Prudential Regulation Authority 20 Moorgate London EC2R 6DA 31 October 2016 A response to the Prudential Regulation Authority s Consultation Paper CP29/16 Introduction Residential mortgage risk weights October 2016 We are pleased to reply to CP 29/6 Residential Mortgage Risk Weights 1 on proposed changes to the methodologies used to estimate the risk of the exposure to residential mortgages. This response represents the views of the members of the British Bankers Association (BBA), the Building Societies Association (BSA) and the Council for Mortgage Lenders (CML). We support the Bank s objective of greater harmonisation in the approaches to modelling probability of default (PD) and loss given default (LGD) resulting in a more consistent approach to the estimation of the risk parameters for UK residential mortgage (owneroccupier and buy-to-let) exposures. However, in light of the strategic importance of lending to this segment of the UK economy, which encompasses a full spectrum of owner-occupier and buy-to-let mortgages we would like the Bank to undertake increased engagement with the industry on its proposals before finalising its revision to Supervisory Statement 11/13. 1. Reaching a consensus on the approach to modelling There was considerable benefit achieved through the meeting between the PRA and BBA/BSA members (industry meeting) following publication of the Consultation. It also highlighted a number of areas where it was acknowledged the proposals required further development. We think further consultation is needed and propose that a series of workshops be held at which the PRA and firms model developers and risk managers would be present. 1 http://www.bankofengland.co.uk/pra/documents/publications/cp/2016/cp2916.pdf
2 We would suggest the following topics: a) Modelling of long-run average PD; b) Modelling of downturn LGD; c) Alignment with EBA guidance and rules on PD and LGD (yet to be published); d) Methodology, calculation and implementation of the cyclicality requirements; e) Data usage and techniques when there is insufficient data; f) Use test expectations; and g) Further discussion of the strengths and weaknesses of the variable scalar approach and ways in which its implementation could be made more robust. We appreciate these are areas where thoughts of firms and regulation should be mature, however with this publication and some ten years since implementation of Basel II think that practical discussion to ensure a consistent understanding of expectations and implement ability are fundamentally important to the effective delivery of this consultation. The following areas are of particular concern: Probability of Default It appears that if a firm has a long run of internal data a cyclicality constraint is not required. This would allow a firm to de facto maintain a Point in Time Rating philosophy and hence pro-cyclical PD estimates. We think that there is the potential for variation in the interpretation as to how the 30% cyclicality cap may be applied across firms over the period in which internal grade level data is not available. We think that further discussion and clarification is required. Firms are uncertain with regard to the practicality of implementation in the event that the 30% cyclicality cap is breached. Firms have carried out a preliminary analysis of the measures of cyclicality and have concluded that these measures will not provide consistent coefficients through time and are also likely to produce negative or extreme values in periods of poor model calibration or economic or operational instability. We recognise the PRA s desire to retain the early 1990s recession as a key downturn reference point within the data series for the calculation of Long Run Average default rates. This creates further scope for inconsistency as firm s approaches to incorporate the inferred grade level default rates will vary given that portfolio Observed Default Rates (ODR), grade concentrations and predicted PDs will need to be historically modelled referencing internal data. It is not clear whether the proposed guidelines are commensurate with the requirements of the ECB or whether a TTC (inc. Variable Scalar) model including simply acyclic variables would be acceptable. Loss Given Default There is uncertainty regarding the application of the 25% minimum floor for house prices. Given that a minimum is proposed it is not clear if the bank has expectations as to when they would expect it to be greater? We challenge the appropriateness of use of the most recent house peak as the reference point potentially due to its pro-cyclical impact.
