Prudential Regulation Authority Rulebook

12.A1

When applying the CRR requirements relating to the estimation of the probability of default, the PRA expects firms to comply with the EBA’s Guidelines on PD estimation, LGD estimation and the treatment of defaulted assets (EBA/GL/2017/16).

12.1

‘Rating philosophy’ describes the point at which a rating system sits on the spectrum between the stylised extremes of a point in time (PiT) rating system and a through the cycle (TTC) rating system. Points (a) and (b) explain these concepts further:

1. (a) PiT: firms seek explicitly to estimate default risk over a fixed period, typically one year. Under such an approach the increase in default risk in a downturn results in a general tendency for migration to lower grades. When combined with the fixed estimate of the long-run default rate for the grade, the result is a higher capital requirement. Where data are sufficient, grade level default rates tend to be stable and relatively close to the PD estimates; and
2. (b) TTC: firms seek to remove cyclical volatility from the estimation of default risk, by assessing borrowers’ performance across the economic cycle. TTC ratings do not react to changes in the cycle, so there is no consequent volatility in capital requirements. Actual default rates in each grade diverge from the PD estimate for the grade, with actual default rates relatively higher at weak points in the cycle and relatively lower at strong points.

12.2

Most rating systems sit between these two extremes. Rating philosophy is determined by the cyclicality of the drivers/criteria used in the rating assessment, and should not be confused with the requirement for grade level PDs to be ‘long run’. The calibration of even the most PiT rating system needs to be targeted at the long-run default rates for its grades; the use of long-run default rates does not convert such a system into one producing TTC ratings or PDs.

12.3

The cyclicality of the rating system is a measure of its degree of responsiveness to economic changes. At one extreme a fully cyclical rating system or ‘point-in-time’ (PiT) would see an economic downturn picked up through migration of exposures to lower rating grades and therefore no increase in default rate within a grade. At the other extreme a non-cyclical or ‘through-the-cycle’ (TtC) rating system does not respond to an economic downturn with grade migration, but the default rate within a grade increases instead. The PRA expects firms to be aware of the cyclicality of their rating systems to enable them to calibrate, monitor and stress test their systems. The PRA would define cyclicality for a rating system as follows:

Where:
PD_t means the long-run average PD at time t
DR_t means the observed default rate at time t

12.4

In the PRA’s experience, firms often have difficulty in practice in understanding the cyclicality of their residential mortgage rating systems. To mitigate the risk of under-calibration of these rating systems due to inaccurate estimation of their cyclicality, the PRA expects that when firms calibrate their residential mortgage rating systems by uplifting internal observed default rates to a long-run average, they should do so on the assumption that the cyclicality of each rating system is no more than 30% in those years where grade level internal observed default rates are not available. This cyclicality cap is the PRA’s expectation of what firms should assume is the maximum level of cyclicality when imputing missing historical default rates. If 30% of the change in portfolio default rates comes from grade migration the remaining 70% would come from change in default rates within grades. Therefore when calibrating the long-run average default rates to assign to each rating grade the PRA expects firms to assume that at least 70% of the portfolio change in default rate reflects grade level changes in default rate. This level reflects the PRA’s current view of an appropriately conservative assumption for rating system cyclicality in light of recent experience. This expectation may be adjusted by the PRA if it judges that there has been a change in the risk of under-calibration.

12.5

When a firm is calibrating or recalibrating a residential mortgage rating system using internal observed default rates taken predominantly from a downturn period (ie the firm is reducing the internal observed default rates to a long-run average) the PRA’s expectation of a 30% cap on cyclicality will not apply. Instead, firms should determine an appropriately conservative adjustment to allow for uncertainty in their estimates of cyclicality in such circumstances.

12.6

As an alternative to the expectations on risk mitigation methodology in paragraph 12.4, the PRA may be satisfied that a firm has taken steps to mitigate these risks if the residential mortgage PD rating system meets the following standards:

1. (a) the firm is able to convincingly articulate how the risk drivers in a rating system will generate the migration into other grades, scores or ratings assumed in its estimates of cyclicality;
2. (b) the firm is able to demonstrate that the assumed changes have occurred in practice across an economic cycle; and
3. (c) the above analysis is able to isolate the impact on the existing exposures covered by the rating system from changes in composition of the portfolio over the period being analysed.

