Quick answer: a liquidity stress test asks whether a bank can keep paying what must be paid as cash leaves faster than expected, assets become harder to sell, collateral demands rise, and normal funding channels weaken. The algorithm is fundamentally a time-bucket problem: project inflows and outflows, apply stress assumptions, haircut the usable value of liquid resources, accumulate the gaps, and identify how long the institution can continue before counterbalancing capacity is exhausted.

Solvency asks whether assets ultimately cover liabilities. Liquidity asks whether cash arrives in time.

Why liquidity is a mathematical problem before it is a banking problem

A bank can own valuable assets and still be in trouble if those assets cannot be converted into usable cash quickly enough. That turns liquidity into a problem of timing, sequences, inequalities and constraints. The same total inflow received on Day 40 cannot solve an obligation due on Day 2. The order of events matters.

This is why liquidity stress testing belongs naturally beside graphs, recurrence relations and queueing ideas. It asks: what enters, what leaves, when does it happen, which resources are genuinely available, and at what point does the cumulative balance cross a critical boundary?

1. Build the cash-flow ladder

The basic representation is a maturity ladder. Divide the future into time buckets — for example overnight, 2–7 days, 8–30 days, 31–90 days and longer horizons — then map contractual and expected cash flows into those buckets.

For bucket t, a simplified net funding gap can be written as NFG_t = stressed inflows_t − stressed outflows_t. A cumulative gap is CNFG_T = Σ NFG_t from the first bucket through horizon T. Counterbalancing capacity — cash, central-bank-eligible collateral, marketable securities and other usable sources — is then added subject to haircuts, encumbrance and operational constraints.

The key word is usable. An asset with a quoted market value of S$100 million is not automatically S$100 million of stress liquidity. It may be pledged already, slow to transfer, subject to a haircut, denominated in the wrong currency or difficult to sell without a large price concession.

2. Contractual cash flow is only the starting point

If all customers behaved exactly as contracts allowed, liquidity modelling would be easier. They do not. Some deposits stay for years even though they can be withdrawn tomorrow. Some wholesale funds disappear quickly when confidence falls. Credit lines that are normally unused may suddenly be drawn. Margin requirements can rise as markets move.

A stress test therefore converts contractual cash flows into behavioural cash flows by applying assumptions such as deposit runoff, drawdown rates, renewal rates, collateral haircuts and monetisation delays.

3. The runoff transformation

Suppose a bank has S$500 million of a deposit category and applies a 10% stressed runoff assumption over the test horizon. The model creates a S$50 million outflow. If evidence suggests the category is less stable than previously believed, changing the runoff factor to 25% creates a S$125 million outflow. A small-looking parameter change has produced a S$75 million difference in required liquidity.

This illustrates model sensitivity. The important question is not merely whether 10% or 25% is the “correct” number. It is whether the assumed number is defensible for that deposit population, scenario and time horizon, and whether management knows how the conclusion changes when the assumption is wrong.

4. LCR, NSFR and cash-flow stress tests answer different questions

Measure	Core question	Horizon
Liquidity Coverage Ratio (LCR)	Is the stock of high-quality liquid assets sufficient relative to stressed net cash outflows?	30 calendar days
Net Stable Funding Ratio (NSFR)	Is available stable funding adequate relative to required stable funding?	One-year structural horizon
Cash-flow stress test	When do stressed cumulative cash gaps emerge, and can available liquidity cover them?	Multiple buckets from intraday/overnight outward

The Basel LCR framework assigns runoff and drawdown assumptions to different categories and compares high-quality liquid assets with stressed net cash outflows. The public Basel LCR cash-inflow and outflow rules show why “all deposits are the same” is not an acceptable model.

These tools are complementary. A bank can meet one regulatory ratio and still have a weak internal stress scenario, a currency-specific problem, an operational bottleneck or a concentration that deserves attention.

5. A simplified 30-day example

Imagine a teaching example with S$120 million of usable high-quality liquid assets, S$150 million of stressed gross outflows and S$40 million of permitted stressed inflows. Ignoring further technical adjustments, stressed net outflows are S$110 million, so a simple LCR-style ratio is about 120/110 ≈ 109%.

Now change just one assumption: a deposit cohort runs faster, adding S$30 million of outflow. Net outflows become S$140 million and the same S$120 million buffer produces about 86%. Nothing happened to the face value of the liquid-asset portfolio. The conclusion changed because behaviour changed.

This is the mathematical reason stress testing should not be reduced to one base-case forecast. The useful information often sits in the slope of the result as assumptions move.

6. Survival horizon

A practical internal concept is the survival horizon: how long the bank can continue meeting stressed obligations before its available counterbalancing capacity is exhausted, assuming the stress path and available actions specified by the model.

The algorithm is sequential. Start with the opening liquidity buffer. For each time bucket, add stressed inflows, subtract stressed outflows, add only funding sources that can actually be executed, apply asset haircuts and transaction delays, and carry the remaining buffer into the next bucket. The first point at which available liquidity becomes insufficient marks the modelled failure horizon.

That makes survival horizon similar to a recurrence relation: the state at time t+1 depends on the state at time t plus the flows occurring between them.

