New Research on Factor Investing in the Bond Market
Factor-driven investing, while highly popular among equity investors, has not been widely adopted in the bond market. But new research shows how to construct highly efficient fixed-income portfolios using factors, as well as the ongoing importance of reducing expenses.
Professors Eugene Fama and Ken French are best known for their 1992 study “The Cross-Section of Expected Stock Returns.” Based on the findings documented in that paper, they proposed that the single-factor (market beta) capital asset pricing model (CAPM) be replaced by a new three-factor model, adding size and value, as their addition significantly improved the ability to explain the differences in returns of diversified portfolios. Less well known, perhaps because bonds are less sexy than stocks, is that in their 1993 paper “Common Risk Factors in the Returns on Stocks and Bonds,” Fama and French also proposed a two-factor (term and default) model to explain bond returns.
Marlena Lee, co-head of research at Dimensional Fund Advisors, conducted perhaps the most rigorous performance study of bond managers in her 2009 study “Is There Skill among Bond Managers?” Lee employed modified Fama-French term and default factors in combination with market, size and value factors to risk adjust alphas. Lee’s sample included 2,353 active U.S. investment-grade, high-yield and government bond funds and covered the period January 1991 to December 2008. Following is a summary of her findings:
- In aggregate, active bond funds underperform appropriate benchmarks by an amount roughly equal to fees.
- All categories of funds (government, corporate, high yield) produce negative net alphas.
- There is no evidence of winner persistence in net returns beyond the randomly expected. We cannot separate skill from luck.
- Collectively, investors in active bond funds lose about 90 basis points per year, or about $1.4 billion in 2008, in underperformance – a triumph of hope over experience.
While the Fama-French three-factor equity model has long ago been replaced by first the Carhart four-factor model (adding momentum), and now by the competing newer models (the Q-factor model, which adds investment and profitability to market beta and size; and the Fama-French five-factor model, which adds investment and profitability to market beta, size and value), the Fama-French two-factor model has basically remained the workhorse model for bonds.
Recently, other factors have been proposed as adding explanatory power to bond returns. In their 2018 study “Style Investing in Fixed Income,” published in The Journal of Portfolio Management, Jordan Brooks, Diogo Palhares and Scott Richardson of AQR Capital Management identified four fixed income style premiums:
- Value: the tendency for relatively cheap assets to outperform relatively expensive assets.
- Momentum: the tendency for an asset’s recent performance to continue in the near future.
- Carry: the tendency for higher-yielding assets to outperform lower-yielding assets.
- Defensive (quality): the tendency of safer, lower-risk assets to deliver higher risk-adjusted returns than their low-quality, higher-risk counterparts.
They found that applying the four style premiums identified in other asset classes would have enhanced returns in various fixed income markets over the past two decades. They concluded: “Our empirical analysis suggests a powerful role for style-based investing in fixed income.” They added: “Style investing can be applied through long-only tilts or through long/short strategies. Both can make sense. Long/short strategies provide better diversification, but investor constraints and limited shorting ability/capacity may make the long-only path more realistic for many investors.”
Peter Mladina and Steven Germani contribute to the literature on asset pricing models and the bond market with their study “Bond-Market Risk Factors and Manager Performance,” which appeared in the September 2019 issue of The Journal of Portfolio Management. They began by noting that just as there has been a proliferation of equity factors, Brooks, Palhares and Richardson had shown that factors other than term and default helped explain returns of bond portfolios. However, they noted: “The objective in science is to find the minimum number of factors that will explain something, not the maximum amount. Additionally, factors with a sound theoretical foundation should be given priority over observed anomalies without a good theory.”
While there have been hundreds of equity factors identified in the literature, only a handful (four or perhaps five) are needed to explain the vast majority of the differences in returns of diversified portfolios. With that in mind, Mladina and Germani were motivated by the following questions. How efficient are bond markets? What compensated risk-factor exposures do bond investors actually bear, and what is the evidence for factor-adjusted alpha? How well do term and default factors explain the return and risk of traded bond portfolios? Can we improve upon the original Fama-French term and default factors, and are other common risk factors present in traded bond portfolios? Are some of the style premiums redundant with term and default, and do they show an effect in real-world portfolios?
