Credit Expected Loss Through the Cycle

Why the boom is when you misprice — and why the loan that feels safest is the one most likely to destroy value

In 2006, point-in-time default estimates for BBB corporates sat near 0.3 percent. Credit was easy, spreads were tight, and the models said risk was low. By 2009, the same estimate was 3.5 percent — more than ten times higher. The loan priced in 2006 was still on the book when the 2009 number arrived. It had been priced for paradise and was now living through the flood. Through-the-cycle pricing charges the long-run average — roughly 1.5 percent — regardless of where you are in the cycle. The boom is not when credit is cheap. The boom is when the mispricing happens.

So what does a bank actually need to know in order to price credit losses correctly? Three quantities, one product, and one philosophical decision about time horizon. The three quantities are probability of default, loss given default, and exposure at default. Their product is expected loss. And the philosophical decision — the one that separates banks that survive cycles from banks that do not — is whether to measure those quantities at a point in time or through the cycle.

PD, LGD, EAD — what each one measures

Probability of Default

How likely is this borrower to stop paying? Not "will they default this year" but "what is the annual probability, averaged across good years and bad?" This is the heart of credit analysis: the assessment that a given counterparty, given its financial condition, industry, country, and capital structure, will fail to meet its obligations within a defined horizon. For a BBB corporate borrower, the through-the-cycle annual PD typically sits in the range of 0.20 to 0.40 percent. It sounds small. Over five years, it compounds into something far less comfortable.

Loss Given Default

If they default, how much do you lose? Default does not mean total loss. After collateral is seized, workouts are negotiated, and recoveries collected, some fraction of the exposure comes back. What remains lost is the LGD. For unsecured senior corporate debt, historical LGDs cluster between 40 and 60 percent — meaning the bank recovers roughly half. For a secured mortgage with 70 percent loan-to-value, the collateral covers most of the exposure; the LGD drops to perhaps 15 to 25 percent. LGD is where collateral earns its keep. It is also where recovery assumptions earn their reputation for optimism — because the recoveries that look generous in calm markets tend to shrink precisely when you need them most.

Exposure at Default

What will the balance be when the default happens? For a fully drawn term loan, this is straightforward: EAD equals the outstanding balance. But for a revolver with undrawn commitment, the picture changes. Borrowers approaching distress tend to draw down their available lines. The credit conversion factor captures this behavior, inflating EAD beyond the current drawn balance to reflect the portion likely to be pulled before the borrower stops paying. A firm with a 10-million commitment and 6 million drawn may well have an EAD of 8.5 million — because the last act of a drowning borrower is to reach for every lifeline available.

The product

Multiply all three together and you get the annual expected loss, expressed in basis points of exposure. For Avelmont's BBB corporate borrower — a through-the-cycle PD of roughly 0.30 percent, an LGD of 45 percent, on a fully drawn term loan — the product lands near 135 basis points. That is the actuarial cost of credit: the amount the loan must earn, each year, just to cover the losses that statistics say will arrive.

Through the cycle vs. point in time — the pricing choice

Here is the question that determines whether the credit charge in the loan price is trustworthy or delusional: are you measuring default probability as it is right now, or as it is on average across the full economic cycle?

Point-in-time

A point-in-time PD reflects current conditions. When the economy is booming, corporate balance sheets are strong, defaults are rare, and the PIT estimate drops. In 2006, it said 0.3 percent for BBB. When the recession arrives, balance sheets weaken, defaults spike, and the PIT estimate surges. In 2009, it said 3.5 percent. PIT is responsive, accurate for this quarter, and violently pro-cyclical.

Through-the-cycle

A through-the-cycle PD reflects the full economic cycle. It asks: over a period long enough to include both booms and busts, what is the average annual default rate for this grade? The answer does not change with the season. In 2006, TTC said 1.5 percent for BBB. In 2009, it still said 1.5 percent. TTC is stable, slower to react to genuine structural change, and appropriate for pricing a loan whose life will span both sunshine and storm.

Why TTC for pricing

Consider Avelmont's five-year loan, originated in 2024 and maturing in 2029. That loan will experience good years and bad. Pricing it at today's low PIT estimate assumes the good times last forever — an assumption that has been tested and found wanting with depressing regularity. Pricing it at the TTC average charges the borrower for the full cost of credit across the cycle. Not too much in booms. Not too little in busts. The right amount, on average, for the loan's actual life.

