A conventional stress scenario asks what happens when specified conditions deteriorate. Reverse stress testing begins with an adverse outcome and investigates circumstances that could produce it. What interests me is the need to connect narrative, data relationships and timing. The result should be more than a dramatic description of a world event: it should establish an explainable chain of effects on a particular bank’s business model, financial position and capacity to respond.
Geopolitical risk can connect borrower income, input prices, payment channels and technology suppliers. I would not place it exclusively in a macroeconomic category. Analysis needs to identify exposures and processes that could be affected together, distinguish observations from assumptions and reveal information gaps. This connects credit assessment with operational resilience without implying that the disciplines are interchangeable or that an analyst can replace the specialist judgement required in each.
A real exercise: euro area, July 2026
On 31 July 2026 the ECB published results of its geopolitical reverse stress test covering 110 directly supervised euro-area banks. Participants designed scenarios producing a 300-basis-point CET1-ratio depletion. The ECB identified improvement areas including scenario consistency, realistic mitigating actions and solvency-liquidity interactions.
This was a supervisory exercise, not a prediction that this decline will occur or an identical capital requirement for every Czech bank. My question is how an institution connects a narrative to financial mechanisms and demonstrates that its proposed defence would work under the same stressed conditions. The following is my illustrative reasoning, not a reproduction of confidential submissions or an assessment of any participating bank’s individual result.
Start by clarifying the outcome
I would establish whether the target is a ratio, an absolute loss or disruption of a service. A capital ratio can decline through its numerator or denominator. If risk-weighted assets increase, the movement cannot be explained entirely as a loss. A calculation may be arithmetically accurate while answering the wrong question. The opening record should therefore specify the calculation basis, reference date and the quantities that are allowed to change.
Timing comes next. Immediate loss of a payment channel, three months of weaker sales and a multi-year business-model shift are different effects. For each material impact I would record its onset, duration and potential recovery. A timeline reveals whether mitigation arrives after the problem has already materialised. It also provides a useful common reference for credit, finance, liquidity and operational specialists whose models may otherwise assume incompatible sequences of events.
An illustrative combined shock
Consider a fictitious bank with CET1 capital of CZK 10 billion and risk-weighted assets of CZK 50 billion. Its starting ratio is 20%. Holding the denominator constant, a CZK 1.5 billion capital decline produces 17%, a fall of three percentage points. This is simplified arithmetic, not a regulatory model or the calibrated position of any real institution, and it deliberately excludes other mechanisms to make the relationship visible.
The denominator may not remain constant. If capital falls to CZK 9 billion while risk-weighted assets rise to CZK 55 billion, the ratio is approximately 16.36%. A narrative cannot therefore be represented only by subtracting losses from capital. I would identify which mechanisms change exposure, risk weights, income and costs, and whether an effect has been counted twice through different model inputs or independently maintained reporting views.
A hypothetical narrative might combine export restrictions, expensive energy and disruption of a critical supply chain. Exporters lose cash flow, other borrowers face higher costs and shared technology dependency limits the bank’s response. These relationships are not automatic facts. Each would be an explicit assumption to test against relevant data with the responsible specialist. A credible narrative should identify where evidence is strong and where a cautious range is more appropriate than a single point estimate.
Examine dependencies beneath the headline categories
A borrower’s registered country is insufficient. A domestic company may depend on exports elsewhere or one overseas supplier. I would seek significant customer and supplier relationships, currency flows and financing needs. Unknown items should stay visible. An exact-looking exposure percentage based on incomplete mapping can imply greater certainty than the evidence supports, especially when the same missing dependency affects several borrowers that appear unrelated in the reporting hierarchy.
For IT I would map services required to execute mitigation. If the plan assumes rapid limit changes, do the relevant applications remain available during supplier disruption or unusual processing demand? I would not assess the dependency independently of the business response. An unavailable support application can delay the very action on which the financial model relies to reduce losses or preserve liquidity, making an otherwise reasonable assumption inconsistent with the operational scenario.
Challenge mitigation under the same conditions
Asset sales, reduced new lending or additional funding may look straightforward in a spreadsheet. I would ask about price, execution time and market conditions within the scenario. When several institutions are affected, counterparty capacity and demand may differ from normal periods. The relevant question is not whether an action appears in a plan but whether it remains compatible with the conditions causing the problem and with other actions assumed simultaneously.
For each action I would create a brief evidence card: trigger, decision-maker, supporting systems, required time and third-party dependencies. An untested assumption should be labelled. A time-limited tabletop exercise could investigate selected steps without carrying out an actual transaction. It can demonstrate whether participants can obtain the necessary information and identify the person authorised to decide, while exposing delays concealed by a high-level recovery-plan description.
Summary and recommendations
I would use reverse testing as a structured investigation of vulnerability rather than a forecast. Align units, separate capital from risk-weighted assets, create a timeline and challenge whether assumptions can coexist. Require an execution argument for mitigation and disclose data gaps. Compare several routes to the outcome and look for shared weaknesses. The most valuable insight may be a dependency absent from separate credit and availability reports, made visible to the people able to assess and address it.
Decisions need a review date
For the proposed scenario, I would separate immediate, conditional and preparatory actions. An immediate action might close a data gap. A conditional action could constrain new exposure after an agreed trigger. A preparatory step might explore alternative funding whose availability has not yet been confirmed. Each category needs an owner and a review date. Without that distinction, an action list can suggest readiness even when some steps cannot be executed under stress. I would explicitly flag dependencies on third parties. The scenario then produces more than a loss estimate: it explains what management should monitor and when assumptions require reconsideration. Evidence of readiness should concern executable steps rather than the length of the action list.