A liquidity stress test asks a single question in many forms: if funding leaves and markets close, how many days does the institution survive before it runs out of cash and collateral it can turn into cash, and what would it have to do, at what cost, to survive longer? The answer depends almost entirely on the scenario (who leaves, how fast, what else happens at the same time), on the horizon (seven days, thirty, ninety, a year), and on a handful of assumptions about behavior that no contract specifies: how quickly uninsured depositors move, how many committed lines get drawn, what haircut a stressed market applies to the securities being sold. A test built on gentle assumptions passes comfortably and teaches nothing; the banks that failed in March 2023 all had stress tests, and all of them passed.
This guide explains what regulation requires of liquidity stress testing in the United States and under the Basel framework, how scenarios are designed and what the 2023 failures changed about them, the horizons and why each one answers a different question, the assumptions that decide the result, how the survival horizon and the liquidity buffer are calibrated from the test, reverse stress testing, the death spiral dynamics that turn a liquidity problem into a solvency problem, how a test should change behavior through limits and the contingency funding plan, internal audit’s program, and a worked example from Lakeshore Bancorp, the $9 billion regional bank used in examples across this site, whose stress testing framework was rebuilt after its FY24 audit found the severe scenario milder than the bank’s own experience. The liquidity risk guide covers the wider subject and the corporate liquidity risk guide the non-bank version; this one stays on the test.
In this guide
- What regulation requires
- Scenario design: idiosyncratic, market-wide, combined
- What 2023 taught about speed
- Horizons and the questions each one answers
- The assumptions that decide the result
- Survival horizon, buffer and limits
- Reverse stress testing
- The death spiral: how liquidity becomes solvency
- How a test should change behavior
- Internal audit’s program
- A worked example: Lakeshore Bancorp rebuilds its framework
- Questions about liquidity stress testing
- Related guides
What regulation requires
Liquidity stress testing became a supervisory requirement after 2008, when Northern Rock, Bear Stearns and Lehman Brothers showed that solvent-looking institutions could fail in days when funding left. The Basel Committee’s Principles for Sound Liquidity Risk Management and Supervision (September 2008, seventeen principles) made regular stress tests over short and protracted institution-specific and market-wide scenarios a core expectation, with results used to adjust strategy, positions and the contingency funding plan. The liquidity coverage ratio that followed (final text January 2013, fully phased in 2019) is itself a standardised 30-day stress test: prescribed run-off rates by funding category against a stock of high-quality liquid assets, with inflows capped at 75 percent of outflows. The United States added its own layers: the 2010 Interagency Policy Statement on Funding and Liquidity Risk Management, which applies to every depository institution and expects stress testing proportionate to complexity; Regulation YY’s enhanced prudential standards for large bank holding companies, which require internal liquidity stress tests at overnight, 30-day, 90-day and one-year horizons, monthly for Category I to III firms and quarterly for Category IV, with a buffer of highly liquid assets sized to the 30-day result; the LCR and NSFR rules for the largest firms under the tailoring framework; and the FR 2052a complex institution liquidity monitoring report. After the 2023 failures the agencies issued an addendum to the 2010 statement (July 2023) on contingency funding plans, stressing operational readiness to borrow from the discount window and the Federal Home Loan Banks, pre-positioned collateral and periodic tests of access. The table summarizes who must do what.
