Bank Supervision & Regulation: CAMELS, Capital, and Expected-Loss Accounting

Every model in this track so far measures the risk of a borrower. This module flips the perspective: how do supervisors measure the risk of the lender? Banks fund long, illiquid, risky loans with short-dated deposits at roughly ten-to-one leverage — a structure that works beautifully until confidence wobbles. Module 1 sketched why the Basel framework exists; here we go one level deeper into the machinery: the grading system, the capital arithmetic, the stress tests, and the accounting revolution that made every bank run an expected-loss model.

CAMELS: the supervisor’s report card

After an examination, US regulators assign each bank a confidential CAMELS rating — six components and a composite, each on a 1 (strong) to 5 (critically deficient) scale:

Composites of 1–2 mean routine supervision; 3 puts a bank on the watchlist; 4–5 mean enforcement actions, restrictions, and, at the bottom, resolution planning. The rating is never published — disclosing that a bank is a 4 could itself cause the run it warns about.

Try the intuition version — one headline ratio per component:

Regulatory capital: the actual arithmetic

“Capital” in regulation is not cash in a vault — it’s the slice of the balance sheet that absorbs losses before depositors and senior creditors take a hit. The hierarchy, from most to least loss-absorbing:

• CET1 — Common Equity Tier 1: common shares plus retained earnings, minus deductions (goodwill, most intangibles, certain deferred tax assets). The only capital that absorbs losses seamlessly while the bank operates. When markets ask “how much capital does it have?”, they mean CET1.
• AT1 — Additional Tier 1: perpetual instruments (CoCos) that convert to equity or write down when a trigger breaches. Loss-absorbing, but discontinuously — as their holders periodically rediscover.
• Tier 2: subordinated debt that absorbs losses only in failure (“gone concern” capital).

The denominator is risk-weighted assets. Every exposure gets a weight reflecting its riskiness, and the requirement applies to the weighted sum:

Two regimes compute the weights. The standardized approach reads them off regulatory tables, largely keyed to external ratings — Module 8’s letters translated into capital. The IRB approach lets qualifying banks feed their own estimated PD (and, for the advanced version, LGD and EAD) into a regulator-fixed formula — the Vasicek single-factor model, which is Merton’s Module 6 logic applied to a portfolio (the full derivation is Module 10). Basel III’s “endgame” output floor caps how far IRB results may fall below standardized (72.5% at full phase-in), a direct response to years of suspiciously optimistic internal models.

On top of the 4.5% CET1 minimum sit the buffers: a 2.5% capital conservation buffer (breach it and dividends and bonuses get restricted — a fail-soft, not a failure), a countercyclical buffer (0–2.5%, switched on in booms), and G-SIB surcharges (1–3.5%) for the systemically important. A large global bank’s practical CET1 requirement lands somewhere between 9% and 13%. And because every risk-weighting scheme can be gamed, a non-risk-based leverage ratio (Tier 1 over unweighted exposure, ≥3%, more for G-SIBs) stands behind it as the backstop. Alongside capital, Basel III added the liquidity pair: LCR (survive 30 days of stressed outflows with high-quality liquid assets) and NSFR (fund long-dated assets with stable money) — the L and S of CAMELS, hardened into ratios.

Stress testing: capital under a story

A capital ratio is a photograph. Stress tests ask the question that matters: what does the ratio look like after two bad years? Supervisors publish a macro scenario — deep recession, unemployment spiking, real estate down 25–40%, a market shock for trading books — and banks must project income, losses, and capital quarter by quarter through it. In the US, the Fed’s CCAR/DFAST exercise makes the result binding through the stress capital buffer: the worse a bank’s projected capital depletion, the more capital it must carry today. The EBA/ECB run the European equivalent. Beneath all of it sits this track’s machinery — stressed PDs (scorecards from Module 2, transition matrices from Module 8 conditioned on the downturn), stressed LGDs, revenue models — thousands of them, which is why model risk management became its own discipline (SR 11-7, covered in Module 12).

The deepest value of stress testing isn’t the number — scenarios are always wrong in the details. It’s that the capability forces banks to know, mechanically, how a macro story propagates through their book. The 2009 SCAP exercise arguably ended the US phase of the crisis not by being severe but by being credible: published, bank-by-bank results that let investors finally price the holes.

IFRS 9 and CECL: the expected-loss revolution

Until the late 2010s, loan-loss accounting ran on an incurred loss standard: you provisioned when a loss event had already happened. In 2008 this produced the absurdity of banks reporting healthy earnings while their books were visibly rotting — “too little, too late,” as the post-mortems put it. The replacements — IFRS 9 (international, 2018) and CECL (US, 2020) — flipped the trigger from evidence to expectation: provision for losses you expect, when you book the loan.

IFRS 9 does it in three stages:

CECL is blunter: lifetime expected loss for everything, from day one, no staging.

Two consequences matter for a modeler. First, the accounting explicitly demands point-in-time, forward-looking, scenario-weighted estimates of PD, LGD, and EAD — Module 1’s equation, now embedded in audited financial statements, using PIT measures precisely where Basel capital wants TTC (Module 8’s distinction, running in production on the same portfolio). The Stage 2 trigger — “significant increase in credit risk” — is a migration statement, so transition thinking decides when the cliff hits. Second, expected-loss accounting is procyclical by design: a worsening forecast raises provisions for the entire book at once, exactly when earnings are weakest. COVID, arriving two years into IFRS 9 and weeks after CECL went live, made every bank re-run lifetime losses under a scenario nobody’s models had seen — and regulators promptly allowed capital-relief transition arrangements, an admission of the tension between accounting truth-telling and prudential stability.

Where this connects

• The supervisor’s toolkit is this track’s toolkit pointed at a bank: A is Module 3’s ratio analysis on the loan book, C prices the cushion against Module 1’s unexpected loss, S is a Merton-style question (Module 6) about the balance sheet’s option exposure to rates.
• The IRB capital formula this module keeps gesturing at — regulator-fixed, Vasicek single-factor, Merton underneath — gets derived properly in Module 10.
• IFRS 9’s scenario-weighted lifetime PDs are the industrial-scale application of everything from scorecards to transition matrices; the models that produce them are exactly what SR 11-7 model-risk governance (Module 12) exists to police.

Try it yourself

The field-notes PDF below is the two-page version of this module, formatted for revision: the CAMELS table with score directions, the capital stack as a single diagram-in-words, the buffer arithmetic with a worked example, LCR/NSFR one-liners, and the IFRS 9 staging table with the classic exam traps flagged (12-month EL is not the loss on loans maturing within 12 months; Stage 2 loans are still performing). Print it, annotate it, and take a CAMELS pass at the next bank 10-K you read using the 10-K guide from Module 3.

Get new posts by email

One email per new article. No spam, no upsells, unsubscribe anytime.