Key takeaways & industry challenges following the ECB TRIM project – a focus on credit risk (Part 2)

The Targeted Review of Internal Models (TRIM) was one of the largest projects by the European Central Bank (ECB) aimed at identifying potential sources of unwarranted or non-risk based variability in Significant Institutions (SIs) risk-weighted assets (RWA) from the use of Pillar 1 internal models such as Probability of Default (PD), Loss Given Default (LGD) and Credit Conversion Factor (CCF). The projects required co-operation from competent national authorities (NCAs or regulators), external consultants, and the SIs themselves. From 2016 until 2019, it initially focused on the high default portfolios (HDP) such as retail mortgages and SME, and low default portfolios (LDP) such as project finance and corporates. Key regulatory texts are the Capital Requirements Regulation (CRR), Regulatory Technical Standards (RTS) and European Banking Authority (EBA) guidelines.

Essential to TRIM is the ability to ensure ‘like-for-like’ comparisons such that relevant conclusions can be drawn. Inspection Techniques and Tools (ITTs) for substantive testing, preparation of a standardised assessment report/work plans and regular touch-points helped achieve this goal.

In April 2021, the ECB published its final project report on the TRIM, which summarises the findings from the project. Below we outline some of the key insights with a focus on credit risk. 

Key insights

A total of 161 onsite TRIMs were conducted on the credit risk models, which were mainly focused on the PD and LGD models due to the limited nature of off-balance sheet exposures for the CCF. The below table shows the split by portfolio type and finding percentage mix across each risk parameter. As expected, the majority of findings involve the PD and LGD, with the latter being the primary concern for the HDP (58% of total findings) and the former for the LDP (52% of total findings).

PortfolioSITRIMFindingsPDLGDCCFOther
HDP5385                 2,00029%52%8%11%
LDP4676                 1,73158%28%9%5%
Total99*161                   3,73142%41%9%8%
* Not mutually exclusive as separate missions for HDP and LDP for an SI may have been conducted

Around 22 findings were raised per investigation, with a third flagged as high severity (F3 or above on the ECB finding scale of F1-F4). Furthermore, 40% of SIs need to remediate findings within 18 months (25% within 12 months), indicating the urgency for some to apply fixes identified.

Notwithstanding the above, the overall conclusion from the ECB TRIM report is that SIs can continue using internal models for their own-funds requirement / Pillar 1 RWA subject to supervisory measures. The ECB has subsequently published guidance to improve clarity on specific requirements.

In the following sections, Mazars highlights just some of the TRIM report’s key weaknesses and provides some useful information worth considering for SIs when planning their remedial actions.

Risk differentiation

The ability to differentiate obligors according to their level of risk has a direct impact on the financial viability of the lender where the underwriting of new obligors and the credit risk management (e.g. payment behaviours) will impact the level of required capital, and hence RWA, commensurate to the level of risk in the portfolio. The process involves scoring/rating each obligor and grouping them into risk pools (rating grades) where obligors in the same grade should be similar enough (i.e. homogeneous with common risk drivers). Those across grades should display sufficient risk differentiation (i.e. heterogeneous where obligors can be rank-ordered by their risk). Some of the key regulatory expectations as relevant to homogeneity/heterogeneity are shown in the table below.

RegulationReference
ECB Guide to Internal Models4.1.2, 5.2.1
Regulation (EU) No 575/2013 (CRR)170
EBA Guideline on PD and LGD5.2.5, 6.2.4
Final draft RTS on IRB models38

The TRIM report noted various approaches to risk differentiation and raised findings on the PD and LGD scoring/ rating functions for most investigations, with many being high severity (F3 or worse). Due to sparse data for LDP, using statistical approaches to assess risk differentiation did not necessarily provide meaningful results; however, SIs should provide more justifications around the judgement-based assumptions (e.g. balance between granularity of rating grade scale and volume). Roughly half of SIs performed the testing during model development, but only a quarter had a process for checking homogeneity/ heterogeneity of grades as part of the model risk governance.

In terms of relative importance, around 10% of findings are related to the lack of risk differentiation and under 20% of findings, including issues related to the risk frameworks around monitoring and governance. Obligations imposed by the ECB to either revise or redevelop the SI’s risk differentiation methodology would have direct impacts on the concerned risk parameters for around 60% of SIs requiring to make changes as it involves updating the relevant scoring functions.

SIs should aim to provide more evidence of meaningful risk differentiation, expand model testing and, where required, update risk parameters which should help reduce non-risk-based variations in RWA.

Realised LGD calculation

In estimating the Loss Given Default, firms would need to measure the observed losses / realized LGD for each exposure. According to CRR Article 4(1)(55), the realized LGD is measured as the ratio of the loss on exposure due to the default of a counterparty to the amount outstanding at default. RWA is calculated using estimates of LRA LGD (long-run average), DLGD (downturn), ELBE (best estimate of expected loss) and LGD in default. Some of the key regulatory expectations as relevant to the calculation of the realized LGD is shown in the table below.

RegulationReference
ECB Guide to Internal Models5.1.3
Regulation (EU) No 575/2013 (CRR)181
EBA Guideline on PD and LGD 7.3.1
Final draft RTS on downturn LGDSeveral

Similarly to the harmonization of the default identification and calculation of the default rate, discrepancies in the calculation of the realized LGD were observed. Mazars note that the regulatory expectations evolved during the TRIM project where, in the first phase, expectations were based on the Draft EBA Guidelines on PD and LGD, which treated interest, fees and the ‘artificial cash flow’ (i.e. balance at the date of the default exit which includes due fees and interest during the default period) differently than in the final Guidelines on PD and LGD estimation. It is now expected that fees and interest after default are not considered as part of the movements increasing the credit exposure (i.e. excluded from the numerator of the realized LGD calculation) and that the artificial cash flow shall be discounted as interest and fees recovered can be used to compensate for the time value for money.

