How is probability of default calculated?

For a mid-market borrower, a bank's rating model combines financial ratios, industry and qualitative judgement into a grade, and the grade carries a percentage calibrated to the default history of borrowers in it. Statistical models such as logistic regression are common underneath. For large listed issuers, an analyst can also infer a figure from credit default swap prices or share-price models. Either way, the result is an estimate that depends on the definition of default used.

What is the difference between point-in-time and through-the-cycle PD?

A point-in-time PD reflects current and expected conditions, so it rises in a recession and falls in a boom. A through-the-cycle PD is averaged over a full economic cycle and moves far less. Internal-ratings capital rules lean towards the long-run view so that capital does not swing with every turn of the economy. Expected credit loss accounting needs the forward-looking view, so banks commonly adjust one to produce the other.

Is probability of default the same as a credit rating?

No, though they are closely related. A rating is a grade on a scale, published by an agency or assigned inside a bank. A probability of default is a percentage, and a bank's master scale attaches one to each of its grades. Agencies publish how often each grade has defaulted historically, which is a record rather than a stated probability, and banks calibrate their internal grades to their own default experience. Two institutions can give the same company the same grade and attach different percentages to it, because their definitions and histories differ.

Can two loans to the same company have different probabilities of default?

Usually not. Default is an event at the borrower, and the Basel framework requires separate exposures to one borrower to carry the same borrower grade, with exceptions for country transfer risk and for guarantees reflected in an adjusted grade. Differences between the loans, such as security or ranking, are captured in loss given default instead. Outside the corporate book the position differs: retail exposures can be defaulted and rated facility by facility.

// WORKED EXAMPLETHE ARITHMETIC A DECISION RESTS ON

Probability of Default: a Worked Example

One in fifty describes the grade. This company will or will not fail.

BY MICHAEL SHANG · SIXTEEN YEARS IN COMMERCIAL BANKING · UPDATED 2026-11-09 · 6 MIN READ

// THE SHORT ANSWER

No credit committee can know whether one company will fail next year. It can say how often companies like this one have failed, and probability of default is that frequency, read from the borrower's risk grade, normally over twelve months. An airport shuttle operator graded at 2.0% carries a one-in-fifty chance over a year and close to one in ten over five. If it is downgraded to a 4.0% grade, the expected loss on its US$8.0m loan doubles with nothing about the loan having changed. The figure is an estimate about a group of borrowers, applied to one.

BASIS Practitioner judgment from sixteen years in commercial banking, set out in How Bankers Think (Highbank Press, 2026). Every borrower on this site is a composite constructed for teaching, not a client. The market is mid-market commercial lending as practised in New Zealand and Australia; where a convention differs elsewhere, the page says so.

// MAKE THE CALL

Issue 09 — The Tuesday List. Eight dashboards. Two intensive-care slots. Portfolio triage, Tuesday 8:30am. Your call first, then the senior banker's.

Read the file →

PART OF When Your Loan Goes to the Workout Team

// 01 — WHAT THE NUMBER SAYS

A probability of default does not predict this company's fate. A borrower either defaults or does not; the 2% describes how a large group of borrowers like it has behaved. It has meaning only beside a definition of default, because a bank that counts 90 days past due and unlikely to pay will see more defaults than one that counts only insolvency.

It belongs to the borrower. The Basel framework requires banks on internal ratings to place separate exposures to one borrower in the same borrower grade, whatever the differences between the facilities, with two exceptions: country transfer risk, and guarantees reflected in an adjusted grade. Security and ranking live in a different number, loss given default.

// 02 — HOW A BANK ARRIVES AT IT

Financial ratios + industry + judgement → risk grade

Risk grade → long-run default rate of that grade = one-year PD

Cumulative PD over n years = 1 − (1 − PD)^n (if PD stays flat)

Expected loss = PD × LGD × EAD

A rating model scores the borrower's accounts and the qualitative facts, and the score places it in a risk grade. The grade is mapped to a percentage by a master scale, calibrated to how many borrowers in that grade have actually defaulted. Banks rating large listed companies can also draw on agency ratings or on what credit default swaps imply; a mid-market borrower has neither, so the internal model carries the weight.

For capital, the Basel framework asks for a long-run average of one-year default rates, built on at least five years of observations for at least one source, with a floor of 0.05% for non-sovereign exposures. Each market writes its own version: Australia's accredited banks work under APRA's APS 113, New Zealand's under the Reserve Bank's IRB requirements, BPR134 among them, and the UK moves to its Basel 3.1 rules on 1 January 2027. A bank on the standardised approach does not put a PD into its capital sum, though many such banks still use grades for approval and pricing.

// 03 — THE EXAMPLE

An airport shuttle operator, constructed for this page like every figure in it, borrows US$8.0m, fully drawn, secured so that the bank expects to lose 35% if it fails. The bank's master scale, invented here and no guide to any real scale, maps three grades to these figures.

Grade One-year PD Five-year cumulative Expected loss a year

4 1.0% 1 − 0.99^5 = 4.90% 28,000

5 2.0% 1 − 0.98^5 = 9.61% 56,000

6 4.0% 1 − 0.96^5 = 18.46% 112,000

Expected loss = PD × 35% × 8,000,000

At grade 5 the chance of default across a five-year loan is close to one in ten, not one in fifty. When falling passenger numbers take the operator to grade 6, the loan, the security and the margin are unchanged, and the expected loss has doubled. The rating migration page shows what the same move does across a whole book.

