Robust regression, which has been very rarely used to determine factor significance and premia, is used to revisit the Fama and French 1992 least squares analysis of the Size, BM, Beta and EP factors. We do so using an optimal bias robust regression estimator that is not much influenced by outliers. The robust Fama-MacBeth regressions for these factors show that small fractions of outliers in the range 2% to 5%, depending on the factor, market capitalization group, and time interval, have substantial adverse influence on least squares based factor conclusions, and reverse the Fama and French 1992 conclusions for Size, Beta, and EP. Specifically, our robust regressions for 1960–1993 and 1980–2018 show that: Size has a significant positive premia, Beta has a significant negative premia, and EP is significant jointly with Size. Further robust regression analysis reveals that BM was no longer a value factor during 2007 to 2018.
R. Douglas Martin (Fri,) studied this question.