# Credit risk provisioning: modeling expected losses through the cycle
A bank reports a bad quarter. Earnings drop sharply. Yet when you read the disclosures, almost no loans have actually defaulted. Borrowers are still paying. So what happened?
The answer is provisioning. Under today's accounting rules, banks must recognize expected losses on loans *before* those losses arrive. When the economic outlook darkens, that forward-looking estimate rises, and the increase hits the income statement immediately. This lesson shows you how that number is built and why it can swing so hard.
Before the 2008 crisis, banks used an "incurred loss" model. They booked a provision (a reserve set aside against future losses) only once there was objective evidence a loan had gone bad. The problem: reserves were "too little, too late." Losses piled up during the downturn exactly when banks were weakest.
Regulators responded with two forward-looking standards:
Both share the same core idea: estimate losses you *expect* over the life of the loan, and reserve for them now. The Bank for International Settlements has a clear primer on the supervisory guidance behind this shift.
Expected loss is built from three inputs. Memorize these; they are the language of credit risk.
The basic formula:
Expected Loss = PD x LGD x EAD
Take a single corporate loan.
Expected loss for the year:
0.015 x 0.40 x 10,000,000 = $60,000
That $60,000 is the provision. It sits on the balance sheet as an allowance and reduces reported earnings, even though the borrower is paying on time.
IFRS 9 sorts every loan into one of three "stages." This is the mechanism that makes provisions lurch during downturns.
Stage 1: performing. The loan is healthy, or credit risk has not increased significantly since origination. You reserve for expected losses over the *next 12 months* only.
Stage 2: significant increase in credit risk (SICR). The borrower has not defaulted, but risk has jumped meaningfully (for example, a two-notch downgrade, or falling 30 days past due). Now you must reserve for expected losses over the *entire remaining life* of the loan.
Stage 3: credit-impaired. Default has occurred or is highly likely. Lifetime losses, and interest income is recognized differently.
The critical jump is Stage 1 to Stage 2. Nothing has defaulted. But the provision horizon explodes from 12 months to the full loan life. For a 5-year loan, that can multiply the reserve several times over.
CECL is structurally simpler: it uses lifetime expected losses for *all* loans from day one, no staging. But it shares the same forward-looking DNA, and it is just as sensitive to the economic outlook.
Here is the scene from the opening, made concrete.
Imagine a bank holds that $10m loan in Stage 1. Life-of-loan is 4 years. Suddenly the macro forecast worsens: unemployment is projected up, the borrower's sector weakens, and internal ratings slip enough to trigger SICR.
The loan migrates to Stage 2. Now assume the lifetime PD (across 4 years) is 6%, not the 1.5% one-year figure.
Lifetime expected loss:
0.06 x 0.40 x 10,000,000 = $240,000
The provision jumps from $60,000 to $240,000. That $180,000 increase is booked as an expense *this quarter*. Multiply across a whole loan book, and you get the earnings hit that appears before a single default.
Both standards require point-in-time estimates that reflect current and forecast conditions, not long-run averages. This is what makes provisions "procyclical," meaning they move with the economy.
Banks build macroeconomic scenarios (base, upside, downside) and weight them. PDs and LGDs are conditioned on variables like GDP growth, unemployment, and house prices.
When the outlook sours, three things happen at once:
1. PDs rise across the book.
2. LGDs rise (collateral values, like property, fall in a downturn, so recoveries shrink).
3. Loans migrate from Stage 1 to Stage 2, extending the loss horizon.
All three push provisions up together. This is why bank earnings can fall fast at the *start* of a recession, well ahead of actual defaults.
Because the forecast drives the number, provisioning involves heavy management judgment. During the sharp uncertainty of early 2020, many banks used post-model adjustments (also called overlays): manual add-ons where the models could not capture unprecedented conditions. These overlays are disclosed and closely watched by auditors and supervisors, because they are a place where earnings can be smoothed.
🎬 [VIDEO: "IFRS 9 Expected Credit Loss Model Explained" — youtube.com — a concise walkthrough of the three-stage model and ECL calculation for non-specialists]
Here is the logic as a short snippet. It shows how staging changes the horizon, not the arithmetic.
def expected_loss(pd, lgd, ead):
return pd * lgd * ead
ead = 10_000_000
lgd = 0.40
# Stage 1: 12-month PD
stage1 = expected_loss(0.015, lgd, ead) # 60,000
# Stage 2: lifetime PD after SICR trigger
stage2 = expected_loss(0.06, lgd, ead) # 240,000
provision_increase = stage2 - stage1 # 180,000 hit to earningsThe formula never changes. What changes is which PD you feed it: a 12-month number in Stage 1, a lifetime number in Stage 2. That single switch is the engine of provision volatility.
Knowledge check
1. A bank reports a sharp drop in earnings for the quarter, yet its disclosures show that almost no loans have actually defaulted and borrowers are still paying. What most likely explains this outcome?
2. Why did regulators move away from the 'incurred loss' model after the 2008 crisis?
3. A mortgage backed by strong collateral is assigned a much lower LGD than an unsecured credit card. What does this difference primarily reflect?
4. Select ALL correct answers about the shared principles of IFRS 9 and CECL.
Select all the correct answers.
5. Select ALL correct answers about the three ingredients of expected loss (PD, LGD, EAD).
Select all the correct answers.
When you review a bank's results, three lines tell the story.
Provision expense (or "cost of risk"). The charge for the period, often quoted in basis points of loans. A rising cost of risk signals the bank sees trouble ahead. One basis point is 0.01%.
Stage 2 balances. Watch the percentage of loans in Stage 2. A rising Stage 2 share is an early warning: risk is building even before defaults show up in Stage 3.
Coverage ratio. Allowances divided by gross loans. Higher coverage means a more conservative reserve. Compare it against peers and across quarters.
Because so much depends on assumptions, two banks holding similar portfolios can report different provisions. Differences come from:
This is why provisioning is a genuine analytical skill, not a mechanical output. The number reflects a view of the future, and views differ.