Against the background of the meaningfulness of a qualitative assessment, inclusion of qualitative templates should be reconsidered. Since a group report is involved and the group has heterogeneous business models, the meaningfulness of the information would be diluted. These distortions became quite clear in the IFRS 9 EBA benchmarking exercise in October 2019. In our opinion, the objective of the exercise can be achieved by a purely quantitative assessment. Especially as, in our opinion, concluding there are differences in modelling and parameterisation sheds no additional insights. Nor should it give any cause for applying the metrics set by the supervisory authority, because this would run counter to the objective of the best possible and unbiased estimate. Even quantitative comparisons are only interpretable to a certain degree. Different outcomes for the amount of risk provision for the same addressees can come about due to the differing quality of data basis for developing statistical approaches, differences in collateral, and differing individual expectations about the future. Likewise, the methods for integrating various economic scenarios in the models do not really lend themselves to comparisons.
Due to the, in our opinion, limited informative value of qualitative assessments, in particular owing to the aforementioned heterogeneity in implementation, we are unable to formulate any suitable responses to this question.
The basis of the consulted ITS of the EBA on benchmarking reporting is the EBA’s mandate arising from Article 78 para. 7 and 8 CRD. The mandate reflects the purpose of Art. 78 CRD, namely ensuring supervisory benchmarking of internal approaches for calculating own funds requirements. Extending the reporting to RWA of Standardized Approach (SA) is not covered by the wording of Art. 78 CRD nor is it required for benchmarking internal approaches. Accordingly, we regard the requirement as being beyond the mandate of Art. 78 CRD.
Moreover, using SA-RWA as a benchmark for internal approaches is unsuitable. RWA of SA, on the one hand, lack sufficient risk sensitivity and, on the other hand, do insufficient justice to the individuality of portfolios in institutions due to the high concentration on a few risk weights. This becomes clear, for example, with retail portfolios, which have no risk differentiation in KSA. In fact, for retail business, only one risk weight is applied.
Furthermore, with the Basel I floor requirement expiring after 31 December 2017, institutions are no longer required to determine RWA of SA. Only up to that date did some institutions determine the Basel I floor on the basis of SA-RWA.
Furthermore, such an extension also contradicts the current efforts and intention of the EU Commission to implement measures to reduce the costs of supervisory reporting systems. We therefore advocate keeping SA-RWA reporting voluntary or doing away with it completely.
The general described approach can lead to smaller deviations between banks over the course of time (for example due to the stipulated dates conflicting with (movable) local public holidays or by various banks interpreting the dates differently). We assume, however, that this risk is low and will be mitigated by the EBA, for example by:
- generally stipulating dates with low potential overlap with (movable) public holidays
- assessing & comparing the dates actually used by banks (in the EBA Benchmark Exercise)
Due to the in part starkly different behaviour of the "new" and "old" reference rates (for example EONIA and ESTER), the risk management models and concepts have to be modified. The "new" reference rates in the overnight segment differ from the "old" reference rates not only in size but also with respect to volatilities. At this point in time, it is too early to tell how the new reference rates will specifically impact the benchmark exercise. The following general statements can be made:
- For EURIBOR, the requirement of an alternative rate seems questionable, because EURIBOR is expected to continue to apply and an assessment with alternative reference rates seems somewhat pointless.
- For LIBOR, the inclusion of an alternative rate could lead to very high costs, because there is still a great deal of uncertainty and no methods have established themselves in the marketplace.
The Consultation Paper states in Article 21, that "The EBA was also aware of the difficulties to fully decompose the changes in variability between the IRB and the IFRS 9 models, in particular due to the lack of definitions related to these intermediate steps".
- We agree that the effort for determining the listed PDs (PD IRB, PD TTC, PD PIT, PD IFRS9) outweighs the additional value. Therefore, we are in favour of only the two PDs stated in the template being assessed.
ANNEX IX, PART II, C 111.01, Column 100: The data field requires "The PD calculated at the counterparty level at the reporting date, representing the probability of a default event within 12 months following the reporting date, as considered in the application of the impairment requirements under IFRS 9.This shall be the PD used to compute the 12 months expected credit loss (ECL amount – 12 months IFRS 9), and associated with the economic scenario 0 in template C111.02."
- The first part requires "the probability of a default event within 12 months", i.e. the 12-month PD.
- The second paragraph defines the PD "as considered in the application of the impairment requirements under IFRS 9. This shall be the PD used to compute the 12-month expected credit loss (ECL amount – 12 months IFRS 9)". Here, it should be borne in mind that banks in part use the residual-term adjusted PD for transactions expiring within the next year to compute the expected loss. For a transaction with a residual term of 6 months, banks inter alia use 6-month PD to compute the 6-month ECL as 1Y EL. In the light of 1YEL computed for IFRS 9, the residual-term adjusted PD (for example 6 months) can also be understood here.
