Martin Blåvarg Risk-weighted capital requirements do they work?

Transcription

1 Martin Blåvarg Risk-weighted capital requirements do they work? Risk-weighted capital requirements give banks incentives to manage their risks and mean that bank lending is steered to the parts of the economy where it provides most benefit. With the correct regulations and follow-up, it is possible to ensure that the capital requirement is not too low and thus does not need to use the leverage ratio requirement and other non-risk-based requirements which steer banks towards taking risks to a greater extent. Handelsbanken s Series of Small Publications No. 30

2

3 Martin Blåvarg Risk-weighted capital requirements do they work? Risk-weighted capital requirements give banks incentives to manage their risks and mean that bank lending is steered to the parts of the economy where it provides most benefit. With the correct regulations and follow-up, it is possible to ensure that the capital requirement is not too low and thus does not need to use the leverage ratio requirement and other non-risk-based requirements which steer banks towards taking risks to a greater extent. October 2014 Handelsbanken s Series of Small Publications No. 30 1

4 Innehåll Introduction... 4 The IRB system how does it work at Handelsbanken?... 7 Control mechanisms to ensure conservative values in the IRB system The IRB regulations restrict the choice of a bank s own model Approval Validation...17 Comparative studies of banks risk weights Reasons for differences in risk weights between banks Leverage ratio the alternative to risk-weighted capital requirements Summarising comments

5 Foreword The new capital adequacy regulations that were drawn up internationally in the early 2000s under the name of Basel 2 were, in many respects, revolutionary. They were generally regarded by banks and regulatory authorities as a very positive step. The main revolutionary aspect was that capital adequacy for traditional lending to companies and households would start to take into account the risk-taking in the lending in an entirely new, far more finely tuned manner than its predecessor, Basel 1. The main positive aspect was that the more risk-based capital adequacy would mean that the capital buffer that banks needed to maintain would be affected by the risk they actually took. This was naturally a positive step for stability in the banking system, since higher risktaking would require considerably more capital. And since the capital adequacy cost determines the banks funding cost for a loan, the risk-based capital adequacy would mean that credits with low risk became cheaper and those with high risk became more expensive. The consequence of this would be a more effective allocation of capital in the society, whereby investments that have a low risk in relation to the return are simpler to carry out, resulting in higher economic growth. However, as soon as Basel 2 came into force in 2007, the global financial crisis began a crisis that we are still feeling the effects of. The crisis exposed a series of shortcomings in the regulation of banks, relating to capital adequacy for market risks, derivatives, securitisations, etc., as well as a lack of quality in the capital that banks were holding as a buffer against their risks. Although the Basel 2 methods for measuring credit risks vis-à-vis standard companies and households did not contribute to the crisis that arose, they have to a large extent been included in the criticism of the regulation of banks in the wake of the crisis. It is fairly obvious that the new rules cannot have contributed to the crisis, given that they had only just come into effect when the crisis began, and no bank could have drastically changed its capitalisation on account of them. And there are hardly any examples of capital adequacy for traditional household and corporate credits being too low as a result of Basel 2, in such a way that excessive risk-taking was induced in any particular bank, with this causing the bank financial problems. The aim of this report is to endeavour to show, as simply as possible, how riskbased capital adequacy for credit risks actually works, and to show that there is every opportunity for supervisory authorities to monitor matters and to ensure that, in capital adequacy calculations, risk is being measured in an appropriate way. It is not a matter of a black box from which banks can conjure favourable results using creative models, but of relatively simple calculations of how well banks internal assessments of counterparties credit risks reflect the actual risk that the counterparty will cause loan losses for the bank. This report also discusses the question of why the risk-based capital requirements are positive for the national economy, and why the introduction of non-risk-based capital adequacy requirements, such as leverage ratio requirements, may be detrimental to economic and financial stability. The report is not a part of Handelsbanken s official financial information. 3

