Liquidity Stress Testing Techniques: Modeling Run-Off, Wholesale Withdrawal, and Contingent Draws

TL;DR

• Liquidity stress testing scenarios are the easy part — every examiner knows idiosyncratic, market-wide, and combined stress. The hard part is the assumptions inside: run-off rates, wholesale maturity ladders, and contingent draw probabilities.
• LCR floor run-off rates (3% stable retail, 10% less-stable retail, 40% non-operational corporate, 100% short-term wholesale) are starting points, not answers — institutions with concentrated or uninsured depositors must apply higher institution-specific rates.
• Wholesale funding cliff risk requires maturity ladder modeling: mapping exactly when FHLB advances, brokered CDs, and repo agreements mature and stress-testing the rollover assumption.
• U.S. bank contingent commitments reached $2.5 trillion in H1 2025 — models that treat undrawn credit facilities as zero outflow are underestimating exposure at the moment it matters most.

The stress test scenarios in your CFP are the easy part. Every interagency guidance document tells you to model idiosyncratic stress, market-wide stress, and combined stress. The hard part is what lives inside those scenarios: what exactly is your deposit run-off rate for non-operational business accounts? What draw probability are you applying to undrawn revolvers extended to financial counterparties? What happens to your FHLB access when collateral haircuts increase and advance availability shrinks?

Those assumptions are where examiners push back. They’re also where stress tests fail quietly — producing survival horizon numbers that look acceptable while the underlying assumptions are optimistic. Silicon Valley Bank’s stress tests showed an $18 billion 30-day deficit eight months before it failed. Management’s response was to change the model assumptions rather than fix the liquidity problem. That’s what happens when the modeling discipline isn’t there.

This post focuses on the techniques: how to build defensible run-off rates, model wholesale funding cliff risk, and quantify contingent draw exposure in a way that holds up under examiner scrutiny and under actual stress.

The Three Modeling Components

According to BIS FSI Insights No. 59, which surveys liquidity stress testing practices across major banking jurisdictions, stress models incorporate three distinct types of assumptions:

1. Run-off rates: What percentage of each funding source leaves under stress?
2. Asset haircuts: What percentage of asset value is assumed impaired or unavailable under stress?
3. Contingent draw probabilities: What percentage of off-balance-sheet commitments will be exercised under stress?

Most practitioners focus their energy on selecting scenarios. The modeling discipline is in getting these three inputs right for each scenario — and being able to defend them when an examiner asks how you got there.

Run-Off Rate Modeling: Starting from LCR Floors

The Basel III Liquidity Coverage Ratio establishes standardized run-off rates that function as regulatory minimums. Examiners expect institutions with concentrated, uninsured, or rate-sensitive deposit profiles to apply higher institution-specific rates.

Where Standard Rates Break Down

SVB’s failure produced the clearest modern data point: approximately 94% of SVB’s deposits were uninsured at year-end 2022. The 3% stable retail floor was completely inapplicable. Even the 40% non-operational corporate assumption didn’t capture the speed of actual outflow, which reached roughly 25% in a single day on March 9, 2023.

Three situations require institution-specific run-off rates that exceed LCR floors:

Concentrated depositors. When a single depositor or affiliated group exceeds 5% of total deposits, model that relationship separately with institution-specific behavioral assumptions. The concentration itself means standard statistical assumptions don’t apply — one relationship’s decision to withdraw creates a cliff, not a gradual outflow.

Uninsured deposit concentration. When uninsured deposits represent a material share of total deposits, floor rates dramatically underestimate potential outflow. Examiners increasingly ask institutions to demonstrate what share of deposits exceeds FDIC/NCUA coverage and whether run-off assumptions reflect that exposure.

Rate-sensitive deposit bases. CD portfolios, brokered deposits, and accounts held by Treasury and investment management clients require explicit behavioral analysis. Historical deposit behavior during prior rate cycles — how quickly did balances migrate when competitor rates diverged — is the primary data source for calibration.

Documentation matters as much as the rate. Examiners don’t just want a number; they want the methodology that produced it, the data that supports it, and the rationale for why it’s appropriate for your institution’s specific profile.

Wholesale Funding Cliff Risk: Maturity Ladder Modeling

Wholesale funding is the most vulnerable component of any balance sheet under combined stress scenarios. The FDIC’s Liquidity and Funds Management guidance treats reliance on volatile funding as a primary safety and soundness concern, and the NSFR framework identifies institutions with more than 40% reliance on wholesale funding as having significant dependency.

Cliff risk emerges from maturity concentration: when a large portion of FHLB advances, brokered CDs, Fed funds purchased, and repo agreements mature in a compressed window, the rollover assumption becomes the entire stress test.

Building the Maturity Ladder

A maturity ladder maps contractual cash outflows day-by-day for the first week, week-by-week through 30 days, then monthly through the 90-day or 1-year horizon. For each wholesale funding category:

FHLB advances: Map advance maturity dates by notional, rate, and pledged collateral type. Under stress, model FHLB access at reduced borrowing capacity — haircuts on pledged collateral increase under market-wide stress, and FHLB advance availability tightens as collateral values decline. Community banks that rely heavily on term FHLB advances often assume full rollover in normal scenarios; the stress assumption should model partial or complete rollover failure in the 30-day combined stress scenario.

Brokered CDs: Contractual maturity is the floor. For stress modeling: treat all brokered CDs maturing within 90 days as running off at maturity with zero rollover assumption. Retail brokered deposits may have early withdrawal provisions that introduce additional behavioral outflow risk.

