A consistent treatment of all aspects (including back-testing/P&L) concerning the market risk internal model is possible when own creditworthiness is taken into account. “Carving-out” own creditworthiness from the P&L would be operationally challenging. Moreover, Article 363 (1) of the CRR envisages that “…competent authorities shall grant permission to institutions to calculate their own funds requirements for one or more of the following risk categories by using their internal models…”, and thus refers to risk categories and not to a “set of positions”, as claimed in the explanatory text box.
On the other hand, the problem exists that relevant IFRS 9 rules clearly go in the opposite direction. These do not allow the inclusion of own creditworthiness, so that there is no coordination between both regimes. The accounting approach, which has also its merit, has, moreover, been adopted by some banks for assessing risk as well. Own creditworthiness is by no means ignored by these institutions either, but is in fact reflected via valuation adjustments. This means that capturing own creditworthiness would be just as challenging and operationally complex for these banks as its exclusion for banks that have adopted the opposite approach.
As a result, we are in favour of also giving banks the option of not taking their “own creditworthiness” into account. We do not believe that the issue is important enough to require harmonisation.
please see Q1
please see Q1
The hierarchy is historically based and generally makes sense.
These three categories could only be used as a broad guide for assessing the complexity of an internal model; at least the materiality of (complex) options needs to be taken into account, too. Moreover, in our view, barrier and, in particular digital options do not qualify a portfolio as being of highest complexity.
Both approaches for assessing “significance” in Articles 11 and 12 of the RTS are acceptable, in our view.
Institutions will normally take all positions in the four mentioned risk categories into account in their internal model. The proposed thresholds therefore do not pose any problems under the current supervisory regime.
We consider it important that, when calculating the thresholds, institutions do not have to compute the current standardised approach for market risk on all positions. Implementation is not currently required for positions in the model and would impose a considerable burden on banks. Furthermore, the application period would probably only be short since, once the Basel “Minimum capital requirements for market risk” come into effect on 1.1.2019, a completely different standardised approach would have to be implemented. Given the experience made with the entry into force of draft RTS, first-time application is unlikely to take place before the end of 2017.
In addition, as an alternative to focusing on the daily P&L, we suggest focusing on the VaR predicted by the internal model. The daily P&L is, in our view, inadequate and unsatisfactory in this regard. In practice, there are, for example, days on which the portfolio-wide P&L is close to zero due to strong compensatory effects from individual positions while, at the same time, the P&L resulting from the excluded positions is small, yet clearly not close to zero. In such a scenario, a proportion would be created that is well above 5% (and in extreme cases is actually infinite) although the proportion of excluded positions is objectively marginal.
With regard to the quarterly own funds requirements, it must be assumed, in conjunction with section 3.1.3 of the guidelines, that the “own funds requirements” calculated under the internal model are the basis here for determining the metric. Should this not be the case and the standardised method is the basis, the standardised method would have to be implemented and calculable for all positions (i.e. also those included in the internal model). We are opposed to this, as it means there would be no difference in comparison to the situation at initial validation and the relief with regard to the use of approved internal models envisaged in section 3.1.3 would not be put into effect.
We believe it is sensible and sufficient to build on the committee structures currently in place. Separating the NPI functions from the functions of the risk committee makes sense, but not necessarily at organisational level. It should thus be possible to handle their respective functions within one committee.
In addition, we would suggest the following changes to the wording of Article 20 of the RTS: Article 20 (1) (a) (suggested additional wording in bold print): “The internal audit of the institution reviews at least annually all internal models, including those used for capital calculation purposes, and reflects the conclusions obtained from this review in a report submitted to senior management and, if appropriate, to the management body or audit committee, as referred to in point (c) of Article 18(1)”.
It does not appear appropriate, completely irrespective of the overall findings of internal audit reports, to make it mandatory for these reports to be submitted in every case to the management body in supervisory function (cf. EBA GL 44). After all, audit committees are made mandatory for systemically important institutions.
Article 20 (1) (c): The word “proportionate” should be deleted. In the context of the requirements for the internal audit, “proportionate” is a new and “unnecessary” legal term. This term appears neither in BCBS 223 and BCBS d328 nor in the SREP Guidelines (paragraph 105); only the UK division of the Chartered Institute of Internal Auditors (IIA) uses it. There is the danger that the resources of the internal audit not only need to be adequate to meet the regulatory requirements but would also have to be geared to the relative share of market risk in the aggregate risk exposure. This is at odds with the principles for the internal audit function laid down in BCBS 223, BCBS d238 and paragraph 105, particularly paragraph 105 (e), of the SREP Guidelines (risk-based approach).
