The Future of IRB CCF Models: Understanding the Impact of the 2025 Draft Guidelines. Part 1

Part 1 provides an overview of the proposed amendments to the calculation of the LRA CCF, focusing on the fixed 12-month horizon and the use of equal weights in the LRA framework. It also discusses the clarified definition of the Region of Instability (RoI) and its implementation.

The 12-month fixed horizon requirement establishes that the reference date for CCF estimation must be selected as 12 months prior to default date. This requirement was introduced under CRR3 Art. 182(g), para. 3 and came into effect from 1^st of January 2025. It was further detailed in the draft CCF GLs.

The requirement implies that risk differentiation and quantification must be based exclusively on the snapshots corresponding to 12 months prior to default. In case of “fast defaults” (defaults occurring within 12 months of origination), a retracted reference date must be applied as these exposures cannot be omitted [CRR3, Art. 182(1)(a)]. In most cases, the retracted reference date corresponds to the earliest date at which both the exposure amount and limit are reliably observed, typically the origination date. When including “fast default” into the model, additional adjustments must be made to prevent bias in the CCF estimates (appropriate adjustment or a MoC).

Figure 1: Observed average CCF by time to default and share of defaults observable in each sample month t

The illustrative example in Figure 1 demonstrates that the average observed CCF consistently decreases when moving from 12 months prior toward ToD. To assess the impact of the 12-month fixed-horizon requirement, this LRA CCF value is compared with the LRA CCFs obtained using two commonly applied methodologies: the random selection approach and the cohort approach. A summary of these results is presented in Table 1. 

                    Table 1: 12-month fixed horizon impact on LRA CCF

This evidence suggests that applying a fixed 12-month horizon, rather than relying on a randomly selected reference date or cohort approach, leads to systematically higher CCF estimates. For the portfolio under consideration, the shift in the LRA computation approach results in an increase of approximately 10% in the LRA CCF. Consequently, the use of the fixed horizon approach is expected to increase the RWAs as well.

Additionally, the example shows that including “fast defaults” in the LRA CCF calculation for the selected retail portfolio does not compromise the conservatism of the resulting CCF estimates.

As a result, banks whose CCF models do not currently rely on 12-month fixed horizon will need to review their frameworks. This will require not only methodological adjustments but also a careful quantification of the potential increase in RWA, to ensure that the models remain fully compliant once the proposed CCF GLs enter into force.

The draft CCF GL provide a clearer definition and specific guidance for the treatment of fully, and nearly fully, drawn facilities (Region of Instability, hereafter, RoI), where the low undrawn amounts in the CCF denominator can lead to unstable or extreme CCF values.

• Fully drawn facilities. Fully drawn facilities at the reference date must always be treated using the LF approach.

The draft CCF GL reiterate that institutions are not permitted to exclude, cap, or truncate extreme or unstable CCF observations. In line with the principles set out in CRR3 (Art. 182(1)(b)), such observations must be incorporated into model calibration, with appropriate segmentation applied where relevant to ensure robust and reliable CCF estimates.

For the individual facility B in the application portfolio, the EAD impact of switching from CCF to LF parameters is neutral. However, at the portfolio level, the effect can differ. Since fully drawn facilities are generally associated with higher realized CCFs, their exclusion from the CCF region may lead to a decrease in the LRA and, consequently, a reduction in EAD. This effect arises from the relatively large exposure within the CCF region in the application portfolio.

Nevertheless, the impact is not unidirectional. Almost fully drawn facilities are often very unstable: small relative changes in exposure between the reference date and the default date can result in substantial variations in observed CCFs.

Consider, for example, the three facilities shown in Table 3. For these facilities, exposure changes of only a few cents between the reference and default dates lead to significant fluctuations in observed CCFs. If such facilities are retained within the CCF region, the LRA CCF may become severely unstable and in this example biased downward due to the presence of low or even negative realized CCFs. In contrast, realized LF values remain relatively stable after small changes in exposure

                                Table 3: RoI example, positive impact.

In conclusion, the direction of the RWA impact stemming from the introduction of the LF factor as a parameter for estimating EAD for nearly or fully drawn facilities depends on the specific structure of the portfolio under analysis. However, in all cases, accurate identification of RoI facilities remains critical to maintaining stable and reliable estimates. From a technical standpoint, defining the rules to correctly isolate RoI facilities may entail substantial adjustments to the model’s specification and implementation.

The requirement establishes the methodology for aggregation of the individual realized CCFs to the calibrated LRA CCF value. In particular, the weighting of each individual contribution when calibrating to the LRA CCF.

Under the draft CCF GLs, the LRA CCF is computed as the arithmetic average of realized CCFs over the full historical observation period, weighted by the number of defaults. Unlike earlier guidelines, unequal weighting of historical periods or annual averaging should not be applied, ensuring that each default contributes proportionally to the LRA CCF estimate. Additionally, it is clarified in the draft CCF GLs, that the length of the calibration sample should be the same as that of the risk quantification sample.

For portfolios with concentrated default periods, the removal of annual averaging can either raise or lower the LRA CCF. The impact depends on the relationship between default frequency and realized CCFs. If years with higher number of defaulted exposures coincide with higher average CCFs, the LRA CCF will increase, if not, the effect may be neutral or negative.