Repo agreements: Map each counterparty relationship, collateral type, and maturity date individually. Under severe market stress, overnight repo counterparties stop rolling. Term repo counterparties may invoke margin calls or early termination provisions. Stress assumption: treat all overnight repo as non-renewable; model term repo with conservative haircuts on collateral values equal to or exceeding the LCR non-HQLA haircuts.

Federal funds purchased: Typically overnight. Model as 100% runoff under any stress scenario above “watch” level — banks that need Fed funds in a crisis aren’t getting them at normal prices or volumes.

The output is a net cumulative cash flow projection showing the largest gap between projected outflows and projected inflows. That gap — its size and the window in which it occurs — is the vulnerability your CFP needs to specifically address.

For the scenario framework that these maturity ladder outputs feed into, see our detailed breakdown of CFP liquidity stress testing scenarios and assumption methodology.

Contingent Draw Modeling: The Hardest Input

Contingent draws are technically the most challenging component. Unlike deposit run-off — which has LCR floor assumptions to anchor from — or wholesale funding maturities — which have contractual dates — contingent draw probabilities require behavioral assumptions with limited historical data at the institution level.

The LCR framework provides regulatory starting points:

These are Basel III regulatory floors. They are not empirically validated behavioral rates for every institution and market condition.

March 2020 as a Behavioral Calibration Point

March 2020 provided the clearest modern dataset on contingent draw behavior under combined stress: major U.S. banks experienced rapid large-scale drawdowns of corporate revolving credit facilities. The mechanism was primarily precautionary — investment-grade companies drawing down revolvers to secure liquidity before credit markets froze, not because they were in immediate financial distress.

The lesson for stress testing: relationship-based revolvers to investment-grade borrowers still get drawn in a crisis. The 10% floor assumption for non-financial corporate revolvers significantly understates behavioral draw rates under combined stress conditions where borrowers are uncertain about future market access.

By H1 2025, undrawn credit lines, liquidity facilities, and other binding commitments at U.S. banks had risen to $2.5 trillion, according to Structured Finance Association research. Growth in bank-to-NBFI committed facility relationships means stress in nonbank channels can transmit to bank liquidity positions faster than pre-2020 models anticipated. For institutions with material NBFI counterparty exposure, the 40% financial institution revolver draw rate may be inadequate under combined stress.

Derivative Collateral Calls Under Downgrade Scenarios

A dimension many institutions underweight: derivative collateral calls triggered by a credit rating downgrade. ISDA master agreements typically include ratings-based threshold provisions — if the institution’s credit rating drops one or two notches, the required collateral posting under existing swap agreements increases.

Model this explicitly: for each notch of downgrade in your idiosyncratic stress scenario, calculate the additional collateral posting required across your derivative book. The Federal Reserve’s proposed 2026 stress test scenarios (released October 2025) incorporate severe market shocks that would trigger ratings-based margin provisions for institutions with significant rate swap or credit derivative positions.

Connecting Stress Test Outputs to CFP Action Triggers

Stress testing techniques produce quantitative outputs — survival horizon in days, funding gap in dollars, collateral shortfall by scenario. Those outputs need to directly drive the CFP’s action triggers and escalation protocols.

Most institutions structure three tiers:

The survival horizon calculation: sum available High Quality Liquid Assets (HQLA) and committed, accessible contingent funding sources, then subtract projected cumulative net outflows day-by-day under the stress scenario. The point where cumulative outflow exceeds cumulative available liquidity is your survival horizon.

Document the inputs to that calculation explicitly. It’s the first thing examiners request.

What Examiners Actually Look For

Based on FDIC and Federal Reserve examination guidance on liquidity risk, examiner focus areas in stress testing reviews include:

Assumption documentation. Can you demonstrate where your run-off rates came from? Historical deposit behavior data? Regulatory floor plus an institution-specific adjustment? External peer benchmarking? “We used industry standards” is not a defensible answer. Document the methodology and the data.

Idiosyncratic scenario specificity. Does the idiosyncratic scenario model risks specific to your institution? A bank with one depositor representing 8% of total balances that doesn’t specifically stress that relationship’s outflow is conducting inadequate idiosyncratic modeling. The scenario should be tailored to your actual concentration profile.

Concentration identification. Does the stress test identify deposit concentrations, wholesale funding maturity concentrations, and contingent exposure concentrations explicitly? Concentrations should be identified, quantified, and stress-tested at the relationship level for material concentrations.

Behavioral vs. contractual assumptions. Are you using contractual floor assumptions in categories where behavioral analysis would produce materially different results? Examiners increasingly push back on institutions that apply the 40% non-operational corporate deposit floor when their actual depositor behavior history suggests higher sensitivity.

CFP integration. Do stress test outputs actually connect to CFP action triggers? Or are the stress test and CFP separate documents that don’t reference each other’s conclusions? Examiners look for the linkage — the stress test survival horizon should map directly to the CFP trigger level that activates at that threshold.

Your early warning indicators framework is the operational bridge: if your EWIs aren’t calibrated to your stress test survival horizon thresholds, the monitoring program is disconnected from the quantitative analysis it’s supposed to support.

So What?

Regulators aren’t looking for a stress test that shows you can survive 30 days. They’re looking for a stress test whose assumptions can be defended when the scenario turns real — because when it’s real, you don’t get to change the inputs.

The modeling discipline isn’t separate from risk management. Getting run-off rates right, building the wholesale maturity ladder accurately, and sizing contingent draw exposure correctly is what makes the difference between a stress test that works as a risk management tool and one that SVB’s experience showed us — where the stress test identified a real problem and the response was to fix the model rather than the liquidity.

The assumptions are the stress test. Build them first.