- From our understanding, the 12-month PD is to be stated here. We recommend that this be made clear, however, such that the 12-month PD is to be used and no residual-term adjusted PD be stated, even if one was used for the ECL calculation.
ANNEX IX, PART II, C 111.01, Column 400: "The sum of the expected credit losses to the counterparty over all facilities for the exposures, at the reporting date, as considered for the application of impairment requirements under IFRS 9."
- Here, it should be borne in mind that under IFRS 9 the 1Y expected loss can be based on a residual-term adjusted PD for transactions expiring within the next year. Thus, the ECL required in the template cannot be replicated from the values Exposure * PD * LGD stated in the template. From our understanding, the IFRS 9 1Y ECL is nevertheless to be stated here. We request that this be made clearer and taken into consideration when interpreting outcomes.
ANNEX IX, PART II, C 111.01, Column 200: The data field requires that the weighted LGD is calculated using the "PD" and "exposure value" (exposure-pd-weighted).
- For the sake of clarity, we recommend the terms "PD - 12 months IFRS 9 as defined in column 100" and "Exposure value - IFRS 9 as defined in column 300" be used. In addition, the scenario perspective should either be applied consistently over all the tabs or be completely excluded for corresponding tabs (e.g. 111.01), so that only the baseline scenario is applied. This would make good sense here, because the baseline expected loss is compared with the RWA, thus making the difference from the various baseline parameters (PD IRB, LGD, etc.) apparent.
ANNEX IX, PART II, C 111.01, Column 100 "PD - 12 months - IFRS 9":
The PDs are to be stated at the counterparty level. In general, one can expect banks to have PDs at the counterparty level. But transactions with guarantors can, however, lead to transaction-specific PDs. In addition, IFRS 9 does not require banks to determine PDs at the counterparty level for IFRS 9 purposes. Therefore, banks might not use counterparty-specific IFRS 9 PDs, but transaction-specific IFRS 9 PDs. Therefore, a method should be specified here for aggregating transaction-specific PDs to a counterparty PD (for example exposure weighted).
ANNEX IX, PART II, C 111.01, Column 410 / 400 / 200:
We regard the comparability of values in an inter-bank comparison to be problematic. For a particular counterparty, the one bank may hold collateral, thus leading to a low LGD, whereas another bank determines a high LGD as there is no collateral for the counterparty. In addition, the regulatory approach AIRB vs. FIRB leads to differing regulatory LGDs, which in turn leads to different regulatory ECLs in an inter-bank comparison. So that the second aspect of different regulatory approaches can be taken into consideration, we recommend an additional "Regulatory Approach" column with 0=STA, 1=FIRB, 2=AIRB.
ANNEX IX, PART II, C 111.01, Column 410 / 110: Column 110 requires a PD adjusted for IRB-specific adjustments. Column 410 requires the IRB EL, hence with the possible conservative adjustments (which are not applied under IFRS9) with PD. We point out that comparability with IFRS 9 EL is systematically not possible.
In this context, we consider the expected loss over Lifetime (ELL) and the residual term of the facility as further important values by way of addition to 111.01, 111.02 and 111.03. Where these values should be entered depends on the intention of the individual tabs. If 111.01 serves only to compare 12M-PD and 12M-EL with the benchmarking for supervisory metrics, then ELL can be dispensed with here. Accordingly, ELL and residual term should be included in a different tab.
ANNEX IX, PART I, Article 5 requests PDs and LGDs to be expressed as values between 0 and 1.
- We point out that LGDs can take on a value greater than 1.
ANNEX IX, PART II, C 111.01, Column 110: "PD - IRB without conservative adjustments":
- The specific determination of the "IRB PD without conservative adjustments" is currently implemented by banks as part of the "Future of IRB" projects. We expect that some banks will find it difficult if not impossible to provide "IRB PD without conservative adjustments" data for End-of-Year 2020. We therefore recommend here that regulatory PD is to be used and not "IRB PD without conservative adjustments". Institutions in Approach 3 are also confronted with the problem of determining the PD 12-month for scenario 0.
The data points are clearly defined. We do, however, see serious challenges in extracting the scenario-specific information (macro-economic variables and PD profiles) from the IT systems in the required form. We recommend that the scale of the data collection be reconsidered under the following aspects:
- Work load for banks: The scenario-specific information cannot be extracted by every institution at the individual transaction level without significant effort. We expect several institutions to be confronted with such challenges. In our opinion, ultimately only the weighted, average PD is relevant for the IFRS 9 ECL calculation. Most banks should on the other hand be able to provide the probability-weighted, average PD with reasonable effort.
- Inter-bank comparison: "Scenario 0" seems to be the suitable comparison level for comparing PD profiles across banks. Most banks ought to be in a position to provide this scenario. Additional scenarios are, on the other hand, hardly comparable in our opinion, because the scenarios are defined by the individual banks.