6 Introduction In the past few years, increasing scepticism concerning risk-weighted capital requirements has been seen in the international process of developing well-functioning capital adequacy rules. Basel 3, which is fundamentally based on developing the concept of risk-weighted capital requirements, is now being implemented. At the same time, increasingly high on the agenda is the desire to supplement or even replace the risk-weighted capital requirements with a leverage ratio requirement which does not take into account the quality of the assets banks have on their balance sheets. Basel 3 contains proposals to implement a leverage ratio requirement, but this has not definitely been decided as a binding requirement and implementation of the ratio is further in the future than the other regulatory changes. In the meantime, some countries, such as the UK and the US, have decided to introduce a binding leverage ratio requirement. The desire to use a non-risk-based ratio instead of the risk-weighted measure developed by the Basel Committee, is largely the result of three different objections. The first is that, as a result of Basel 2, banks were given the right to use their own models to measure their capital requirements and that it seems unhealthy for banks to decide themselves how their capital requirements are to be calculated. This objection reflects the view that the banks can, by means of their work with models, to a large extent determine the capital requirements they will be subject to. The second objection stems from the observation that the capital adequacy regulations imply major differences in banks risk weights. In other words, some banks have a far lower proportion of risk-weighted assets on their balance sheets, relative to their assets, than other banks. According to this view, this difference per se would indicate something suspect in the risk-weighted capital requirements, or that there is something unfair about how the regulations are applied among the various banks. The third objection is based on the fact that risk-weighted capital ratios have not proved to be a better indicator of whether a bank risks failing than non-risk-weighted capital ratios. This would indicate that a leverage ratio requirement could be better than a risk-weighted capital ratio. One problem in the debate about risk-weighted capital ratios is that, in fact, there are few arguments or studies showing that the risk-weighted capital ratios do not work, at least regarding the capital adequacy regulations for credit risk which were introduced through Basel 2. It is seldom heard of anyone pointing to an actual bank failure where a contributory reason has been that the risk was underestimated as a result of implementing the Basel 2 regulations for credit risk. Much of the scepticism seems to stem from the fact that the regulations are complicated, difficult to understand and even more difficult to evaluate, especially for those who only have access to public information about the banks. The purpose of this report is to discuss as simply as possible how the system for calculating the capital requirement for credit risk is constructed at Handelsbanken, and also to demonstrate that the supervisory 4

7 authority is well able to check that the capital requirement does not underestimate the risk. We also discuss why there are major advantages in risk-based capital requirements compared to a non-risk-based leverage ratio requirement. By focusing on the most important components and providing actual examples in figures, the description will hopefully make the IRB system 1 somewhat easier to understand. We hope that the report will help to allay the scepticism which is based on the first two objections mentioned above, i.e. that the IRB system leads to the banks determining their own capital requirements to too great an extent, and that the mere fact that risk weights vary significantly shows that risk-weighted capital requirements should be challenged. The third objection that risk-weighted capital requirements have not proved to be a better indicator of risk for a bank than a leverage ratio requirement is worth reflecting on 2. This objection is based on the assumption that one purpose of the capital adequacy regulations is for the capital adequacy measurements to provide a signal about a bank s stability. But this has never been the purpose of the capital adequacy regulations. The purpose of capital adequacy is that there must be a buffer of equity to protect against major losses where depositors and other lenders, including tax-payers in certain cases, lose the money they have lent. If all banks had the same level for their risk exposure, a bank with better capitalisation would have a lower risk of failing. But of course this is not the case, and it is precisely the risk in the assets which determines the size of the losses for a bank. A bank can choose to have high capitalisation because it wishes to show that it has a large buffer which can cover losses and protect the lenders. But the bank can also decide to have high capitalisation because its operations entail risk, which its lenders realise and therefore demand high capitalisation for in order to lend money to the bank. All experience shows that it is the risk in the assets that determines whether or not a bank will have problems. It is difficult to see why capitalisation would be a good indicator of the risk in a bank, particularly if it is measured using a non-risk-based leverage ratio requirement. If capital adequacy is measured using a leverage ratio requirement, it is likely that a bank with very low-risk assets would not need such high capitalisation, since the investors would not require as high capital buffers. The studies that have shown that good capitalisation in terms of a risk-based measurement is not a high indicator of a low risk of default in a bank, have principally used capital measurements from Basel 1 3. The Basel 1 regulations did not factor in the risks that caused the banking problems during the major international financial crisis. Examples 1 IRB stands for internal ratings-based and the term IRB system is used in this report to designate the rules, calculation methods and practical application used for internal calculation of capital requirements for credit risks implemented in Basel 2. 2 This objection has been discussed in a publicised speech by Andrew Haldane (Executive Director at Bank of England), The Dog and the Frisbee, August See, for example, Aikman et al Taking uncertainty seriously: simplicity versus complexity in financial regulation, Financial Stability Paper No. 28, Bank of England, May

