The Development of Mean-Variance-Efficient Portfolios in Japan and the U.S.

Abstract

Stock selection models have been, and can be, effectively employed in Japan to deliver excess returns. In 1992, in the initial year of this Journal’s publication, Guerard and Takano (1992) reported mean-variance efficient portfolios for the Japanese and U.S. equity markets, and show that the use of a regression-weighted composite model of earnings, book value, cash flow, sales, and their relative variables outperformed their respective equity benchmarks by approximately 400 basis points annually. Markowitz and Xu (1994) tested the composite model strategy and found that its excess returns were statistically significant from a variety of models tested, and the composite model strategy was not the result of data mining. Guerard (2006) updated the Guerard and Takano stock selection models are updated through 2003 using a similar Japanese-only database. The sophisticated regression model continued to produce the highest information coefficients in Japan and the US. Guerard included a consensus analyst derived forecast, revisions, and breadth variable, and show its contribution to stock selection. Several Japanese-only analyst forecasting model results are similar to US-only results. The use of the Global Compustat database allowed the construction of global variables that produce similar stock selection models. The effectiveness of these quantitative models had not lessened during the 1992-2003 period. In this update to celebrate 25 years of the Journal of Investing, the author uses a commercially available global database, FactSet, for the 2002-June 2016 time period to address stock selection composite models, mean-variance efficient portfolios in Japan and the U.S., and reports three results: (1) the original stock selection continues to be effective in Japan and the U.S. in 2002-6/2016; (2) the mean-variance efficient portfolios outperformed in Japan and the U.S. in 2002-6/2016; and (3) the Guerard and Takano stock selection model is not the result of data mining. The author reports three additional results: (1) the weighted latent root regression Guerard and Takano can be enhanced by using the Tukey optimal influence function; (2) the mean-variance efficient portfolios failed in Japan and the U.S. in 1996-2000 with the PCAP and Global Compustat databases and 2006-2007, 2012-2013, and YTD 2016 with the FactSet databases, but work in the vast majority of the years; and (3) the Guerard and Takano portfolios outperformed in 14 of 17 years in backtest, 82.4%, of the yeas whereas the Japan model portfolios outperformed in 9 of 14.5 years, or 66.7%, of the years, post-publication and the U.S. model outperformed in 9.5 of 13.5 years, or 70%, post-publication, respectively!