- Taking these aspects into consideration, we recommend reducing the choice of scenarios to "Scenario 0".
In general, the three categorisations described cover the lion’s share of the approaches. Nevertheless, we would like to raise the following aspects:
- Approach 1 describes the use of several scenarios. Approach 1 defines the final IFRS 9 ECL as probability-weighted ECL. Aggregation here is at the ECL level. However, scenarios can also be aggregated earlier at the PD level, so that a probability-weighted PD profile is determined. In this case, the final ECL calculation uses the probability-weighted PD profile. From our understanding, this approach is not listed as such but could be subsumed, however, by extending the specifications under approach 1.
- Hybrid forms are not fully covered with the three approaches. Hybrid forms are a combination of two of the described approaches. For example, implementation can permit Approach 2 for ECL determination, as long as no material deviations arise vis-à-vis Approach 1. In so far as material deviations are apparent, Approach 1 is used. This means that the specific approach (Approach 1 or 2) depends on whether the differences are material. Such a method is not addressed in the consultation paper.
In general, we feel it is difficult to provide meaningful reporting of PD profiles for Scenario 0 in Approach 3. In a standard case, only a part of the effects on the ECL can be attributed to changes in the PD curves, e.g. due to changes in collateral values, etc. Accordingly, reporting a probability-weighted average PD curve, whilst possible, does not capture the full effect of the scenario on the ECL, however. We request that a fundamental look be taken again at the link between the scenarios, the PD curves and the impact on the ECL.
We feel that it is difficult to handle macro-economic variables for cross-bank comparisons. GDP is only one factor for macro-economic development and for some banks will have no relevance in their models. We would instead suggest extending the templates to include a "Macro-economic scenario (yes / no)" or "GDP-relevant scenario (yes / no)" field in order to boost their meaningfulness. This makes it apparent, for example, why the PD curves may not or not consistently change compared with other (GDP-relevant) scenarios.
ANNEX IX, PART II, C 111.03, Column 300/400/500: We point out that using annualised PDs for determining level assignment is not meaningful or correctly classifiable in all implementation models. In particular, a development of the annualised PD over the term is already anticipated on the basis of the typical structure of PD profiles. Hence, for very good engagements, for example, the average PD will typically tend to rise during the course of the transaction. This does not necessarily lead to a change of level, however. On the other hand, the PD tends to drop for initially poor transactions. In the latter case, it might even prove to be meaningful to set an annualised PD that is lower than the initial PD as a transfer threshold.
ANNEX IX, PART II, C 111.03, Column 300/400/500: Technically the fields can be filled. Using annualised PDs to assess transfers in cross-bank comparisons is, in our opinion, difficult. The main reasons for this are:
- A Lifetime-PD comparison (ACTUAL-Lifetime PD / ORIG-Lifetime PD) leads to a different ratio than comparing annualised Lifetime PDs. Even if column 500 is also migrated to an annualised value, the transfer thresholds were set for the originated Lifetime PDs and not the annualised values. In a comparison with a bank basing its transfer logic on 1YPDs, this difference may lead to methodically grounded, different transfer levels. In inter-bank comparisons, this circumstance distorts meaningfulness, because at that level it is no longer apparent whether the levels were derived from a Lifetime PD or 1Y PD assessment.
- Banks have to enter the "annualised originated Lifetime PD" in column 300. When using 1YPD as a proxy, this is to be supplied. Hence, both transfer logics are explicitly mapped. Not explicitly mapped are transfer logics which, for example, use Forward Lifetime PD (cumulative PD from "reporting date" to "date of maturity" subject to the condition that the contract has not defaulted as of the "reporting date") or use rating comparisons.
- In addition, in the context of scenario assessments the question arises as to whether it might not be meaningful to also include a scenario perspective in this template. In particular, the "Annualised PD at reporting date", which is needed to assess the SICR, changes when FLI or macro-economic scenarios are taken into consideration. Is the PD in Scenario 0, 1 or for the yet to be defined presentation of the PD for institutions in "Approach 3" to be stated here?
ANNEX IX, PART II, C 111.03, Column 300/400/500: The current methodology does not envisage that transfers only have qualitative triggers. In this case, it must be possible to enter 'ZERO' in the fields. If the type of transfer logic (qualitative / quantitative / qualitative & quantitative) is to be apparent, an additional "SICR approach" column ought to be included in the template with possible valid entries of 1 = quantitative trigger, 2 = qualitative trigger and 3 = quantitative & qualitative trigger.
The number of facilities does not, in our opinion, lead to any reduction in the workload for collecting data. Admittedly, the data volume is lower, but groups with several institutions will still be facing an additional burden in consolidating the data.
We regard the proposed qualitative triggers to be sufficient. At the same time, we would point out that - whereas 30DPD and Forbearance are defined uniformly across banks - the "watch list" criterion is defined subjectively by banks and therefore not uniformly implemented. A potential comparison should bear that circumstance in mind.