Why Credit Risk Has Paid Off for Bond Investors

Academics have long debated whether corporate bonds pay investors sufficiently for the credit risk they bear, or whether virtually all the extra yield over Treasuries just compensates investors for duration (interest rate) risk. Recent academic work using post-1986 data has leaned toward the latter conclusion, finding that investment-grade corporate bonds offer little to no reward for default risk once the term premium is stripped out. The authors of the April 2026 study “Reconstructing a Century of U.S. Corporate Bonds: Credit Risk in Historical Perspective,” argue that this conclusion is an artifact of looking at too short and too unusual a sample period. When they extend the data back to 1926 (and even 1895), the credit risk premium turns out to be sizable, statistically significant, and monotonically related to credit risk.

Building a ‘CRSP for Corporate Bonds’

The core contribution of the paper is a new database spanning 128 years (1895–2022), built by hand-collecting monthly bond quotes from three archival print sources—the Commercial and Financial Chronicle, Standard & Poor’s Bond Guide, and Mergent/Moody’s Bond Record—and combining them with existing modern datasets from Lehman Brothers and Wharton Research Data Services. The result is a panel of more than 100,000 unique bonds and more than seven million observations, digitized from more than 80,000 pages of archival material with double-blind data entry to ensure accuracy. The authors also hand-collected bond characteristics (coupon, maturity, collateral status, ratings, call features) from Moody’s Manual going back to 1917.

By building returns bond by bond from actual transaction prices, the authors produce what is arguably the first ground-up, nonsynthetic long-run corporate bond return series.

The Central Finding: Sample Period Changes the Answer

Using the post-1986 sample—the period examined in prior influential work (such as van Binsbergen, Nozawa, and Schwert, 2025)—the authors replicate the now-familiar result: The credit risk premium is small and statistically insignificant for investment-grade bonds. AAA/AA bonds earned a 3.75% annualized excess return over the period, but only 0.25 percentage points of that was attributable to credit risk; the rest was term premium, reflecting the multidecade decline in interest rates that happened to occur during this window.

Extend the sample back to 1926, and the picture changes substantially. The credit risk premium for AAA/AA bonds rises to 0.62%, and for BBB bonds to 1.4%. For high-yield bonds, the gap is even starker: B-rated bonds show a credit risk premium of 5.88% in the long sample versus 3.06% in the post-1986 sample. Across every rating category, the long-run premium is larger, more statistically significant, and increases monotonically as credit quality declines—the pattern you’d expect if investors are being compensated for risk.

Reconciling the Credit Spread Puzzle

The paper also tackles a well-known tension in the literature: Average credit spreads (the promised-yield measure) have stayed fairly stable across different eras, while estimated credit risk premiums (the realized-return measure) have swung wildly by comparison, exceeding spreads in 1947–85 and falling well below them in 1986–2022. The authors show mathematically that this “wedge” is driven mostly by a spread duration effect: When credit spreads are generally falling over a sample period (as they were, on net, in the earlier postwar decades), bond prices get an extra realized-return boost; when spreads are generally rising or volatile (as in the recent decades, on net), realized returns understate the “true” spread compensation. Over a long-enough sample, spread increases and decreases roughly wash out, and the estimated credit risk premium converges toward the average credit spread, which is exactly what the centurylong data shows.

A related and more technical finding is that much of the existing literature ignores embedded call provisions when matching corporate bonds to comparable-duration Treasuries. During the multidecade decline in interest rates since the 1980s, many callable bonds had their calls move into the money, shrinking their effective duration well below their stated (Macaulay) duration. Matching these bonds to long-duration Treasuries overstates their interest rate exposure and, mechanically, understates the credit risk premium—a bias that’s particularly acute in exactly the post-1986 period most previous studies rely on.

Credit Risk Premia Track Systematic Risk Exposure

Sorting bonds by their sensitivity to stock returns, aggregate bond market returns, industrial production shocks, and inflation shocks, the authors find that bonds with higher exposure to these systematic factors earn meaningfully higher returns, and that this entire spread is explained by the credit risk premium component, not the term premium. That’s a textbook signature of a priced risk factor rather than a market inefficiency.

The paper also extends the Gilchrist-Zakrajšek credit spread measure back to 1926. The extended spread robustly predicts future corporate bond excess returns across multiple horizons (consistent with spreads reflecting time-varying risk premia, not just default expectations) and predicts National Bureau of Economic Research recession probabilities, though its power to forecast industrial production and employment growth weakens once the Great Depression and other prewar regimes are folded in.

Their findings led the authors to conclude: “Over the full historical period, corporate bonds earn a sizable and statistically significant credit risk premium that increases monotonically with credit risk.”

Investor Takeaways

• Don’t extrapolate the credit risk premium from the post-1986 era. That period featured a historic, multidecade decline in interest rates, which inflated realized term premiums and happened to coincide with generally tightening (rising and falling) spread dynamics that depressed realized credit premiums. It’s a short, unusual sample, not a representative one.
• Credit risk has historically been compensated, and the size of that compensation scales with credit quality. Over the long run, high-yield bonds have earned several percentage points more in annualized credit risk premium than investment-grade bonds, consistent with basic risk/reward logic, not a market anomaly.
• Be skeptical of “the term premium explains everything” narratives built on short samples. Duration-matching methodology matters. Ignoring embedded call features can systematically shift compensation from the “credit” bucket into the “term” bucket in a falling-rate environment, precisely the environment most investors have lived through for the last several decades.
• Elevated credit spreads still carry useful predictive information. This is true for subsequent corporate bond returns and, to a degree, for recession risk, supporting the case that credit spreads are meaningful signals of time-varying risk premia rather than noise.

What This Means for Market Efficiency

This paper is useful for how we think about “market efficiency” debates in credit markets. The prior finding—that investment-grade credit risk premiums looked negligible in recent decades—had been read by some as evidence that corporate bond investors weren’t being rewarded for bearing default risk at all, raising the specter of mispricing or a “credit spread puzzle” that couldn’t be squared with observed spreads. This paper’s evidence points the other way: Markets have, over the long run, priced credit risk sensibly and in a manner that lines up with exposure to systematic (undiversifiable) risk factors—the hallmark of an efficient, risk-based explanation rather than a market failure.

What looked like a puzzle wasn’t really a pricing failure; it was a small sample-and-measurement problem. Realized returns over any single, historically unusual multidecade window (particularly one dominated by secular declines in rates) are a noisy and potentially biased estimate of the true, ex ante expected risk premium investors demand. That’s a broader lesson that extends well beyond credit markets: Short samples, however lengthy they may seem in the moment, can lead researchers and practitioners to draw conclusions about risk and return that don’t hold up once a longer, more representative history is available. It’s also a reminder that market efficiency should be judged over full cycles, not over spans of time distorted by one persistent regime.