Guide

Loan loss provisions and CECL explained

Harbor Credit Union's wholesale desk tracked a $2.4B regional lender through 2024 with green lights on every screen: five-year CDS near 90 bp, a passing CCAR scorecard, and CET1 at 10.4%. What the desk missed was the allowance trajectory. The bank's allowance for credit losses (ACL) rose from 1.05% of loans to 1.48% in three quarters — almost entirely from forward-looking CECL builds on commercial real estate, not from loans that had already defaulted. Net charge-offs stayed modest at 0.32% annualized. Earnings fell 18% on provision expense alone while capital ratios looked fine. Harbor kept exposure limits flat until CDS widened six weeks later.

Loan loss provisions are the quarterly expense that adjusts the allowance for credit losses on the balance sheet. Under CECL (Current Expected Credit Loss, ASC 326 in the U.S., effective 2020 for most institutions), banks must estimate lifetime expected credit losses at origination and update them each period as forecasts change — not wait for a loss to be “probable.” Harbor rebuilt its counterparty watchlist around ACL coverage ratios, provision-to-net-interest-income, and CRE concentration overlays alongside capital metrics. False-positive downgrades on healthy-but-provisioning banks fell from 38% to 14% over two quarters. This guide covers the incurred-loss vs CECL shift, lifetime ECL mechanics, staging and macro scenarios, purchased credit deteriorated (PCD) loans, the Harbor refactor, a technique decision table, pitfalls, and a production checklist.

From incurred loss to lifetime expected credit loss

Before CECL, U.S. GAAP largely followed an incurred loss model: you booked a reserve only when a loss was probable and estimable. That meant allowances often lagged deteriorating credit cycles — reserves stayed thin until defaults spiked, then banks rushed to catch up in a few painful quarters. The 2008 crisis exposed this procyclical lag: earnings and capital looked stable until charge-offs arrived in bulk.

CECL flips the timing. At loan origination (or acquisition), the bank estimates lifetime expected credit losses over the contractual term, adjusted for expected prepayments and extensions. Each quarter, management revises that estimate using current conditions and reasonable, supportable forecasts. A worsening macro outlook can increase the ACL even when no loan has missed a payment yet — exactly what Harbor's counterparty missed.

Provision expense = change in ACL (plus charge-offs net of recoveries flow through the allowance, not always through provision in the same line on all disclosures — read the reconciliation table in 10-Q/K).
ACL coverage ratio = ACL ÷ total loans (or by segment: CRE, C&I, consumer). Rising coverage with flat charge-offs signals forward-looking pessimism.
Reserve release = negative provision when forecasts improve; boosts earnings but can mask underlying credit softening if charge-offs are still rising.

IFRS 9 uses a similar expected-loss framework with three stages; CECL does not use explicit stages for most instruments but still distinguishes collateral-dependent and troubled debt restructuring (TDR) measurement. Investors comparing U.S. banks to European peers should normalize stage-2 builds on IFRS banks to ACL coverage, not headline provision swings alone.

How banks estimate CECL: PD, LGD, EAD and macro overlays

CECL does not mandate one model. Banks use discounted cash flow, loss-rate methods, vintage analysis, or probability-of-default (PD) × loss-given-default (LGD) × exposure-at-default (EAD) frameworks. The regulator expects documentation, governance, and data that support “reasonable and supportable” forecasts — typically 1–2 years of explicit macro paths, then reversion to long-run trends.

Historical loss rates — roll-rate or vintage curves by product and risk grade; fast to implement but slow to react to new risks (e.g., office CRE post-pandemic) unless overlaid.
PD/LGD models — econometric scores tied to unemployment, GDP, property prices, and sector indices; flexible but model-risk heavy.
DCF for collateral-dependent loans — when repayment depends on collateral sale, fair value of collateral minus costs to sell can cap the allowance.
Qualitative factors (Q-factors) — management overlays for concentrations, policy changes, or data gaps not captured in models; scrutinized by auditors and supervisors.

Macro scenarios are where peer banks diverge. Two institutions with identical loan books can report different ACL builds if one weights a soft-landing baseline at 70% and another assigns 40% to a mild recession. Harbor now requests disclosed scenario weights from large counterparties when available in investor presentations, and proxies with provision sensitivity tables when not.

Segments, PCD loans, and earnings vs capital impact

ACL is measured at the pool level for loans sharing similar risk characteristics — not usually loan-by-loan except for large criticized credits. Common segments: residential mortgage, auto, credit card, C&I, owner-occupied CRE, non-owner-occupied CRE, construction, and agriculture. Segment trends matter more than headline ACL: a bank can hold flat total coverage while CRE coverage jumps 40 bp and consumer releases offset it.

Purchased credit deteriorated (PCD) loans — acquired loans with more-than-insignificant credit deterioration at acquisition — get an initial allowance baked into purchase price accounting. Subsequent changes hit provision expense like any other CECL loan. M&A-heavy banks can show volatile provisions in the quarter of close; normalize for PCD when comparing year-over-year.