| Requirement | Who it reaches | What it requires of stress testing |
|---|---|---|
| Basel Principles for Sound Liquidity Risk Management (2008) | Internationally active banks; adopted into most supervisory frameworks | Regular stress tests of short and protracted, idiosyncratic and market-wide scenarios; results used in strategy, limits and the contingency funding plan |
| Liquidity coverage ratio (Basel 2013; US rule 2014) | Category I to IV firms in the US (full or reduced calibration); most large banks worldwide | A prescribed 30-day stress: run-off rates by funding type, inflows capped at 75 percent, HQLA by level with haircuts; ratio at least 100 percent |
| Net stable funding ratio (Basel 2014; US rule effective July 2021) | Same population | A one-year structural test: available stable funding at least equal to required stable funding |
| Regulation YY internal liquidity stress tests | US bank holding companies with $100 billion or more in assets (Categories I to IV) | Firm-designed tests at overnight, 30-day, 90-day and one-year horizons; monthly (Categories I to III) or quarterly (Category IV); a liquidity buffer of unencumbered highly liquid assets covering the 30-day net outflow; contingency funding plan; limits; board-approved tolerance |
| 2010 Interagency Policy Statement on Funding and Liquidity Risk Management | All US depository institutions | Stress testing proportionate to size and complexity, covering institution-specific and market-wide events over multiple horizons; results feed the contingency funding plan and the liquidity cushion |
| 2023 addendum on contingency funding plans | All US depository institutions | Operational readiness: borrowing arrangements established in advance, collateral pre-positioned, access tested periodically, the discount window as part of the plan |
| FR 2052a | US firms subject to the LCR; frequency by category (daily for the largest) | Granular cash flow and asset data that let supervisors run their own stress analyses |
| Supervisory liquidity reviews (CLAR; horizontal reviews) | The largest US firms; periodic horizontal reviews of others | Examiners compare firms’ stress assumptions and results against each other and against peers |
Below the $100 billion line, where most banks by count sit, nothing prescribes the horizons or the frequency; the 2010 statement’s proportionality is the test, and examiners judge the program in the CAMELS liquidity component. In practice a mid-size bank runs the test at least quarterly, with 30-day and 90-day horizons at a minimum and a one-year view for structure, and the programs that satisfy examiners after 2023 also run a short, severe seven-to-ten-day test because that is how long the runs took.
Scenario design: idiosyncratic, market-wide, combined
Every program needs at least three scenario families, because they stress different things. An idiosyncratic scenario is a loss of confidence in this institution alone: a large loss announced, a rating downgrade, an enforcement action, a rumor on social media. Deposits leave (uninsured and non-operational balances first), wholesale counterparties decline to roll funding, rating-linked collateral triggers fire, but markets stay open, so assets can be sold or pledged at normal haircuts and the Federal Home Loan Bank and discount window remain available. A market-wide scenario is a systemic event in which this institution is not the problem: funding markets freeze, securities values fall and haircuts widen, correspondent banks hoard, but the institution’s own depositors are relatively calm and may even bring in flight-to-quality balances. A combined scenario, usually the severe one, stacks both: the institution is under suspicion in a market that is also under stress, which is what Signature Bank and First Republic faced in March 2023. Each family needs at least two severities (adverse and severe), and the severe case must be severe enough to fail at some horizon; a severe scenario the institution survives indefinitely has not been calibrated. Scenario narratives should name the trigger, the sequence of events by day, who leaves and why, which contingent obligations crystallize and which funding sources remain, so that assumptions can be traced to the story rather than to a spreadsheet default.
| Element | Idiosyncratic | Market-wide | Combined (severe) |
|---|---|---|---|
| Trigger | Loss announcement, downgrade, enforcement action, rumor | Systemic funding freeze; asset price shock | Both at once, or a systemic event that exposes this institution’s weakness |
| Uninsured deposits | Heavy run-off, fastest in non-operational and concentrated balances | Modest run-off; possible inflows from weaker peers | Heaviest and fastest run-off |
| Insured retail deposits | Light run-off | Minimal | Light to moderate |
| Wholesale funding | Unsecured rolls off at maturity; secured continues with wider haircuts | Unsecured markets closed to all; secured markets strained | Unsecured gone; secured only against the best collateral |
| Commitments | Borrowers draw to lock in access before the bank restricts it | Borrowers draw because their own markets are closed | Maximum draws |
| Securities | Saleable at near-normal haircuts | Saleable only at stressed prices; haircuts wide | Sale crystallizes losses; pledge preferred |
| Central bank and FHLB | Available; stigma concern | Available; heavy demand | Available if collateral is pre-positioned |
| Rating and collateral triggers | Fire | Do not fire | Fire |
What 2023 taught about speed
Before March 2023 the industry’s severe scenarios were calibrated to 2008, when runs took weeks: Northern Rock’s depositors queued for days, and wholesale funding rolled off at maturity. Silicon Valley Bank lost about $42 billion of deposits on 9 March 2023, roughly a quarter of its deposit base, in a single day, with about $100 billion more queued for the next morning when it was closed; the LCR’s 30-day assumptions (and SVB was not subject to the LCR) would have spread that outflow over a month. The Basel Committee’s October 2023 report on the turmoil, the Federal Reserve’s review of SVB and the FDIC’s reviews of Signature and First Republic drew the same lessons: uninsured, concentrated, digitally mobile depositors who talk to each other move in hours; social media and mobile banking removed the frictions that older assumptions relied on; held-to-maturity securities with unrealized losses were not usable liquidity because selling them would have breached capital; and the discount window was not operationally ready at any of the failed banks (SVB had not pre-positioned enough collateral and could not move it fast enough on the day). Credit Suisse, which met its LCR to the end, lost about CHF 110 billion of client assets in the fourth quarter of 2022 after a social media rumor and was gone by March 2023, which showed that a franchise can erode past the point of recovery while every regulatory ratio is satisfied.
