Value, Momentum, Carry, and Defensive Signals for Corporate Bonds

Book: Systematic Fixed Income: An Investor’s Guide Author: Scott A. Richardson, Ph.D. ISBN: 9781119900139 Publisher: John Wiley & Sons, 2022

This is the second part of Chapter 6, covering the actual investment signals for corporate bonds and what happens when you put them together. If the first part was the setup, this is the payoff.

The Four Investment Themes

Momentum

Two measures here. First, “own” momentum: the trailing six-month credit excess return of the bond. Companies with better recent credit performance are expected to keep outperforming. Second, “related” momentum: the trailing six-month stock return of the bond issuer. This cross-market signal was first documented by Kwan back in 1996.

One interesting detail: unlike equity momentum, you don’t skip the most recent month for the equity signal when applying it to credit markets. There’s no evidence of short-term cross-market reversal, and recent equity returns are actually more predictive for credit.

Of course, equity momentum only works for public companies. Private issuers get left out. But the broader point is that momentum in credit can extend to many other measures: changes in default forecasts, supply-chain linkages between firms, and various fundamental indicators.

Carry

Same concept as government bonds, but applied to credit. Carry is the return you get if nothing changes but time passes. The simple measure here is the option-adjusted spread (OAS) of the corporate bond. Higher OAS means more carry. It’s not a perfect measure (a full carry calculation would need the entire credit curve for each issuer), but it’s straightforward and widely available.

Defensive

This theme has two parts:

Low risk: Use spread duration as the measure. Shorter-duration bonds have historically delivered better risk-adjusted returns across multiple asset classes. Note that some people use DTS (Duration Times Spread) as a risk measure, but Richardson argues this mixes two conflicting ideas. Duration is your low-risk signal, and spread is your carry signal. Keep them separate.

Quality: Prefer issuers with lower market leverage and higher gross profitability. There’s a vast literature on financial statement analysis that feeds into quality measures. Default probability could also work here as an unconditional quality signal (as opposed to its conditional use in the value signal where it’s compared against spreads).

Performance Results: Almost Too Good

Using a combined US IG and HY universe from 1997 to 2020, Richardson builds simple academic portfolios: long the most attractive 20% of issuers, short the least attractive 20%, market-cap weighted. These are explicitly academic. No liquidity constraints, no transaction costs, no position limits.

The results:

Value stands out with a Sharpe of 1.71. The equally weighted combination beats any individual signal because the themes have low pairwise correlations.

Here’s what really matters: after controlling for traditional risk premiums (credit, term, equity) and equity factor returns (size, value, momentum, quality, betting-against-beta), the value signal has an information ratio of 1.99 and is not significantly correlated with any of these factors.

And credit value is uncorrelated with equity value (HML). This might seem weird since both are about finding “cheap” assets tied to companies. But the signals measure fundamentally different things. Equity value is about price relative to book value and long-term earnings growth. Credit value is about spreads relative to default risk. Different claims, different drivers.

The combination portfolio’s regression intercept has a t-statistic of 12.7. Richardson himself notes this is “too good to be true” for an academic exercise, but it shows the potential if you could implement it.

IG and HY Separately: Even Better

When you split the universe into IG and HY and use improved portfolio construction (sector demeaning, beta-neutralizing), the results look different but still strong.

Momentum works better in HY. Carry works better in IG. This makes sense: riskier bonds have more volatile spreads with stronger momentum effects, while safer bonds trade closer to par with more predictable spread mean-reversion.

The adjusted R-squared for the factor regression drops to 12-26% for HY and 8-19% for IG. Much lower than the combined universe. This means the systematic signals are genuinely diversifying and not just repackaging known risk premiums.

The Liquidity Myth

People often ask about harvesting a “liquidity premium” in credit markets. The theory sounds good: hold less liquid bonds, earn a premium for bearing that risk.

Palhares and Richardson (2019) tested this with six different liquidity measures: issue size, bid-ask spreads, market impact, daily trading volume, percentage of no-trading days, and bond age. They looked within the same issuer to control for credit risk, comparing the least liquid bond to the most liquid bond from the same company.

The result on spreads: there is a small positive spread difference for less liquid bonds. So liquidity is priced, at least a little.

But the result on returns: basically nothing. No statistically significant evidence that holding less liquid corporate bonds generates positive risk-adjusted returns. And that’s in an academic setting with zero transaction costs. In the real world, where you have to actually source these illiquid bonds and pay wider bid-ask spreads, the picture would be even worse.

This is a big deal for investors. If you can’t earn a liquidity premium even in theory, chasing illiquid bonds in practice is probably a losing game. And holding illiquid bonds in a fund with frequent redemptions creates additional risk that appears uncompensated.

Machine Learning: Cautious Optimism

Richardson is refreshingly honest about machine learning. He’s “no expert” but has “used them successfully.” His main concern: you need a lot of data, and corporate bond trading data is not abundant.

Where ML might actually help is in default forecasting. The default experience of thousands of companies over 50+ years gives you a large enough dataset. Correia, Kang, and Richardson (2018) used random forests (a type of decision tree ensemble) to forecast bankruptcy.

The example of a pruned decision tree is illuminating. It starts with all firms and splits on the variable that best separates defaulters from survivors. The first split is on leverage (enterprise value over total obligations). The second is on volatility. This data-driven approach naturally discovers the leverage-volatility interaction that’s central to the Merton structural model. But it can also find further interactions that linear models would miss.

The risks are real though. Data mining is always a concern with flexible models. Cross-validation (train on one dataset, test on another) and randomization across many trees help, but you need to be disciplined.

IG vs. HY: Don’t Segment Unnecessarily

Richardson dislikes unnecessary market segmentation. IG and HY are treated as completely separate asset classes by most institutional investors. This reduces your investment opportunity set.

If you can, be downgrade tolerant. Let bonds cross the IG/HY boundary without forced selling. Even if you can’t fully combine IG and HY, consider allowing spillover into neighboring rating buckets. This avoids the bad-selling practices we discussed in Chapter 4 and expands your security selection opportunities.

Communication Matters

One last point that surprised me. Richardson emphasizes the importance of explaining your investment process to clients. Systematic managers can’t tell a story about each bond in the portfolio. But they can build visual tools that show how valuation measures drive positioning and how different themes contributed to performance.

This matters because poor communication can accelerate redemption risk even when performance is fine. It’s not enough to have good returns. You need your investors to understand what you’re doing and why.

The evidence across both government and corporate bonds is compelling. Systematic investment approaches using value, momentum, carry, and defensive signals can produce attractive, diversifying returns. The challenge now shifts to actually implementing these ideas in real portfolios with real liquidity constraints. That’s what the later chapters are about.

Previous: Corporate Bond Investing 101 | Next: Emerging Market Bonds in Hard Currency