12.7

Highly cyclical PiT models do not always adequately capture risks over the long-run and this is particularly an issue for residential mortgage portfolios where default rates are highly cyclical. The PRA therefore expects firms not to use an artificial highly cyclical PiT approach achieved through dynamic recalibration of the score to PD relationship in their application and behavioural scorecards for residential mortgage models.

12.8

We use the term ‘variable scalar’ to describe approaches in which the outputs of an underlying, relatively PiT, rating system are transformed to produce final PD estimates used for regulatory capital requirements that are relatively non-cyclical. Typically this involves basing the resulting requirement on the long-run default rate of the portfolio or segments thereof.

12.9

CRR Article 169(3) allows the use of direct estimates of PDs, though such a measure could be assessed over a variety of different time horizons which CRR does not specify. Accordingly, the PRA considers it acceptable in principle to use methodologies of this type in lieu of estimation of long-run averages for the grade/pool/score of the underlying rating system where conditions set out below are met. Meeting these conditions would require firms using the variable scalar approach to have a deep understanding of how and why their default rates varied over time.

1. (a) firms meet the following four principles which address the considerable conceptual and technical challenges to be overcome in order to carry out variable scalar adjustments in an appropriate way:
   1. Principle 1: both the initial calculations of and subsequent changes to the scalar should be able to take account of changes in default risk that are not purely related to the changes in the cycle;
   2. Principle 2: a firm should be able accurately to measure the long-run default risk of its portfolio; this must include an assumption that there are no changes in the business written;
   3. Principle 3: a firm should use a data series of appropriate length in order to provide a reasonable estimate of the long-run default rate referred to in paragraph 10.13; and
   4. Principle 4: a firm should be able to demonstrate the appropriateness of the scaling factor being used across a portfolio.
2. (b) stress testing includes a stress test covering the downturn scenarios outlined by the PRA, based on the PDs of the underlying PiT rating system, in addition to the stress test based on the parameters used in the Pillar 1 capital calculation (ie the portfolio level average long-run default rates); and
3. (c) firms are able to understand and articulate upfront how the scaling factor would vary over time in order to achieve the intended effect.

12.10

The PRA has found in its experience that for residential mortgage portfolios, firms are unable to distinguish sufficiently between movements in default rates that result from cyclical factors and those that result from non-cyclical reasons, and this results in risks not being sufficiently captured. The PRA therefore expects that firms should not use variable scalar approaches for residential mortgage portfolios.

12.11

The PRA will not permit firms using a variable scalar approach to revert to using a PiT approach during more benign economic conditions.

12.12

Principle 1 is the most important and challenging to achieve as it requires an ability to be able to distinguish movements not related to the economic cycle, from changes purely related to the economic cycle, and not to average these away. This is because a variable scalar approach removes the ability of a rating system to take account automatically of changes in risk through migration between its grades.

12.13

Accordingly, the PRA expects firms using a variable scalar approach to adopt a PD that is the long-run default rate expected over a representative mix of good and bad economic periods, assuming that the current lending conditions including borrower mix and attitudes and the firm’s lending policies remain unchanged. If the relevant lending conditions or policies change, then we would expect the long-run default rate to change.

(CRR Article 180(1)(a), 180(1)(b) and 180(2)(a))

12.14

The PRA considers that until more promising account level arrears data are collected, enabling firms to better explain the movement in their arrears rate over time, the likelihood of firms being able to develop a compliant variable scalar approach for non-mortgage retail portfolios is low. This is because of the difficulty that firms have in distinguishing between movements in default rates that result from cyclical factors and those that result from non-cyclical reasons for these portfolios.

12.15

For the purposes of this subsection ‘non-mortgage retail portfolios’ refers to non-mortgage lending to individuals (eg credit cards, unsecured personal loans, auto-finance) but does not include portfolios of exposures to small and medium-sized entities (SMEs in the retail exposure class).

12.16

The PRA considers that one variable scalar approach, potentially compliant with the four principles set out above, could involve:

1. (a) segmenting a portfolio by its underlying drivers of default risk; and
2. (b) estimating separate long-run default rates for each of these segmented pools.

12.17

A firm that applied a segmentation approach properly could satisfy both Principle 1 and Principle 4. The choice of the basis of segmentation and the calibration of the estimated long-run default rate for the segments would both be of critical importance.

12.18

The PRA expects segmentation to be done on the basis of the main drivers of both willingness and ability to pay. The PRA expects firms to:

1. (a) incorporate an appropriate number of drivers of risk within the segmentation to maximise the accuracy of the system;
2. (b) provide detailed explanations supporting their choices of drivers, including an explanation of the drivers they have considered but chosen not to use; and
3. (c) ensure that the drivers reflect their risk processes and lending policy, and are not chosen using only statistical criteria (ie a judgemental assessment of the drivers chosen is applied).

