Follow the Money: How a Card Portfolio Produces a Return
The subject here is the income statement of a specialized credit card bank: not the APR printed on one offer, but the way a large portfolio turns assets into pretax earnings. The question is which streams explain the rise in reported profitability from 2023 to 2024. Keeping every stream on the same average-asset denominator gives readers a practical way to see what increased, what offset it, and why revenue is not the same thing as profit.
The central result is easy to state and easy to misread. In 2024, credit card banks reported a 3.87% pretax return on average quarterly assets, compared with 3.31% in 2023 after a revision. That outcome was produced by several lines moving together: interest income grew faster than interest expense, provisions for loan losses declined, and net noninterest income stayed approximately flat. Following those movements is more informative than asking whether a single rate or a single loss statistic “explains” the business.
Start with the denominator: what a return on assets means
A credit card bank holds a portfolio of assets: balances that customers have not repaid, along with the other exposures recorded on its balance sheet. The Federal Reserve measures pretax income before extraordinary items as a percentage of average quarterly assets. The denominator matters because it puts income and expenses on a common scale. A bank with a larger balance sheet may collect more dollars while still producing a lower return relative to the assets required to generate those dollars.
This is a bank-level measure, not a consumer-level price. If a card carries a 25% purchase APR, that rate describes the contractual price of borrowing under specified conditions. It does not say how much the issuing bank will retain after paying for deposits or other funding, servicing, marketing, fraud controls, rewards, staff, technology, compliance, and credit losses. Conversely, a 3.87% return on assets does not mean the bank charges 3.87% to borrowers. The two percentages belong to different parts of the system and use different denominators.
The Federal Reserve’s definition also limits the population being measured. A “credit card bank” in this report is a bank with more than 50% of its assets in consumer lending and at least 90% of that consumer lending in credit cards or related plans. Eight banks met that definition at the end of 2024. Their reported results provide a concentrated view of card operations, but diversified banks that issue cards without specializing in them may have different customer mixes, cost structures, pricing behavior, and profitability.
Within that population, the return can be understood as a sequence of offsets. Interest income is the largest positive stream in the reported table. Interest expense is the cost of funding the bank’s earning assets. Subtracting the latter from the former produces net interest income. Noninterest income adds fees and other revenue that are not classified as interest, while noninterest expenses capture costs outside interest paid. Provisions for loan losses recognize the expected cost of credit deterioration under the accounting framework. The remainder is the pretax return on average assets.
The relationship is not a claim that these categories are independent. A pricing decision can affect interest income, borrower behavior can affect both balances and losses, and a larger account base can bring both revenue and servicing work. The point of the decomposition is to identify where each effect appears in the reported result. It prevents the common shortcut of treating all money collected from customers as profit or all money set aside for losses as a complete measure of the bank’s risk.
The denominator also helps separate a flow from a stock. Interest income, interest expense, operating results, and provisions accumulate over the reporting period, while assets are the resources supporting the portfolio. Using average quarterly assets rather than only the balance on the last day of the year makes the comparison less sensitive to a single end-of-period snapshot. It still does not answer every question about efficiency: two banks can report the same return on assets with different funding mixes, customer profiles, or costs. It provides a common measuring surface, not a complete explanation.
There are also interfaces between the lines. The customer’s unpaid balance is an asset for the bank and a liability for the customer. The bank’s pricing and credit-limit decisions influence the balance, while the bank’s funding conditions influence the cost of carrying it. If payment behavior worsens, the effect may first appear in delinquency data, later in charge-offs, and through accounting estimates in provisions. The income statement records the combined outcome after these processes interact; it does not show the full chain inside every account.
The Federal Reserve’s table makes the scale of the streams visible. In 2024, total interest income equaled 13.94% of average quarterly assets. Total interest expense was 3.78%, leaving net interest income of 10.16%. Net noninterest income was negative 2.69%: noninterest income of 6.34% was smaller than noninterest expenses of 9.03%. Provisions for loan losses were 3.62%. After these offsets, the reported return was 3.87%.
A reader should not force those rounded figures into an exact identity. Adding 10.16% and negative 2.69%, then subtracting 3.62%, produces a result close to, but not exactly, 3.87%. The difference is consistent with rounding and smaller items in the reported measure. The important fact is structural: a large net-interest stream can coexist with substantial operating costs and provisions and still leave a positive return.
That structure also explains why a bank can have a high return even when one expense category is large. Provisions at 3.62% of average assets are not trivial. Neither is a negative net noninterest contribution of 2.69%. But profitability is the residual after all relevant streams are combined. A large subtraction does not answer whether the final residual is positive; it must be interpreted relative to the income available to absorb it.
The same logic works in reverse when conditions turn less favorable. If provisions rise while every other line stays fixed, the reported return falls. If interest expense rises faster than interest income, net interest income narrows even if customers are still paying interest. If noninterest expenses grow faster than noninterest income, the negative contribution from operations becomes larger. These are accounting implications of the bridge, not forecasts about what will happen next. They show why the phrase “profitable despite losses” is shorthand for a relative relationship: the question is whether the portfolio’s positive streams are expanding, stable, or shrinking compared with the costs attached to them.
The bridge also clarifies what a headline comparison cannot tell us. A higher return might reflect a stronger revenue environment, a lower expected-loss estimate, a different balance mix, or some combination. The aggregate table records the result after those forces meet. It is therefore a starting point for investigation, not a bank-by-bank explanation of which decision created each percentage point.
