The Missing Factor: Why Nobody Can Explain Credit Spread Moves

By 2001, credit modelling had a rich theory. Merton had connected default to firm value. Black and Cox had added a default barrier. Longstaff and Schwartz had brought in interest rates. Every one of these models makes clear, checkable predictions about what should move a credit spread: the company's stock price, its leverage, the volatility of its assets, the level of interest rates, the slope of the yield curve.

Pierre Collin-Dufresne, Robert Goldstein and Spencer Martin decided to simply check. They took actual corporate bonds, measured how their spreads moved month to month, and ran the theory's variables against them.

The results were, to put it gently, humbling.

The problem: does the theory actually explain anything?

The test is straightforward in concept. Theory says a company's credit spread should widen when its stock falls (it is closer to trouble), when its volatility rises (its assets are wobblier), when leverage climbs, and it should tighten when interest rates rise (as Longstaff and Schwartz predicted). Add in a few sensible controls for the state of the economy and market liquidity.

So: gather a lot of straight industrial corporate bonds, compute the monthly change in each bond's spread over Treasuries, and regress those changes on the theoretical drivers. If the models are right, you should soak up most of the movement. If they are half right, you should soak up a decent chunk.

They soaked up remarkably little.

The variables had the signs theory predicted, which is at least something. Spreads did widen when equity fell. They did tighten when rates rose. The theory was not backwards. But the amount of month-to-month spread movement these variables collectively explained was small. Most of what credit spreads do, they do for reasons that structural credit theory has no name for.

The key idea via analogy: the leftovers move as one

The truly striking finding was not the low explanatory power. Models being incomplete is normal. It was what the authors did next.

They took the residuals, the part of each bond's spread change that the theory failed to explain, and looked at them across bonds. If the leftover movement were just noise, idiosyncratic wobble specific to each issuer, the residuals would be unrelated to each other. Some bonds would have positive leftovers, some negative, all scattered.

Instead, the leftovers were highly correlated across bonds. A single common factor drove most of them. When one bond's spread moved for an unexplained reason, nearly every other bond's spread moved for the same unexplained reason, at the same time, in the same direction.

Think of it like this. You are trying to explain why a fleet of boats bobs up and down. You carefully model each boat's cargo, hull shape and engine, and you find your models are poor. Fine. But then you notice that the errors in your models all move together: whenever your model under-predicts one boat's height, it under-predicts every boat's height. That is not a fleet of badly modelled boats. That is a tide, and you have not modelled the tide.

So there is a tide moving credit spreads, and it is not any of the things credit theory says should move them.

The authors then went hunting for it. They threw a wide net of macroeconomic and financial variables at this common factor: measures of the business cycle, market returns, market volatility, standard proxies for liquidity, and more. None of them captured it. The mystery factor was not just missing from credit models, it was missing from the obvious candidate list too.

Their conclusion, offered with appropriate caution, was that monthly credit spread changes seem to be driven mostly by local supply and demand shocks in the corporate bond market itself, forces that have nothing to do with the credit risk of the issuers and are not captured by the usual liquidity measures. In plain terms: spreads move because someone big needs to buy or sell bonds, not because anything changed about whether the companies will pay you back.

Why it mattered

• It punctured a very confident theory. Structural credit models were the intellectual centrepiece of the field. Showing that they explain only a minority of actual spread movement was a serious, well-executed challenge that the literature could not brush off.
• It reframed the credit spread. If most spread movement is not about default, then the yield on a corporate bond is not mostly a default premium. It is a bundle: some default risk, plus a large and time-varying charge for illiquidity, market segmentation, and whatever else the common factor is. This directly set up the later research (Longstaff, Mithal and Neis, among others) that used the CDS market to prise those pieces apart.
• It has real consequences for hedging. If you are running a credit book and you believe spread moves are driven by issuer fundamentals, you will hedge with equity and volatility and be repeatedly surprised. If most of the risk is a market-wide, non-fundamental factor, your hedge is against the wrong thing, and every position in your book is quietly exposed to the same hidden driver at once. That is a recipe for a book that looks diversified and is not.
• It gave a name to something traders already knew. Desk-level intuition has always held that "the credit market has its own weather." This paper measured the weather, showed it was the dominant force, and showed nobody could explain it.

The honest limitations

• Monthly changes are a specific lens. The test is about changes in spreads at a monthly frequency. Structural models might do considerably better at explaining spread levels, or at longer horizons, and the authors were careful not to overclaim.
• The data are difficult. Corporate bonds trade infrequently, and the study relied on dealer quotes rather than a stream of executed prices. Stale and noisy quotes make any relationship harder to detect, so some of the "unexplained" movement may be measurement error rather than a real hidden factor.
• Liquidity was proxied, not measured. The liquidity variables available at the time were crude. It is entirely possible that the mystery factor is liquidity all along, just liquidity that the era's proxies could not see. Later work using CDS data suggests a substantial non-default, illiquidity-driven component, which points in exactly that direction.
• The paper names a puzzle rather than solving it. Identifying a large common factor and then failing to explain it is honest science, but it leaves the field with an unlabelled box. Precisely what the box contains is still argued about.

The one-line takeaway

Collin-Dufresne, Goldstein and Martin showed that the variables credit theory says should drive spreads explain surprisingly little of what spreads actually do, and that the unexplained part moves as one giant common factor across every bond, which means corporate credit is driven far more by market-wide supply and demand than by anything happening at the companies themselves.