Replicating Leveraged Small Value Equities in Japan

Scientific researchers are growing increasingly worried that most published research findings are false. In 2012, Amgen researchers were able to replicate only six out of 53 landmark cancer research studies. Amgen argued that bad science was leading to medical research that just didn’t work.

The problem is arguably worse in academic finance, where financial incentives are strong and a proliferation of data creates rife opportunity for data mining and over-fit studies. Duke researcher, Campbell Harvey, is at the forefront of a move to challenge the factor soup that academic finance research has become, arguing, “most claimed research findings in financial economics are likely false.”

We share this skepticism, and that’s why we place such a high importance on the academic rigor of our quantitative work. This is also why we think it’s so important to replicate our core findings out of sample. Replication in out-of-sample data is the gold standard for scientific proof, and we demand — and our investors deserve — the highest quality research to support the implementation of investment strategies.

Our out-of-sample test was conducted using Compustat data on Japanese equities from 1991 to 2014. Portfolios were formed on June 30 of each year, using accounting information that is as of March 31 or earlier. We are delighted to report that we were able to replicate the findings of our paper Leveraged Small Value Equities in the Japanese markets. We applied our ranking algorithm to our relevant universe of leveraged value stocks in Japan and found the following in terms of value-weighted returns:

Figure 1: Replicating Leveraged Small Value Equities Ranking Algorithm in Japan

Without using regression results, we applied the U.S. ranking algorithm in Japan and it sorted stocks effectively in this out-of-sample market. The best ranked stocks had the highest returns. Lowest ranked stocks had the lowest returns. Quartile one far outperformed quartile four. And our algorithm produced far better results than the Fama-French Small Value Benchmark. Note that our algorithm selected a much smaller sample, roughly 10% of the small value benchmark.

Below, we show a graph comparing the returns of the value-weighted top quartile of our ranking algorithm with the Nikkei 225 Index. The Verdad time series had an average annual return of 15.7% and a CAGR over the full time period of 12.4% relative to the Nikkei 225, which was essentially flat over the time period.

Figure 2: YoY Performance and Cumulative Performance of Verdad vs. Nikkei 225

The two time series are 75% correlated, and the Verdad strategy generates an alpha of 13.7% per annum with a beta of 0.83.

Below we show a regression of the key factors underlying our ranking algorithm.

Figure 3: Regression of Key Factors vs. Returns in Japan 1991–2014


In figure 3, we regress the Next 1 Year Return of universe stocks on the factors in our model and we control for time using a trend by portfolio year. The t-statistic of each coefficient is shown in parentheses.
 

The first seven factors listed above, including Prior Year Debt Paydown, are statistically significant at the 5% level. Comparing this with the U.S. regression, there are five statistically-significant factors in common:  Prior Year Debt Paydown , Debt/EV, Asset Turnover, Prior Year Return below Median and the interaction of EBITDA/EV and Debt/EV. Interestingly, these factors had significantly higher predictive power in Japan than in the United States.

Size, which was a powerful factor in the U.S. regressions, did not show up as an important variable in Japan. This is in line with other studies that have failed to find a size premium in international markets.
In short, the strategy of focusing on leveraged small value equities has even more statistical power in Japan than it does in the United States, generating even more significant alpha relative to broader benchmarks (13% outperformance of the Nikkei 225 vs. 10% outperformance of the S&P500).
Our work shows not only that Japan is an ideal market for investing in leveraged small value stocks, but that our model works out of sample with strong predictive power. Our analysis is replicable.

If you’re investing in quantitative managers whose core models can’t be replicated across markets, you’re taking risk you shouldn’t be taking. If you’re investing in fundamental managers with a non-systematic approach, you’re betting on a black box. Investors who pay for active management deserve the best: systematic, replicable models that generate significant alpha in an intellectually elegant but operationally simple fashion.