The practical changes to scenario design that followed: a short severe horizon (five to ten business days) in which the bulk of the run-off occurs on days one to three; uninsured deposit run-off assumptions segmented by relationship depth, operational use, concentration and sector, with the most mobile segments assumed to move in days rather than weeks; held-to-maturity securities excluded from the usable buffer or counted only at pledge value; explicit modeling of the capital consequence of selling securities at a loss; and a contingency funding plan that assumes the discount window will be needed on day one. The bank runs guide gives the history, and the IRR vs IRRBB guide the interest rate side of the SVB story.
Horizons and the questions each one answers
| Horizon | Question answered | What dominates the result | Management response it informs |
|---|---|---|---|
| Intraday | Can every payment be settled today? | Timing of inflows and outflows within the day; settlement balances; intraday credit | Intraday collateral; payment throttling; correspondent arrangements |
| Overnight to 7 or 10 days (acute) | Do we survive the first week of a run? | Uninsured deposit run-off speed; discount window and FHLB readiness; cash and overnight-saleable assets | Pre-positioned collateral; the crisis communication plan; the first calls to counterparties |
| 30 days | Do we survive a month without market access? | Run-off by segment; wholesale roll-off; commitment draws; HQLA stock and haircuts | Buffer size (the LCR and Regulation YY anchor); limits |
| 90 days | Can we restructure funding before the buffer is exhausted? | Term wholesale maturities; loan inflows; asset sales at stressed prices; franchise effects on deposit pricing | Funding plan actions: term debt, deposit campaigns, asset sales, balance sheet shrinkage |
| One year (structural) | Is the balance sheet fundable through a prolonged stress? | Maturity profile of assets versus stability of funding (the NSFR view); earnings effect of higher funding costs | Strategy: funding mix, asset growth, securities duration |
The horizons are not alternatives. An institution can pass the 30-day test and fail the ten-day one (front-loaded run-off exhausts cash before the slower inflows arrive), or pass both and fail the one-year test (stable in a crisis, unfundable as a going concern). Regulation YY requires four horizons for a reason, and a mid-size program that runs only a 30-day test has adopted the LCR’s convention without the LCR’s population of banks, whose depositors are more granular than a regional commercial bank’s.
The assumptions that decide the result
A stress test is a set of assumptions applied to a balance sheet, and five of them decide the outcome. Deposit run-off by segment is the largest: the model must segment deposits by insured status, retail or commercial, operational or non-operational, concentration (the top depositors individually) and sector, and assign each segment a run-off rate and a speed under each scenario. The LCR’s rates (3 percent for stable insured retail, 10 percent for less stable retail, 25 percent for operational wholesale deposits, 40 percent for non-operational deposits of non-financial corporates, 100 percent for unsecured funding from financial institutions, all over 30 days) are a floor for a market-wide scenario and too gentle for an idiosyncratic one at a bank with concentrated uninsured deposits; the bank’s own experience and the 2023 cases are the calibration evidence. Wholesale funding roll-off assumes unsecured funding is not renewed at maturity and secured funding continues against eligible collateral at wider haircuts; brokered deposits and listing-service deposits should be treated as wholesale. Contingent outflows include draws on committed credit and liquidity lines (the LCR assumes 10 percent of retail and small business commitments, 30 percent of non-financial corporate credit lines, 100 percent of liquidity lines to financial institutions), collateral calls on derivatives and secured funding after a downgrade, and the funding of assets the bank feels obliged to support (sponsored vehicles, customer repurchases). Asset liquidity and haircuts decide what the buffer is worth: cash and central bank reserves at par; Treasuries and agencies at small haircuts; other securities at stressed haircuts; loans at pledge value only through the FHLB or discount window, after the collateral has been delivered and valued; held-to-maturity securities at pledge value, never at sale value, unless the capital consequence is modeled. Inflows are the assumption most often too generous: contractual loan repayments continue in stress but renewals do not generate cash, new deposits do not arrive, and the LCR’s 75 percent cap on inflows exists because institutions that assumed inflows would fund outflows did not survive to collect them.