Article 20 (2) (d): this should be deleted. The independence of the internal audit is already ensured by paragraph 105c of the SREP Guidelines, in conjunction with Article 16 of Regulation 1093/2010.
Article 22 (1): The assumption in paragraphs (b) and (c) that the validation function is organisationally separated from the risk control unit is in contradiction to Article 368 (1) (b) of the CRR, which states that the risk control “unit shall conduct the initial and on-going validation of any internal model used for purposes of this Chapter”. Paragraphs (b) and (c) should thus be deleted.
The additional requirement for G-SIBs under (d) (ii) is not contained in the CRR and should be deleted.
Not all requirements are relevant and properly stated, and many elements are too burdensome:
- In Article 22 (3) (c) “validation manuals” should be replaced by “validation concept”, which is market standard.
- The requirements in Article 23 (2) should only be considered comprehensively – if at all – for the initial validation process. Moreover, the proportionality principle should be taken into account by including aspects of materiality and/or appropriateness in the individual requirements.
- The requirement in Article 23 (2) (a) to use “other alternative methodologies” should be reworded since it should not lead to the compulsory implementation of different VaR models.
- The background to the requirement in Article 23 (2) (b) to also analyse the 1% “gains” percentile is unclear, particularly since for non-linear portfolios the absolute value of the upper 1% percentile (gains) is different from the lower (losses).
- The requirement in Article 23 (2) (b) to assess “specific risk” in the context of back-testing is only relevant for models which cover specific risk.
- The requirements in Article 23 (2) (f) are not relevant for all risk categories and thus not for all internal models.
- In our view, the requirement in Article 23 (2) (h) is not clearly worded.
- It is unclear to us what is meant by “percentage marginal contribution of these time series” in Article 23 (2) (i) (i).
- The requirement to perform validation at least annually tightens the CRR requirement under Article 369 (1) of the CRR, which merely refers to a “periodic validation”. This wording should also be adopted in the RTS.
- The requirement in Article 23 (3) (c) in regard to “large pricing discrepancies with counterparties” is a pricing model issue, not a risk model issue.
The requirement in Article 25 (1) (d) that all internal limits, including the granular limits defined in (c), that could be non-VaR limits like sensitivities should be approved by the “committee” is inappropriate and too burdensome. In this area, sub-delegation by the risk committee should be possible. This should also be possible, for example, in the case of temporary limits within a framework set by the risk committee.
In this context, the requirement in Article 25 (1) (e) concerning the consistency of sensitivity or loss- based limits with the global VaR limit is not feasible.
The requirement in Article 25 (1) (f) for a formally approved inventory of authorised instruments is a general requirement not bound to IMA and should be deleted.
The requirement in Article 25 (2) (a) is essentially the obligation to introduce RWA limits for market risk.
Flexibility for the introduction of new products is essential to avoid unduly hampering the innovation process. The option in Article 29 (d) should not be dropped.
The option to delegate provided for in Article 29 (c) should not be dropped either, since it makes sense. Not every NPI measure should necessarily require board approval.
The back-testing overshooting history of 250 business days is necessary for calculating the capital requirements. This history should therefore be available upon approval of the internal model, but not already when a request for approval is made. A three-month history to demonstrate the stability of a model should be sufficient; it is also standard supervisory practice in many European countries. Longer histories would seriously hamper and delay the approval process. Assuming the average period between a model approval request and actual approval to be 9-12 months, the required history would be available when approval is granted. We suggest that, in line with the more realistic requirement in Article 34, Article 31 should stipulate that a history of 250 business days should be available when approval is granted.
This would also ensure that such a back-testing history requirement is not considered relevant for a request for approval of a (material) model change or extension within the meaning of Delegated Act (EU) 2015/942.
The EBA should be guided in this area by the new FRTB requirements. Page 73 thereof says: “The implementation of the P&L attribution and backtesting programme must formally begin on the date that the internal models capital requirement becomes effective”.
These benchmarking results are only partially useful for validation purposes since the internal models are always tailored to the business strategy of the individual institution, for example with respect to capturing product features or market data, thus leading by definition to results differing among the individual institutions.
The stated requirements for stress testing are too detailed and do not reflect the principle of proportionality:
- Stress test scenarios should not automatically be run more frequently than monthly (Article 32 (1) (b)), but only where appropriate. For a higher stress-testing frequency in regard to the current portfolio, stressed VaR is already calculated at least weekly in any case. The requirement should follow the CRR, which merely refers to “frequently” and not tighten it (Article 368 (1) (g) of the CRR).