8 of types of exposure that were not risk-weighted in a reasonable manner in Basel 1 were exposures between banks, securitisations, credit risks in traded bonds, derivatives and liquidity lines for securitisations. Consequently, the risk-weighting did not function well at all, but this has largely been rectified through changes made as part of Basel II and Basel III. In fact, the Basel 1 regulations were essentially a leverage ratio measurement for loans to households and companies, which are the main areas where the IRB system has a major impact, since all such loans were risk-weighted at 100 per cent, except for residential mortgages, which had a 50 per cent risk weight. In the light of this, proceeding from studies that use Basel 1 capital requirements, and using these as arguments against risk-weighting exposures to household and corporate loans, is difficult to understand. Furthermore, many banks which have had problems and received government support during the financial crisis and are thus regarded as having failed, have not reported losses to the extent that a large part of the capital has been consumed. In many cases, the losses have suddenly increased and created uncertainty, thus causing the banks financiers to start pulling out and subsequently needing government assistance to avoid being hit by acute liquidity problems. In this situation, the size of the capitalisation does not matter; merely the fact that a bank is on a slippery slope where the losses can grow to an amount which is impossible to assess, can cause it to fail. Confidence in a bank is of such importance to its ability to survive that banks risk running into difficulties and needing a state bailout, regardless of whether they have financial problems that are visible in the reported figures. There are also plenty of examples of where the valuation of the assets is not tested on a regular basis in a fair way, with the result that the capital adequacy does not correspond to a correct valuation of a bank s financial position 4. In the light of these arguments, it is difficult to see the indicator value of capitalisation as an important argument for challenging risk-weighted capital requirements. First of all, it should be stressed that the discussion in this report deals with the use of internal models for the capital requirement for credit risk, not market risk or operational risk. In many respects, the internal models in these areas function differently from credit risk. They entailed major problems during the latest crisis (at least the models for market risk) and since Handelsbanken does not use internal models in these areas, it is not relevant to describe them here. In this context, it is also worth noting that the IRB system is mainly used for lending to households and companies. How it works is therefore of major importance for how well the banks manage to fulfil their central task from a socio-economic point of view: that of providing the real economy with credit. The capital requirement for market risk mainly covers banks 4 A fairly recent example of this is the Dutch SNS Reaal, which had a common equity tier 1 ratio of 10.4 per cent, comfortably above the regulatory requirements, in its last public report before the bank was taken over by the state of the Netherlands in At the time of the takeover, it was declared that all the capital had been consumed, with the consequence that the shares, as well as the subordinated loans that made up the banking group s capital base, were expropriated by the state. 6

