Banks don’t guess when they set aside money for loans that might go bad. They book a provision for credit losses, a line item that moves before a single account charges off. Under the current expected credit loss (CECL) standard, issuers must estimate lifetime losses the moment a new credit card account is opened. That estimate gets updated every quarter, pulling in unemployment forecasts, wage data, and the bank’s own early-stage delinquency buckets. The result is a forward-looking signal that often turns three to six months before the headline delinquency rate peaks. Right now, that signal is flashing amber—not red—and the shape of the build tells us more than the absolute level.
This piece walks through how loss provisions are constructed, what the latest Q4 2024 and Q1 2025 earnings calls revealed, and why the relationship between the allowance coverage ratio and 30+ day past-due rates matters more than any single number. If you track consumer credit, this is the plumbing you need to understand.
The Provision Engine: CECL, Vintage Curves, and the 30-Day Lag
To read the signal, you have to know how it’s generated. CECL forces issuers to forecast lifetime losses for every loan they book. That forecast isn’t a static percentage. It’s a dynamic model that ingests macroeconomic scenarios and updates quarterly. The provision expense on the income statement is the change in that allowance, plus actual net charge-offs. When an issuer says it “built reserves,” it means its models are predicting higher future losses, and it’s recognizing that cost today.
The most sensitive input is the vintage performance of recent originations. A credit card account opened in Q3 2023 will have a distinct loss curve compared to one from Q3 2021. Issuers slice their portfolios by vintage, FICO band, and channel—direct mail, digital, balance transfer—to see which cohorts are deteriorating first. Typically, 30-day delinquencies in a vintage start to rise two to three quarters before charge-offs. The provision follows the delinquency signal, not the charge-off. That’s why the allowance coverage ratio—the reserve balance divided by total loans—is a better leading indicator than the charge-off rate itself.
Consider the math. If an issuer holds a 10% allowance on a $100 billion card portfolio, that’s $10 billion set aside. If early-stage delinquencies in the 2023 vintage jump 50 basis points in a single quarter, the model will demand a higher reserve for that entire vintage, not just the delinquent accounts. The provision expense spikes, and the coverage ratio climbs. The actual charge-offs may not appear for another six months. This is the mechanism that turned JPMorgan Chase’s card provision from $1.4 billion in Q3 2022 to $2.2 billion in Q4 2024, even as net charge-offs remained within historical norms.
Reading the Q4 2024 Earnings Prints: A Divergence Story
The most recent earnings season revealed a split between prime and subprime issuers that’s worth dissecting. Synchrony Financial, which skews toward store-branded cards and lower-FICO borrowers, increased its allowance coverage ratio to 10.8% in Q4 2024, up from 10.2% a year earlier. Its 30+ day delinquency rate on the core portfolio hit 4.7%, above the pre-pandemic 2019 average of 4.3%. Management cited pressure in the 660-700 FICO band, a cohort that had been resilient through 2023. The provision for credit losses rose 18% year-over-year, outpacing loan growth of 9%. That gap is the signal: reserves are building faster than the book, implying the model sees a higher loss content in new vintages.
Contrast that with American Express, whose cardmember loans are concentrated in the super-prime space. Amex’s consolidated provision was $1.8 billion in Q4 2024, up from $1.5 billion a year prior, but its allowance coverage ratio actually ticked down to 2.6% from 2.7%. The difference is loan growth: Amex’s cardmember loans grew 14% year-over-year, absorbing the higher dollar provision. The write-off rate for the U.S. consumer segment was 2.3%, still below 2019 levels. Amex isn’t seeing the same early-delinquency pressure in its high-FICO vintages. The provision build is a function of portfolio expansion, not credit deterioration.
This divergence tells a layered story. The aggregate credit card delinquency rate, as reported by the Federal Reserve, ticked up to 3.23% in Q4 2024, but that headline masks a bifurcation. Prime and super-prime borrowers are managing; near-prime and subprime are straining. The provision data from individual issuers confirms it. For readers tracking rate sensitivity, this connects directly to how a rate hold affects different tranches of card borrowers—a topic we unpacked in What a Rate Hold Actually Means for Credit Card Borrowers.
Allowance Coverage Ratio as a Leading Indicator
The allowance coverage ratio deserves its own spotlight. When it rises, it means the issuer expects a higher percentage of its outstanding balances to eventually charge off. Because CECL requires a lifetime loss estimate, the ratio embeds assumptions about the entire economic cycle. A ratio that climbs for three consecutive quarters—as it has for Capital One’s domestic card portfolio, moving from 7.8% in Q2 2024 to 8.4% in Q4 2024—suggests the model sees sustained deterioration, not a one-quarter blip. Capital One’s 30+ day delinquency rate was 4.5% in Q4, but the rising coverage ratio implies the model expects those delinquencies to cure at a lower rate than in 2023. That’s a subtle but critical shift: it’s not just about more accounts going bad; it’s about fewer bad accounts recovering.
Historically, a coverage ratio that rises while the unemployment rate is below 4.5% has preceded a delinquency peak by roughly two quarters. The logic is straightforward: issuers build reserves when the labor market is still tight because they know the marginal borrower is already stretched. By the time unemployment actually rises, the reserves are already in place, and the provision expense can actually fall even as charge-offs rise. That pattern played out in 2018-2019 and appears to be repeating.
The Data Trail: What Public Filings Reveal
Publicly traded issuers disclose detailed vintage data in their 10-K filings, usually in the “Loan Portfolio Analysis” section. For the 2024 fiscal year, several trends stand out. First, the 2022 vintage is performing worse than the 2021 vintage at the same point in its lifecycle across all FICO bands. At Capital One, the 2022 vintage had a cumulative gross charge-off rate of 6.8% after 24 months, compared to 5.1% for the 2021 vintage. The 2023 vintage is tracking closer to the 2022 cohort than the 2021 cohort, suggesting the normalization isn’t yet complete.