3 2. Sufficient time to implement the revisions It was agreed at the industry meeting that the proposals to be published by the Basel Committee anticipated for year-end will not impact this consultation. Although the EBA has recently published its guidelines on the definition of default, there are two further relevant consultations due to be published by year end Long Run PD and Downturn LGD. We are keen to ensure that guidance and rules published by the Bank / PRA are consistent with EBA RTS and without requiring two sets of model development activity due to the PRA requirements being applied in advance of the EBA. Whilst mindful of the PRA s objectives, and considering the impact upon resources at the Bank, PRA and firms, we think that there should be a consistent implementation date of 1 January 2021. This would also allow further time to determine the requirements and sufficient time for testing and calibration of the models and the PRA to review and approve them. 3. Expectations of current developments Some of our members are currently developing new models for their residential mortgage portfolios and have been given the impression that they should develop new models and amend existing models to take into account the draft rules set out in the CP. This was discussed at the meeting and we would appreciate confirmation that our members will only be required to comply with the new rules after the revised Supervisory Statement is published and in line with the finalised timelines. 4. Implications for other models We note that the PRA opines that the revised proposals should also be applicable to non-uk mortgages. We are concerned that for the small set of our members that have non-uk mortgage portfolios, this may result in the modelling of those portfolios being inconsistent with the rules of the jurisdictions that require local prudential regulatory capital reporting. We urge the PRA to allow our members to implement models consistent with the rules where they operate. We note the Bank thinks that similar deficiencies to those already identified, may exist in other models and that firms should consider the matters set out in the CP when redeveloping other models. When asked during the meeting, the PRA advised that it was not expected to be rolled out to other portfolios at this time, due to a lack of reliable external data source. We would appreciate the PRA confirmation of this Conclusion We appreciate the initial publication of this consultation and the consequent engagement with industry. We believe that further dialogue is critical to ensuring the final expectations to be contained in the finalised SS11/13 address the challenges both methodological and with implementation so actually achieve the PRA s objectives. Furthermore, we also believe the implementation dates conflict with the activity required to meet EBA RTS changes and would strongly recommend that the PRA agree to the series
4 of workshops proposed by industry to support conclusion of the SS, with this corresponding to an implementation date which aligns to that of the RTS. We once again thank the Bank and the PRA for the opportunity to provide views and comment on this initial consultation and look forward to engaging on this topic over the coming months. Yours faithfully John Perry Jeremy Palmer Jon Saunders Senior Consultant, Head of Financial Policy Senior policy adviser (funding) Prudential Capital and Risk The British Bankers Association Building Societies Association Council for Mortgage Lenders T 020 7216 8862 0207 520 5912 0207 438 8934 john.perry@bba.org.uk Jeremy.Palmer@bsa.org.uk Jon.Saunders@cml.org.uk
5 A response to the Prudential Regulation Authority s Consultation Paper CP29/16 Residential mortgage risk weights October 2016 Detailed Comments Summary of the context to the Bank s proposals The UK has experienced two primary housing crises in the past thirty years. The first was in the early 1990s and the second in the years 2007-2009. Fiscal and monetary policies coupled with the roles and responsibilities of the Bank of England and the Government were profoundly different during these periods. They are not comparable. In addition, the approach to the measurement and pricing of risk by firms and the prudential regulatory and capital framework in the UK has evolved, with this period seeing the implementation of Basel I, independence of the Bank of England and the use of the IRB framework since 2008 (Basel II). This framework evolved further from 2010, as significant reforms have been put in place and continue to be implemented by the Bank and firms to improve the safety and soundness of the financial system. These measures have included an increase in the estimation of risk, improved quality in loss-absorbing capacity, back-stop measures to limit leverage (resulting in higher capital ratios for firms with lower risk weight %), higher levels of core funding and enhanced liquidity risk management. Yet also during the same thirty years the housing market itself has undergone changes and continues to do so to adapt to the changing economy. The current evidence is that there is a shortage of housing, material differences between house prices in different parts of the UK and a growing trend in the buy-to-let market as many younger people choose to rent and share accommodation instead of buy, often due to their inability to buy. The Bank estimates that the current stock of UK retail mortgages embracing owner-occupier and Buy-to-Let is c. 1,300b of which a growing percentage c. 15% of the total is accounted for by the latter. 2 Retail mortgages are the single largest real economy asset exposure class for UK banks and building societies (firms). For many their principal business model is originating and retaining retail mortgages, and for commercial banks these portfolios represent a material proportion of their overall risk. Revisions to the methodology for modelling the risk of residential mortgages are therefore potentially very impactful requiring very careful consideration before finalising rule changes. 2 http://www.bankofengland.co.uk/publications/documents/fsr/2015/fsr37sec4.pdf
