Verified development
The CNB explains the CRR3 output floor as a lower bound for total risk exposure relative to standardised calculations, introduced gradually. Model-using institutions therefore also need reliable standardised calculations; the floor is not simply a feature of one credit model. [1]
The following sections are my analytical commentary and hypothetical examples, not findings about a specific bank.
A number can rise for different reasons
A risk-weighted result may change because of new business, borrower quality, corrected data or methodology. Separate these drivers. An increase does not automatically prove deterioration, and a decrease does not necessarily demonstrate lower economic risk.
Use a comparable scope. Changes in consolidation or classification need explanation before interpreting percentages. Identify the rules and transitional provisions actually used in the calculation.
Hypothetical example: an input correction
Part of a portfolio is misclassified in the standardised view. Model assessments remain unchanged, but correcting the input alters the comparison calculation. Explain the source and scope rather than present it as a sudden change in borrower behaviour.
I would trace source fields, classification and totals before and after correction. Also identify other reports using the same information. This connects capital calculations with data quality and change management.
Where further study is needed
Specialist work requires the applicable rules and level of application. This article describes an analytical principle rather than a calculation guide for a particular bank. My foundation is comparing data journeys and documenting differences; regulatory detail requires separate study and expert verification.
A real example: measure framework changes separately
Since 2025, the EBA has used its CRR3/CRD6 dashboard to monitor the new framework’s impact. This is a concrete European implementation-monitoring example, not a failure at a named bank. It connects reported information with projections over the output-floor phase-in and under full implementation.
I would distinguish the published methodology, aggregate monitoring and an institution’s particular calculation. A sector-level exercise cannot by itself establish one bank’s capital requirement. The regulation also contains conditions and transitional provisions. The following illustrations explain an analytical comparison; they are neither a legal interpretation nor a complete instruction for calculating regulatory requirements.
A hypothetical calculation with two inputs
Consider an invented comparison mechanism. The internal result is 55 units and a standardised reference base is 100 units. For illustration, apply 60%; this is not presented as the coefficient applicable in a particular year. Comparing 55 with 60 produces 60. If the reference base is corrected to 110, its comparison value becomes 66. The outcome changes even though the internal result remains 55.
The example shows why both calculation branches matter. It does not establish that the bank has originated riskier loans or that borrowers’ repayment capacity has suddenly deteriorated. The change might reflect corrected classification in the second data stream. For an actual calculation I would first confirm the consolidation level, exposure categories and transitional treatment with the appropriate specialist. Correct arithmetic without that context can still lead to a misleading conclusion.
I would retain original and corrected values in the working table. Each change would identify its source, date and reason: new business, maturity, methodology, correction or an unexplained difference. “CRR3 impact” is too broad a label for this purpose. It cannot distinguish an expected consequence of implementation from a technical discrepancy that still requires remediation, nor help another analyst reproduce the comparison.
How I would prepare the validation trail
I would begin by describing the two underlying populations. Do they share a reference date and cover the same entities? Are amounts expressed in comparable currencies and units? How are post-close corrections handled? These questions precede percentage comparisons. Different underlying populations do not become comparable simply because their results share a chart or use the same number of decimal places.
For a sample of exposures I would trace the source information, classification and contribution to the reference calculation. If an attribute is wrong, I would identify other reports that use it. Correcting only one table can improve that number while creating a new inconsistency elsewhere. Remediation should therefore address the actual origin of the discrepancy and identify which historical and future outputs require reprocessing.
Explaining the outcome to management
I would structure the explanation in three layers. The first states what changed and its scale. The second identifies the principal drivers and their contributions. The third provides traceable calculations and assumptions. This allows readers to understand the message without choosing between an impenetrable technical appendix and a vague headline that cannot be checked or challenged.
I would explicitly distinguish confirmed facts from points awaiting specialist review. If a change depends on interpretation of a rule, it should not be presented as a completed data correction. An analyst can prepare alternative calculations and their consequences, while the relevant specialist confirms the applicable treatment. That preserves accountability without stopping technical work that can be carried out independently of the final interpretation.
Data corrections and rule changes need different treatment
A data correction needs evidence of the wrong original value and its authoritative replacement. A rule change requires a rationale and an effective date. Combining both into an undifferentiated technical request risks applying a later rule to an earlier period, or leaving an error in a report that should be restated. I would map affected outputs, periods, owners and required specialist confirmations before implementation.
I would also compare old and new implementations using identical inputs before changing the data. A second comparison could then isolate the effect of corrected inputs. This gives reviewers an intelligible bridge and helps establish what a rollback would actually reverse. A result that changes as expected is more persuasive when the team can explain each contribution rather than accept only the final total.
Summary and recommendations
For a material capital-result movement I would ask for a clear explanatory bridge and a comparable population. Separating economic changes, framework implementation and data corrections matters more than quickly producing a single percentage. My contribution would be tracing the data, checking relationships and documenting alternatives. The regulatory conclusion should rest on a confirmed application of the relevant provisions to the institution in question.
Sources and links
- ČNB — Výstupní práh v CRR3, 15 November 2024
- EBA — CRR3/CRD6 dashboard
- EUR-Lex — Regulation (EU) 2024/1623
Foreign cases provide analytical lessons; their legal frameworks and measures are not automatically applicable to Czech banks. Hypothetical examples and proposed methods do not describe inspections I personally conducted.