Loan Loss Provisions and CECL: How Banks Account for Credit Risk

Introduction

The provision for credit losses is arguably the most important and most misunderstood line item on a bank income statement. It is the bridge between the income statement and the balance sheet, the primary source of earnings volatility across credit cycles, and the analytical focus point for FIG bankers assessing bank quality, valuation, and M&A feasibility. Understanding how provisions work, how they flow between financial statements, and how the CECL accounting framework changed the mechanics is foundational to FIG analysis.

This article covers the complete framework: what the provision is, how it connects to the allowance on the balance sheet, what drives the provision higher or lower, how CECL differs from the old incurred-loss model, and why all of this matters for FIG valuation and M&A.

Three interconnected concepts define how banks account for credit risk. Understanding their relationship is critical.

The Provision for Credit Losses (PCL) is an income statement expense. It is the charge that management records each quarter to build or adjust the reserve against future loan losses. Like any expense, it reduces pre-tax income and net income. When a bank increases its provision, earnings decline. When a bank releases reserves (a negative provision), earnings increase.

The Allowance for Credit Losses (ACL) is a balance sheet contra-asset. It sits against the gross loan portfolio and reduces it to its estimated net realizable value. If a bank has $100 billion in gross loans and a $2 billion ACL, the net loan figure on the balance sheet is $98 billion. The ACL represents management's best estimate of the lifetime expected credit losses embedded in the loan portfolio.

Net Charge-Offs (NCOs) represent actual losses. When a loan is deemed uncollectable (the borrower has defaulted and collection efforts have been exhausted), the loan is charged off: it is removed from the gross loan portfolio and the corresponding amount is deducted from the ACL. When money is subsequently recovered on a previously charged-off loan, that is a recovery. NCOs equal gross charge-offs minus recoveries.

The ACL Rollforward

The fundamental accounting relationship that connects the income statement provision to the balance sheet allowance: Beginning ACL + Provision for Credit Losses - Net Charge-Offs = Ending ACL. This rollforward is the mechanical link between credit expense on the income statement and the reserve balance on the balance sheet. If a bank starts a quarter with a $2.0 billion ACL, records a $500 million provision, and experiences $400 million in NCOs, the ending ACL is $2.1 billion ($2.0B + $0.5B - $0.4B). The provision builds the reserve; charge-offs draw it down. This framework is one of the most frequently tested concepts in FIG interviews.

The provision serves two purposes within this framework. First, it replenishes the ACL for actual losses that have occurred (matching the NCO drain). Second, it adjusts the ACL for changes in expected future losses based on updated economic forecasts, changes in loan portfolio composition, or shifts in credit quality indicators. A provision that exceeds NCOs results in a reserve build (the ACL grows). A provision below NCOs results in a reserve release (the ACL shrinks).

In 2016, the Financial Accounting Standards Board (FASB) issued ASU 2016-13, which introduced the Current Expected Credit Losses (CECL) methodology. CECL became effective for large public banks in January 2020 and for all remaining banks (including community banks) by fiscal years beginning after December 15, 2023. As of 2024, all US banks operate under CECL.

The Incurred Loss Model (Old Framework)

Under the previous incurred-loss model, banks recognized credit losses only when they were probable and estimable. This meant the bank would wait until clear evidence of impairment existed (a borrower missed payments, a credit rating was downgraded, or economic conditions deteriorated enough to make losses probable) before recording a provision. The loss estimation horizon was typically 12-18 months.

The weakness of this approach was its backward-looking nature: losses were recognized late in the cycle, after they had already begun to materialize. During the 2008 financial crisis, this delayed recognition meant that bank reserves were inadequate heading into the downturn, forcing massive provision spikes that devastated earnings precisely when the economy could least absorb them. The G20 and the Financial Stability Board identified this "too little, too late" provisioning as a systemic vulnerability that amplified the crisis, prompting the development of both CECL in the US and IFRS 9 internationally.

The CECL Model (Current Framework)

CECL replaced the "probable and estimable" threshold with a lifetime expected loss requirement. Under CECL, banks must estimate the total expected credit losses over the remaining life of every loan at origination (or at each reporting date for existing loans). This means:

Key Differences: CECL vs. Incurred Loss

Importantly, CECL does not change the total amount of credit losses a bank will experience over the life of a loan. It changes the timing of when those losses are recognized in the income statement. Losses are front-loaded under CECL: the provision is higher at origination and during forecast deterioration, but lower when actual losses materialize because the reserve was already built.

European and most non-US banks operate under IFRS 9, which was introduced by the International Accounting Standards Board (IASB) and became effective in January 2018, two years before CECL reached large US banks. Like CECL, IFRS 9 replaced an incurred-loss model with an expected credit loss framework, but the implementation differs in ways that matter for cross-border FIG analysis.

