Time Instability of the Fama-French Multifactor Models: An International Evidence

Koichiro Moriya,Akihiko Noda

2024-06-01

Abstract:This paper investigates the time-varying structure of Fama and French's (1993; 2015) multi-factor models using Fama and MacBeth's (1973) two-step estimation based on the rolling window method. In particular, we employ the generalized GRS statistics proposed by Kamstra and Shi (2024) to examine whether the validity of the risk factors (or factor redundancy) in the FF3 and FF5 models remains stable over time, and investigate whether the manner of portfolio sorting affects the time stability of the validity of the risk factors. In addition, we examine whether the similar results are obtained even when we use different datasets by country and region. First, we find that the effectiveness of factors in the FF3 and FF5 models is not stable over time in all countries. Second, the effectiveness of factors is also affected by the manner of portfolio sorting. Third, the validity of the FF3, FF5, and their nested models do not remain stable over time except for Japan. This suggests that the efficient market hypothesis is supported in the Japanese stock market. Finally, the factor redundancy varies over time and is affected by the manner of portfolio sorting mainly in the U.S. and Europe.

Statistical Finance,General Economics,Pricing of Securities

What problem does this paper attempt to address?

This paper explores the time stability of the Fama-French multi-factor model, particularly in the international market. The study uses a rolling window approach and Fama and MacBeth's two-step estimation method, as well as the generalized GRS statistic suggested by Kamstra and Shi, to test the effectiveness (or redundancy) of the risk factors over time. The paper finds that the effectiveness of the FF3 and FF5 model factors is not stable in any country, and the ranking method affects the time stability of these factors. Additionally, except for the Japanese market, the effectiveness of the FF3, FF5, and their nested models is also unstable in other markets, indicating support for the efficient market hypothesis in the Japanese stock market. In the US and Europe, the factor redundancy varies with time and ranking method. The study also highlights that the effectiveness of the FF3 and FF5 models may differ due to different sample periods and ranking methods.