Positive return premia in Japan

Abstract

This paper examines Jensen's [J. Finance, 1968, 23, 389–416] alphas and the time-varying return premia unexplained by standard risk factors in Japan and presents several new findings. First, in contrast to the US experience, positive alphas remain after Fama and French's three factors are applied to excess stock returns in Japan. Second, positive alphas remain in Japan, even if the Fama–French three factors combined with momentum and reversal factors are applied to excess stock returns. Third, the positive return premia unexplained by these five factors bear little relation to the dynamics of the Japanese macroeconomy. Fourth, the time series evolution of the positive return premia indicates autonomous dynamics with at least three regimes. Fifth, we can predict or time the acquisition of the positive return premia for small-size portfolios in Japan by observing the direction and effect of the return premia of large-size portfolios and high-book equity to market equity (BE/ME) portfolios. Finally, application of the self-exciting threshold autoregressive (SETAR) model shows that the size effects are stronger than the BE/ME effects in Japan, given that the return premia from small-size portfolios in the SETAR model are bounded by positive thresholds, while the return premia from high-BE/ME portfolios are bounded by negative thresholds.

Keywords:

• Fama–French model
• Jensen's alpha
• Markov switching model
• Momentum
• Reversal
• SETAR model

Acknowledgements

The authors acknowledge the generous financial assistance of the Japan Society for the Promotion of Science, the Zengin Foundation for Studies on Economics and Finance, and the Nihon Housei Gakkai. They would also like to thank Michael Brennan, Jason McQueen, Ike Mathur, Giorgio Szego, Geert Bekaert, Nick Wade, and Takao Kobayashi for their useful suggestions. The authors are also grateful to the editors of this journal for their professionalism and for encouraging the revision of an earlier version of this paper, and two anonymous referees for their careful, constructive, and helpful comments.

Notes

†Other work using alphas in the performance evaluation of mutual funds includes Brown et al. (1992), Grinblatt et al. (1995), Ferson and Schadt (1996), Gruber (1996), Carhart (1997), Chevalier and Ellison (1997), Daniel et al. (1997), Christopherson et al. (1998), Chen et al. (2000), Carhart et al. (2002), Pastor and Stambaugh (2002a, b), Cohen et al. (2005) and Jones and Shanken (2005).

†In terms of existing international work, Cai et al. (1997) analysed the performance of Japanese mutual stock funds using Jensen's alpha. Korkie and Nakamura (1997) also investigated the effects of capital market liberalization on Japanese equity markets using Jensen's alpha. However, none of these studies undertook the analysis presented in this paper.

‡We believe that the differences in the economic structures of the US and Japan are self-evident. However, we can readily illustrate some differences in their financial markets. To start with, Hamao (1988) points out that, in Japan, bond markets are less developed than stock markets. As a result, we consider that the relations between equity and bond markets and the macroeconomy in Japan differ from those in the US. Furthermore, Tsuji (2005) revealed that term spreads in Japan have almost no forecastability for real GDP growth, unlike experience in the US. In addition, Tsuji (2007) found that the macroeconomic factors priced in Japanese stock markets differ from the US. Hence, we naturally conjecture that the differences in financial markets between the US and Japan will produce divergence in the risk factors that explain excess stock returns in Japan.

§For example, Brealey et al. (2006) argue that the value of financial claims, household financial asset allocation, and the portfolio allocations of financial institutions and non-financial corporations differ greatly between the US and Japan.

¶In Japan, the Financial Services Agency recently recognized the importance of financial education for senior and junior high school students with the publication in June 2005 of a short report. This report details the differences and deficiencies in financial education in Japanese senior and junior high schools in comparison with those in the US and the UK.

⊥Hoshi and Kashyap (2004) point out that differences in the financial systems of Japan and the US can be seen in: (1) household asset holdings, (2) corporate financing, (3) banking services, and (4) corporate governance. Furthermore, Becht et al. (2003) conclude that, broadly speaking, there are two different systems of corporate governance: an Anglo–American market-based system and a long-term large investor model as found in, say, Germany and Japan. We argue that these differences will directly or indirectly produce different stock return structures in the US and Japan.

†They suggested that weak firms with persistently lower earnings tend to have higher BE/ME and positive slopes on HML, while stronger firms with persistently higher earnings have lower BE/ME and negative slopes on HML. They also explain that the motivation to use HML to explain returns is given by the evidence of Chan and Chen (1991) of the covariation in returns relating to relative distress that is uncaptured by the market return and compensated for in average returns.