9 trading operations, which are normally not considered to have the same socio-economic importance. The report is structured as follows: first we describe how Handelsbanken produces its internally calculated capital requirement. The next section discusses the control mechanisms which exist to ensure that the capital requirement results in a reasonably fair measure of the risk in the exposures. The third section discusses the studies by the international authorities of differences in risk weights, followed by a discussion on the principles as to why it is reasonable to expect banks to have different risk weights. The fourth section deals with the disadvantages of non-risk-based capital requirements compared to risk-based capital requirements. The report ends with some summing-up conclusions. The IRB system how does it work at Handelsbanken? The debate about how banks capital requirements should be designed often presents the internal risk classification models as black boxes models which the banks design at their own discretion, with a great deal of scope for producing the desired risk weights. This is definitely not the case. The IRB system is subject to strict, comprehensive regulations which allow very little opportunity for internal choice of method and where there are a large number of control mechanisms to ensure that the risk parameters are not underestimated. In this context, it is clear that the often used concept of internal models is misleading, since it suggests that these are risk models with a large degree of freedom which are internally constructed at the bank. In fact, it is not a model at all. It is a pre-defined matematical formula called the Basel formula, which uses certain internally estimated risk parameters to calculate a capital requirement for a bank. The IRB system measures these central risk parameters based on a bank s historical data and then uses these parameters in the Basel formula. The following describes at an overarching level how the IRB system works at Handelsbanken. Risk-weighting of lending in the IRB system is done by producing three risk parameters for each credit agreement. These three parameters are put into the Basel formula, which then calculates a capital requirement. The three risk parameters are: Probability of Default (PD), which indicates the probability of a counterparty in the credit agreement not paying interest and amortisation when due. Loss Given Default (LGD), which indicates how large the loss, as a proportion of the exposure, can be expected to be for a counterparty which has defaulted. Exposure At Default (EAD), which indicates how large the exposure can be expected to be if the counterparty defaults. For a normal loan, EAD is the loan 7

10 amount, but for off-balance sheet exposures such as derivatives, guarantees and committed loan offers, an expected exposure must be calculated. In most cases, EAD is calculated as a proportion of a nominal exposure where this proportion is called the Credit Conversion Factor (CCF) and is a risk parameter which is calculated in a similar way as PD and LGD. In addition to these three central risk parameters, there is also a maturity factor (M) for each credit agreement. The purpose of taking into account the maturity is that it is more difficult to assess the credit risk over a longer time horizon, so if the bank ties up its exposure for a longer period, the risk will be higher than for a short-term credit agreement. The maturity factor is not, however, a parameter based on a statistical estimate. It comprises the actual remaining time to maturity for the credit agreement. Another concept which appears in this context is Expected Loss (EL), which for a certain credit exposure or portfolio of credits shows the amount of the statistically expected loss for the exposure, based on the risk parameters in the IRB system. EL is calculated as PD times LGD times EAD. The IRB system can be applied either as a foundation approach or an advanced approach. In the foundation approach, only the PD values are calculated internally. The other parameters are determined by the regulations and the M factor is assumed to be 2.5 years for all exposures. In the advanced approach, internal calculations are also used to estimate LGD and EAD, and the M factor is taken individually for each exposure. The following description and discussion focus on how Handelsbanken works with the PD values for corporate exposures. There are several reasons for this. Firstly, the purpose of the description is to provide an instructive and easily understood description of how the IRB system works. This means that the description must be restricted to the most important components. Secondly, corporate exposures calculated using the IRB approach account for 59 per cent of Handelsbanken s total capital requirement. Consequently, how these methods work is by far the most important factor in terms of Handelsbanken s capital requirement. Naturally, the capital requirement for private individual exposures is also important and the IRB system contributed to significantly reducing this capital requirement. In Sweden, however, the Financial Supervisory Authority has decided to have a higher capital requirement for mortgage loans by introducing a floor for how low the risk weights are allowed to be. This floor was initially set at 15 per cent, but the Supervisory Authority recently raised it to 25 per cent. The main justification for the risk weight floor is not an assessment of a high risk of loan losses on mortgages. Rather, it is justified by a concern that high indebtedness in the household sector will make household consumption vulnerable to shocks, such as falling house prices or rising interest rates. The floor works in such a way that the reported capital adequacy values will continue to be based on the risk weights from the IRB system, but the 8

