Periods, Regression Methods, and Database Selection

Abstract

We try to reconcile the findings of prior size effect studies by re-examining the issue with different time periods, regression methods, and database selection. We test whether the size effect varies in relation to the time period, whether extreme observations cause the size effect, by experimenting with different regression methods, and whether the survivorship bias in the COMPUSTAT database induces the size effect. It is found that the size effect is highly significant for data from the earlier time period, but its significance is noticeably reduced for the later time period. Extreme returns cannot fully account for the size effect, because the effect remains strong in the earlier time period even when the extreme observations are trimmed. Finally, we do not find any evidence indicating that the survivorship bias in the COMPUSTAT database is responsible for the size effect.