‡We thank Kenneth French and Eugene Fama for permitting us to use their data.

§For the purposes of comparison and because of data availability, the analysed sample horizons are from October 1981 to April 2005 in tables 1, 2, 3, 4, 10, and 11 and figures 2 and 3, October 1981 to July 2004 in tables 5 and 6, January 1982 to December 2004 in table 7, the fourth quarter of 1981 to the first quarter of 2005 in tables 8 and 9 and figure 1, and from October 1981 to December 2004 in table 12.

¶Data frequency is generally important in analysing asset returns, as shown in many existing studies. For example, see the interesting work of Cont (2001).

†Newey–West (1987) heteroskedasticity and autocorrelation consistent covariance matrixes are used to obtain the t-values. Although not stated, when estimating regressions (3) and (4), we always apply the Newey–West (1987) covariance matrix.

‡We again note that because of limits in data availability, this is the longest period over which we can replicate the analysis in Japan.

§To parallel the analysis of Fama and French (1996), we analyse the case in Japan on a monthly basis in tables 2–6.

†The average values of 0.68 and 0.79% are calculated using all of the alphas in tables 6 and 5, respectively.

‡In terms of research on stock markets and macroeconomic factors, many studies follow Chen et al.'s (1986) representative study. For more recent work, see Avramov (2002, 2004), Jagannathan and Wang (2002), Campbell and Taksler (2003), Conrad et al. (2003), Goyal and Santa-Clara (2003), Vassalou and Xing (2004), Avramov and Chordia (2006), and Bali (2008), among others.

†In a related study, Hwang and Salmon (2004) construct a monthly time-varying beta by applying the Fama–French three-factor model to daily return data. We consider this an interesting future extension to the work in this paper.

‡Although our time series return premia as defined are not strictly the time series version of Jensen's alpha, our interest lies in clarifying the characteristics of this additional return source. This kind of analysis—that is, investigating any unexplained parts including the residuals left by factors or models—is often undertaken in financial research, as shown by McElroy and Burmeister (1988), Eom et al. (2004), Petkova (2006), and Brennan and Wang (2007). Furthermore, as generally recognized, the five factors in model (4) do not strongly capture the macroeconomic dynamics. Hence, there is the possibility that our time series return premia are related to the macroeconomic dynamics, and again, as our return premia provide an important additional source of profitable returns, it is therefore meaningful to clarify the relation between our return premia and real GDP growth in Japan. We thank an anonymous referee for commenting on this matter.

§For details of the Markov switching model, see, for example, Hamilton (1989, 1990). Related representative studies using this approach include Turner et al. (1989), McQueen and Thorlay (1991), Hamilton and Susmel (1994), and Gray (1996), among others.

†After examining the statistical significance of the model parameters, we determined order 1 in model (5) for the four kinds of return premia, and order 2 in model (6) for real GDP growth.

†We also tested all of the seven macroeconomic factors of Chen et al. (1986), namely industrial production, unanticipated inflation changes, expected inflation changes, credit spreads, term spreads, consumption growth, and oil price changes. However, these macroeconomic factors little explained the positive return premia in Japan. In the interests of brevity, the results are available from the authors upon request.

‡In this study, our interest lies much more in the quantitative modeling of the positive return premia in Japan than in the search for the fundamental firm characteristics that might help explain the return premia.

§These models were first introduced by Tong (1978, 1983, 1990), and are based on a simple relaxation of standard linear autoregressive models. Recent studies applying this model include Teräsvirta (1994), Tiao and Tsay (1994), Clements and Smith (1997), van Dijk and Franses (1999), Gallagher and Taylor (2001), and Taylor and Taylor (2004), among others.

†As with the Markov switching model (5) of the four kinds of return premia, the lag order is determined to be 1 for the SETAR model after considering the statistical significance of the model parameters.

‡We determined the fifth order of the VAR model for uniformity.

§Why then do big-size (high-BE/ME) portfolios' return premia lead small-size portfolios' return premia? We suggest that this is because of differences in the recognition of investors. More generally, there is relatively little information on small companies, and investors' recognition of small companies is relatively weak (Hou and Moskowitz 2005). Hence, the response of small stocks to important information is delayed. Consequently, small-size portfolios' return premia lag big-size (high-BE/ME) portfolios' return premia.