11 Supervisory Authority is adding an extra capital requirement as part of Pillar 2 corresponding to the difference between the floor level and the risk weights calculated according to the IRB system. This means that for private individuals, the IRB system will be of less importance for Handelsbanken s total capital requirement. Thirdly, it is the PD dimension which has had the greatest impact for Handelsbanken (and probably most other banks) in terms of the difference in capital requirement between Basel 2 and Basel 1. In the transition from the foundation approach to the advanced approach, when Handelsbanken also got approval for calculating LGD and EAD using its own estimates, the overall change in the capital requirement was only marginal (although naturally there could be major changes for individual credit agreements in both a positive and negative direction). It is also the case that the principles for making estimates based on historical data are mainly the same for PD as for LGD and EAD. The development and use of the PD values can be summarised in a number of steps. The basic idea is to measure how common defaults have been historically and to make some necessary adjustments to ensure that the historical data do not underestimate the risk. The adjusted historical default rates are then used as the PD values in the Basel formula. The various steps are: 1. Sort all counterparties into different risk classes. 2. Measure how common defaults have been historically in each risk class, in other words calculate the historical default rate. 3. Calculate statistical safety margins to ensure that the default rates are translated to PD values in a conservative way. 4. Possibly make adjustments to take account of where in the business cycle the Bank is, relative to the business cycle situation in the period of the historical data. 5. The historical PD values, after security margins and any business cycle adjustments, are then used for all counterparties in each risk class when calculating the capital requirement for each credit agreement based on the Basel formula. These steps are each described in the following section. 1. Risk classification. Handelsbanken s risk classification of corporate counterparties is based on an internal rating methodology used at the Bank for more than 30 years. The internal rating is based on each counterparty being assessed in two aspects: the risk of financial strain and the financial powers of resistance. The rating is set by the person at the local branch office who is responsible for granting the credit and it includes both quantitative data and qualitative assessments. 9

12 In conjunction with the decision on a credit or a credit limit, the rating is also assessed by a superior level in the credit process. The rating of the counterparty is dynamic and it is tested continuously whenever new information is received concerning the counterparty. The internal rating leads to the counterparties being categorised in the risk classes 1 to 9, where 1 is the lowest risk and 9 is the highest risk. When the rating system is used for capital requirement calculations, in general there is an aim for the rating to represent the risk over a business cycle, which is usually referred to as through-the-cycle. In practice it is more or less impossible to design a rating system that provides a stable rating over the business cycle while simultaneously capturing changes in a counterparty s risk. This is because deterioration of a counterparty will normally be expressed in the financial data, such as weaker earnings or a fall in the value of the counterparty s assets. It is often very difficult to determine whether this type of deterioration is due to normal cyclical variations, or because the counterparty is experiencing other problems which undermine its creditworthiness in the long term. A rating system which completely disregards cyclical variation and where the rating must represent a certain probability of default regardless of where the counterparty is in the business cycle, is usually called point-in-time. Handelsbanken s rating system can be viewed as a hybrid of a through-the-cycle system and a pointin-time system. 2. Measuring historical default frequencies. This calculation is done by simply measuring how many counterparties have defaulted in a given risk class during a given year. Since the measurement is performed for a large number of years and because the number of counterparties in each risk class is relatively large, a statistically sound estimate of the historical default frequency is obtained. Handelsbanken has used the internal rating system for a long time and it is therefore possible to use long time series, which is important for achieving statistically acceptable values. Handelsbanken has split its corporate counterparties into four groups, where the default frequency in each risk class and group is calculated separately. This is because the default frequencies for each risk class in the different groups can be expected to differ, and this has also been confirmed by the data. In other words, the ability of the risk system to estimate risk of default is improved by means of this division. The four groups are large companies, medium-sized companies, property companies and housing co-operative associations. 3. Calculating statistical safety margins. When historical data is used to estimate a risk parameter for forward-looking use, there is reason to take account of the size of the statistical uncertainty in terms of the access to and quality of the historical data. Thus, a statistical safety margin is always added to the historical default rate. The size of the safety margin depends on the number of observations which 10