- The requirements in Article 33 (1) (c) are too detailed and should, at a minimum, be restricted to “where relevant”.
- To our knowledge, there is no reference in the CRR to “ad-hoc stress testing”. Thus, in our view, the detailed requirements set in Article 33 (2) for ad-hoc stress testing are not covered by the EBA’s mandate.
Stress testing scenarios must be used to assess the reasonableness of VaR, IRC and CRM (Article 32 (2)). Does this really make sense, given that stress scenarios do not usually have a confidence interval associated with them and are usually not as comprehensive in risk factors as VaR? Article 32 (2) should therefore be deleted.
Scenarios must consider jump-to-default risk of the four largest specific interest rate risk positions and the four largest equity long positions with zero recovery in scenarios (Article 33 (2) (b) (ii)). It does not make sense to have a scenario where, for example, the US, UK and German governments default with zero recovery. The requirement should be adjusted.
Scenarios must consider “any other events different from those captured in point ii” (Art. 33 (2) (b) (iii)). It is impossible to prove that an institution has considered every possible event – this makes no sense in the proposed RTS in the current wording and should be reworded.
Article 34 (2) should be reworded to read: “… that the institution reconciles on a frequent basis, all internal model positions…” and “The reconciliation process shall ensure that position differences between front office and…”.
Of the three options outlined in the explanatory text box on p. 70f, only the second option is sensible and flexible enough, in our view, to deal with the problem of different time zones. The first option would undoubtedly be the ideal option in an ideal world. In reality, however, banks face a situation in which market liquidity differs widely in different time zones over the course of the day, which leads to distortions. It would therefore be appropriate to consistently gear to the end-of-day positions in the different time zones. The third option is not a sensible one, as no consolidation in the real sense takes place; instead, positions are aggregated without taking into account hedging and diversification relationships between locations in different time zones. This does not produce any realistic picture of the risks at consolidated level.
The requirement in Article 36 (3) (d) “…that the two daily P&L calculations are computed at the same time as the VaR…” is not operationally feasible – even when considering paragraph (4) – since these are often based on different processes.
The proposed ad-hoc update of market data sets should be restricted to the exception triggered by (volatility) events in market prices, as stated in Article 39 (3). Any update requirements going beyond this would be too burdensome, in our view.
In our view, a distinction should be made between “partial use” and “full use” banks.
The inclusion of all risk factors in hypothetical (clean) and actual (dirty) P&L back-testing contradicts the very idea of allowing a partial use of risk category inclusion. By using only those risk categories in back-testing for which the model is designed and approved, the model is only tested and “punished” for its actual purpose. Otherwise the model will – due to the capital multiplier – also have to account for risk categories it is not supposed to cover, since for these other risk categories the standardised method is used for capital purposes.
In our view, the argument given in the explanatory text box, referring to the requirement under Article 367 (1) of the CRR that “…the model shall capture accurately all material price risks…”, ultimately aiming at inclusion of more material risk factors, is not valid in this context, since it should refer only to risk factors within an approved risk category (otherwise the partial use and the “assessment of significance” (Articles 11 and 12) would not be meaningful at all). Our proposed treatment would also be in line with Article 40 (12) (b) and (c).
The fact that a bank makes partial use of an internal model for regulatory capital purposes does not mean that the bank has no model for the other types of risk. For internal market risk reporting purposes, banks normally have risk models which include the types of risk outside the scope of partial use. What is more, these banks conduct not only regulatory back-testing geared to the partial-use VaR but also internal back-testing geared to the VaR encompassing all types of risk. The basis for this is naturally a clean P&L which, in turn, includes all market factors. Senior management is also informed about the results of such back-testing.
In the case of full use, on the other hand, it makes good sense – for the P&L attribution process as well – to include all risk factors to obtain the required “reality check” and to be able to make improvements to the model.
please see Q22
please see Q22
In our view, if there is concern that intra-day/new trades constitute a relevant source of risk for an institution, this is captured by the use of the actual P&L for the back-testing multiplier (larger intra-day losses lead to outliers which increase the capital requirements via a higher back-testing add-on factor). Capital requirements for these risks – where relevant – thus do in fact apply. The CRR does not allow any further discretion in this area, in our view.
We also assume that the importance of intra-day risk-taking is limited for many institutions, so that no general requirement for a higher multiplier is needed.
In any case, it should not be assessed by introducing a third kind of back-testing P&L (hypothetical P&L plus intra-day/new trade effects) or even the re-inclusion of actual P&L back-testing through the “back door”, as stated in Article 40 (11) and Article 48 (2) (a).
