Verified development
On 4 August 2026, the CNB published supervisory stress-test results for thirteen banks representing approximately 92% of sector assets. The tested group’s capital ratio was 20.7% at end-2025 and would fall to 18.4% under the adverse scenario. The CNB assessed resilience as strong. [1]
The following sections are my analytical commentary and hypothetical examples, not findings about a specific bank.
What the aggregate result does not establish
A favourable sector assessment does not describe every lending process within each bank. It does not prove that all inputs are flawless or that no segment has weaknesses. Separate overall resilience, institution-specific results and the quality of underlying evidence.
The reference date matters too. End-2025 data does not automatically describe an autumn-2026 portfolio. Subsequent changes may be commercial, economic or methodological. A current review should explain the link between the tested starting point and more recent information.
Hypothetical example: a resilient bank, a sensitive segment
A well-capitalised bank can still have a growing group of borrowers dependent on one market. The aggregate test does not itself explain how that group is monitored. I would trace its identification, exposure changes and the scenario assumptions applied.
Separate results before and after management responses. A modelled risk reduction means something different if the action needs lengthy preparation or external approval. This is an analytical question, not a claim about a deficiency found at a particular Czech bank.
Usefulness for review work
I would treat stress results as the start of a focused discussion about sensitivity, data and responses. A useful output connects scenario, evidence and the explanation of capital movements. Traceability helps specialists distinguish an economic vulnerability from a data-processing discrepancy.
A real event and the boundaries of its conclusion
The CNB’s results published on 4 August 2026 are a specific Czech supervisory case. Their scope matters: they are not a prediction of the actual future or an assurance that every bank can absorb any combination of events. The analysis below is my proposed way of examining the result, not an account of a confidential inspection.
I would distinguish four layers: the starting position, the assumed economic path, its translation into the bank’s balance sheet and the interpretation of the outcome. A weakness in one layer may not be visible in the final percentage. Even an arithmetically correct calculation can have limited meaning when it covers a different population from the one the specialist team believes it is discussing.
A hypothetical example: a capital ratio has two components
Consider an invented bank with CZK 20 billion of capital and CZK 100 billion of risk-weighted assets. The ratio is 20%. Under a scenario, capital falls to CZK 18 billion and risk-weighted assets increase to CZK 120 billion. The resulting ratio is 15%. The movement therefore reflects more than the CZK 2 billion reduction in capital; the larger denominator also matters.
Keeping the original denominator would produce 18%, a significantly different picture. I would therefore compare more than the finished ratios. Both capital and risk-weighted assets need separate bridges from the starting position through the relevant changes. This is a simplified analytical illustration, not a reconstruction of a CNB or bank calculation; the actual capital framework includes further rules, adjustments and constraints.
I would prepare the bridge for each period. Credit losses may dominate one year, followed by income developments or changes in risk weights later. An end-point ratio can conceal a substantially weaker intermediate position. I would examine the minimum over the full horizon and the assumed timing of management action. A sufficient final position does not automatically resolve a shortfall arising earlier in the scenario.
How I would check the inputs
I would begin with an inventory of datasets and reference information. For every material input I would identify the date, population, owner and connection to its source system. Contract counts and aggregate amounts would be reconciled before applying the scenario. Where reconciliation fails, I would describe the difference and its possible effect. A manual balancing entry without an explanation would not establish that the input is correct.
For a selected sample I would trace records from the contract to the segment used in the test. I would check whether transformations confuse missing values with zero and whether currency conversions use comparable assumptions. A weakness would not immediately justify rejecting the entire exercise. First I would establish its scope, materiality and the possibility of rerunning the affected calculation, keeping the conclusion proportionate to the evidence.
Discussing management actions
I would want proposed action separated from the mechanical effect of the scenario. Selling an asset, reducing expenditure and obtaining new funding have different conditions and lead times. Who can approve the action, what preparation is needed, and does it depend on a counterparty affected by the same stress? Its contribution to the result should be explained rather than embedded as an invisible adjustment.
I would also look for double counting in an improving intervention. If a plan both reduces a lending portfolio and increases the associated income, the timing needs to explain that combination. It is not automatically proof of error, but it is a relationship worth clarifying. Such questions illustrate the benefit of combining economic reasoning, data validation and documentation instead of evaluating each spreadsheet independently.
Sensitivity is a different question
Before adding a more elaborate scenario, I would examine one influential assumption while holding the others constant. This can show whether losses, funding costs or portfolio size drive the result. I would clearly label it as sensitivity analysis, explaining what changes and what remains fixed. It should not be compared with the published supervisory exercise without accounting for different purposes, assumptions and constraints.
When presenting my conclusion, I would also distinguish a sector-wide finding from a conclusion requiring evidence about an individual institution. Aggregate resilience does not mean every bank responds identically. For a specific bank, I would want to understand the starting portfolio, sensitivity to individual assumptions and explanations for differences. That comparison would guide further questions rather than establish a problem in advance.
Summary and recommendations
I would use aggregate results to guide further questions, not to replace individual assessment. A useful working paper connects initial data, the scenario, a bridge explaining the change and a realistic response. A strong result is one that a colleague can follow and reproduce, with an honest explanation of the situations covered by the published conclusion and the situations beyond its scope.