To answer these questions, they conducted asset pricing tests on traded portfolios – bond mutual funds from 1983 to 2017. This allowed them to both evaluate factors and manager performance from the perspective of market efficiency. For the default factor, they used the difference in returns between high-yield bonds and Treasury bonds of similar duration. Their sample was broader than the one used by Marlena Lee in that they also included municipal and asset-backed funds, which represent approximately 25% of total assets in their sample in December 2017. Their final sample included 2,243 bond funds. Following is a summary of their findings:
- The 0.94% term and 0.12% default betas are highly statistically significant, but the 0.24% alpha is statistically insignificant (t-stat of 1.49) at the 5% confidence level. The R-squared value of the two-factor model was 0.95.
- The term and default factors are independent sources of systematic risk that are compensated on average.
- There is evidence of two-factor-adjusted alpha in only a small subset of active bond managers – about 9% compared to the 2.5% randomly expected. The excess could be explained by either skill, luck or other factors. Importantly, they found that the subset clustered in just the asset class of mortgage backed securities (MBS), which has unique prepayment risks. Relative to their three-factor model (adding prepayment), the% of active managers outperforming fell to less than 5%.
- The prepayment risk inherent in MBS is an independent risk factor, separate from term and default. However, the term factor contributes 77% to total MBS risk.
- The U.S. investment-grade bond market is almost perfectly explained by the three common risk factors of term, default and prepayment (many corporate bonds and municipal bonds have call provisions).
Importantly, they also found: “The default factor is redundant with the market factor of equities – it is merely a linear combination of the Fama-French market factor. The default factor has a 0.34 market beta that is highly statistically significant (t = 13.64), whereas its -0.45% alpha is insignificant (t = -0.34). In other words, the compensated returns of the default factor are replicated by holding 34% equities and the rest in T-bills. Despite this redundancy, we choose to retain the default factor because it captures more common variation in bond returns than does the market factor.”
Mladina and Germani concluded that their overall results indicated that the three-factor (term, default and prepayment) model “is the better asset pricing model and more consistent with the conditions of bond-market efficiency.” I would add that the finding that the default factor is well explained by the exposure to market beta, along with the very small default premium, calls into question the prudence of investing in publicly traded high-yield bonds.
Mladina and Germani also noted that in the period before the great financial crisis (GFC), they found fewer than 2% of funds with statistically significant positive alphas using the three-factor model. This percentage is less than might occur by chance, consistent with the findings of Marlena Lee’s study. However, in the post-GFC period, they found a much higher percentage of funds with statistically significant positive alphas (12.3%) that did not systematically cluster in any particular subcategory. “Perhaps this increase in alphas is a result of government and Federal Reserve interventions, but we leave that to future research.”
Examining the style factors
Mladina and Germani tested the style premiums to determine if they were truly independent factors, adding explanatory power. They found that the return to carry was well explained by the default factor, making it redundant. They also found that the defensive factor was well explained by the value and momentum factors, making it redundant. Value and momentum survived as independent factors. However, they also found that when adding these factors to their three-factor model, the explanatory power was basically unchanged, explaining only 0.08% more systematic risk. Since bonds (other than Treasury bonds) generally have higher trading costs than stocks, the hurdle to add value after implementation costs is higher for these factors that increase turnover. Given the very small increase in explanatory power provided by the styles, they concluded that their more parsimonious three-factor model was the superior choice.
The reported results have implications for investors in terms of portfolio construction, risk monitoring and manager selection. First, because the term, default and prepayment factors explain almost all the returns of bond portfolios, investors should construct their bond portfolios with these factors in mind and then carefully monitor their exposure to these systematic risks.
Second, because the findings show little evidence of persistent ability to generate alpha beyond the randomly expected, investors should choose their fixed income managers based mainly on expense ratios and exposure to term risk (the default premium was not only well explained by market beta, but it has also been very small, even before costs).
Because factor betas drive compensated return and risk in bonds, it is far more important to capture desired factor betas efficiently than to seek alpha.