| Assumption | Gentle (passes comfortably) | Calibrated after 2023 | Evidence to calibrate against |
|---|---|---|---|
| Uninsured non-operational commercial deposits | 25 to 40 percent over 30 days, evenly spread | 40 to 60 percent over 30 days with half in the first five days; higher for the top depositors and for concentrated sectors | The bank’s own 2023 outflow by segment; SVB, Signature and First Republic disclosures; peer run-off in supervisory horizontal reviews |
| Uninsured operational deposits | 25 percent | 25 to 35 percent, with documentation that the accounts truly are operational | Account-level activity analysis; re-performed classification |
| Insured retail deposits | 3 to 5 percent | 5 to 10 percent in an idiosyncratic scenario, because insured customers also leave when the bank is in the news | 2023 experience at banks with heavy media coverage |
| Brokered and listing-service deposits | Treated as deposits | Treated as wholesale: no renewal at maturity | Contractual terms |
| FHLB advances | Fully available up to the credit line | Available only against collateral already delivered and valued; new advances take days | FHLB collateral statements; the 2023 addendum |
| Discount window | Assumed available for the full pledgeable loan book | Available only for pre-positioned collateral; operational test within the last quarter | Reserve Bank collateral records; test documentation |
| Securities sales | Available-for-sale and held-to-maturity securities saleable at market | HTM excluded or pledge-only; AFS sales modeled with the capital effect of realized losses | Unrealized loss reports; capital triggers |
| Commitment draws | 5 to 10 percent | Segmented: higher for borrowers whose own funding is stressed; 100 percent for lines to financial institutions | Draw behavior in March 2020 and March 2023 |
| Inflows | Contractual repayments plus renewals and new deposits | Contractual repayments only, capped at 75 percent of outflows | LCR convention; prepayment behavior in stress |
Survival horizon, buffer and limits
The output that boards understand is the survival horizon: the number of days under a scenario before cumulative net outflows exceed the cash and the liquidity that can be raised from unencumbered assets and tested facilities. It is computed by projecting daily cash flows under the scenario, adding the liquidity value of the buffer (with haircuts) and facility capacity (with operational constraints), and finding the day the cumulative position turns negative. The board sets a tolerance, which is a minimum survival horizon under each named scenario (commonly 30 days under the combined severe scenario and 90 under the adverse ones), and treasury sizes the buffer so that the horizon holds with a margin. Limits then translate the tolerance into operating constraints: minimum liquid assets, maximum wholesale funding share, maximum uninsured deposit concentration, maximum unencumbered collateral pledged, and early-warning indicators whose breach escalates before the limits do (the KRI program guide covers indicator design and the risk appetite, tolerance and limits guide the structure). The survival horizon is only as honest as the scenario; a 92-day horizon under a gentle severe scenario is less information than a 41-day horizon under a calibrated one, which is the lesson of the Lakeshore example below.
Reverse stress testing
A conventional test starts from a scenario and finds the outcome; a reverse stress test starts from the outcome (the institution fails: the buffer is exhausted, or a regulatory ratio is breached, or the franchise is lost) and works backward to the combination of events that would produce it, then asks how plausible that combination is and what early warnings would precede it. For liquidity, reverse testing answers questions the forward tests cannot: what deposit run-off rate, over how many days, exhausts the buffer; which three depositors leaving would breach the tolerance; what securities price decline makes the buffer unusable because sales would breach capital; what combination of a downgrade, a commitment draw spike and a market freeze breaks the funding plan. The result is a map of the institution’s breaking points, and the management use is to compare them with early-warning indicators: if the breaking point is 15 percent uninsured run-off in five days and the indicator fires at 5 percent in a day, the plan has time; if the indicator fires only on a monthly report, it does not. Regulators expect reverse stress testing of larger firms and increasingly of mid-size ones after 2023, and auditors should ask to see it even where no rule requires it.
