Priced risk in corporate bonds

This paper challenges recent findings supporting multifactor models by demonstrating that, with the sole exception of traded liquidity, proposed bond risk factors lack incremental explanatory power over the corporate bond market factor, thereby suggesting that the bond CAPM remains robust against competing factor models.

The Gist

Imagine the world of corporate bonds as a massive, bustling marketplace where companies borrow money from investors. For decades, financial experts have been trying to build a "GPS" for this market—a set of rules (or factors) that can predict which bonds will be winners and which will be losers, and why.

Recently, a popular group of researchers (Bai, Bali, and Wen, or "BBW") claimed to have found a new, super-accurate GPS. They said that to understand bond prices, you need to look at four specific things: the general market, "downside" risk (fear of crashes), credit risk (how likely a company is to go bust), and liquidity risk (how hard it is to sell the bond quickly). They argued that just looking at the general market wasn't enough; you needed these extra four ingredients to get the recipe right.

This paper is like a group of skeptical mechanics who decided to take that "super-accurate GPS" apart and check the engine.

Here is the story of what they found, explained simply:

1. The "Fake" Data Problem (The Broken Compass)

The first thing the authors noticed was that the data BBW used was glitchy. It was like trying to navigate a car using a map that was printed with the dates wrong.

• The Glitch: For most of the time period they studied, the "Downside" and "Credit" risk factors were actually showing you what happened next month instead of what happened this month. This is called "look-ahead bias." It's like a weather app telling you it's going to rain tomorrow, but you're using that info to decide whether to bring an umbrella today.
• The Fix: When the authors fixed the dates and rebuilt the factors correctly, the "magic" disappeared. The factors that looked so powerful suddenly looked much weaker and noisier.

2. The "One-Size-Fits-All" vs. The "Special Sauce"

The authors asked a simple question: Do we really need these four special ingredients, or is the "General Market" factor enough?

• The Analogy: Imagine you are trying to predict the weather.
  • The Old Way (BBW): You need a thermometer, a barometer, a humidity sensor, a wind gauge, and a crystal ball.
  • The New Way (This Paper): The authors say, "Actually, if you just look at the general temperature trend (the Bond Market Factor), you can predict the weather just as well as the fancy multi-sensor kit."
• The Result: When they tested it, adding the "Downside," "Credit," and "Liquidity" factors didn't actually help predict returns any better than just watching the overall bond market. It was like adding extra spices to a stew that was already perfectly seasoned; it just made the pot messier without making the food taste better.

3. The "Liquidity" Exception (The One Spice That Works)

There was one tiny exception. The authors found that Liquidity Risk (how hard it is to sell a bond) did seem to have a little bit of extra power.

• The Metaphor: Think of liquidity like the "ease of exit" from a party. If you can leave a party easily, you feel safer. If you are stuck, you pay a "risk premium" (you want more money to stay).
• The authors found that this specific factor does matter, but only slightly. However, for everything else (credit risk, downside risk, macro uncertainty), the "extra" factors were basically useless noise.

4. The "Spurious" Factors (The Ghost in the Machine)

The paper also looked at other complex models that tried to use things like "macroeconomic uncertainty" or "long-term consumption risk."

• The Analogy: This is like a detective trying to solve a crime by blaming the butler, the gardener, and the cook, when the evidence shows it was just the wind blowing the door open.
• The authors showed that many of these fancy factors were "spurious." They looked like they were doing something important, but it was just a statistical illusion caused by bad data or the way the math was set up. When you corrected for these errors, the factors vanished into thin air.

5. The Big Conclusion: Keep It Simple

The main takeaway is a bit of a reality check for the financial world.

• The Verdict: The complex, multi-factor models that everyone was excited about are likely over-engineered. They are like a Swiss Army knife with 50 tools when you only need a screwdriver.
• The "Bond CAPM": The simplest model—just looking at how the whole bond market moves (the Bond CAPM)—is actually the most reliable. It explains almost everything the fancy models claim to explain, without the headache of trying to calculate complex, error-prone extra factors.

