Rating Methodology Governance: HKSI Paper 4
A CRA should use rigorous, systematic rating methodologies that can be objectively validated where possible.
- It should analyze relevant known information, maintain supportable records, review methods and controls at least annually through an independent function, monitor published ratings, and refrain from rating when information is insufficient for a credible result.
On this page10 sections
- A rating methodology is a governed process
- Use relevant information and record the reasoning
- Validate models and review them independently
- Change a methodology with control and transparency
- Continue surveillance after publication
- When there is not enough reliable information
- Structured finance requires specific attention
- Worked scenario: a new product with thin data
- Worked scenario: the model passed last year’s review
- Methodology governance checklist
A rating methodology is a governed process
For HKSI Paper 4, rating methodology is not just a model formula. The public syllabus includes the rating process and the CRA’s responsibilities under its Code of Conduct. The SFC CRA Code expects written procedures, rigorous and systematic methodologies, qualified representatives, sufficient information, review of models and controls, ongoing monitoring and appropriate disclosures. A sound rating process connects data, assumptions, analysis, committee judgment, publication and later surveillance. If one link fails, the resulting rating may no longer be credible or properly supported.
The distinction between a methodology and a rating decision is useful. A methodology describes the framework the CRA applies to a type of rating target. The rating is the result of applying that framework to the facts available for a particular issuer, security or agreement to provide credit. Governance keeps the methodology current and applied consistently; the rating team examines the specific evidence and records how it reached its conclusion. An analyst should not quietly improvise a new method to reach a preferred outcome.
- Method quality
- Rigorous, systematic and objectively validated where possible, including through historical experience or back-testing
- Evidence
- Analyze all information known to the CRA that is relevant under its published methodology
- Review
- Independent formal function periodically, and at least annually, reviews methodologies/models and systems/controls
- Surveillance
- Monitor published ratings on an ongoing basis and review at least annually unless clearly designated otherwise
- Insufficient information
- Do not assign or continue a rating when the CRA lacks quality information to support a credible rating
Use relevant information and record the reasoning
The CRA Code says a rating should be based on a thorough analysis of information known to the CRA and relevant to its analysis under its published methodologies. That requires more than collecting a large data set. The team should determine whether the information is relevant, current, reliable and sufficient for the rating target. It should test important assumptions, consider contrary evidence and explain how the method treats unusual facts. A pile of documents does not support a rating if the key forecast or collateral data is missing or unreliable.
Records make the process reviewable. The Code says a CRA should keep records supporting ratings and maintain them for at least seven years in writing in Chinese or English, or in a format readily accessible and convertible into one of those written forms. The record should let an independent reviewer understand what information was considered, which methodology version applied, who participated, what issues the committee debated and why the final rating followed from the analysis. The article does not replace the statutory Keeping of Records Rules; consult the applicable rule and current manual for detailed requirements.
A good record is not a verbatim transcript of every conversation. It is an adequate audit trail. For example, if a committee rejects management’s optimistic forecast because historical conversion rates and current contract data do not support it, the file should explain that comparison. If the rating depends on an assumption that cannot be independently verified, the file should say how the CRA assessed the limitation and whether it affects the rating’s scope or disclosure.
Validate models and review them independently
The CRA Code states that methodologies should be rigorous and systematic and, where possible, produce ratings that can be subjected to some form of objective validation based on historical experience, including back-testing. Validation asks whether the method behaves as intended when compared with outcomes and whether the model remains suitable for the rating class. It does not promise that a model can predict every default or that one historical sample proves future accuracy.
The Code calls for a formal review function, independent from the business lines principally responsible for rating the relevant classes of target. That function should periodically, and at least annually, review methodologies and models, significant changes, and the adequacy and effectiveness of systems and internal controls. The review findings should be comprehensively recorded in a written report, a copy provided to the SFC forthwith upon completion, and deficiencies addressed. Operationally, this means the model owner should not be the only person deciding whether the model and controls work.