The difference between hypothetical and actual P&L should be more a validation issue; this is in line with current practice.
We endorse the proposed treatment. Effects of these adjustments do not result from deficiencies in the internal model and thus should not be considered for the number of overshootings.
As is market practice, the Theta effect is not included in VaR models, since it does not really constitute a “risk” (deterministic, not stochastic effect). Thus, Article 43 (1) (b) is not relevant. Theta should not be part of hypothetical P&L back-testing either. There should therefore be no requirement to take Theta effects into account.
Should, however, supervisors come to the conclusion that Theta must be taken into account in the VaR and P&L, it should be ensured that Theta risk only has to be captured for options. Where other products are concerned, this would otherwise create additional model complexity when it comes to separating theta and/or trend effects clearly from the net interest income.
In our view, the “correlation risk” in Article 367 (1) (b) of the CRR clearly refers to correlation as a price/risk factor which is thus relevant for pricing. Article 367 (1) (b) of the CRR does not refer to the correlation between distributions of risk factors; this is treated in Article 367 (3) of the CRR. Thus, there is no need to consider the use of stochastic correlations. Stochastic correlations are only an issue for pricing/valuation, not for risk modelling. The inclusion of correlations in risk factor modelling calls, moreover, for the availability of liquid daily time series. This is normally not the case for correlations.
please see Q30
Concerning Article 48 (2) (a), see reply to Q26. Article 48 (2) (d) is redundant to Article 43 and should thus be deleted.
We see no reason why the SVaR multiplier should be higher than the VaR multiplier.
- Article 40 (1): The reference should probably be point (l), not (g) of Article (18) (1).
- Article 40 (2) (c): The back-testing requirement here refers to portfolios not relevant for the capital multiplier. Thus, this is a validation issue that is already covered in Article 23 (2) (b) and should be deleted here.
- Article 40 (6): The comparison of the two back-testing calculations is a validation issue and should be moved to the corresponding validation section.
- Article 40 (10): It is possible to report overshootings within three working days, but not together with an analysis and any conclusions drawn from this analysis in regard to, for example, model improvements. For this, a longer period of at least one month is essential.
- Article 41 (1) (a): The requirement to assess distributional/stochastic assumptions of the stochastic processes does not seem to be covered by Article 367 (1) (a) of the CRR: “the model shall capture accurately all material price risks”, and thus should be deleted.
- Article 41 (1) (b): Only historical data is allowed for calibrating the risk model. The mention of “market implied data” should be deleted.
- Article 41 (2) (b): To allow for the principle of proportionality and appropriateness, it should read “…performing statistical tests which assess, for example, the autocorrelation…”
- Article 42 (2): Appears to be redundant to Article 42 (1).
- Article 42 (3): To allow for the principle of proportionality and appropriateness, it should read “Competent authorities may also accept the exclusion of a risk factor …”
- Article 42 (4): Appears to be redundant to Article 42 (1) and Article 40 (8) (d).
- Article 43 (2) (b) (ii): The inclusion of the ‘moneyness’ effect has just been discussed in the context of the FRTB and ultimately been dropped; consequently, it should also be deleted here. To allow for the principle of proportionality and appropriateness, there should be at least the restriction of “where relevant”.
- Article 43 (3): Appears to be redundant to Article 43 (1)/should be integrated into Article 43 (1).
- Article 45: Reflecting illiquidity in the valuations is not a risk modelling requirement; (a) should thus be deleted. (b) is redundant to Article 44 and should thus also be deleted.
- Article 46 (1) (a): The envisaged automatism “…; this shall be material where institutions hold material positions in instruments included in Category 3 of Article 7;” should be deleted since the pricing of, for example, plain digital options does not depend on correlations.
- Article 46 (1) (b): This requirement does not seem to be covered by the cited Article 367 (1) (a) of the CRR, reading “the model shall capture accurately all material price risks”. Moreover, this point is redundant to Article 367 (3) of the CRR and Article 46 (2) here. Article 46 (1) (b) should thus be deleted.
- Article 46 (2) (b): In our view, there is no need for such a requirement since the issue of different correlation levels is already covered by stressed VaR and the regular validation of the historical period used for stressed VaR.
Yes. There are no arguments against this, in our view.
- Article 49 (2): To allow for the principle of proportionality and appropriateness, it should read “…the institution’s current portfolios, which may be determined applying the results obtained…”
- Article 49 (2): Concerning the ad-hoc scenarios, see reply to Q17.