IFRS 9 uses a three-stage model that determines how much expected loss is recognized:

The practical difference between CECL and IFRS 9 is how provisions behave at origination and during early deterioration. CECL requires lifetime losses from day one on all loans, creating a larger upfront provision. IFRS 9 starts with only 12-month losses (Stage 1), which is lower, but creates a "cliff effect" when loans migrate to Stage 2 and the bank must suddenly recognize full lifetime losses. During the COVID pandemic, this cliff effect caused significant provision volatility at European banks as large loan pools shifted from Stage 1 to Stage 2 based on deteriorating economic conditions.

Provision expense across the US banking industry illustrates CECL's real-world impact on earnings volatility. In the first half of 2025, US banks reported approximately $22.5 billion in credit loss provisions for Q1 alone, with reserves exceeding net charge-offs by $1.2 billion, indicating continued reserve building. By Q4 2025, industry-wide provisions settled at approximately $20.9 billion for the quarter, with provision expense roughly equaling total net charge-offs, suggesting the industry had reached a steady-state reserve level after the post-pandemic build-release cycle.

The current credit environment shows a bifurcation. Consumer credit (particularly credit cards and auto loans) is showing signs of strain as household balance sheets weaken, with credit card charge-off rates rising above pre-pandemic levels at several large banks. Commercial credit quality remains broadly stable, though community banks with concentrated CRE exposure face elevated risk as office vacancies persist above 20% nationally. This bifurcation creates uneven provision pressure across the banking industry: consumer-heavy lenders (Capital One, Synchrony, Discover before its acquisition) face rising provisions while commercial-focused regional banks maintain relatively benign credit costs.

The provision for credit losses has outsized importance in FIG analysis for three reasons.

Normalized Earnings and PPNR

The provision is the most volatile line on the bank income statement. It can swing from near-zero during benign credit environments to billions of dollars during downturns, overwhelming the relatively stable NII and non-interest income lines. This volatility makes reported earnings an unreliable measure of a bank's underlying profitability.

FIG analysts address this by calculating Pre-Provision Net Revenue (PPNR): total revenue minus non-interest expense, before the provision. PPNR isolates operating profitability from credit cycle noise and is the primary metric for assessing a bank's normalized earnings power. A bank with strong PPNR can absorb elevated provisions during a downturn while still remaining profitable; a bank with weak PPNR may swing to losses even with moderate credit stress.

Credit Quality Assessment

The provision level, in combination with credit quality metrics (non-performing loans, net charge-off rates, coverage ratios), tells FIG analysts whether a bank's credit risk is improving or deteriorating.

Key indicators include:

Stress Testing and Capital Adequacy

The provision is the central variable in bank stress tests. Under the Federal Reserve's Comprehensive Capital Analysis and Review (CCAR), banks must model their provision expense under severely adverse economic scenarios (deep recession, unemployment spike, asset price collapse) to determine whether they can maintain minimum capital ratios through the stress horizon. The projected provision under stress directly determines the bank's Stress Capital Buffer (SCB), which sets the minimum capital the bank must hold above regulatory minimums.

For FIG bankers building bank models or projecting acquisition targets, CECL introduces several modeling considerations.

Forecast sensitivity: Under CECL, the provision is sensitive to macroeconomic forecast assumptions (GDP growth, unemployment, housing prices). Changes in the forecast alone, without any change in actual credit performance, can cause significant provision swings. Models should include scenario analysis showing how different economic forecasts affect provision expense and earnings.

Day-one provision on loan growth: When a bank grows its loan portfolio, CECL requires an immediate provision for the lifetime expected losses on new originations. This means loan growth has a higher income statement cost under CECL than under the incurred-loss model, which recognized losses later. FIG models should capture this day-one cost when projecting provision expense for growing banks.

Qualitative adjustments: CECL allows management discretion in qualitative adjustments to model-derived loss estimates. These adjustments are often the largest single driver of provision changes and create uncertainty in provision forecasting. FIG analysts typically model a base case provision rate (based on historical NCO trends) and layer on scenario adjustments for economic forecast changes.

Portfolio mix effects: Different loan types carry vastly different expected loss profiles. Credit card portfolios have the highest loss rates (NCO rates of 3-6% in normal environments, 7%+ during stress) but also the shortest duration (revolving credit). Residential mortgages have lower loss rates (0.1-0.5%) but much longer duration, meaning the lifetime expected loss under CECL can be significant despite the low annual rate. When modeling a bank acquisition, the provision projection must reflect the target's specific loan mix: acquiring a consumer-heavy lender will carry a structurally higher provision rate than acquiring a commercial bank with a C&I-dominated portfolio.

M&A purchase accounting: In a bank acquisition, the acquirer does not inherit the target's existing ACL. Instead, the target's loans are marked to fair value at close (incorporating a credit discount for expected losses), and the target's ACL is eliminated. The acquirer then builds a new ACL on the acquired portfolio from day one under CECL. This purchase accounting treatment means the provision impact of an acquisition is front-loaded: the acquirer records a day-one provision on the acquired portfolio that flows through its income statement, which must be modeled in the accretion/dilution analysis.