(CRR Article 179(1)(d))

2.19

To the extent that the basis of segmentation is not sufficient completely to explain movements in non-cyclical default risk, the long-run default rate for that segment will not be stable (eg a change in the mix of the portfolio within the segment could change the long-run default rate). In such cases, we expect firms to make a conservative compensating adjustment to the calibration of the long-run average PD for the affected segments and to be able to demonstrate that the amount of judgement required to make such adjustments is not excessive. Where judgement is used, considerable conservatism may be required. The PRA expects conservatism applied for this reason not to be removed as the cycle changes.

12.20

The PRA expects firms to review and amend as necessary the long-run default rate to be applied to each segment on a regular (at least an annual) basis. When reviewing the long-run default rate to be applied to each segment, the PRA expects firms to consider the extent to which:

1. (a) realised default rates are changing due to cyclical factors and the scaling factors need to be changed;
2. (b) new information suggests that both the PiT PDs and the long-run PDs should be changed; and
3. (c) new information suggests that the basis of segmentation should be amended.

12.21

The PRA expects that over time the actual default rates incurred in each segment would form the basis of PD estimates for the segments. However at the outset the key calibration issue is likely to be the setting of the initial long-run default rate for each segment, as this will underpin the PD of the entire portfolio for some years to come. The PRA expects firms to apply conservatism in this area and this is something on which the PRA is likely to focus on in particular in PRA model reviews.

12.22

The PRA expects firms to put in place a governance process to provide a judgemental overlay to assess their choices of segments, PD estimates and scalars, both initially and on a continuing basis. Moreover, where the basis of their estimation is a formulaic approach, we would consider that the act of either accepting or adjusting the estimate suggested by the formula would represent the exercise of judgement.

12.23

The PRA expects firms to consider what use they can make of industry information. However, we would expect firms to seek to measure the absolute level of and changes to their own default risk, rather than changes in default risk relative to the industry. Given the potential for conditions to change across the market as a whole, the PRA expects a firm should not draw undue comfort from the observation that its default risk is changing in the same way as the industry as a whole. Doing so would not allow them to meet Principle 1.

12.24

The PRA expects firms to be able to demonstrate that they have adequate information and processes in order to underpin the decisions outlined above on choice of segmentation, source of data, and adequacy of conservatism in the calibration, and that this information is reflected in the reports and information being used to support the variable scalar governance process. Given that, for retail business, these decisions would be likely to affect only the regulatory capital requirements of the firm and not the day-to-day running of its business, we will be looking for a high level of reassurance and commitment from firms’ senior management to maintain an adequate governance process.

12.25

Where a firm has not chosen to apply the definition of default at the level of an individual credit facility in accordance with CRR Article 178(1), the PRA expects it to ensure that the PD associated with unsecured exposures is not understated as a result of the presence of any collateralised exposures.

12.26

The PRA expects the PD of a residential mortgage would typically be lower than the PD of an unsecured loan to the same borrower.

(CRR Article 178(1))

12.27

Where a firm chooses to apply the definition of default at the level of an individual credit facility in accordance with CRR Article 178(1) and a customer has defaulted on a facility, then default on that facility is likely to influence the PD assigned to that customer on other facilities. The PRA expects firms to take this into account in its estimates of PD.

(CRR Article 178(1))

12.28

In order to ensure that a rating system provides a meaningful differentiation of risk and accurate and consistent quantitative estimates of risk, the PRA would expect firms to develop country-specific mid-market PD models. Where firms develop multi-country mid-market PD models, we would expect firms to be able to demonstrate that the model rank orders risk and predicts default rates for each country where it is to be used for regulatory capital calculation.

12.29

The PRA expects firms to have challenging standards in place to meaningfully assess whether a model rank orders risk and accurately predict default rates. These standards should specify the number of defaults that are needed for a meaningful assessment to be done.

12.30

We would expect firms to assess the model’s ability to predict default rates using a time series of data (ie not only based on one year of default data).

12.31

In our view a model is not likely to be compliant where the firm cannot demonstrate that it rank orders risk and predicts default rates for each country regardless of any apparent conservatism in the model.