Provisions reduce retained earnings and therefore flow into regulatory capital over time, but they are not a direct CET1 deduction like goodwill. A large CECL build hurts ROA and ROTCE immediately; CET1 falls only through lower earnings unless dividends are maintained unsustainably. Pair ACL trends with net interest margin: rising provisions plus compressing NIM is a double hit to pre-provision earnings power.

Harbor Credit Union refactor (worked example)

After the 2024 CRE provision surprise, Harbor's counterparty team added ACL-centric fields to its bank monitor:

ACL coverage delta: quarter-over-quarter change in ACL ÷ loans; amber at +15 bp, red at +25 bp without matching charge-off rise.
Provision / NII ratio: provision expense annualized divided by net interest income; red above 35% for two consecutive quarters.
CRE concentration overlay: non-owner-occupied CRE > 300% of capital triggers limit cap regardless of CDS.
Criticized asset trend: classified loans rising > 10% QoQ pairs with ACL review even if pass-rated book is stable.
Reserve release flag: negative provision in a quarter when peer CRE charge-offs are rising triggers manual review — not auto upgrade.

On a 12-bank watchlist heavy in Sun Belt CRE, the refactor cut false-positive downgrades (banks flagged then stabilizing without CDS widening) from 38% to 14% while catching two lenders that built ACL 30+ bp ahead of rating-agency outlook cuts. Harbor's counterparty forecast error on one-year line renewals improved from 19 bp to 7 bp on spread.

Technique decision table: when CECL metrics beat alternatives

Question	Best primary signal	CECL / ACL role	Weak alternative
Will earnings disappoint before capital breaks?	ACL coverage trend + provision/NII	Forward-looking; leads charge-offs	CET1 alone (lags via retained earnings)
Is credit stress sector-specific?	Segment ACL and charge-offs	CRE vs consumer split in 10-Q	Headline net charge-off ratio only
Counterparty default risk (wholesale)	CDS + ACL trajectory + classified loans	ACL builds often precede CDS by weeks	CDS alone in low-liquidity names
Severe tail loss absorption	DFAST/CCAR projected CET1 trough	Stress tests complement CECL base case	CECL base macro (not tail by design)
Acquisition accounting noise	PCD allowance reconciliation	Isolate day-2 provision from M&A	Raw YoY provision compare
Peer relative credit quality	ACL coverage vs charge-offs scatter	High coverage + low NCO = cautious bank	P/E only

Common pitfalls

Treating provision as cash — ACL is an accounting reserve, not money set aside in a vault; liquidity is LCR territory.
Ignoring segment mix — total ACL/loans hides CRE builds offset by card releases.
Comparing CECL to pre-2020 history — level shifts at adoption broke long-run coverage series; use post-CECL vintages.
Assuming reserve release is good news — can reflect optimistic macro reversion while classified assets grow.
Double-counting with stress tests — CCAR losses and CECL base-case ECL are different constructs; do not sum them as one loss number.
Overweighting Q-factors — large qualitative overlays without documentation draw SEC comment letters and auditor pushback.
Missing PCD in M&A quarters — distorts provision growth rates.
Using ACL for fair-value marks — AFS securities unrealized losses hit AOCI and CET1 through other channels, not CECL loan ACL.

Production checklist

Pull ACL reconciliation table from latest 10-Q/K: beginning balance, charge-offs, recoveries, provision, ending balance.
Calculate ACL ÷ total loans and by disclosed segment (CRE, C&I, consumer).
Track QoQ change in coverage (basis points) alongside annualized net charge-off rate.
Compute provision ÷ NII and provision ÷ pre-provision net revenue (PPNR) if disclosed.
Read management discussion for macro scenario assumptions and Q-factor narrative.
Cross-check criticized/classified loan trends and CRE concentration vs capital.
For acquisitions, separate PCD initial allowance from recurring CECL expense.
Pair with CET1 headroom and stress test results for tail risk.
Monitor CDS and bond spread moves as market validation lagging ACL builds.
Maintain peer scatter: ACL coverage vs NCO for 8–12 regional banks in the same footprint.
Flag reserve releases in deteriorating peer charge-off environments for manual review.

Key takeaways

CECL requires lifetime expected credit losses at origination, updated each quarter — provisions can rise before defaults do.
ACL coverage trend and provision-to-NII often lead CDS and rating actions on regional banks.
Segment ACL (especially CRE) matters more than headline allowance ratios.
Harbor Credit Union cut false-positive counterparty downgrades from 38% to 14% by pairing ACL metrics with capital screens.
Stress tests measure tail CET1; CECL measures base-case forward loss — use both, do not conflate.
PCD accounting and reserve releases are common sources of misread earnings and credit signals.