Does Factor Investing Work in Bonds?
There are now hundreds of funds devoted to factor investing in equities. Whole papers are devoted simply to processing the “factor zoo” of competing ideas for quantitative equity investing.
But corporate credit is different. The number of factor-focused bond funds—or even research papers on factor investing in bonds—can be counted on one hand. While equity researchers and investors have relied on the Compustat-CRSP merged database for decades to run studies on equities, no equivalent database even exists for bonds.
Our collaborator on this series, Greg Obenshain, has built just such a database to test whether the insights from factor investing in equities can be applied in corporate credit. Does factor investing work in bonds?
In this piece, we use Greg’s database to evaluate how three very simple and well-established quantitative factors work in bonds and how they might be combined to produce an effective multi-factor model. We look at value, momentum, and quality before diving into how to combine those factors—and use leverage—to test a factor approach to bond investing.
Value
As we established in our piece on Fool’s Yield, simply buying the highest yielding bonds is not the right strategy for investing in corporate bonds. Yields are too closely related to default rates. Rather, investors should focus on bonds in the universe of B1 to BB1, where returns are the highest and default rates are low.
But what about looking within these ratings categories and selecting the bonds with the highest yield within their category? We call this metric “yield-to-rating.”
Figure 1: Return & Risk Characteristics of Corporate Bonds by Yield-to-Rating
Source: Proprietary database. Data from 1997 to April 2019. B1-BB1 market-implied rating universe.
As you can see from Figure 1, yield-to-rating does improve total returns (unlike fool’s yield, where returns go down at higher yields) but at the cost of higher drawdowns and lower Sharpe Ratios. This is a clear risk-return trade-off: investing based on yield-to-rating does not generate alpha.
Momentum
As with value, we need to make some adjustments to momentum to make it work for debt. Rather than just look at return momentum, we look at the change in the relative pricing of bonds over the previous three months. We rank bonds by whether their yields have tightened or widened relative to bonds within the same credit rating category.
Figure 2: Return & Risk Characteristics of Corporate Bonds by 3-Month Momentum
Source: Proprietary database. Data from 1997 to April 2019. B1-BB1 market-implied rating universe.
Applying a simple momentum screen also effectively ranks bonds by future total return. Unlike with yield-to-rating, this does not necessarily come with increased risk. The risk metrics do not go up linearly as momentum increases.
Quality
We have written extensively about Stanford professor Joseph Piotroski’s work on accounting quality and how combining his quality metrics with value investing produces better returns at lower risk. We rank bonds by their Piotroski F-Score below.
Figure 3: Return & Risk Characteristics of Corporate Bonds by Piotroski F-Score
Source: Proprietary database. Data from 1997 to April 2019. B1-BB1 market-implied rating universe.
Piotroski’s F-Score works miracles in corporate debt: returns increase linearly while volatility and drawdowns decrease linearly as you move toward the top of the screen. This is a pure alpha driver and the best of these three signals. This makes theoretical sense, since a company’s accounting quality should be a good future predictor of credit upgrades or downgrades and future defaults.
Combining the Factors
These factors are not perfectly correlated. The quality score has a negative correlation to value, while the momentum score has low correlations to both value and quality. Combining these three approaches should therefore produce better results. And indeed it does: even the simplest approach of taking an equal-weight average of the three scores produces excellent results.
Figure 4: Return & Risk Characteristics of Corporate Bonds by Equal-Weighted Blended Rank
Source: Proprietary database. Data from 1997 to April 2019. B1-BB1 market-implied rating universe.
This blended ranking model works better than any of the three individual signals while also producing an effective linear ranking by return/risk and drawdowns. And this is just equal weighting three relatively basic quantitative factors, this model could likely be improved further with machine learning and additional factors.
How does this compare to equity and high-yield benchmarks? We backtested this strategy, comparing the historic returns of the top quintile on this blended ranking model to the broader high-yield index (as measured by Vanguard’s high-yield bond fund) and to equities (as measured by Vanguard’s Total Stock Market Index Fund).
Figure 5: Blended Model Returns vs. High Yield and Equities
Source: Proprietary Database, Capital IQ. Data from 1997 to April 2019. B1-BB1 market-implied rating universe.
Factor investing in high-quality high-yield corporate debt is an attractive alternative to reaching for yield. The blended model produced higher returns and lower drawdowns than the high-yield index, resulting in a superior Sharpe Ratio.
Today, this blended model produces a portfolio with a current yield of 5.1%. Historically, the blended model has outperformed its yield by 1.22x, which would imply that buying this portfolio today would return about 6.2%.
But many investors have return hurdles that are higher than the expected return of this strategy. Because this strategy has a high Sharpe Ratio and limited drawdowns—lower than both high yield and equities—it is ripe for leverage. Adding margin leverage to this portfolio could enhance the strategy’s portfolio benefits and its attractiveness to investors looking for high returns.
Below, we show a backtest of this strategy at low, medium and high leverage (1.2x, 1.6x, and 2.0x). This analysis is shown inclusive of the costs of leverage, which is why, for example, the 2x leveraged portfolio does not produce 2x the returns. We compare the results to high-yield, equities, and an unleveraged version of our model run on lower-rated B-CCC bonds.
Figure 6: Returns at Low, Medium, and High Leverage
Source: Proprietary Database, Capital IQ. Data from 1997 to April 2019. B1-BB1 market-implied rating universe.
Leveraging the blended model produces portfolios with expected returns in the 10–14% range, with results that have superior Sharpe ratios to running this model unleveraged on lower-rated credit.
The superior results of leveraging lower-risk, higher-quality bonds will not come as a surprise to quants. Cliff Asness of AQR wrote, “If some investors are averse to leverage, low-beta assets will offer higher risk-adjusted returns, and high-beta assets will offer lower risk-adjusted returns.” Ray Dalio and his colleagues at Bridgewater argue in their piece “The Biggest Mistake in Investing” that over-investment in equities at the expense of bonds is the greatest mistake in investing and that using leverage to enhance the returns of fixed income is the solution. “Many people still confuse leverage with risk, but the reality is that levering up low-risk assets so you can diversify away from risky investments is risk reducing.”
Conclusion
We set out to test whether factor investing works in bonds. We adapted some of the workhorse factors from factor investing in stocks and applied them in a simple way to corporate bonds. We found significant alpha. Much more is possible, and these results are just the tip of the iceberg.
Though this work would be obvious in stocks, in bonds it is cutting edge. Today, there are almost no funds doing quantitative investing in corporate bonds, even though the data suggest that factor investing works very well in this market. There is no DFA equivalent for corporate bonds—a significant untapped opportunity set for those who believe in the promise of quantitative approaches.