We have the following comments on the overall approach:
• With regards to point 11, we believe that if the objective is to make comparisons across ECAIs “when risk is very low” then a period greater than 3 years should be used. ECAI grades from AAA to AA would qualify as “very low risk”, but rarely contain default events.
• In fact, the length of the period should be defined by the ECAI which should demonstrate that the chosen length (3 or more years) is adjusted to the frequency of defaults observed within the rating grade.
• It must be noted that rating migration should be considered with a particular attention in case of a multi-year observation period is used to observe and calibrate default rate at a rating grade.
• Point 19 – If the default rate breaches the benchmark, we believe that action should be taken within 2 years. If the default rate is being assessed over a rolling 6 month cohort then the breaches can be evidenced before the 2 year period is up. However, the underlying issue is the economic cycle when the default observations are made and that consideration needs to be given to the fact that default rates fluctuate during the cycle.
• Article 2(a) – Our members believe that the items should not be limited to “Corporates” only. We recommend that all observations should be used, based on the fact that ECAI’s ratings are equivalent across the sectors (e.g. Corp AAA = Banks AAA = Sovereign AAA).
• Article 3 (2) – Similarly to our comment regarding Point 19 above under Q1, AFME members believe that a period longer than 3 years should be used.
• Article 4 (b) – Further clarity is required on the quantification of “sufficiently numerous”
• Article 5 – The article proposes to count withdrawn ratings at 50% in the denominator. AFME believes that if there are grounds to think that the ratings were withdrawn due to imminent default, they should be considered as defaults. This would imply that the withdrawal should count at 100% in the numerator. Additionally, more clarity is sought why 50% has been chosen for the calculation. We recommend a more accurate weighing methodology to better reflect when in the observation period the withdrawal was made.
AFME believes that there should be a principle that guides the firm specific methodologies rather than a closed approach such as proposed in this ITS (short-run equals 3 years, long-run equals 10 H years).
The industry proposes that the guideline should state: “a short-run and a long-run default rate must be calculated. The time horizons and length of data used for calibration must lead to statistically strong results. If the data quality is poor or the history is short, then the portfolio segmentation and other risk drivers should be adjusted so to adapt the granularity of the calibration to the objective of robustness. Finally a fully documented use of expert judgment and rationale can complement the calibration process if there is a need for management adjustment of calculation output.”
The method for calibration of the default rate by the ECAI and in particular the suggested rule to assess the minimum size of the pool used for calculation of the default rate should not be limited to the unique approach “number of rated items to be greater or equal to the inverse of the expected long-run default rate”. In this particular case various approaches are possible and should be allowed, such as:
• Bayesian inference to take properly into account the prior estimate (“expected long-run default rate”) and the observed distribution;
• Monte Carlo simulations or boot-strapping to calculate proper confidence intervals around values in the available data history.
• Article 12 – We believe that the minimum standard for assigning creditworthiness should not be limited to just size, sector and geography. AFME believes that the metrics for size should be better defined and that historical and projected financial performance should also be considered.
We would like to confirm that the internal mapping for specialised ratings, such as short US municipal ratings, would also be captured by this provision.
• On back testing and monitoring of the mapping breach, we believe that the principal objective of the ITS should state: “an objective breach-criteria (for example default rates breaching the benchmark by a magnitude that is material in the context of the firm’s own portfolios and over a notably long and stable period) must be defined by the institution. The mapping must be back-tested regularly against the breach-criteria as described in the regulatory approval process.
• Regarding the formula for confidence interval, we understand that it’s a binomial confidence interval. However, we recommend that the EBA seeks for a more robust approach than the addition of two defaults to the populations. This methodology will lead to higher tolerances for the strongest rating categories which seems inappropriate.
We agree with paragraph 34 of the Cost Benefit Analysis that the main impact of this regulation will be that on the capital requirements of banks. We would note that the impact will not be confined to the standardised approach, since ratings are used in other parts of the regulatory framework – for example the ratings based approach for securitisation, collateral and guarantor eligibility and large exposures exemptions.
Without the publication of the associated mapping tables, it is therefore impossible for us to comment on the impact of the draft ITS.