The death spiral: how liquidity becomes solvency
Liquidity and solvency are taught as different problems: a solvent bank has assets worth more than its liabilities and merely cannot turn them into cash fast enough; an insolvent one has nothing to turn. In a run the two merge through a sequence that every stress test should model explicitly. Funding leaves; the bank sells its most liquid assets first (cash, Treasuries), which concentrates what remains in less liquid, longer-duration or lower-quality assets; the next sales realize losses because the market applies stressed prices and because the bank is a known forced seller; realized losses reduce capital, and the disclosure of losses (SVB’s $1.8 billion after-tax loss on an $21 billion sale) tells depositors and counterparties the bank is in trouble; the run accelerates; rating agencies downgrade, collateral triggers fire, and secured lenders demand more collateral for the same funding; the bank is now selling illiquid assets at fire-sale prices into a market that knows it must sell, and the losses consume the capital that would have made it solvent. The spiral has three accelerants that assumptions should capture: the capital constraint on asset sales (a bank that cannot sell securities without breaching a capital trigger has a smaller usable buffer than its balance sheet suggests), the information effect of visible actions (a securities sale, a capital raise, an emergency borrowing disclosed in a filing), and the correlation of contingent outflows with the stress itself (downgrade triggers, commitment draws by stressed borrowers). A stress test that treats the buffer as a fixed stock of saleable assets and ignores these feedbacks will overstate survival, which is why the better programs model the combined scenario path by path, with the capital ratio and the market’s reaction recomputed each day.
| Stage | What happens | Feedback | Modeling implication |
|---|---|---|---|
| 1. Trigger | A loss, downgrade, rumor or market event starts outflows | Information spreads faster than in past cycles | Front-load run-off; model the first three days separately |
| 2. Liquid assets used | Cash and best securities go first | Remaining portfolio is less liquid and longer duration | Sequence asset use; recompute buffer quality daily |
| 3. Forced sales | Securities sold at stressed prices | Realized losses reduce capital; the sale is visible | Model the capital effect; add an information shock to run-off after a disclosed sale |
| 4. Downgrade and triggers | Rating action; collateral calls; covenant triggers | Secured funding costs more collateral; unsecured disappears | Contingent outflows conditional on the downgrade |
| 5. Fire sale and capital breach | Illiquid assets sold at deep discounts; capital falls below triggers | Resolution or rescue | The reverse test’s breaking point |
| Circuit breakers | Pre-positioned collateral at the central bank; term funding; a credible capital backstop; clear communication | Borrowing instead of selling avoids stages 3 and 5 | Model the contingency funding plan’s actions as explicit alternatives to sales |
How a test should change behavior
A stress test that does not change anything is a reporting exercise. The results should drive four things. The buffer: its size, composition (how much in cash and Treasuries versus other securities) and encumbrance are set from the severe scenario’s cumulative net outflow at the tolerance horizon, plus a management margin. The contingency funding plan: each scenario’s funding gap by day is matched to specific actions (draw the FHLB line, borrow at the discount window, sell named securities, run a deposit campaign, stop lending growth, sell loans), each with an owner, a lead time, a realistic amount and a precondition (collateral delivered, documentation signed, access tested), so that the plan is a sequence rather than a list. Limits and indicators: the structural limits (wholesale share, uninsured concentration, encumbrance) are set where the test shows the balance sheet becomes fragile, and the early-warning indicators are set to fire before the breaking points the reverse test identified. Strategy: a test that shows the one-year structural position unfundable is an argument about the business model (securities duration, deposit gathering, growth pace), not about the buffer. The ten ways institutions manage liquidity risk guide covers the management toolkit and the twelve ways internal audit improves liquidity risk management guide the audit angles on it.
Internal audit’s program
The audit of a stress testing program tests whether the test is honest and whether it matters. The program below assumes the function has, or co-sources, someone who can read a cash flow model; a review that confirms the test exists and was presented to ALCO has not audited it. The liquidity risk guide gives the wider liquidity audit, the model risk management guide the governance of the model the test runs on, and the bank audit universe guide where the entity sits among a bank’s auditable units.