6 Summary of the Bank s proposals The Bank has concluded that the current range of approaches to modelling PD and LGD has resulted in a lack of comparability of UK banks residential risk weights. This is a consequence of using highly point in time (PiT) or highly through the cycle (TtC) PD models and a wide variation in firm s house price fall assumptions embedded in LGD. Pro-cyclicality of capital requirements has also been observed due to the use of PiT rating systems by some firms. PD Firms would be required to understand the degree of cyclicality within their rating system in a consistent manner and assume a constrained degree of model cyclicality where they do not have grade level data. In doing so it is proposed that firms may no longer use the variable scalar approach to transform the PiT PD into TtC PD or undertake frequent recalibration to maintain PiT PD models. The proposals for PD can be summarised as follows: Require the use of hybrid models 3 to provide the grading structure; containing all relevant data Define a pool or grading structure Produce a long run average based on observed data where models could be scored out and internal data is held Where data is not held to the early 1990 s, the default rate by grade will require imputation from available external sources For the period over which imputed data is required, a 30% cyclicality constraint must be applied. The cyclicality of the rating system would be required to be measured and understood by firms A key proposal is that in measuring cyclicality the economic cycle must include the economic conditions prevailing in the early 1990s in the United Kingdom. 4 LGD Regarding the estimation of LGD, the Bank concluded that variability of existing approaches requires a harmonisation of assumptions. The existing expectation that the LGD model includes at least a 40% reduction in property sale prices from the peak will remain. This is supplemented by an assumption for the fall in the value of a property due to house price deflation of at least 25%. 3 During the meeting the PRA observed their expectation that models would not require development in the main (footnote following their thematic review of mortgages) but was a calibration issue 4 Refer to proposed paragraph 10.14 that includes the statement: The PRA is setting this expectation in light of recent economic experience and may revise it in the future as appropriate.
7 Specific Comments The following outlines concerns from the industry regarding the proposals and forms the basis for our request for a series of workshops. a) Methodology for PD We think that the proposed methodology for modelling PD may be even more challenging to implement than the existing approaches. Although conceptually it may appear that there will be a consistent methodology we think that this may not lead to consistent outputs. We recognise that there are some limitations with the variable scalar approach. However, we think that it would be beneficial to consider introduction of minimum periods (for example) after which a firm must review the calibration, rather than dismiss outright use of variable scalar approaches. We would like to explore the possibility of an improvement to this methodology that could overcome the Bank s concerns instead of requiring the revised approach. We think that these are important matters to consider when agreeing a new consistent methodology. We would like to explore techniques for modelling synthetic data especially for the buy-to-let mortgage market. b) Measurement of Cyclicality To achieve the Bank s objective requires the development of a sound and consistent approach to the measurement of cyclicality to calibrate the revised models. The Bank has set out its proposed approach for a method to measure this. 5 We support the establishment of an approach to measure the cyclicality of a model. However, we have many concerns that the approach proposed by the Bank is too theoretical, simplistic and stylised and will not be useful in practice. We would also observe material variation in outcomes. We have neither seen nor understood the basis for setting of a limit of 30% cyclicality. The proposed cyclicality measures will not in our view provide consistent coefficients through time and are also likely to produce negative or extreme values in periods of poor model calibration or economic or operational instability. Further work is required to determine whether either proposed formula can be augmented to provide a meaningful measure of cyclicality. Firms are keen to work with the PRA to develop a measure of cyclicality to be used in the calibration of the revised hybrid models. In Appendix A we set out in more detail our comments and concerns about the two approaches that the Bank has proposed. 5 Paragraphs 12.3 and 12.4 Rating philosophy