Why Does This Matter?

If you are an investor or a fund manager, this paper is telling you: "Stop chasing the shiny new, complicated models."

• They are likely built on shaky data.
• They don't actually help you make more money.
• They might even cost you money because trading these complex strategies involves high fees and transaction costs (bonds are hard to trade, unlike stocks).

In short: The authors peeled back the layers of the "fancy" bond pricing models and found that underneath, the market is much simpler than we thought. The "Bond Market Factor" is the king, and the other factors are mostly just court jesters trying to look important.

Technical Summary

1. Problem Statement

The paper addresses the ongoing debate regarding the existence and pricing of common risk factors in the corporate bond market. While recent literature (notably Bai, Bali, and Wen (2019), henceforth BBW) claims that multifactor models (incorporating downside, credit, and liquidity risks) significantly outperform the single-factor Capital Asset Pricing Model (CAPM) in explaining the cross-section of corporate bond returns, the authors argue that these findings are flawed.

The core problem is that empirical asset pricing in corporate bonds suffers from:

• Data construction errors: Potential look-ahead biases and lead/lag errors in factor construction.
• Model misspecification: The use of Ordinary Least Squares (OLS) in cross-sectional regressions, which can yield high $R^2$ values even when the fundamental pricing relation is violated.
• Identification issues: Weak factors and small time-series samples relative to the number of test assets.
• Over-reliance on ad-hoc factors: The search for factors based on empirical relevance rather than theoretical grounding.

The authors aim to rigorously test whether factors beyond the bond market factor (MKTB) possess any incremental explanatory power for corporate bond returns.

2. Methodology

The authors employ a comprehensive suite of statistical tools, moving beyond standard OLS regressions to robust inference methods.

A. Data Construction and Replication

• Data Source: Enhanced TRACE (intraday prices/volumes) and Mergent FISD (bond characteristics) for the period July 2002 – December 2016.
• Replication of BBW Factors: The authors strictly replicate the BBW four-factor model (MKTB, Downside Risk Factor [DRF], Credit Risk Factor [CRF], Liquidity Risk Factor [LRF]).
• Critical Discovery: They identify lead/lag errors in the original BBW factors:
  • DRF and CRF: Suffer from a one-month look-ahead bias (returns for month $t$ actually reflect month $t+1$) for the majority of the sample (2004–2014).
  • LRF: Suffers from a one-month lag error in the final two years (2015–2016).
  • Correction: The authors construct "replicated" factors free of these errors and use these for the primary analysis.

B. Econometric Framework

The paper utilizes three levels of analysis:

1. Time-Series Analysis (Portfolio Level):
   • Mean-Variance Frontier & Sharpe Ratios: Comparing the maximum squared Sharpe ratios of single-factor (CAPM) vs. multifactor models.
   • Pairwise and Multiple Model Comparison: Using the methodology of Barillas, Kan, Robotti, and Shanken (2020) to test if one model dominates another in terms of the maximum Sharpe ratio.
2. Cross-Sectional Regression (Portfolio Level):
   • Test Portfolios: 32 portfolios sorted by rating, maturity, credit spread, and Fama-French industries.
   • OLS vs. GLS: The authors contrast Ordinary Least Squares (OLS) with Generalized Least Squares (GLS). Following Lewellen, Nagel, and Shanken (2010), they argue GLS is superior for corporate bonds because it downweights the informational content of portfolios with high return heterogeneity, providing a more reliable goodness-of-fit measure ($R^2$).
   • Robust Inference: They employ misspecification-robust $t$-statistics (KRS) and identification-robust methods to handle weak factors and model uncertainty.
3. Bond-Level Analysis:
   • Fama-MacBeth Regressions: Using post-ranking factor betas (to reduce attenuation bias) on individual bond returns.
   • Gagliardini, Ossola, and Scaillet (2016) Method: An alternative bond-level estimation technique used for robustness checks.