7. The algorithmic pipeline

Inventory every material source and use of cash. Include assets, liabilities, off-balance-sheet commitments, derivatives, collateral and intraday obligations.
Assign contractual timing. Put each payment, maturity, reset and contingent event into a time bucket.
Segment funding. Retail, insured, uninsured, operational, corporate, financial-institution and secured funding can behave differently.
Apply scenario-dependent runoff and drawdown assumptions. Behaviour must be stressed, not merely copied from normal periods.
Haircut the buffer. Reduce assets to an amount realistically monetisable under stress.
Respect encumbrance and currency. Pledged assets or trapped liquidity cannot be counted as if freely transferable.
Model collateral and margin. Market moves can create additional cash demands.
Calculate bucket and cumulative gaps. Find when the gap becomes material.
Add executable contingency actions. Funding sources count only if legal, operational and timely.
Compute survival horizon and key ratios. Then identify the assumptions that dominate them.
Reverse-stress the system. Ask what combination of outflows, haircuts or market closure would consume the buffer.
Validate against real behaviour. Deposit movements, collateral usage and funding execution should update future assumptions.

8. What the model must not quietly assume

Every liquid asset can be sold instantly. Market depth can shrink when many firms sell simultaneously.
Every funding line will be available. A committed-looking source may have conditions, collateral requirements or operational delays.
Depositors act independently. Digital banking and shared information can make withdrawals correlated.
Yesterday’s runoff behaviour describes tomorrow. Confidence regimes can change abruptly.
Liquidity is transferable across currencies and entities. Legal, operational or market constraints can trap it.
End-of-day liquidity is enough. Payment systems create intraday peaks that can matter even if the day closes with positive cash.
Collateral value and collateral usability are the same. Haircuts, settlement times and pre-positioning matter.

9. Failure modes and counterexamples

A counterexample is useful because it attacks the hidden rule. Consider the statement, “The bank owns enough government securities, therefore it cannot have a liquidity problem.” A counterexample could be a bank whose securities are long-duration, deeply underwater, operationally unprepared for central-bank borrowing and facing withdrawals faster than assets can be monetised. The assets exist, yet timing and execution can still fail. This is the same reasoning habit explored in One Counterexample Can Be Enough.

The 2023 US banking turmoil made this interaction visible: banks with greater reliance on uninsured deposits and larger unrealised securities losses experienced larger outflows, while emergency facilities were used to meet funding needs. See the Federal Reserve’s public research on the 2023 banking turmoil and the Bank Term Funding Program.

10. Diagnostics: where should a reviewer press hardest?

Which deposit groups drive most of the 30-day outflow?
How much of the liquidity buffer is already encumbered?
What happens if securities haircuts double?
How quickly can collateral be moved to the central bank or another lender?
Are there currencies in which the bank fails earlier than the consolidated result suggests?
How much unused committed credit is assumed to be drawn in stress?
Does the model include derivative margin and collateral calls?
What is the largest intraday net payment requirement?
Which contingency source has never been operationally tested?
How much faster would deposits have to run for the survival horizon to halve?

11. Verification must include operations, not just arithmetic

A model can say that a bank has S$1 billion of contingent borrowing capacity, but if documents are incomplete, collateral is not pre-positioned or systems cannot execute the transaction fast enough, the mathematical capacity is fictional at the moment it is needed.

Public supervisory guidance therefore links liquidity models to contingency funding plans, stress testing and operational readiness. The FDIC’s contingency-funding guidance explicitly stresses actionable plans and regular testing of access to contingent sources; the OCC Liquidity handbook similarly emphasises multiple horizons and bank-specific as well as market-wide stresses.

12. When should assumptions be updated?

deposit concentrations change materially;
the share of uninsured or rate-sensitive funding changes;
new digital channels alter withdrawal speed;
collateral haircuts or market depth change;
funding markets become more concentrated or volatile;
new commitments or derivative positions increase contingent outflows;
actual deposit behaviour repeatedly exceeds model assumptions;
contingency-funding tests reveal operational delays;
regulatory liquidity standards or local supervisory expectations change.

13. Connections to other mathematical banking systems

Payment-system graphs, queues and real-time settlement — where intraday timing becomes operational.
Loan amortisation and recurrence relations — contractual asset cash flows entering the liquidity ladder.
Fraud detection and decision thresholds — another example of a model whose errors have asymmetric costs.
A Gradient Is Not Finished Until We Know What Is Changing — useful for thinking about the rate at which a liquidity buffer is being consumed.

Research anchors

The deeper lesson

Liquidity models teach a powerful form of mathematical humility. Totals are not enough. Timing matters. Behaviour matters. Constraints matter. A resource matters only if it can reach the required place before the deadline. The strongest liquidity algorithm therefore does not ask simply, “How much do we own?” It asks, “What can actually be converted into usable cash, in this scenario, in this currency, at this time?”

Educational note: This article explains banking mathematics and risk-management concepts. It is not financial advice, investment advice, a recommendation regarding any bank or security, or institution-specific regulatory guidance.

Small Group Tutorials

How Banks Stress-Test Liquidity: Cash-Flow Buckets, Runoff Assumptions and Survival Horizons