12.32

We would expect firms using rating agency grades as the primary driver in their IRB models to be able to demonstrate (and document) compliance with the following criteria:

1. (a) the firm has its own internal rating scale;
2. (b) the firm has a system and processes in place that allow it continuously to collect and analyse all relevant information, and the ‘other relevant information’ considered by the firm in accordance with CRR Article 171(2) reflects the information collected and analysed by the firm when extending credit to new or existing obligors;
3. (c) the ‘other relevant information’ considered by the firm is included in an IRB model in a transparent and objective way and is subject to challenge. We would expect the firm to be able to demonstrate what information was used and why, and, how it was included; and if no additional information is included, to be able to document what information was discarded and why;
4. (d) the development of final grades includes the following steps:
   
   1. (i) the firm takes into account all available information (eg external agency grades and any ‘other relevant information’) prior to allocating obligors to internal grades. The firm does not automatically assign obligors to grades based on the rating agency grade;
   2. (ii) any overrides are applied to these grades; and
   3. (iii) the firm has a system and processes in place that allows it to continuously collect and analyse final rating overrides.
5. (e) the grades to which obligors are assigned is reassessed at least annually. The firm is able to demonstrate how the grades are reassessed on a more frequent than annual basis when new relevant information becomes available; and
6. (f) firms can demonstrate that a modelling approach is being applied, both in terms of the choice of the rating agency grade as the primary driver and, where information is found materially and consistently to add to the accuracy or predictive power of the internal rating grade, that they have incorporated this information as an additional driver. The PRA expects this work to be analytical (rather than entirely subjective) and could form part of the annual independent review of the model.

12.33

In the PRA’s view, if a firm does not have any additional information to add to the external ratings for the significant part of its portfolio then the PRA expects it will not meet the requirements for using an IRB approach.

12.34

The PRA expects a firm to estimate PD for a rating system in accordance with this section where a firm’s internal experience of defaults for that rating system was 20 or fewer, and reliable estimates of PD cannot be derived from external sources of default data including the use of market price related data. In PD estimation for all exposures covered by that rating system, the PRA expects firms to:

1. (a) use a statistical technique to derive the distribution of defaults implied by the firm’s experience, estimating PDs (the ‘statistical PD’) from the upper bound of a confidence interval set by the firm in order to produce conservative estimates of PDs in accordance with CRR Article 179(f);
2. (b) use a statistical technique to derive the distribution of default which takes account, as a minimum, of the following modelling issues:
   
   1. (i) the number of defaults and number of obligor years in the sample;
   2. (ii) the number of years from which the sample was drawn;
   3. (iii) the interdependence between default events for individual obligors;
   4. (iv) the interdependence between default rates for different years; and
   5. (v) the choice of the statistical estimators and the associated distributions and confidence intervals.
3. (c) further adjust the statistical PD to the extent necessary to take account of the following:
   1. (i) any likely differences between the observed default rates over the period covered by the firm’s default experience and the long-run PD for each grade required by CRR Articles 180(1)(a) and 180(2)(a); and
   2. (ii) any other information that indicates (taking into account the robustness and cogency of that information) that the statistical PD is likely to be an inaccurate estimate of PD.

12.35

The PRA expects firms to take into account only defaults that occurred during periods that are relevant to the validation under the CRR of the model or other rating system in question when determining whether there are 20 defaults or fewer.

12.36

The PRA expects firms to assign exposures to the risk-weight category for specialised lending exposures based on the criteria set out in the tables in Appendix A.

12.37

The PRA expects that, for residential mortgages, where a firm’s internal experience of defaults for a rating system is low, it may use external data to supplement internal data for rank-ordering different borrowers by credit quality and to help adjust for seasoning as credit quality changes with loan vintage. This is in addition to use of external data for calibration purposes. The PRA expects that firms attempting to evidence comparability with third-party data should include a comparison of default rates.

12.38

The PRA believes internal data may be considered to be the ‘primary source’ for residential mortgages where a firm assigns sufficient weight to internal data, including security (loan-to-value), loan (arrears history) and borrower (applicant information) factors, as inputs into their rank ordering but uses external data to achieve greater discrimination.

12.39

The PRA expects firms to apply appropriate margins of conservatism at every step to account for uncertainty in their estimates and to mitigate against incomplete data and where external data are not wholly representative.

12.40

Where firms lack sufficient internal defaults to evidence rank ordering or a reliable calibration, firms may use models that rank order on an early arrears definition (which tends to be correlated with default), provided they are calibrated with sufficient conservatism.

(CRR Articles 171, 179 and 180)