| Area | Tests | Evidence | Finding that recurs |
|---|---|---|---|
| Scenario design | Scenarios cover idiosyncratic, market-wide and combined cases at adverse and severe levels; narratives trace to assumptions; the severe scenario fails at some horizon; scenarios were reviewed after 2023 | Scenario documentation, approval records, change log | Severe scenario unchanged since before 2023 and passed every quarter |
| Assumption calibration | Run-off rates by segment compared with the institution’s own stress experience, peer disclosures and supervisory expectations; speed as well as amount; HTM treatment; facility availability conditional on operational readiness | Calibration memos, back-testing against actual outflows, assumption inventory with owners and review dates | Assumed uninsured run-off below the bank’s own experience; HTM counted at sale value |
| Data and model | Balance sheet positions reconcile to the ledger; deposit segmentation re-performed on a sample; the model is in the model inventory and validated; manual adjustments logged | Reconciliations, segmentation logic, validation report, adjustment log | Operational deposit classification undocumented; segmentation stale |
| Horizons and frequency | Horizons match the policy and the institution’s vulnerabilities (a short acute horizon exists); frequency as required; ad hoc runs in stress | Run schedule, ALCO packs | Only a 30-day test at a bank with concentrated uninsured deposits |
| Survival horizon and tolerance | Computation re-performed for one scenario; tolerance board-approved and tied to the appetite statement; margin to tolerance reported and trended | Model output, board minutes, limit reports | Tolerance defined but never compared with results in board reporting |
| Contingency funding plan linkage | Each scenario’s gap matched to actions with owners, amounts and lead times; collateral pre-positioned to the amounts assumed; discount window and FHLB access tested within the period; triggers link to indicators | CFP, collateral statements from the FHLB and Reserve Bank, test records | Capacity assumed for collateral never delivered; window never tested by current staff |
| Reverse stress test | Exists; identifies breaking points; compared with early-warning indicators | Reverse test documentation | Absent at mid-size banks |
| Use of results | Evidence that results changed the buffer, limits, the plan or strategy; failed scenarios led to action | ALCO and board minutes, buffer and limit history | Results presented and noted; nothing changed |
| Integration | Interest rate risk (unrealized losses and capital triggers) in the liquidity test; capital stress and liquidity stress scenarios consistent | Integrated scenario documentation | Liquidity buffer counts securities whose sale would breach capital |
A worked example: Lakeshore Bancorp rebuilds its framework
Lakeshore Bancorp, the $9 billion publicly traded regional bank holding company with 60 branches used in examples across this site, sits below the thresholds for the LCR, the NSFR and Regulation YY and is supervised against the 2010 interagency statement. In March 2023 it watched $410 million of deposits, about 6 percent of its then $7.1 billion deposit base, leave in eleven business days, concentrated among uninsured commercial customers in two metropolitan markets, with 19 percent of the most affected segment gone before an insured sweep product stabilized the outflow. Its FY24 liquidity risk management audit (480 hours by the 22-person internal audit function) found that the severe scenario still assumed 12 percent uninsured run-off evenly spread over 30 days, that $340 million of the $1.2 billion of assumed Federal Home Loan Bank capacity was supported by collateral never delivered, that the discount window had last been tested in 2019 with a $1 million borrowing against an assumed $600 million of capacity, and that 14 of a sample of 60 commercial accounts classified as operational no longer met the definition. The liquidity risk guide tells the audit; this section tells what treasury built afterward and what the FY26 follow-up found.
| Element | Before (FY24 audit) | After (FY26 framework) |
|---|---|---|
| Scenario families | One severe scenario (market-wide flavor), run quarterly | Idiosyncratic, market-wide and combined, each at adverse and severe; combined severe is the tolerance scenario; run monthly, with an ad hoc protocol when any early-warning indicator fires |
| Horizons | 30 days | 10 business days (acute), 30 days, 90 days, one year |
| Uninsured run-off (combined severe) | 12 percent over 30 days, evenly spread | Non-operational uninsured 45 percent over 30 days with 25 percent in the first five business days; top-20 depositors modeled individually at 60 percent; operational uninsured 30 percent; insured retail 7 percent; brokered deposits not renewed |
| Securities | All securities counted at market value as saleable | Treasuries and agencies saleable at a 3 percent haircut; other AFS pledge-only; HTM pledge-only through the FHLB and discount window; realized-loss capital effect computed for any modeled sale |
| Facilities | $1.2 billion FHLB and $600 million discount window assumed | Only delivered and valued collateral counted: $1.05 billion FHLB (after pre-positioning) and $740 million discount window (loan collateral pre-positioned in FY25); both tested quarterly with real borrowings |