Where does 2.0% come from? From a history like this one:

Year Borrowers in grade 5 Defaults Default rate

1 400 3 0.75%

2 420 5 1.19%

3 410 14 3.41%

4 390 11 2.82%

5 400 7 1.75%

Average of the five years ≈ 2.0%

No single year was 2.0%. Years 3 and 4 were a downturn, and they are why a capital PD is taken over a long run rather than from last year.

// 04 — HOW IT DIFFERS FROM THE FIGURES BESIDE IT

The default rate is what happened; probability of default is what the bank expects. A risk grade is a label, and the PD is the number attached to it. An agency rating is an outside opinion with its own definition of default, and a bank maps it to its own scale rather than adopting it. Expected loss is the product of three factors, of which PD is one.

The same borrower can also carry two PDs in the same bank. The capital figure is deliberately stable. The accounting figure, under IFRS 9 and its New Zealand and Australian equivalents, is forward-looking and moves with the economy, measured over twelve months for a performing loan and over the remaining life once credit risk has risen significantly.

DEFINED IN ONE PARAGRAPH → · EXPECTED LOSS, WORKED → · THE IN-TRAY, ONE MINUTE →

// 05 — WHERE IT DECIDES A FILE

Much of a file follows from the grade: who has authority to approve, how often the file is reviewed, what margin is needed to cover the expected loss and the capital. A borrower's arguments about its prospects are arguments about the grade, so the useful place to contest it is in the inputs, by showing cash flow, contract cover or a history the model did not see.

Bankers can override a model, and for banks on internal ratings the Basel framework requires overrides to be identified and their performance tracked. A run of upward overrides in one sector is telling a credit team something about the model or about the sector. A downgrade is the other pressure point: at renewal, a lower grade can reopen margin, tenor and limit on a loan that has done nothing wrong yet.

RISK-WEIGHTED ASSETS, WORKED → · HOW BANKS PRICE BUSINESS LOANS →

// SOURCES

The documents the statements on this page rest on. Each was opened and checked when cited.

Basel Committee on Banking Supervision, The Basel Framework, CRE32 and CRE36 — paragraphs 32.4, 36.12, 36.44, 36.63 and 36.79 — The 0.05% floor, one borrower grade per borrower with two exceptions, tracking of overrides, and PD as a long-run average of one-year default rates over at least five years — sections 01 and 02.

// QUESTIONS PEOPLE ASK

How is probability of default calculated?: For a mid-market borrower, a bank's rating model combines financial ratios, industry and qualitative judgement into a grade, and the grade carries a percentage calibrated to the default history of borrowers in it. Statistical models such as logistic regression are common underneath. For large listed issuers, an analyst can also infer a figure from credit default swap prices or share-price models. Either way, the result is an estimate that depends on the definition of default used.
What is the difference between point-in-time and through-the-cycle PD?: A point-in-time PD reflects current and expected conditions, so it rises in a recession and falls in a boom. A through-the-cycle PD is averaged over a full economic cycle and moves far less. Internal-ratings capital rules lean towards the long-run view so that capital does not swing with every turn of the economy. Expected credit loss accounting needs the forward-looking view, so banks commonly adjust one to produce the other.
Is probability of default the same as a credit rating?: No, though they are closely related. A rating is a grade on a scale, published by an agency or assigned inside a bank. A probability of default is a percentage, and a bank's master scale attaches one to each of its grades. Agencies publish how often each grade has defaulted historically, which is a record rather than a stated probability, and banks calibrate their internal grades to their own default experience. Two institutions can give the same company the same grade and attach different percentages to it, because their definitions and histories differ.
Can two loans to the same company have different probabilities of default?: Usually not. Default is an event at the borrower, and the Basel framework requires separate exposures to one borrower to carry the same borrower grade, with exceptions for country transfer risk and for guarantees reflected in an adjusted grade. Differences between the loans, such as security or ranking, are captured in loss given default instead. Outside the corporate book the position differs: retail exposures can be defaulted and rated facility by facility.

// NEXT TO IT

// ON WHOSE DESK

The seats that read this term in an ordinary week. Each page says what the seat is for, who arrives in it, and the shape of the round it runs.

Credit risk analystSets and monitors the standard the lending line works to — policy, appetite, ratings and the models underneath them.
Portfolio managerOwns the credit on a book after the money is out — reviews, covenants, ratings and early warning.
Model validatorIndependently checks that the bank's models do what they claim, before anyone relies on them.

THE IN-TRAY · ONE MINUTE

Make the call on one file.

A short file on the ground this page covers. Every option opens with how far it sits from a senior banker’s read.

SIGNAL OR NOISE · THE BANK · ASSET QUALITY

A regional bank's three largest borrowers are all dairy processors. Each one, read alone, is sound.

Signal or noise?

5 MORE ON THE BANK · ASSET QUALITY →WHERE DO YOU READ FROM? THE PLACEMENT →

THE READING ROOMIN ONE PARAGRAPHProbability of default (PD) What is probability of default in credit risk?