Stress Testing in Banks: Scenarios, Severity and ICAAP

A bank that only measures risk under normal conditions is measuring the easy half of the problem. Stress testing in banks is the discipline of asking what happens when conditions stop being normal — when GDP contracts, when the yield curve shifts 250 basis points, when the three largest borrower groups default in the same quarter. For IIBF Risk Management candidates, this is one of the most reliably examined areas because it cuts across credit, market, liquidity and capital planning in a single framework.

Examiners rarely ask for definitions alone. They ask which technique suits which question, how severity is calibrated, who owns the scenario, and where the output finally lands. This article walks the full chain — from shock design to board table — with the vocabulary the paper expects.

🧭 What Stress Testing in Banks Really Tests

A stress test is a forward-looking assessment of the impact of exceptional but plausible adverse events on a bank's earnings, asset quality, liquidity and capital. It is deliberately not a forecast. A forecast asks "what is most likely"; a stress test asks "what would hurt, and could we survive it".

Three properties separate a genuine stress test from a routine sensitivity report:

• Severity beyond experience. The shock should sit in the tail, not in the ordinary volatility band that Value at Risk already captures.
• Plausibility. A shock nobody can explain economically gets dismissed by the board and produces no management action.
• Actionability. Every run must end in a decision — capital, limits, provisioning, funding or pricing.

The relationship with VaR is a favourite exam hook. VaR quantifies loss at a confidence level under assumed distributions; it says nothing about the shape of the loss beyond that cut-off, and it degrades badly when correlations break. Stress testing is the complement that probes exactly that region — which is why the chapter on Value at Risk and the stress-testing material are almost always taught back to back. Read them as one unit: model-based measurement plus judgement-based challenge.

The broader framing — risk identification, measurement, monitoring and control as a continuous cycle — is set out in Risks and Risk Management in Banks, and stress testing sits squarely in the "control" leg of that cycle.

💡 Exam Tip: If a question contrasts VaR with stress testing, the discriminator is almost always probability. VaR attaches a confidence level to a loss; a stress scenario usually carries no assigned probability at all.

🧪 Sensitivity Tests vs Scenario Tests

The first classification every candidate must hold is sensitivity versus scenario. They answer different questions and are graded differently in the exam.

Sensitivity analysis

A sensitivity test moves one risk factor by a defined amount and holds everything else constant. Shift the entire yield curve up by 200 bps and read the change in economic value of equity. Depreciate the rupee by 10% and read the revaluation impact on the net open position. Raise the probability of default of a single rating grade by one notch and read the provisioning gap.

Sensitivity tests are cheap, fast, easy to repeat monthly, and easy to explain. Their weakness is that they are economically incoherent — in the real world a 200 bps rate shock does not arrive without credit spreads, deposit behaviour and collateral values also moving.

Scenario analysis

A scenario test moves a coherent bundle of factors together according to a narrative: a slowdown drives GDP down, unemployment up, corporate cash flows down, property prices down, and policy rates in whichever direction the story supports. Because the factors move jointly, second-order effects appear that no single-factor test can produce.

Scenario work is where the analyst adds value, and where the exam sets its harder questions. The technique is closely related to the loss-estimation approach described in scenario analysis in operational risk, where expert workshops replace statistical fitting because loss data is thin.

Single-factor vs integrated tests

The second classification is single-factor versus integrated. A single-factor test isolates one risk type — say, a credit shock only. An integrated stress test runs the same macro narrative through credit, market and liquidity books simultaneously and adds up the impact on capital. Integration is what surfaces the compounding problem: rating downgrades raise risk weights while mark-to-market losses erode the capital those weights are measured against, and both happen while deposit outflows force asset sales at distressed prices.

🌪️ Historical, Hypothetical and Reverse Stress Tests

Scenarios are sourced in three ways, and the exam expects you to know the trade-off of each.

Historical scenarios replay an actual episode — the 2008 global financial crisis, the 2013 taper-driven rupee and bond sell-off, the 2020 pandemic shutdown. Their great advantage is credibility: nobody can call the shock implausible because it happened. Their weakness is that the balance sheet, product mix and counterparties of the bank today may bear little resemblance to the portfolio that suffered then.

Hypothetical scenarios are constructed by economists and risk managers to capture vulnerabilities the historical record has not yet tested — a concentrated exposure to one infrastructure sector, a new digital deposit base with untested behavioural stickiness, or a climate-transition shock to a carbon-heavy loan book. They are forward-looking but attract the challenge "why this severity and not another".

Reverse stress testing inverts the whole exercise. Instead of starting with a shock and computing the loss, it starts with the failure outcome — capital falling below the regulatory minimum, or the bank being unable to meet obligations — and works backwards to identify the combinations of events that would produce it. Its value is diagnostic: it exposes the specific concentrations and dependencies on which the business model actually rests, which a menu of standard scenarios may never touch.