13 are available, where a lower number of observations gives rise to larger safety margins. Other safety margins can be added if the data material indicates that the default rate behaves in an unreasonable manner and its reliability hence can be questioned. One example is if we observe a lower default rate in a poorer risk class than in higher risk classes. A better risk class should always have a lower default rate than a poorer risk class. In other words the default rates should continuously increase, the poorer the risk class is. If the data does not show this pattern, Handelsbanken raises the PD value for the poorer risk class by a safety margin which is sufficiently large for the PD value to be at least as high as for the risk class which is one step better. 4. Performing business cycle adjustments. The principle of the PD values used to calculate the capital requirement in the Basel formula is that they should represent an average expected frequency of default over a business cycle. The reason for this is that the Basel formula is designed as a simple credit risk model where the capital requirement generated by the formula for a credit portfolio must correspond to the risk that the capital covers losses within a 99.9 per cent confidence interval. In other words, the capital can risk being insufficient in only one year out of 1,000. The risk in a credit portfolio naturally varies over the business cycle. The Basel regulations have chosen to allow this risk to be captured by the statistical characteristics of the Basel formula instead of demanding that the risk parameters included must vary over time. If the time series were long enough, covering several business cycles, there would be no need for a business cycle adjustment of the historically measured PD values. But often the supply of long time series is limited, so the PD values may need to be adjusted so that they as far as possible correspond to the values over an average business cycle. If the historical period contains default frequencies which are expected to be lower than this historical average, the PD values will need to be raised, and conversely if they can be expected to be higher, the PD values may need to be lowered. In this context, Handelsbanken has chosen a conservative approach and never lowers the PD values below the PD values measured historically, with the addition of statistical safety margins. 5. Using the final PD values to calculate the capital requirement. The historical PD values, with the addition of safety margins and any business cycle adjustment, are used in the Basel formula to calculate the capital requirement. The table below shows the average PD values as at 31 December 2013 for each risk class for Handelsbanken s corporate exposures in the advanced approach, together with the average values for EAD and LGD and the resulting average risk weight for each risk class. 11

14 Average risk parameters and exposure volume for Handelsbanken s corporate exposures within the advanced approach, as at 31 December 2013 Risk class Total exposure (EAD) (mkr) Exposure weighted PD % Exposure weighted LGD % Average risk weight % Control mechanisms to ensure conservative values in the IRB system When the supervisory authorities in the Basel 2 regulations allowed banks to use their own data to calculate the capital requirement, obviously it was important to find methods to ensure that banks do not abuse the system. It was necessary to ensure that it would not be possible for banks, based on historical data, to use risk parameters which do not reflect the risk in the portfolio. There are many control mechanisms and comprehensive regulations which must be followed when developing the IRB system in a bank. There is not space to report on all these aspects here, but there are some parts which are particularly relevant and which are worth highlighting. The IRB regulations restrict the choice of a bank s own model In the debate about capital requirements, it is often said that the IRB system allows banks to use their own models to calculate the capital requirements. This is not the case. In fact, the banks have been given the opportunity to use internal data to estimate certain risk parameters, which are then used in a standardised credit risk model known as the Basel formula. The choices to be made when measuring default frequencies and loss rates are clearly specified in the regulations and in most cases have no effect on the final outcome. The historical defaults and losses which a bank has experienced will turn up somewhere in the data material and will thus affect the internal risk weights. One problem which was discussed in detail ahead of the implementation of Basel 2 is that many banks lacked sufficiently long time series of internal data. Furthermore, 12