8 c) Data and period covered by the models We note that the Bank has explicitly proposed that low historical data and prime portfolios would not normally be combined in the same rating system, unless a firm was able to demonstrate that the approach met the proposed expectations to be contained within paragraphs 10.15-10.17 of SS11/13. Yet the Bank considers that data (if available) from the period 1990 1995 should be included in the modelling of PD and LGD. And if this data is unavailable and or portfolios do not cover this period then firms will be expected to model how book level default rates would have performed under the economic conditions experienced in a representative economic cycle. We recognise and acknowledge the need for a comprehensive set of data. The time series used to define the cycle is also critical. We think that the proposed definition of a cycle to include early 1990s data either actual or synthetically created could lead to a central tendency (CT) that may be inappropriate for some portfolios, leading in turn to inconsistent and incomparable outcomes, precisely the opposite of what the Bank seeks to achieve. The Bank will be aware that this is a challenge to determine and yet the value is central to the modelling of a long run average PD. We would like to have further discussion with the Bank on the use of data and acceptable approaches for creation of a synthetic portfolio in order to ensure that the Bank s objectives are met. d) Loss Given Default We recognise that the 25% floor defines minimum reduction in house prices, but that the proposed changes to SS 11/13 provides little additional guidance as to when a larger or smaller reduction may be appropriate. There has been material variation in the nature of house price peaks experienced by region ahead of the recessions of the 1990s and mid- 2000s. It is not clear whether the floor must be applied at regional or portfolio level to allow this to be recognised. Given the variation in the price of houses across the UK, the levels of indebtedness, and variations of the economy during the last recession and variation in the business models of firms (established and challenger banks and building societies) that may focus on different parts of the UK and different segments (owner-occupier, buy-to-let) and have differing lending criteria with respect to factors such as LtV. The imposition of a minimum national average fall in house prices may be inappropriate and not risk-sensitive. We recommend a more flexible risk-sensitive approach that better reflects firms risk profiles could be considered. We would query the sole use of peak to trough. Referencing the latest peak in a period of rising prices is likely to lead to cyclicality in downturn LGDs and hence RWAs. Consideration should be given to alternate reference points for example long run real house prices by region or house price to earnings ratio that would damp downturn LGD sensitivity to short term house price volatility.
9 Should the peak to trough reference point remain, and taking into account the above comments on the challenge to determine the cycle and estimate cyclicality, this returns us to the question of how to calculate the peak to trough benchmark of 25%. 2. Proposed timeline Should the PRA accept our proposal for a series of workshops to address the many challenges proposed by this consultation, we would propose the following plan: Hold a series of workshops between December 2016 March 2017 to reach a consensus on the approaches to modelling PD, LGD and measuring cyclicality Publish a revised consultation by mid 2017 - inclusive of the other matters set out in this paper c. 6 months after the publication of EBA RTS on PD and LGD Allow for a two-year period after the finalisation of the various EBA RTS and revised PRA supervisory statement for firms to submit to the PRA revised models Set aside a period of 18 months for the PRA to review models, and Firm-wide implementation of the revised models a year later. This supports the later implementation date of 1 January 2021 instead of 31 March 2019 as proposed in the consultation paper. Appendices Appendix A Cyclicality
10 Appendix A Cyclicality At the heart of the Bank s proposal is for firms to recalibrate their revised hybrid models to assess the cyclicality of the model. The approach is set out in sections 12.3, 12.4 and 12.5 of the proposed revised SS 11/13. The Bank requires a firm to segregate the recalibrated portfolios into those using internal observed default rates taken predominantly from a downturn period and those that have not. The assumption is that in the former case the cyclicality is taken into account and thus calculation of the cap on cyclicality will not apply, whereas in the latter case the application of a cyclicality cap may be required. In both cases it will be necessary to measure the cyclicality. The cyclicality of the rating system is a measure of where a system lies on the PiT / TtC spectrum. 