| Survival horizon, combined severe | 92 days (passed comfortably) | 38 days at the June 2026 run against a 30-day board tolerance; margin trended monthly |
| Reverse stress test | None | Annual: the buffer is exhausted at 31 percent total deposit run-off in ten days, or if the top-20 depositors and 15 percent of other uninsured balances leave in five days; indicators set at 2 percent daily segment outflow and any top-20 departure |
| Contingency funding plan | List of sources | Day-by-day action sequence per scenario with owners, amounts and preconditions; crisis communication plan; semi-annual tabletop exercise |
| Use of results | Presented to ALCO | Buffer target raised from $900 million to $1.3 billion of cash and Treasuries; uninsured non-operational concentration limit set at 22 percent of deposits (was unlimited); securities purchases limited to a five-year average duration |
The FY26 follow-up (220 hours) validated the four FY24 findings closed and tested the new framework. It re-performed the June 2026 survival horizon from the model’s inputs and reproduced 38 days within a day; it traced the FHLB and discount window capacity to collateral statements and borrowing confirmations dated within the quarter; and it re-performed the deposit segmentation on 60 accounts with two exceptions. Its one new finding was that the combined severe scenario’s commitment draw assumption (10 percent across all borrower types) had not been segmented, while the bank’s own March 2020 experience showed middle-market borrowers drawing 28 percent of availability in three weeks; treasury re-segmented the assumption, and the survival horizon moved from 38 days to 35, still inside tolerance but with a margin the committee now watches. The committee’s view, recorded in the minutes, was that it preferred a 35-day answer it believed to a 92-day answer it did not.
Questions about liquidity stress testing
What is a liquidity stress test?
A projection of an institution’s cash inflows and outflows under a defined adverse scenario (a loss of confidence in the institution, a market-wide funding freeze, or both), compared with the cash and unencumbered collateral it can raise, to determine how many days it survives and what actions would extend that horizon.
What horizons should a liquidity stress test use?
Regulation YY requires overnight, 30-day, 90-day and one-year horizons for large US bank holding companies, and the LCR is a 30-day test. After 2023, well-run programs at any size add a short acute horizon of five to ten business days, because the runs that failed banks in March 2023 took days, not weeks.
What is a survival horizon?
The number of days under a scenario before cumulative net outflows exceed the liquidity available from cash, unencumbered assets after haircuts and tested facilities. Boards set a minimum survival horizon as their liquidity tolerance, and treasury sizes the buffer to meet it with a margin.
What is reverse stress testing in liquidity?
Starting from the failure outcome (buffer exhausted, ratio breached) and working backward to the combination of run-off, market and contingent events that would cause it, then judging its plausibility and setting early-warning indicators to fire before those breaking points.
What is the liquidity death spiral?
The feedback loop in which funding outflows force asset sales, the sales realize losses that reduce capital and signal distress, the signal accelerates outflows and triggers downgrades and collateral calls, and the institution ends up selling illiquid assets at fire-sale prices until capital is gone. Borrowing against pre-positioned collateral instead of selling is the main circuit breaker.
What should internal audit test in a liquidity stress testing program?
Whether the scenarios are severe enough to fail somewhere, whether the assumptions are calibrated to the institution’s own and peers’ stress experience, whether the data and model are controlled, whether the contingency funding plan’s capacity is backed by delivered collateral and tested access, whether a reverse test exists, and whether results have changed the buffer, the limits or the plan.
Related guides
- Liquidity risk — the wider subject and the FY24 Lakeshore audit.
- Corporate liquidity risk — the non-bank version: cash forecasting, covenants and committed lines.
- Twelve ways internal audit improves liquidity risk management — audit angles on the whole framework.
- Ten ways financial institutions manage liquidity risk — the management toolkit.
- Bank runs in history — the pattern the scenarios are calibrated to.
- IRR vs IRRBB vs SIRR — the interest rate side of the 2023 failures.
- EVE vs NII — the measures that interact with the liquidity test.
- Designing a KRI program — early-warning indicators that fire before the breaking points.
- Risk appetite, tolerance, capacity and limits — where the survival horizon tolerance sits.
- How to audit model risk management — governance of the model the test runs on.
- Internal audit in banking — the supervisory context.
- A bank audit universe worked example — where liquidity sits among 126 entities.
- How to audit business continuity management — the operational side of crisis readiness.
- MRAs, MRIAs and consent orders — what a liquidity finding becomes.
- The Risk Library — every risk type on this site.
Leave a Reply