Bank of England Tightens IRB Mortgage Models

The UK Prudential Regulation Authority (PRA) proposes to set out a revised approach to IRB risk weights for residential mortgage portfolios and guidance as to how firms model probability of default (PD) and loss given default (LGD) for these exposures. The effect will be in some cases to lift the amount of capital held against mortgages.

This follows a review of the causes of variability of residential mortgage risk weights for firms with permission to use the IRB approach to calculate credit risk capital requirements which showed that first firms’ approaches to modelling PD vary. The majority of firms either use a highly point-in-time (PiT ) approach or a highly through-the-cycle (TtC) approach. In both cases a deficiency in risk capture was identified. Secondly, firms’ house price fall assumptions for UK residential mortgage LGD models vary widely.

House-and-ArrowSo the PRA proposes that firms would be expected to adopt PD modelling approaches that avoid the deficiency in risk capture identified in the PiT and TtC models currently used by firms, and calibrate their models using a consistent and appropriate assumption for the level of model cyclicality.

The PRA also proposes to expect firms not to apply a house price fall assumption of less than 25% in their UK residential mortgage LGD models.

The PRA expects, in general, that these changes will result in firms having to recalibrate existing models rather than develop new ones. The PRA proposes that they will come into effect by 31 March 2019, though the PRA may on a case by case basis allow a longer period for firms to meet these expectations.

By way of background:

in December 2014, the Financial Policy Committee (FPC) raised concerns about excessive procyclicality and lack of comparability of UK banks’ residential mortgage risk weights in the 2014 UK stress test. The FPC mentioned in December 2015 that work was underway to try to investigate these issues, stating that in “the United Kingdom, the FPC and PRA Board are also considering ways of reducing the sensitivity of UK mortgage risk weights to economic conditions. The 2014 stress test demonstrated that the risk weights on some banks’ residential mortgage portfolios can increase significantly in stressed conditions”.

In implementing PiT models in the United Kingdom, firms’ residential mortgage models estimate a PD for the next year based upon the previous year’s default rate. This means that PDs are based only on very recent experience.

The PRA believes that for residential mortgages, this approach leads to capital requirements that are excessively procyclical. This is because under this approach mortgage assets, which are long term and cyclical, are calibrated based only on short term experience. This can lead to Pillar I capital requirements which are too low in an upturn and too high in a downturn, because a short term change in default rates leads directly to a change in the capital requirement for what is a long term asset. In turn this means that capital ratios may also appear too good in an upturn and too bad in a downturn.

A procyclical capital framework, where capital requirements are high in a downturn and low in an upturn, can encourage credit exuberance in a boom and deleveraging in a downturn. With major UK firms holding around £1 trillion of UK residential mortgage exposure, this is an asset class where excessive variability of capital requirements can be detrimental to financial stability.

TtC models, as implemented by UK firms, adopt a static approach whereby the PD does not vary with changes in the general economy. These models tend to use a relatively limited number of inputs, that do not change with time, to estimate average default rates for each borrower over an economic cycle. The borrower’s PD does not therefore change with economic conditions, and capital requirements vary much less than with PiT models.

In the UK firms use a form of TtC approach known as ‘variable scalar’ that use as inputs the PDs derived from relatively PiT models. Variable scalars then transform the average PiT PD for a portfolio into a static TtC PD, by using a multiplier, or scalar, that varies through time.

The PRA has found that, for residential mortgage portfolios, firms using TtC approaches, including variable scalar approaches, are unable to distinguish sufficiently between movements in default rates that result from cyclical factors (for example, factors that impact the economy in general) and those that result from non-cyclical reasons (for example, the specific performance of one borrower). These approaches only take account of a small number of risk drivers that do not change with time, and the PRA has found that this results in risks not being sufficiently captured. For example, if a particular portfolio deteriorates due to poor underwriting (rather than due to a downturn), then capital requirements calculated using variable scalar approaches may not increase as they should.

 

 

Author: Martin North

Martin North is the Principal of Digital Finance Analytics

Leave a Reply