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: (PDt CT) Cyclicality % = -------------- * 100 (DRt - CT) or (PDt PDt-1) Cyclicality % = -------------- * 100 DRt - DRt-1) Where: PDt means the long-run average PD at time t CT means the central tendency, or portfolio average default rate over a cycle DRt means the observed default rate at time t We have taken t to be the PD and DR estimates obtained from long run averages of oneyear default rates in each rating grade (refer CRR Article 180.1.a) and have assumed t-1 to be the previous year. The Bank has concluded that from these formulas an outcome of 30% is an acceptable upper limit in those years where grade level internal observed default rates are not available. This level reflects the PRA s current view of an appropriately conservative assumption for rating system cyclicality in light of recent experience. We agree that it is challenging to estimate the cyclicality of a rating system but also agree that it is prudent to establish a consistent approach to doing so. However, we think that Bank s proposals will not result in a consistent approach. It has shortcomings that will result in an unreliable measure of cyclicality so that the measure will be theoretically questionable and practically quite challenging. The following sets out our specific points of concern. 1. The proposed formula assumes an idealised situation, like the one depicted in this Figure below:
11 However, because of sample variation in Default Rates (DR) and estimation errors in PDs, we do not think that it is possible to observe such a consistency in PD and DR movements as shown in the Figure. Instead, what is likely to happen in practice is that the individual year-t realisations of the proposed ratio will often be either negative or above 100%. Our conclusion is that it is then questionable whether such numbers carry any useful information regarding the model s cyclicality (especially the negatives), and whether it makes sense to compute an average over such year-t observations in order to obtain a model-specific value of cyclicality. The Bank acknowledges that many firms will not have a long history of observations over a given rating model. 2. We think that any attempt to compute cyclicality based on a comparison between PD and the corresponding DR faces the following problem. Imagine that a firm has a highly PiT rating system. This knowledge may come from the bank s own declaration that it intends to predict portfolio-level PD as precisely as it can, using all available information as well as forecasts, including macro variables, financial market data, expert opinions, etc. It may be that if one computes a time series of cyclicality values coming from such a model, an erroneous conclusion may be that the (average) cyclicality of this model is low. However, this may be a consequence of a low correlation between PD and DR. Hence, the Bank s proposed cyclicality metric measures a mixture of two effects, namely, model s true cyclicality and its predictive power. We think that this introduces unacceptable model risk and an unacceptable measure of cyclicality. 3. We also think that the first formula, i.e. the one that uses CT, has an additional problem in that its outcome is sensitive to the firm s estimate of the CT. For illustration imagine that in the Figure above, the LRADF (i.e. CT under the Bank s terminology) is mis-estimated. Let s say, the LRADF line in the Figure is set at the estimated 3%, instead of the correct level of 2%. Then, during the early periods (1 through 5) the calculated cyclicality value will be very high; then in period 6 it will be above 100%; then it will suddenly become negative in period 7; then will be positive but very small during periods 8-11; and finally negative again in period 12. It will be quite hard for a firm to aggregate such volatile values into a meaningful result. 4. The Bank envisages that the rating grade PDs should be time-invariant and set at the level of the observed 1-year DRs within each rating grade. In a rating model calibrated so, we think that the only way for the credit cycle to be reflected in portfolio-level PD changes is through rating migration. Hence, the Bank tries to propose a formula that would essentially measure the extent of rating migration due to changing macroeconomic conditions. But in our opinion this does not do because of the reasons set out above.
12 Summary comments Some of our members have calculated the ratio using the formulas proposed by the Bank and have observed outcomes that are volatile, unreliable and random. That in turn leads to concerns regarding model risk and thus to the reliability of outcome of the risk weighted assets. We are also concerned about the proposal that for portfolios where there is a lack of sufficient external default data, the proposals involve modelling these assets using synthetic cycles, which can be done with reference to prime mortgage data, and as such this does not impose any additional data requirements. We dispute the suggestion that this will not impose additional data requirements. Our opinion is that the development of a model to estimate the risk of a portfolio with historical poor data and insufficient default data using a synthetic cycle that may not be representative of current and or historical cycles requires a robust process, methodology and validation that involve establishing a database that can be modelled. In conclusion, we believe the proposed definition of cyclicality has severe shortcomings and will increase model risk rather than reduce it. We urge the Bank to conduct a QIS and engage with the industry to develop an appropriate measure of cyclicality, as well as approach to calibration and data requirements back to 1990, In this context the large literature on TTC and PIT in the wholesale space should be considered, as well as surveys and papers on the topic by International Institute of Finance (including mortgage modelling). END