⚠️ Common Mistake: Candidates describe reverse stress testing as "a more severe stress test". It is not a severity setting — it is a reversed direction of analysis. The endpoint is fixed and the scenario is the unknown being solved for.

Severity calibration is the quiet skill underneath all three. A useful discipline is to define a graded ladder — mild, moderate, severe — anchored to observable history, so that the severe case is defensibly worse than anything in the recent record without becoming a doomsday fantasy that management dismisses. Different risk types also need different clocks: a market shock is calibrated over days, a credit-cycle scenario over eight to twelve quarters, and a liquidity stress over an overnight-to-30-day survival horizon.

📊 Translating Shocks Into PD, LGD, NII, MTM and CRAR

A scenario narrative is worthless until it becomes numbers on the balance sheet. This translation layer is where most marks are won.

• Credit: macro variables feed satellite models that push up probability of default and loss given default. Downgrades raise risk-weighted assets under the standardised approach; higher expected loss raises provisions, which hit the profit and loss account and therefore retained earnings. Both effects squeeze the capital ratio from opposite ends. The underlying mechanics are set out in the Credit Risk chapter.
• Market: rate and spread shocks produce mark-to-market losses on the trading book and AFS portfolio, and a change in economic value of equity on the banking book. The capital consequence is covered in Capital Allocation Against Market Risk.
• Earnings: net interest income compresses when deposits reprice faster than assets, or when the reference index on assets and the funding index diverge — the mechanism explained under basis risk in banking.
• Liquidity: outflow rates on deposits and undrawn commitments are stressed, haircuts on collateral are widened, and the survival horizon is recomputed — the trigger set that activates the contingency funding plan for banks.

The following illustrative table shows how a single macro narrative cascades through a hypothetical bank. The figures are teaching illustrations, not regulatory thresholds.

📌 Remember: A stress test result is a path, not a point. The board wants the quarter-by-quarter capital trajectory and the lowest point it touches, because the trough is what determines whether a capital raise is needed and when.

🏛️ Governance, ICAAP Linkage and Management Actions

Stress testing in banks fails most often not on modelling but on governance. A technically elegant run that nobody owns, challenges or acts on adds no resilience.

Who owns what

• Board: approves the stress-testing policy, the risk appetite it defends, and the capital plan that follows from the results.
• Risk management committee / ALCO: approves scenarios and severity, challenges assumptions, and signs off management actions.
• Risk function: designs and runs scenarios, maintains models, and reports results independently of the business lines.
• Business units: supply portfolio data and behavioural assumptions, and own the mitigation actions in their books.
• Internal audit / independent validation: tests data lineage, model assumptions and whether prior actions were actually executed.

The ICAAP connection

Stress testing is the analytical engine of the Internal Capital Adequacy Assessment Process. Pillar 1 sets minimum capital for credit, market and operational risk on standard rules. ICAAP asks whether that minimum, plus buffers, is enough for this bank's actual risk profile — including concentration, interest rate risk in the banking book, and strategic risk that Pillar 1 does not price. The stressed capital trajectory is the evidence base for the answer, and the gap between the stressed trough and the regulatory floor is what sizes the internal capital buffer and the capital plan's timing.

Results must also survive the supervisory conversation, which is one reason the framework has to be re-examined whenever rules change — the horizon-scanning habit described under regulatory risk in banks belongs inside the scenario design process, not beside it.

Management actions

Every reported result should carry a credible action set: slowing growth in stressed sectors, tightening sanction criteria, reducing single-borrower and sectoral concentration, hedging duration, lengthening funding tenor, raising collateral standards, conserving capital by adjusting dividends, or triggering the contingency funding plan. Actions assumed in the model must be ones the bank could realistically execute in the stressed environment — assuming an equity raise in the middle of a systemic crisis is exactly the kind of assumption supervisors reject.

Frequency should match volatility: full integrated runs at least annually with the ICAAP cycle, liquidity stress far more often, and ad-hoc runs whenever a shock is developing. Wider study material across the paper is collected on the Risk Management article hub, and current policy numbers can be checked against the RBI rates reference before you quote any figure in an answer.

🧠 Practice MCQs: Stress Testing and Scenario Design

Want chapter-wise mock tests with 100+ MCQs? Start practising free →

❓ Frequently Asked Questions

🎯 Final Word

Treat stress testing in banks as one continuous chain — narrative, calibration, translation into PD, LGD, NII, MTM and CRAR, then governance and action. Questions in the paper almost always test one link of that chain, so knowing where each technique sits is worth more than memorising definitions. Practise the numeric translation until the direction of every impact is automatic, then work the governance layer until you can name the owner of each step.

Ready to test yourself under exam conditions? Work through the risk chapters and mocks on the CAIIB course page, then attempt a timed set at iibf.store mock tests.