Why PIT for provisioning

IFRS 9 and CECL require provisions that reflect current and forward-looking conditions. This is a different question entirely. Provisioning asks: "How much should the bank set aside now?" Pricing asks: "How much should the loan earn over its life?" The first question demands sensitivity to current conditions. The second demands indifference to them.

The pro-cyclical trap

Using PIT for pricing produces exactly the wrong behavior at exactly the wrong time. In booms, the PIT estimate is low, the credit charge is small, and the bank originates aggressively — undercharging for risk precisely when the bad vintages are being created. In busts, the PIT estimate is high, the credit charge is punitive, and the bank retreats — refusing creditworthy borrowers precisely when the economy needs lending most. Pro-cyclical pricing amplifies the cycle instead of pricing through it. It is the mechanism by which good intentions produce bad loans.

The transition matrix — BBB doesn't jump to default

Default is rarely sudden. A BBB borrower does not wake up bankrupt. It drifts. BBB becomes BBB-minus. BBB-minus becomes BB-plus. BB-plus becomes BB. The descent is gradual, punctuated by rating actions that mark each step downward. Each step takes time. Each step has a probability. And the tool that captures those probabilities is the transition matrix.

Picture a grid. Each row is a starting grade. Each column is the grade one year later. Most of the mass sits on the diagonal — a BBB borrower is most likely to still be BBB twelve months from now. Small probabilities fan out to adjacent grades: a few percent chance of upgrading to A, a few percent chance of downgrading to BB. And at the far right column, a tiny probability reaches Default directly. For BBB, that direct-to-default probability is roughly 0.25 percent in any given year.

But over five years, the story changes. The cumulative PD is not five times the annual PD. It is higher, because the borrower can drift downward through multiple grades before defaulting. Year one PD for BBB: roughly 0.25 percent. Cumulative five-year PD: roughly 2.5 percent. The curve is convex — each additional year adds more default risk than the last, because the borrower has had more time to drift into weaker territory where the annual default probability is steeper.

For weaker grades, this convexity is starker. A B-rated borrower faces a year-one PD of about 3 percent. Its cumulative five-year PD: roughly 15 percent. That is five times the annual rate, not five times three percent. The transition matrix reveals why tenor matters so profoundly in credit pricing — because time gives gravity a chance to pull weak credits further down.

The credit ladder — how expected loss varies by grade

Line up the rating grades from AAA to CCC and plot the expected loss for each. The shape is not a straight line. It is a curve — nearly flat at the top and vertiginously steep at the bottom.

• AAA: 1–2 bps. Near zero. These are the most creditworthy borrowers on earth, and the expected loss barely registers.
• AA: 5–10 bps. Still negligible in the context of a loan rate measured in hundreds of basis points.
• A: 15–25 bps. Noticeable but modest. The credit charge is a minor layer in the all-in price.
• BBB: 35–50 bps. The boundary of investment grade. Here the credit charge begins to have teeth.
• BB: 80–120 bps. The first speculative grade. Expected loss roughly doubles relative to BBB.
• B: 135–200 bps. Avelmont's SME loan sits here. The credit charge is now a major component of the loan rate.
• CCC: 400+ bps. Distressed territory. Default is not a tail event — it is approaching a central scenario.

The convexity is visible in the increments. Moving from A to BBB adds roughly 25 basis points. Moving from BB to B adds roughly 80. Moving from B to CCC adds 200 or more. Credit risk is not linear with grade. It accelerates sharply at the weak end of the spectrum, and any pricing model that treats the credit ladder as a set of evenly spaced rungs will undercharge the bottom grades catastrophically.

The double-count trap

Here is an error that sounds too obvious to make and yet persists across banking systems worldwide. IFRS 9 requires the bank to provision for expected credit losses. That provision is an expense; it reduces earnings. The FTP expected-loss charge also covers expected credit losses — the same economic loss. If both are charged to the business unit, the same loss is counted twice: once in the provision expense and once in the transfer price.

How does this happen? Usually through organizational fragmentation. The credit risk team runs the provisioning models and books the IFRS 9 expense. The treasury team runs the FTP framework and includes an EL charge in the transfer price. Neither team checks whether the other has already billed the business unit for the same cost. The business unit pays twice and wonders why apparently profitable loans fail to clear the hurdle rate.

The FTP expected-loss charge IS the provision, economically. The business unit pays the EL charge through the transfer price. The provision is funded by that charge. They are not additive — they are two views of the same cost.