C. Scope of Factors Tested

• Traded Factors: BBW four-factor model, DEFTERM (Default + Term), HKM (Intermediary Capital), CAPM (Stock Market), and various liquidity proxies.
• Non-Traded Factors: Macroeconomic uncertainty (UNC), Aggregate Liquidity (PS/AM), Systematic Volatility (VIX), and Long-Run Consumption Risk (LRC).

3. Key Contributions

1. Data Integrity Correction: The paper exposes critical data errors (lead/lag biases) in the widely cited BBW factors, demonstrating that the original results were driven by these construction flaws.
2. Methodological Rigor: It advocates for the use of GLS cross-sectional regressions and misspecification-robust inference in corporate bond pricing, arguing that standard OLS results are economically misleading.
3. Re-evaluation of Factor Pricing: It provides a definitive rebuttal to the claim that non-market factors (downside, credit, liquidity) offer significant incremental pricing power over the bond market factor.
4. Distinction between Traded and Non-Traded: It systematically tests both traded and non-traded factors, concluding that while traded liquidity has marginal power, most other proposed factors are redundant or spurious.

4. Key Results

A. Traded Factors (BBW Model and Others)

• Factor Construction: Corrected replicated factors show significantly higher volatility (standard deviations) than the original BBW factors due to the removal of the attenuation caused by lead/lag errors.
• Sharpe Ratios: The single-factor bond CAPM (using MKTB) yields a bias-adjusted squared Sharpe ratio that is statistically indistinguishable from, and often slightly higher than, the BBW four-factor model.
• Cross-Sectional Performance:
  • OLS Results: Show high $R^2$ (0.83–0.92) for all models, falsely suggesting good fit.
  • GLS Results: When using GLS (which accounts for portfolio heterogeneity), $R^2$ values drop drastically (0.002 for CAPM to 0.185 for BBW). All models are rejected at the 1% significance level.
  • Incremental Power: Only the Liquidity Risk Factor (LRF) shows marginal incremental explanatory power over MKTB in GLS tests. DRF and CRF are spanned by MKTB and offer no additional pricing ability.
• Conclusion: The BBW four-factor model does not dominate the bond CAPM.

B. Non-Traded Factors

• Mimicking Portfolios: When non-traded factors (Uncertainty, Liquidity, Volatility, Consumption) are projected onto traded assets to create mimicking portfolios, they generally fail to produce significant alphas or Sharpe ratios.
• Spanning: Most non-traded factors are spanned by the bond market factor (MKTB).
• Risk Premia: In cross-sectional regressions, none of the non-traded factors command statistically significant risk premia. The Long-Run Consumption (LRC) factor, previously thought to be powerful, is found to be spanned by MKTB when proper estimation error in mimicking portfolio weights is accounted for.

C. Bond-Level Analysis

• Results at the individual bond level largely confirm portfolio-level findings.
• MKTB and LRF continue to exhibit nontrivial pricing power.
• Other factors (DRF, CRF, and non-traded factors) add no explanatory power to the models.

5. Significance and Implications

• Paradigm Shift: The paper challenges the consensus that corporate bond pricing requires complex multifactor models. It suggests that the Bond CAPM (MKTB) remains the dominant benchmark, and the search for additional factors has largely yielded spurious results due to data errors and methodological weaknesses.
• Investment Implications: For investors, the results imply that attempting to construct portfolios based on DRF, CRF, or other non-market factors is likely suboptimal. The transaction costs of rebalancing these factors (especially in the OTC bond market) would likely outweigh the marginal, statistically insignificant alpha they might generate.
• Future Research Directions:
  • Researchers must prioritize data cleaning and avoid look-ahead biases.
  • Empirical studies should rely on GLS and robust inference rather than OLS.
  • Future work should explore risk dimensions beyond standard factors, such as frequency (high-frequency risk) or explicitly incorporate transaction costs into performance metrics.

In summary, Dickerson, Mueller, and Robotti demonstrate that the "excess" pricing power claimed by recent literature is an artifact of data errors and inappropriate statistical methods. Once corrected, the bond market factor alone explains the cross-section of corporate bond returns as well as, or better than, any proposed multifactor model.