A conditional calculation, not a prediction
A credit stress test asks what could happen to a specified portfolio if a specified adverse environment occurred. The Federal Reserve's final 2026 scenario includes unemployment reaching 10%, house prices falling about 30% and commercial real estate prices falling 39%. The published paths begin in the first quarter of 2026 and extend through the first quarter of 2029. These are hypothetical conditions, not the Federal Reserve's economic forecast or measurements of what actually happened during those dates. [1]
That distinction is essential when reading a large dollar loss number. A stress loss is conditional on a constructed set of circumstances and modeling assumptions. It does not assign the scenario a probability, establish the most likely next year's loss, or predict that each individual bank will experience exactly the same shock. A useful result reveals a vulnerability or capacity to absorb losses under the experiment's terms.
From unemployment and prices to individual exposure
Three quantities provide an intuitive starting point: probability of default, loss given default and exposure at default, commonly shortened to PD, LGD and EAD. PD describes the likelihood of the defined failure event. LGD describes the fraction of exposure lost after recoveries and relevant costs. EAD describes the amount outstanding when failure occurs. Multiplying them offers a simplified expected-loss calculation, as described in the Basel Committee’s historical 2005 explanatory note; that note is used here for the mathematics, not as a statement of current capital rules. A full supervisory exercise uses different models for different portfolios rather than one universal multiplication rule. [5][2]
In an invented portfolio with $100 million of exposure, a 2% default probability and 40% loss severity imply $800,000 of expected losses over a specified horizon. If stress raises default probability to 8%, severity to 55% and exposure to $110 million, the same simplified calculation gives $4.84 million. The larger exposure could represent customers drawing committed lines before default. Every amount and assumption in this example is hypothetical.
The stressed amount is more than six times the starting estimate even though no single input rose sixfold. Default frequency quadrupled, severity increased and exposure expanded. This illustrates why shocking only the default rate can miss an important part of the mechanism. Falling collateral prices and greater borrowing can occur alongside weaker repayment capacity. Conversely, a guarantee or recovery process may absorb some losses, although its effectiveness depends on its own terms and performance.
Timing and common exposures change the story
Two loans can share the same annual default probability but create different cash and earnings patterns if one fails immediately and the other near the end of the year. A downturn that lasts several quarters can affect refinancing before it affects final recovery. A balance outstanding when a borrower misses payment may also differ from the balance that originally entered the portfolio. These timing distinctions explain why a sequence of quarterly paths is more informative than a single recession label.
Consider another invented portfolio containing loans to a hotel, a restaurant supplier and a local transport company. Their different industry names might suggest diversification, yet all three could depend heavily on the same tourist destination. A disruption there could weaken all three together. Treating each exposure as independent would understate the chance of concentrated losses, even if the average individual default estimates were sensible.
Collateral introduces a related common exposure. Several borrowers can owe money against the same kind of property, and recovery estimates can decline at the same time that more borrowers fail. The broad economic lesson is that a portfolio is not merely a list of independent averages. Its vulnerability depends on links among borrowers, collateral, funding and the environment, including links that a simple industry classification does not reveal.
Credit losses reach capital through an income statement
Loan losses are only part of a bank-wide stress result. Interest income, funding costs, fees, operating expenses, provisions and other gains or losses all affect earnings. The Federal Reserve's methodology separately models pre-provision net revenue and loan-loss provisions, among other components. Its calculation of regulatory capital also includes adjustments beyond ordinary net income. Therefore, a gross loss figure cannot simply be subtracted from reported capital without reconciling the intervening accounting. [2]
A deliberately simplified illustration starts with $12 billion of capital and $100 billion of , giving a 12% ratio. Suppose the bank earns $3 billion before provisions and has $5 billion of provision expense plus $1 billion of other losses. Ignoring taxes, distributions and regulatory adjustments, the bank records a $3 billion net loss and capital ends at $9 billion. With the denominator held at $100 billion, the ratio becomes 9%.
If the same invented bank instead had $90 billion of risk-weighted assets at the end, the ratio would be 10%. If the denominator were $110 billion, it would be approximately 8.2%. The dollar capital amount is identical in all three variants. This arithmetic is a sensitivity illustration, not a reproduction of the Fed's denominator assumptions. It shows why a published capital ratio must be read with its balance-sheet convention.
Provisions and are also different events. A provision recognizes expected credit expense and adds to the allowance; a charge-off removes an amount deemed uncollectible and uses that allowance. Subtracting both from capital as though each were a new independent loss would double count the same economic damage in a simplified bridge. Differences in horizon, allowance coverage and treatment of new lending can explain why an accounting estimate and a stress estimate are not equal.
Frozen portfolios and management responses answer different questions
A test can hold many balance-sheet quantities steady to improve comparability, or allow lending, repayment and other behavior to change. The 2025 Fed methodology, largely carried into 2026, holds and leverage denominators unchanged over the projection horizon except for changes primarily tied to capital deductions. That convention is different from assuming management can immediately shrink any risky business at its original value. [2][3]
A dynamic internal illustration might assume a lender stops new originations, sells assets or raises equity. Each assumption changes the question being answered. Selling assets at face value during a market shock can create an unrealistically easy escape; forbidding every response can omit genuine flexibility. There is no contradiction in producing different results under different conventions, but comparison becomes misleading when the conventions are left unstated.
Capital and remain distinct. A bank might retain a positive capital cushion while needing cash faster than assets can be sold. Alternatively, ready cash does not repair a loan portfolio whose recoverable value is too low. A credit-capital exercise can inform resilience without constituting a comprehensive simulation of depositor behavior, collateral calls, payment obligations or access to emergency funding.
The dated 2026 implementation and result
The February 2026 methodology note says the Fed generally used the 2025 models while reviewing feedback on proposed changes. It identifies targeted changes to fair-value-option loans, largest-counterparty selection and domestic credit-card sharing agreements, along with a compensation-model phase-in. The proposed model documentation is therefore not interchangeable with the implemented 2026 methodology. [3]
On June 24, 2026, the Fed reported more than $708 billion in hypothetical total losses and an aggregate capital-ratio decline of 1.6 percentage points, with all 32 tested banks remaining above their minimum requirements. The release also said those results would not change large-bank capital requirements, which would remain in place until 2027. These are dated supervisory findings under the exercise's assumptions, not a guarantee against every future financial disruption. [4]
The lasting value of a stress test is the chain connecting the scenario to the outcome. A loss estimate becomes understandable when the reader can identify the affected exposures, default and recovery mechanisms, revenue offsets and capital arithmetic. A favorable result establishes resilience within that defined chain; uncertainty remains about different shocks, imperfect data, behavioral changes and risks outside the model.
Sources
- Federal Reserve, final 2026 Stress Test Scenarios; checked October 4, 2026Official sourceBack to text: ↑
- Federal Reserve, 2025 Supervisory Stress Test Methodology, June 2025; incorporated with changes in 2026Official source · PDFBack to text: ↑1↑2↑3
- Federal Reserve, 2026 Supervisory Stress Test Methodology, February 2026Official source · PDFBack to text: ↑1↑2
- Federal Reserve, 2026 annual bank stress-test results announcement, June 24, 2026Official releaseBack to text: ↑
- Basel Committee, An Explanatory Note on the Basel II IRB Risk Weight Functions, July 2005; historical mathematical explanationSource · PDFBack to text: ↑