ASSET RISK AND STRESS TESTING – HOW TO DELIVER MEANINGFUL CONCLUSIONS
The Bank of England has just announced the details of the scenarios underlying its 2016 Stress Test. The declines in property values being factored in are certainly eye-catching and will lead to interesting results. Those results will only be truly meaningful however if all participants incorporate a deep understanding of underlying asset performance and risk throughout the process.
The 2016 Stress Test Scenario
At the end of March, the Bank of England published the details of the 2016 Stress Test to be undertaken by participating organisations, comprising the seven most significant (by PRAS-regulated lending) banks and building societies (Barclays, HSBC, Lloyds, Nationwide, RBS, Santander and Standard Chartered). The Stress Test scenario is not a forecast of the likely path of the economy in general or individual sub-markets in particular; rather it is a set of potential extreme conditions intended as a backdrop against which to assess the exposure of the banking system in general to “tail risk” events.
Nevertheless, the numbers assumed for the commercial real estate (“CRE”) sector are eye-catching to say the least; under the Stress scenario, CRE values are assumed to decline by -42% in aggregate, with prime values falling even further, by -49%. This is a significant departure from the BoE’s Base scenario, where growth is a very bullish +22% over the next five years (versus the IPF Consensus forecast of just +4%). Figure 1 shows the path of the BoE scenarios. It is apparent from this chart that given historical growth and the timescale of decline and subsequent recovery assumed by the BoE, no five-year lending period would witness anything like a -40% decline in values; Figure 2 shows the market-level declines in CRE values implied by the BoE Stress scenario for lending carried out on a five year basis at the end of 2011, 2012, 2013, 2014 and 2015.
For lending undertaken in 2011, the BoE Stress scenario implies only very slight declines in CRE values over the course of the loan – of -4.3% on an aggregate basis, -1.7% for secondary and -8.9% for prime. For lending from 2012 onwards however, the declines are more significant: of the order c-25% to -30% at the aggregate level, c-34% to -39% for prime and c-14% to -20% for secondary.
Forecasting Key CRE Lending Metrics – Probability of Default and Expected Loss
Whatever the level of average decline forecast, it is vital that this figure be treated correctly. When faced with a market average decline scenario of say -30%, there can be the temptation to assume that all lending at say 80% LTV will be in default; similarly, that no lending at say 60% LTV would default. The reality of course is that because of the heterogeneous nature of CRE assets, and the significant volatility of individual asset performance both above and below a market average, not all 80% LTV loans will default, while a significant number of 60% LTV loans will be at risk of default.
The extent to which this is the case has historically been poorly understood. Through our Debt Analytics service, we at CBRE have attempted to help lenders and Regulators quantify the risk of default of specific assets or entire markets under various growth (or decline) scenarios and for various lending parameters. Figure 3 shows, for the set of BoE Stress scenarios relevant to 2015 lending, this methodology in action.
- First, probability distributions are plotted around the set of average market decline scenarios. The distribution of individual asset performance around the mean is the product of extensive interrogation of CBRE’s database of historic individual asset performance, covering tens of thousands of assets over the last 20 years.
- The shape of the distribution varies according to the market and asset, taking into account factors such as sector, geography, lease length, vacancy, reversion and asset quality.
- In this instance for example, it is assumed that “prime” assets will have longer leases, lower vacancy and be rack rented, all of which factors reduce asset-level volatility resulting in a “tighter” distribution of potential outcomes.
- Similarly, secondary assets are assumed to have shorter leases, some vacancy and some over-rentedness, all of which factors tend to increase asset-level volatility widening the spread of potential outcomes.
- From here, it is possible to apply a given LTV at origination; the example in Figure 3 is of 60%, but any LTV from 0% to 100% (or even higher) could be chosen.
- The intersection of the chosen LTV with the probability distribution gives us the Probability of Default (PD) for that scenario at that LTV. So, for 2015 lending under the BoE Stress scenario, PD is 15% in aggregate, 5% for secondary and 48% for prime assets.
From here, Expected Loss (EL) can be calculated. Our approach to calculating EL first extracts from the probability distributions the end value of assets in default (e.g. £50) and applies a haircut to those end values (e.g. 15%) to produce a recovery value (e.g. £42.50) from which loss can be calculated relative to the LTV at origination (e.g. £60 – £42.50 = £17.50). Under normal circumstances, we would assume the recovery rate was higher from prime assets in default than for secondary assets – i.e. that the additional haircut to end value on prime might be as little as 5-10% but that on secondary could be as much as 20%. However, given that the BoE Stress scenario specifically assumes a more significant fall in prime CRE asset values as a result of reduced liquidity from overseas investors, we have not made such a distinction here.
Figure 4 shows both PD and EL figures for 2015 lending under the BoE Stress scenarios for three different LTVs – 50%, 60% and 70% at origination. As would be expected, there is significant variation in PD of lending at 50% LTV and 70% LTV at origination.
- At the aggregate level, under this scenario of average -30% value decline, PD rises from 2% on 50% LTV lending to 15% on 60% LTV lending to 58% on 70% LTV lending.
- With a higher average value decline of -38% in this scenario, PD on prime lending is even higher, at 6%, 48% and 94% on 50%, 60% and 70% LTV lending respectively.
- Meanwhile, secondary PDs are generally much lower, as a result of a significantly lower average value decline of -20%. 50%, 60% and 70% LTV lending produces PDs of 1%, 5% and 22% respectively.