The TRIM report noted various approaches used by SIs to calculate the realized LGD and raised findings in all investigations, with many being high severity (F3 or worse). Around 40% are not using directly observed recovery flows but are instead making inferences from changes in the outstanding balance (EAD) between two dates, as an example. Furthermore, the ECB acknowledged the lack of necessary information to compute the realized LGD, the definition of economic loss not being comprehensive enough (including potential for not including material discounting effects), allocation method of recoveries and costs to individual defaulted exposures, discounting rates and treatment of multiple defaults could lead to bias in LGD. It is worth mentioning that with Basel III decommissioning LGD models for some asset classes, concerns on realized LGD for the LDPs could be mostly abated.

In terms of relative importance, the findings on the calculation of realized LGD account for over 10% of total findings (17% of findings for HDP and 6% of findings for LDP). Recall that the majority of HDP findings concern the LGD parameter, and for LDP, it concerns the PD parameter. Obligations imposed by the ECB to either revise the SI’s calculation of realized LGD is expected to have direct impacts on the concerned risk parameters for around 90% of SIs required to make changes.

SIs should ensure the realized LGD is calculated in accordance with regulatory guidance as many compliance issues were raised, paying close attention to concepts of economic loss and discounting.

Margin of conservatism

The regulatory expectation for SIs using IRB models when estimating the risk parameters for PD, LGD and CCF are to identify any data & methodology issues (category A), changes in policies (underwriting, recovery, etc.) that could lead to a bias in risk parameters (category B) and any general estimation errors to account for statistical uncertainty (category C). The final MoC, according to the EBA Guideline, is calculated as a sum of A-C categories and added to the central / best estimate of the relevant risk parameter. It is worth noting that these explicit MoCs are in addition to the conservative adjustments within SIs’ modelling approaches (i.e. implicit conservatism). Some of the key regulatory expectations as relevant to MoC is shown in the table below.

RegulationReference
ECB Guide to Internal Models7
Regulation (EU)No 575/2013 (CRR)179(1)(f), 180(1)(e), 182(1)(c)
EBA Guideline on PD and LGD36-52
Final draft RTS on IRB models47
 

The TRIM report found that less than half of SIs are applying an explicit MoC, and with 10% of SIs not accounting for any MoC (explicit or implicit). When an explicit MoC was considered, these were found to mainly account for data & methodology issues and general estimation errors (60-80% of SIs across LDP, HDP and risk parameters) and around a quarter of SIs accounted for an explicit MoC for representativeness issues contrary to paragraph 34 of the EBA Guideline where representativeness issues should be adjusted for by an MoC. Adjusting for lack of representativeness is of particular importance. As the case for LDP, the SIs faced challenges in finding representative data or adapting externally sourced data for computing long-run averages (LRA). Another finding was the absence of a framework for identifying deficiencies and quantification of the respective MoC by SIs, which should contain clear criteria for identifying, computing and applying the MoC.

In terms of relative importance, around 8% of findings are related to the MoC across both HDP and LDP. Obligations imposed by the ECB had a bigger direct impact on risk parameters for LDP (80%) than HDP (40%) required to make changes due to much greater reliance on implicit MoC for LDPs.

SIs should aim to expand their risk framework to include MoC and the relevant drivers such that it complies with the regulatory expectations around explicit conservatism on IRB model parameters.

Other findings

Thus far, we have covered several key themes within the TRIM report but feel it worthwhile to mention some of the inappropriate measures/methodology inconsistencies for HDP as follows.

#TopicFinding
1Multiple Defaults60% are not counting from the return to normal of the first default.
2Cures50% did not compute realised LGD for cures but assumed 0% loss.
3In-default Drawings40% have inconsistent treatment between LGD and EAD.
4LRA PD40% did not sufficiently justify the calibration approach (e.g. weightings).
5LRA LGD36% not calculated at portfolio / calibration segment level.
6LRA LGD34% not calculated using default / facility-weighted average.
7RestructureOver 30% could not connect facilities before and after a restructure.
8Haircut11% did not account for a haircut on repossessed collateral in the LGD.
9DLGD11% did not adjust the LGD parameter for a downturn.

Similarly, some of the inappropriate measures / methodology inconsistencies for LDP is shown below.

#TopicFinding
1Discounting96% are not using the discount rate for LGD as per EBA Guidelines.
2In-default Drawings74% have inconsistent treatment between LGD and EAD.
3Multiple Defaults68% are not counting from the return to normal of the first default.
4DLGD29% did not adjust the LGD parameter for a downturn.
5LGD22% are have neither ELBE or in-default LGD.
6LRA LGD18% not calculated at portfolio / calibration segment level.

Conclusion

The ECB TRIM was an extensive investigation into the RWA variability of SIs. The ECB report provides very useful insights on each of the IRB model components that may not have been clear from the regulation with a clear direction of travel and further guidance to be released. These come at a good time as firms prepare for the upcoming regulatory changes (Basel III and new definition of default). In this paper, Mazars elaborated three of the key themes: lack of risk differentiation, calculation issues of the realised LGD and limited scope of the margin of conservatism, whilst highlighting gaps in the frameworks applicable to each of these areas. With a high number of high severity findings where over a quarter of SIs have a relatively short timeline for remediation, it is crucial that firms fully understand the expectations and prioritise these model updates. As a silver lining, Basel III provides some simplifications for LDP. The ECB supports the continued use of IRB models pending significant effort by SIs to achieve full compliance and increase harmonisation.