Getting this wrong inflates the apparent cost of lending and drives the bank away from creditworthy borrowers who should be profitable. The double-count does not create safety. It creates the illusion of unprofitability, which is arguably more dangerous — because a bank that misprices in the other direction at least knows it has a problem, while a bank that double-counts simply believes it has discovered that lending is unprofitable. It stops competing for the right business and wonders, years later, why its market share has eroded.

Expected vs. unexpected — two different things, two different treatments

This distinction is the fault line along which most pricing confusion occurs. Expected loss and unexpected loss are not two sizes of the same thing. They are two entirely different economic concepts, and confusing them is one of the most persistent errors in loan pricing.

Expected loss is a cost

It is the actuarial average of defaults over the cycle. It is as certain as any statistical average — not in the sense that it will occur precisely, but in the sense that over a large portfolio and a long time horizon, actual losses will converge to it. It belongs in the price, just like operating costs or funding costs. It is the price of doing credit business. A bank that does not charge for expected loss is not lending profitably; it is making charitable donations with depositor money.

Unexpected loss is a risk

It is the possibility that actual defaults exceed the expected average — the variance around the mean, the tail that the average does not capture. In a bad year, losses do not politely stop at their expected level. They overshoot. That overshoot is what equity absorbs. It is the reason banks hold capital. And its cost belongs not in the credit spread but in the capital charge — the return that shareholders demand for bearing the risk that reality will be worse than the average.

The cost of capital — why it sits on top of credit EL

If expected loss were the only credit-related charge in the loan price, the arithmetic would be simple and the job would be done. But expected loss is only the average. What about the years when losses exceed the average? What about the scenario where Avelmont's BBB portfolio experiences not the expected 1.5 percent default rate but a 4 or 5 percent spike? That excess is unexpected loss, and it is what capital exists to absorb.

Credit capital

The regulatory framework assigns a risk weight to the exposure: 100 percent for a typical corporate. The bank must hold CET1 capital equal to the risk weight times the CET1 ratio — say 10.5 percent, including buffers. And shareholders demand a return on that capital — a pre-tax hurdle rate of, say, 14.45 percent. The product: 100 percent times 10.5 percent times 14.45 percent, which rounds to roughly 152 basis points. This is the cost of the equity tied up against credit risk.

Interest-rate risk capital

A fixed-rate loan creates duration mismatch. Even if the credit is pristine — even if it were government-guaranteed — the rate risk ties up capital under the IRRBB framework. For Avelmont's five-year fixed bullet, this adds roughly 49 basis points.

Total capital charge

152 plus 49 equals 201 basis points. In many loan pricing stacks, this is the single largest layer — larger than the credit expected loss, larger than the operating cost, larger than the term liquidity premium. It is the price of the variance, the cost of the tail, the return shareholders demand for bearing the risk that the average will not hold.

And here is the point that catches many pricing teams off guard: a zero-credit-risk government-guaranteed loan still needs rate capital if it is fixed-rate. Capital is not just about credit. It is about every risk that creates unexpected losses. A loan that is credit-riskless but duration-heavy still consumes equity, and equity still demands its return.

The boom is when the mispricing happens

The boom feels like safety. Credit spreads are tight, defaults are rare, the models say risk is low. But the boom is precisely when the five-year loan is being originated — the loan that will live through the next downturn. Every vintage of lending carries the pricing assumptions of its origination date into the future, and those assumptions will be tested by conditions the originator did not foresee and the model did not reflect.

Through-the-cycle pricing refuses to pretend that the calm will last. It charges the cycle's average, not today's snapshot. It separates expected loss — a cost, embedded in the price — from unexpected loss — a risk, embedded in the capital charge. And it does this not because regulators demand it, but because the alternative is how banks discover, too late, that they were lending below cost all along.

The discipline is unglamorous. It produces loan rates that feel too high in booms and too low in busts. Relationship managers will complain that the pricing is uncompetitive. Competitors using point-in-time models will undercut on price. But those competitors are not offering a better deal — they are offering a deal that has not yet revealed its true cost. And when the cycle turns, as it always does, the bank that priced through the cycle will discover that it was not being conservative. It was being accurate.

Pricing for paradise is not optimism. It is a measurable, quantifiable error — the difference between the point-in-time estimate and the through-the-cycle average, multiplied by every dollar of exposure originated during the boom. That product, denominated in basis points during origination, returns denominated in write-offs during the bust.