Independent review should look at more than whether the code ran without an error. It may test input quality, data transformations, exceptions, override practices, performance on historical cases, consistency between rating teams and the governance around model changes. If a model was built for one type of issuer, a new product or borrower population may require a feasibility review before the CRA applies it. The Code requires a formal senior review of whether a rating is feasible for a financial product materially different from the products the CRA currently rates.
Change a methodology with control and transparency
A methodology change can improve a rating process, but it can also alter outcomes or comparability. A controlled change should have a documented reason, impact analysis, review, approval, version history and implementation plan. The CRA should consider whether the change affects existing ratings, how teams will apply it, what disclosures or communication are required, and whether retraining is needed. If a change is significant, it belongs in the formal review described by the CRA Code.
A candidate should not infer that every change triggers the same notice or rating action. The answer depends on the methodology, the materiality of the change and the CRA Code’s applicable requirements. But a CRA should not alter its published method informally for one high-revenue client and leave investors unaware of a material process change. Consistent application helps users compare ratings and helps the firm identify unusual overrides.
Overrides should be visible in governance. If an analyst or committee departs from the model output, the file should explain which judgment was applied and why. A model output is not a substitute for professional analysis, but undocumented overrides can hide bias or inconsistent treatment. Repeated overrides in one direction may signal that the model is stale or that commercial pressure is affecting decisions. Review teams should examine patterns, not only individual cases.
Continue surveillance after publication
A rating is not necessarily a one-time publication. The CRA Code expects ongoing monitoring and updating, with a review of the target’s creditworthiness at least annually, unless the rating clearly indicates that ongoing surveillance does not apply. When the CRA becomes aware of information that might reasonably be expected to require revision or termination, it should initiate a review consistent with the applicable methodology and update the rating in a timely way where appropriate.
This obligation is not a promise that every market movement changes a rating immediately. A price change can be a signal, but the CRA should assess the underlying credit factors under its method. Conversely, the CRA should not wait for an annual review when it learns of a material event that may require action. Monitoring uses appropriate personnel and financial resources, and the record should show what information triggered a review and how the outcome was decided.
Surveillance also connects to conflicts and information controls. An issuer may ask to delay a downgrade until after a financing, or a large subscriber may argue that a rating should remain unchanged. These pressures do not replace the methodology. The CRA should assess relevant evidence, preserve the independence of the decision and make any required rating communication in a timely and accurate manner.
When there is not enough reliable information
The Code expects the CRA to assess whether it has sufficient skilled personnel and access to enough information to make a proper assessment. It should take reasonable measures to ensure that information is of sufficient quality to support a credible rating. If the CRA lacks adequate information, it should refrain from assigning a rating and ensure that an existing rating is withdrawn. An innovative product with limited historical data may still be assessed in some cases, but the Code requires clear disclosure of the rating’s limitations in a prominent place in the rating report.
This distinction avoids two errors. One is treating uncertainty as a reason to publish any rating as long as a disclaimer is added. If the available information cannot support a credible rating, the CRA should not issue or continue it. The other is claiming that a new product can never be rated. The Code instead calls for a feasibility review when the product is materially different, and requires the CRA to refrain if complexity, structure or lack of robust data raises serious questions about whether a credible rating can be determined.
Structured finance requires specific attention
A structured finance rating can depend on an asset pool, transaction structure, payment waterfall, credit enhancement, servicing assumptions and data about underlying borrowers. The CRA Code says the agency should reassess whether its existing methodologies and models remain appropriate when the risk characteristics of the underlying assets change materially. If the complexity or structure of a new type of product or the lack of robust asset data raises serious questions about the ability to determine a credible rating, the CRA should refrain from issuing it.
The methodology review should therefore ask whether the model assumptions reflect the asset pool actually being rated. A performance history from one market, borrower type or underwriting regime may not transfer to another. A change in underwriting quality, concentration, servicing or legal structure can undermine an assumption even if the spreadsheet still calculates a score. The committee should understand the exposure and the data limitation, not only the rating symbol produced by a model.