- Article 49 (2) (a): In line with the EBA guidelines on stressed VaR, it should read “…judgement-based approach, they should include quantitative elements…”
- Article 49 (2) (c): The part “…rather than merely corresponds to the period of highest volatility;” should be deleted; “highest volatility” in the context of a (non-linear) portfolio depending on several risk factors is not defined anyway.
We welcome the explanatory text in the box as helpful clarification. However, Article 56 (4) is redundant since an internal model for the risk category “Specific risk of debt instruments” is only approved if it also contains an IRC model. So, by definition, debt event risk is covered by a model approved for this risk category.
- Article 55 (1): To allow for the principle of proportionality and appropriateness, it should read “…scope of the model which produce material interest rate sensitivity.”
- Article 55 (3): In our view, the requirement to validate and assess yield curve modelling against alternative methods is in contradiction to Article 367 (2) (a) of the CRR, which allows “…using one of the generally accepted approaches.” See also replies to Q10 and Q11.
- Article 55 (4): Appears to be a repetition of the validation/assessment aspect of Article 55 (3). See corresponding comment above.
- Article 55 (5): In our view, the requirements (a) and (b) are in contradiction to Article 367 (1) (b) of the CRR reading “the model shall capture a sufficient number of risk factors, depending on the level of activity of the institution in the respective markets” and Article 367 (2) (a) of the CRR reading “For material exposures to interest-rate risk in the major currencies and markets, the yield curve shall be divided into a minimum of six maturity segments”. Article 55 (5) should thus be deleted.
- Article 55 (7): See reply to Q16.
- Article 56 (3) (b) and (c): See reply to Q16.
- Article 63 (2) (e) (i): The requirement that the IRC model should specify changes “which are attributable to changes in credit spreads other than changes resulting from rating migrations and defaults” is in contradiction to Article 374 (3) of the CRR, where it is stated that “The impact of diversification between, on the one hand, default and migration events and, on the other hand, other risk factors shall not be reflected.” Article 63 (2) (e) (i) should thus be deleted.
- Article 63 (4) (a): In line with Article 375 (1) of the CRR, it should read “…positions refer to the same financial instrument.”
- Article 63 (4) (b): It is unclear to us why maturity mismatches should not be material for the portfolio (last part of point b) as long as the model treats them appropriately.
No, any hierarchy should be acceptable as long as there is a properly documented process around it. This is in line with the EBA guidelines on IRC. Yet not all IMA banks are necessarily also IRBA banks. Banks that use the CRSA should not be disadvantaged by this requirement.
No, we do not agree. It is not possible to say anything general in this respect.
No. We do not agree with this requirement. The IRBA floor for the PDs is 0.035 only in the “Banks” and “Corporates” asset classes, whilst there is no such floor for the “Sovereigns” asset class. It should be possible at least for AAA-rated bonds to set the PD equivalent to zero. This reflects the will of the CRR rule-makers and cannot be nullified by the EBA. Furthermore, positions with a PD equalling zero usually have a risk higher than zero since migration risk also has to be taken into account in the IRC model. There is therefore no need for such a requirement to comply with Article 373 of the CRR.
The losses due to default should be consistent with the definition of the LGD/recovery rate. Since it is market convention that these refer to the principal, the treatment of losses by using LGDs/recovery rates should also be based on principal.
- Article 66 (1) (a) and (b): This article should also take into account that migration matrices (including PDs) for liquidity horizons of less than one year are not necessarily based on 1-year matrices, but could instead be extracted directly.
- Article 69 (2) (b): It should be taken into account that the assessment of the relevance/ impact of different copulae is not always possible and/or reasonable.
- Article 70 (3) (b): Appears to be redundant to (d).
- Article 70 (3) (c): Like PDs, the LGDs/recovery rates do not change that quickly, in particular not weekly. Thus, there is no reason for an – at least – weekly update, which would be the consequence of an alignment of the update frequency with the IRC calculation frequency.
- Article 70 (5) (a) and (b): Appear to be redundant to Article 63 (2) (e).
With regard to allowance for the principle of proportionality and appropriateness, the requirements are too detailed, particularly in Articles 65 (2) and 70 (6) (a). See also comments above.
- Article 58 (1): In line with Article 367 (2) (c) of the CRR, it should read “…for each of the equity markets in which the institution holds significant positions.” Since it is redundant to this CRR article, it could also be deleted.
- Article 59 (4) (a): Should be deleted since it is redundant to Article 370 (d) of the CRR.
- Article 60 (1) (a) and (b): Should be deleted since they are redundant to Article 367 (2) (d) of the CRR.