- These differences in PD are reflected in forecasts of EL. EL is negligible for 50% LTV lending, but becomes more significant for 60% LTV and 70% LTV lending, especially for prime. EL is 10.6% and 26.2% for 60% LTV and 70% LTV prime lending, compared with just 1.2% and 5.0% for secondary lending. Aggregate EL figures are 3.3% and 13.7% respectively.
Figures 3 and 4 show PD and EL analysis for 2015 lending, but of course it is also possible to perform the same calculations for the 2011, 2012, 2013 and 2014 lending periods. Figure 5 shows an example of this, comparing PD and EL for the five lending periods for 60% LTV originations for aggregate property. For this combination of parameters:
- PD and EL are negligible for 2011 CRE lending.
- PD and EL are generally lower for 2013 CRE lending than for 2012, 2014 and 2015 CRE lending, as the BoE Stress scenario predicts less severe value decline for this five year period.
- For CRE lending at 50% LTV, PD peaks at only 2.2% in 2015, and EL at 0.5%.
- At 60% LTV, CRE lending remains defensive. PD reaches low double digits and EL approaches 3%.
- Only at 70% LTV does CRE lending begin to endure significant losses. PD for CRE lending climbs to 31% for 2013 lending, 43% for 2012 lending, and above 50% for 2014 and 2015 lending. EL also rises significantly, to 7% for 2013 lending, 10% for 2012 lending and 12-14% for 2014-15 lending.
Applying an Understanding of Asset Behaviour to Risk Assessment of CRE Lending
Up to now, we have just considered PD and EL at all property level, with the only variant being quality. This approach is very broad-brush, ignoring a huge amount of the idiosyncrasies that contribute to the volatility of individual asset performance – which in turns drives variable performance of loans. It is though possible to quantify the extent to which asset and loan performance will be influenced by a range of factors, and to build in this understanding into an assessment of PD and EL. At CBRE, we have interrogated our unrivalled database of tens of thousands of repeat valuations over the last 20 years to help lenders understand the risk of individual loans given the nature of the assets underpinning those loans. Instead of relying on broad all property averages therefore, lenders can build precise bespoke forecasts of key metrics on a loan-by-loan basis.
Figure 6 shows how PD varies by segment, comparing, for the BoE Stress scenario for 2015 lending (an average decline of -30%), PD on City offices, industrial and retail warehouses. (Of course, all sectors of the market can be considered, but for simplicity’s sake we have chosen just three). For lending at 60% LTV at origination, there is considerable variation, as a result of the different levels of asset volatility at the segment level.
- Historically, industrials have seen low volatility, and PD is just 8% in this scenario.
- Historically, City offices have seen high volatility, and PD is 19% in this scenario.
- Historically, retail warehouses have seen middling volatility, and PD is 11% in this scenario.
It is perhaps worth dwelling on this a little. In this scenario, the average asset falls in value by -30% to £70. For lending at 60% LTV, there is thus a £10 “cushion”; individual assets need to fall by an additional £10 to trigger a default. Industrials are protected by their low volatility; the tighter spread of asset-level performance results in just 8% of assets falling by £10 more than the average decline. City offices show much greater volatility at the asset level, and as a result, thanks to a wider spread of potential outcomes, 19% of assets fall by the additional £10 over the average to trigger a default.
The sector to which an asset belongs is not the only asset characteristic that can be taken into account. It is an important one, and arguably the form of classification that most people immediately think of, but in many cases it is not even the most important determinant of volatility. Arguably, lease length is of equal importance in determining risk at the asset level. Figure 7 demonstrates the impact that varying lease length can have on risk, by showing PD for City offices under the BoE Stress scenario for 2015 lending (an average decline of -30%). Three lease lengths are illustrated – 5 years, 10 years and 15 years – which clearly show how asset level volatility increases as lease length shortens. For this scenario, PD is almost three times as high on short-lease assets, at 29%, than on long-lease assets, at 11%, as the spread of potential outcomes on the former is much wider.
Figure 8 shows the distribution of EL for the above assets. In addition to lending at 60%, results for lending at LTVs of 50% and 70% have also been calculated. What is clear from this chart is the defensiveness of conservative lending:
- At 50% LTV, in an average decline scenario of -30%, EL is generally less than 1%, and even for short-lease City offices is under 2%.
- EL begins to climb and vary for 60% LTV lending; industrials, retail warehouses and long-lease City offices, with EL of around 2%, see half the level of loss as medium- and short-lease City offices.
- At 70% LTV, the level of value decline begins to dominate – the differences in EL across the assets begins to decline as the scale of decline overwhelms the smaller cushion built in by more aggressive lending, and EL bunches at around 11-15%.
Improving Understanding of Risk in CRE Lending
The Stress Testing of CRE lending has improved significantly in recent years, becoming more frequent, more rigorous and more exacting. Nevertheless, there are still huge areas for potential improvement in the process to bring debt-side understanding of asset risk, and the impact this has on potential performance, up to the level seen on the equity-side. Part of this can be achieved through greater depth, standardisation and centralisation of CRE loan data, and it is thus to be hoped that the Regulator follows through on commitments to support the industry in this undertaking. However, the greater part of improvement must come from an enhanced understanding of the drivers of asset-level risk and a quantification of those risks throughout the loan appraisal process – from initial pricing, through to ongoing portfolio monitoring (and re-shaping) and Stress testing. We hope this article makes helpful suggestions as to how this might be accomplished.
For further information on any of the topics covered in this article, or to find out more about CBRE’s Debt Analytics service, please contact Dominic Smith, Head of Real Estate Debt Analytics, t: +44 20 7182 2369, e: firstname.lastname@example.org.