What Happens Inside a Bank When a Loan Goes Bad

Take a bank constructed for this page — the same one the other pages in this order use, with US$8.0bn of loans, US$1.0bn of shareholders' equity and a net profit last year of US$92.2m. One of its loans is US$5m lent to a business for five years, repayable at the end, and priced at 2.0% over what the money costs the bank. That margin, US$100,000 a year, is what the loan earns.

When the loan goes bad, three numbers move, and it helps to hold them apart. The credit impairment charge is an expense in the year's profit and loss. The provision is the running total of those charges, held on the balance sheet against the loans. And CET1 capital — the shareholders' equity a regulator counts — falls by whatever the charges take out of profit. The charge is the flow, the provision is the stock, and the capital is what pays.

The borrower's side of the same months — the letters, the new banker, the standstill — is When Your Loan Goes to Workout. This page stays inside the bank.

IFRS 9, the international accounting standard for financial instruments, has required since 2018 that a bank provide for a loan from the day it is written. While the loan is performing, the provision is twelve months of expected credit loss: the loss that would follow a default in the coming year, weighted by the chance of one. Banks call this stage 1.

The bank's grade for this borrower gives a 1% chance of default in the year. If it did default, the bank expects to lose 40% of what is owed once the security has been sold.

A fifth of the first year's margin is charged on day one, against a loan doing everything it promised. It is not a loss; it is the price of the risk, recognised up front — the same three factors the bank used to price the loan, which Expected Loss: a Worked Example sets out.

A year later the business loses its largest customer. Every payment is still on time, but the bank's reading has changed: the chance of default over the four years left is now put at 20%. IFRS 9 calls this a significant increase in credit risk since the loan was written, and it changes the horizon. The provision now covers the expected loss over the loan's whole remaining life, not twelve months. This is stage 2.

Nothing has been missed, and the bank has charged nearly four years of the loan's earnings against its profit. That is the design: losses are recognised as the risk rises, not when the cash stops. A real model discounts the future losses and weighs several economic scenarios; the sum above leaves both out. The standard also sets a backstop — once a payment is more than 30 days late, the risk is presumed to have risen significantly unless the bank can show it has not. Interest is still recognised on the full US$5m.

Six months on, the payments stop. Under the Basel Committee's 2017 definitions the loan is now non-performing: more than 90 days past due or, earlier, judged unlikely to be repaid in full without the bank selling the security, however current the payments. Under IFRS 9 it is credit-impaired, which banks call stage 3, and the file has usually moved by now to a workout team.

The chance of default is no longer a chance. The provision becomes the bank's best estimate of what it will not get back: the debt, less what the security will bring after the costs of selling it.

Two other things change. Interest is now recognised only on the loan net of its provision — on US$3.0m, not US$5m — and little of it arrives in cash, while the bank goes on paying for the US$5m of funding behind the loan. And the security, which keeps the loss to 40%, did nothing to keep the loan performing: the Basel definitions say collateral should have no direct influence on whether a loan is classed as non-performing. Security decides how much is lost, not whether the borrower pays.

Rescheduling the debt does not make it performing again either. Under the same definitions a non-performing loan is reclassified only after the borrower has paid on time for a period, and a loan restructured because the borrower is in difficulty is reported as forborne for at least a year of payments on the new terms.

Eighteen months later the business's assets have been sold, and they bring US$3.0m after costs, as estimated. The bank applies the US$3.0m to the debt and writes off the remaining US$2.0m: IFRS 9 requires a loan to be written down once there is no reasonable expectation of recovering it. The write-off uses up the provision held against it, the loan leaves the non-performing figure, and nothing new reaches profit, because the loss was charged as it was recognised. Had the assets brought US$2.7m, the bank would have charged another US$300,000; had they brought US$3.3m, it would have released US$300,000 back to profit. The provision was an estimate all along.

Provisioning decided when the loss reached profit, not how large it was. The size is the arithmetic of the trade: one loss of this kind eats twenty years of the loan's margin — or a year's margin on US$100m of loans like it. And margin is not profit. Out of its 2.0%, a loan pays for its own running and for the capital behind it, so in profit the payback is longer still. Read across a whole book, the same movements are what the provision coverage ratio measures: how much of the bad part is already paid for.

Inside the constructed bank, this loan is one line among thousands. The year's credit impairment charge of US$32m is the net of movements like these, up and down, across the whole book. What matters to a shareholder or a depositor is what stands in front of them when that total grows.

The first absorber is the year's earnings. Before credit losses the bank earns US$160m, so its loans could lose 2.0% of their value in a year and it would still break even. Beyond that, losses come out of shareholders' equity — the US$780m of it that counts as CET1 capital — at US$60m for every point of the ratio, and a bank whose capital falls far enough is limited in the dividends it may pay; Who Regulates Banks works that test.

Only when both are gone do losses reach the people who lent the bank money, and even then the US$120m of subordinated bonds, sold to investors on terms that rank them behind depositors, take the loss first. The depositors' US$7.0bn stands behind all of it.

So a single loan cannot hurt a bank this size; this one moved its CET1 ratio by about three hundredths of a point. What hurts a bank is the same downturn reaching a hundred borrowers at once, which is why the machinery around the book watches concentrations as closely as it watches loans.