The diagram is a map of accounting relationships, not a claim that the denominator itself causes every line. Average assets are the reference scale; the streams are the income and expenses measured against that scale. Credit-card operations are profitable only to the extent that the positive streams exceed the negative ones during the period being measured.
What changed between 2023 and 2024
The year-over-year comparison shows why the result improved. Total interest income rose from 13.30% of average quarterly assets in 2023 to 13.94% in 2024, an increase of 0.64 percentage points. Total interest expense also rose, from 3.38% to 3.78%, but by 0.40 points. Because income grew faster than the corresponding expense, net interest income rose from 9.92% to 10.16%, a 0.24-point improvement.
That is the first part of the answer to the profitability puzzle. A rising interest-expense line does not necessarily mean the spread is shrinking. It depends on the relative movement of the two lines. If interest income increases by more than the cost of funding, the net contribution from interest can still improve. The Federal Reserve says that this is what happened in 2024, describing interest income as expanding slightly faster than interest expenses amid rising interest rates.
The second part is provisioning. Provisions for loan losses fell from 3.92% of average quarterly assets in 2023 to 3.62% in 2024, a decline of 0.30 percentage points. In the return calculation, that decline works in the opposite direction from a rise in expenses: it leaves more of the other income streams available in the reported result. Combined with the 0.24-point improvement in net interest income, it explains why the return could rise even though credit stress remained visible elsewhere in the data.
Noninterest items did not provide a hidden second engine. Noninterest income increased slightly, from 6.28% to 6.34%, while noninterest expenses rose from 8.96% to 9.03%. Their net effect remained approximately the same: negative 2.68% in 2023 compared with negative 2.69% in 2024. This matters because it keeps the explanation disciplined. The Federal Reserve did not attribute the improvement to a sudden transformation in every fee or operating line. It pointed mainly to higher interest income and lower provisioning, partly offset by higher interest expense.
It is tempting to convert the table into a story about pricing alone. That would go beyond the evidence. The Federal Reserve’s figures show the aggregate interest-income line, but they do not identify how much of its increase came from higher rates, larger balances, or a change in the composition of accounts. They also do not match one product’s APR with that product’s funding, servicing, rewards, fraud, and loss costs. A card’s posted price and a bank’s portfolio return are connected, but the table does not provide a product-level profit calculation.
Nor should the 13.94% interest-income figure be described as a 13.94% margin. The bank must pay 3.78% in interest expense before arriving at net interest income. It must then absorb the net cost of noninterest operations and loan-loss provisions. The fact that total interest income is large is important because it supplies room for those deductions; it is not evidence that the deductions are unimportant.
Fees fit into the same logic. Credit-card pricing includes more than purchase interest: annual fees, cash-advance and balance-transfer fees, rebates, minimum finance charges, over-limit fees, late-payment charges, and the grace period can all affect what customers pay and what issuers receive. But the presence of a fee does not tell us its net contribution. A fee may come with rewards, servicing, acquisition, or compliance costs, and the available aggregate table does not separate those economics by product. Noninterest income and noninterest expenses must therefore be read together.
This is especially important when comparing a card’s terms with a bank’s return. A customer may focus on the purchase APR because it is easy to quote, while the bank manages a bundle of prices, costs, and risks over the life of an account. A product with a lower advertised rate could have a different fee structure or customer mix; a product with a fee could attract payment activity or benefits that are not visible in the fee alone. The Federal Reserve’s aggregate categories are broad enough to show the system’s direction, but not narrow enough to rank individual products by profitability.
The timing of provisions adds another reason to avoid a one-line explanation. A provision is an accounting recognition of expected loan losses, not a synonym for every delinquency or charge-off occurring in the same month. Its level can reflect expectations about future losses as well as current portfolio performance. In 2024, the Federal Reserve reported that delinquencies rose in the first half of the year and then ticked down in the second half, while charge-offs remained elevated. Those facts can coexist with a lower annual provision rate because the measures describe different moments and use different accounting concepts.
The comparison is strongest for the 2023–24 movement, but even there the report includes qualifications. The 2023 return was revised from 3.33% to 3.31% because of Call Report revisions. More broadly, the Federal Reserve warns that changes in accounting rules affect comparisons across time: the consolidation of previously off-balance-sheet items in 2010 makes earlier and later profitability measures not necessarily comparable, and the adoption of the current expected credit losses methodology in 2020 changed how provisions are estimated. The report also notes that sample composition can change as banks reorganize, although no included-bank changes occurred between 2023 and 2024.
Those limits do not weaken the immediate conclusion. They tell us what kind of conclusion is justified. The data support saying that the defined credit card bank sample earned a higher pretax return in 2024 because interest income rose faster than interest expense and provisions declined, while net noninterest income was roughly stable. They do not support saying that every issuer experienced the same change, that rates alone caused it, or that the accounting result forecasts the next credit cycle.
The income statement also leaves an important question unanswered: why is the interest-income stream so large, and why is credit exposure not evenly distributed across all cardholders? The answer lies in the difference between using a card as a payment device and carrying a revolving balance. Once that borrower mix is visible, the portfolio-level numbers become less abstract. One can see not only how the streams add up, but also which account behaviors place money and risk into them.