Default Risk or Just Hard to Sell? Splitting the Corporate Bond Spread

A corporate bond yields more than a Treasury bond. Everyone agrees. The question that consumed credit researchers for thirty years is: more for what, exactly?

The obvious answer is default risk. The company might not pay you back, so you demand extra. But there is a nagging problem with that answer, known as the credit spread puzzle: if you take historical default rates and recovery rates for, say, investment-grade companies, and work out what compensation those losses justify, you get a number far smaller than the spread the market actually charges. Investment-grade companies almost never default, yet their bonds yield noticeably more than Treasuries. Something else is in the price.

The leading suspect was always liquidity. A Treasury bond is the most tradeable financial instrument on earth. A specific corporate bond might not trade for weeks. You should get paid something for holding an asset you cannot easily sell. But suspects are not convictions, and nobody could cleanly separate the two, because both effects show up in the same number: the yield.

Longstaff, Mithal and Neis found a way to separate them.

The problem: one observation, two causes

Imagine you know that a room is at 30 degrees, and you know the temperature is a combination of the sun coming through the window and a radiator on the wall. You want to know how much each contributes. With one thermometer, you cannot. You need a second measurement that responds to only one of the two sources.

That is exactly the credit spread's problem. The bond yield contains default compensation plus liquidity compensation, hopelessly mixed. Every earlier attempt to split them relied on modelling one side and calling the residual the other, which means the answer depends entirely on how good your model is. That is not evidence, it is arithmetic on top of an assumption.

The key idea via analogy: a second thermometer that only reads default

Then the credit default swap market grew up, and it gave researchers exactly the second thermometer they needed.

A credit default swap is an insurance contract. You pay a regular premium, and if the company defaults, you get compensated. Nothing else. It is a pure bet on the credit event.

Here is why that is so useful. A CDS is a contract, not an asset. There is no bond to hold, no principal to fund, no illiquid security sitting in your inventory. If you want protection on a company, you and a counterparty simply agree terms and one party pays a premium. The size of the market is not limited by how many bonds the company happened to issue. The CDS premium therefore reflects, in a much purer form, what the market thinks about the probability of default and the loss given default, without the funding and inventory frictions that make the bond expensive to hold.

So Longstaff, Mithal and Neis took CDS premiums as their reading of the default component. They used them to back out the market's view of default risk for each company, worked out what corporate bond yield that default risk alone would justify, and compared it with the yield the bond actually offered. The gap, whatever it is, is the non-default component.

The results were a genuine surprise to both camps.

Most of the corporate spread is default risk. Across every rating category, the default component was the dominant piece. That was a real vindication of credit modelling, and a partial answer to the credit spread puzzle: the market really is charging mostly for default risk, and if simple historical-default calculations say otherwise, it is because the market prices default risk with a hefty risk premium on top of the raw statistical odds. Investors do not just want to be compensated for expected losses. They want to be compensated for bearing the risk that losses cluster in bad times, and that premium is large.

But the non-default component is real, substantial, and it moves. It is not a constant tax. It swells and shrinks over time, and when it does, it lines up with measures of how hard that specific bond is to trade (the bid-ask spread, how much of the issue is outstanding and available) and with market-wide measures of bond-market liquidity. In other words, the leftover is not random noise. It behaves exactly like an illiquidity premium should behave.

Why it mattered

• It settled a long argument with data rather than models. The default-versus-liquidity debate had been going in circles for years because everyone was arguing from a model. This paper brought a second, largely independent price series to bear, which is how you actually resolve these things.
• It made the credit spread puzzle less puzzling. The gap between historical default losses and observed spreads is not mostly liquidity. It is mostly a default risk premium: the market charges far more than the raw statistical odds, because defaults cluster in recessions, and investors hate losing money precisely when everything else is going wrong.
• It legitimised the CDS-bond basis as a trading concept. Once you accept that a bond's spread and its CDS premium measure overlapping but different things, the difference between them becomes a tradeable, measurable quantity. The basis trade, buying a bond and buying protection on it, is essentially a bet on that non-default component. Many funds ran that trade. In 2008 it blew apart spectacularly, which was itself the loudest possible confirmation that the non-default component is real and can move violently.
• It reframed liquidity as a priced, time-varying risk. The non-default piece is not a fixed haircut. It is a premium that expands in stress. Any model that treats a corporate bond's spread as pure credit will misprice it exactly when it matters most.

The honest limitations

• The CDS thermometer is not perfect. CDS premiums have their own frictions: counterparty risk (your insurer might fail too, a lesson learned brutally in 2008), the "cheapest to deliver" option that can make protection worth more than pure default compensation, and their own liquidity problems in less-traded names.
• The sample is a specific window. The study used the early years of the CDS market, when it was growing, dominated by dealers, and covered mostly large, liquid, investment-grade issuers. Whether the split looks the same for small high-yield issuers, or in a post-2008 world of central clearing and constrained dealer balance sheets, is a fair question.
• "Non-default" is a residual, not a measurement. Anything the default model fails to capture lands in the liquidity bucket, including taxes, modelling error, and whatever the mysterious common factor is that Collin-Dufresne and co-authors identified. The evidence that it correlates with liquidity proxies is strong, but it is a bucket, not a clean reading.
• Extracting default risk from CDS still needs a recovery assumption. As always, the spread gives you the product of the default probability and the loss, so you have to assume one to get the other.

The one-line takeaway

By using credit default swaps as a clean, second reading of pure default risk, Longstaff, Mithal and Neis showed that most of a corporate bond's extra yield really is compensation for default, but that a genuine and time-varying illiquidity premium sits on top of it, and that premium is exactly the thing that explodes when markets seize up.