Second, the “delinquency roll rate”—the percentage of accounts moving from current to 30 days past due—has been elevated since mid-2023. For Discover Financial Services, the roll rate for accounts with FICO scores below 660 hit 8.2% in Q4 2024, up from 6.9% a year earlier. That’s the highest level since Q1 2020, but the context is different: in 2020, stimulus checks and forbearance programs artificially suppressed roll rates. Today, there’s no such backstop. The roll rate is a pure read on household cash flow, and it’s saying that lower-income households are running out of buffer.
Third, the “payment rate”—the percentage of balances paid down each month—has been declining since early 2023. At Synchrony, the payment rate fell to 18.2% in Q4 2024 from 19.5% a year earlier. A falling payment rate means borrowers are revolving more of their balances, which increases interest income for issuers but also signals that households are relying more on credit to manage expenses. When the payment rate drops and the roll rate rises simultaneously, it’s a classic setup for higher charge-offs six to nine months later.
What the Provision Build Tells Us About the Next Six Months
If we map the current provision trajectory against historical patterns, a few conclusions emerge. First, the pace of reserve building is decelerating. The aggregate allowance coverage ratio for the top six card issuers rose from 7.8% in Q3 2024 to 8.1% in Q4 2024, a smaller increase than the 40-basis-point jump seen between Q2 and Q3. This suggests the models are approaching their peak loss estimates. If the labor market holds steady, provisions could plateau by mid-2025, and the allowance coverage ratio could begin to decline as charge-offs eat into the reserve without new provision builds.
Second, the composition of provisions is shifting. In early 2024, most of the build was driven by higher expected losses on new originations. By Q4 2024, a growing share came from “qualitative adjustments”—management overlays that reflect uncertainty about the economic outlook. These overlays are inherently more volatile and can reverse quickly if conditions stabilize. For example, if the Federal Reserve signals a rate cut in response to softening labor data, issuers may reduce their qualitative reserves, creating a provision release that boosts earnings even as charge-offs remain elevated. This dynamic is already priced into bank stocks, which is why card issuers have outperformed regional banks in recent months.
Third, the timing of peak delinquencies depends on the vintage. The 2022 vintage will reach its peak loss period in mid-2025, while the 2023 vintage will peak in early 2026. If the economy avoids a recession, the 2024 vintage—which was underwritten more conservatively—could perform better than both, creating a “wave” pattern where aggregate charge-offs rise, fall, and then rise again as different vintages mature. This isn’t a double-dip; it’s the natural lifecycle of a card portfolio that grew aggressively in 2022-2023 and then tightened underwriting in 2024.
FAQ: Credit Card Loss Provisions and Delinquency Timing
What is the difference between a provision and a charge-off?
A provision is an accounting entry that reduces earnings today to reflect expected future losses. A charge-off is the actual removal of an uncollectible balance from the books, typically after 180 days past due. Provisions lead charge-offs by three to six months because they are based on early delinquency signals and economic forecasts. When you see a provision spike, charge-offs are likely to follow. When provisions decline while charge-offs remain high, it means the issuer believes the worst of the deterioration has already been reserved for.
Why do issuers build provisions even when unemployment is low?
Under CECL accounting, issuers must estimate lifetime losses using “reasonable and supportable” forecasts that incorporate economic scenarios. Even with a 4% unemployment rate, if the model assigns a 30% probability to a recession scenario, the weighted-average loss rate will be higher than the current charge-off rate. Additionally, issuers track early-stage delinquencies and payment rates that deteriorate before unemployment rises. The provision is a reflection of those leading indicators, not the current unemployment rate.
How can I track provision trends for specific issuers?
Quarterly earnings releases and 10-Q filings include the income statement line “provision for credit losses” and the balance sheet line “allowance for credit losses.” The allowance divided by total loans gives the coverage ratio. Most large issuers also disclose delinquency roll rates, payment rates, and vintage performance in their investor presentations. The Federal Reserve’s G.19 report provides aggregate data on credit card charge-offs and delinquencies, but it lags issuer reports by one to two months. For the timeliest signal, monitor the earnings calls of Synchrony, Capital One, and Discover, as their portfolios are most sensitive to consumer stress.
What does a rising allowance coverage ratio actually predict?
A rising coverage ratio predicts that net charge-offs will increase over the subsequent two to four quarters. The magnitude of the increase depends on the vintage composition of the portfolio. If the coverage ratio rises because of deterioration in recent vintages, the charge-off peak will be higher and arrive later than if the increase is driven by qualitative adjustments. The coverage ratio is most useful when compared across issuers and against each issuer’s own historical range. A ratio above the 2015-2019 average suggests the issuer is pricing in a recessionary scenario, even if one hasn’t materialized.
What This Means for the Consumer Credit Cycle
The provision data from Q4 2024 points to a consumer that is bifurcated but not broken. Prime and super-prime borrowers continue to manage their debt loads, supported by low unemployment and real wage growth. Near-prime and subprime borrowers are showing clear signs of stress, with rising delinquencies and falling payment rates. The aggregate numbers will likely show further deterioration through mid-2025, but the pace is slowing, and the peak may be lower than the 2019 cycle for prime-focused issuers.
For readers who follow the mechanics of consumer finance, the key takeaway is this: the provision isn’t a prediction of doom. It’s a mechanical response to data that’s already visible in the delinquency rolls. The signal is in the rate of change, the vintage splits, and the qualitative overlays. Watch the coverage ratio, the roll rates, and the payment rates. They’ll tell you when the cycle is turning before the charge-off headlines catch up.