The Code also expects relevant structured-finance information to be disclosed so investors and other CRAs can make their own independent analyses. Rating announcements should state whether the issuer has told the CRA that all relevant information is public or whether some remains non-public. This transparency is distinct from the CRA’s own duty to reach a well-supported opinion. A public data package does not excuse weak analysis, and a strong analysis does not justify obscuring important limitations.
Worked scenario: a new product with thin data
A CRA is asked to rate a product backed by a new class of loans. The proposed methodology adapts a model from a related asset class, but the new loans have shorter histories, different repayment behavior and incomplete performance data. The issuer offers additional collateral analysis but asks for a rating before a large offering. What should the CRA consider?
- Ask whether the product is materially different from those the CRA currently rates and send the feasibility question to the formal senior review function.
- Test whether the existing method and model remain appropriate for the changed risk characteristics and whether the available data is robust enough for a credible assessment.
- Consider whether the team has enough skilled staff and information, and document material limitations and assumptions.
- If a credible rating cannot be supported, refrain from assigning it. If the rating can be supported but limitations remain, disclose them prominently as required.
- Do not let the issuer’s timetable or fee override the methodology review and analysis.
The best answer does not jump straight to “publish with a disclaimer” or “all new products are prohibited.” It follows the Code’s sequence: feasibility, method suitability, data quality, credible analysis and clear disclosure where appropriate. If serious data or complexity issues mean the CRA cannot determine a credible rating, it should not issue one.
Worked scenario: the model passed last year’s review
The independent methodology function approved a model twelve months ago. Since then, the underlying asset pool has shifted toward a borrower group with a higher delinquency rate. A committee member argues that the model should remain in use until the next scheduled annual review because it has not yet been twelve months. What is the flaw in that reasoning?
The annual review is a minimum recurring control, not a reason to ignore material changes in risk characteristics. The CRA should assess whether the methodology remains appropriate when the underlying assets change materially and initiate a rating review when new information may require revision or termination. A previous approval does not certify that a method will remain suitable regardless of new data. The committee should document the assessment and take timely action consistent with the method.
Methodology governance checklist
- Use the published method consistently and apply only relevant factors to the rating decision.
- Check that the CRA has skilled staff, sufficient information and data quality to support a credible rating.
- Retain an adequate record of sources, assumptions, committee reasoning, model version and overrides.
- Validate models objectively where possible and review methodology, significant changes and controls through an independent function at least annually.
- Review a materially different product before deciding whether the CRA can credibly rate it.
- Monitor published ratings and review material new information promptly, while maintaining at least annual creditworthiness review where surveillance applies.
- Reassess structured-finance methods when underlying risk characteristics materially change.
- Disclose rating limitations and structured-finance information accurately and prominently where required.
The public syllabus identifies the rating process and the CRA Code as Paper 4 material. The SFC Code supplies the operational duties summarized here, while the assigned HKSI manual gives the examinable detail and terminology. Use current SFC sources for live compliance decisions, since an exam manual and an online code may not have identical publication dates.
Common questions
How often must a CRA review its methodology?
The SFC CRA Code calls for a formal independent review function to review methodologies, models and controls periodically and at least annually. Significant changes are also within the review requirement.
How often must a published rating be reviewed?
The Code expects ongoing monitoring and, unless the rating clearly indicates no ongoing surveillance, a review of the target’s creditworthiness at least annually. Material new information may require an earlier review.
Can a CRA publish a rating when data is incomplete?
It depends on whether the information is sufficient for a credible rating. The CRA should refrain if it lacks adequate quality information. For a product with limited historical data, the Code calls for prominent disclosure of limitations where a rating is supportable.
Does back-testing prove that a rating model will predict defaults?
No. Back-testing can help objectively validate a method against historical experience, but it cannot guarantee future performance or eliminate uncertainty.