15 the period which preceded the implementation had been relatively favourable in terms of losses in the banks credit portfolios. As a consequence of the global financial crisis, this problem has become far less important since most banks experienced increased defaults and losses during the crisis. This means that the IRB portfolios which were approved for the banks, particularly in 2007 and 2008, have been subject to a reality-based stress test. The following section describes the options and degrees of freedom which actually exist in the different steps when producing the PD values. 1. Risk classification. When implementing the IRB system, a bank must decide which types of exposures are to be measured as a portfolio and how the different exposures in the portfolio are to be split into different risk classes. The rough portfolio segmentation is strictly regulated in the regulations exposure classes. The main exposure classes for Handelsbanken are Corporate exposures, Institutions (banks), Sovereign and Retail exposures, which in turn is divided into Private individuals and Small companies. The demarcation lines between these exposure classes are important, since different variants of the Basel formula are used for different exposure classes. The Basel formula assumes certain determined correlations between various exposures and the correlations are higher for Corporates and Institutions but lower for Private individuals and Small companies. This means that a given PD value for an exposure leads to a higher risk weight if it is classified as a Corporate exposure than if it is classified as a Small company. The strict rules for the exposure classes in the regulations mean that banks have not been able to choose which exposure class the different exposures will belong to. One exception is housing co-operative associations, where at least one Swedish bank classifies its exposures as Retail exposures, while Handelsbanken has a more conservative approach, classifying these as Corporate exposures. As regards the risk classification in the respective exposure classes, each bank can choose how the risk classification will be carried out. Here the bank should decide how best it can distinguish low-risk exposures from high-risk exposures and find a way of grading the counterparties according to a scale which is as exact as possible. As discussed above, Handelsbanken is able to use the internal rating structure applied for over twenty years before the IRB system was implemented almost ten years ago. The choices made will affect the bank s risk weights, since a rating system which is better at distinguishing low risk from high risk generates lower risk weights than a rating system with poorer prediction ability. It is not per se negative that there are some options and it does not in itself mean that manipulation is possible. Actually, it provides an incentive for banks to design a rating system which is as precise as possible and to devote large resources to ensuring that the system continuously classifies risks in the best way possible. 13

16 There are two circumstances which reduce banks opportunities to affect the risk weights through their own choices. The first is that the bank cannot declassify defaults and losses which exist in the historical data. They will always affect the exposures in some part of the portfolio. The second is that when these choices are made, the data which is added year after year will affect the risk weights out of sample. In other words, the new data will verify or repudiate whether the risk classification works well and whether the PD values measured continue to be representative (see the section below concerning validation). If the risk classification does not work well, or defaults increase, the PD values will become worse and worse over time. When the bank has built up its risk classification system, it cannot easily be changed. All changes must always be examined and approved by the supervisory authority, which of course will always pay attention to whether the changes have been carried out in order to strengthen the risk classification system or for other reasons. 2. Measuring historical default frequencies. When measuring defaults, there is in fact no choice in terms of how the measurement is to be performed. What constitutes a default is clearly defined in the regulations. An exposure has either defaulted if a counterparty is more than 90 days late in paying or if there are other clear signs that the counterparty will not pay, for example, if the counterparty has been made bankrupt or negotiations on reconstruction have started. Despite the clear regulations, it has been noticed that defaults are sometimes measured differently in Europe, for example that counterparties which have been given payment concessions are not regarded as in default. In view of the clarity of the regulations, it is rather strange that these differences exist. Handelsbanken applies the regulations strictly, with counterparties with payment problems being registered as in default at the latest when their payment is 90 days past due, or earlier if there are other indications of payment problems. In other respects there are certain degrees of freedom in terms of the length of the time series which are used. These are usually limited by how far back in time the bank has default data available, and the extent to which historical data concerning counterparty characteristics is saved. A minimum requirement is that at least five years of data must be saved. When the supervisory authority examines the IRB measurements, it is obviously an advantage to have as long time series as possible, since this increases the statistical reliability. If the time series is short, the supervisory authority requires that greater statistical safety margins are to be used to compensate for the lack of historical data. 3. Calculation of statistical safety margins. The regulations are clear in terms of the requirement for statistical safety margins. But the requirement is not as specific regarding the methods to be used when calculating the safety margins. However, 14

17 the Basel regulations provide a good deal of guidance on calculating statistical safety margins. In its approval process, each supervisory authority has also been able to exercise control over the banks it is responsible for, so that equivalent methods are applied. It is possible that some differences have evolved between countries, but there is no reason why banks within a country should not have the same requirements from the supervisory authority. It is also important to bear in mind that safety margins can only lead to differences between banks in terms of how conservative the applied PD values are relative to the historically measured average values. Thus the PD values may only be higher after the safety margins have been added. 4. Performing business cycle adjustments. Business cycle adjustments are perhaps the area where there is the greatest degree of freedom in terms of the banks choice of method. The opportunity to apply business cycle adjustments to the PD values is not defined in the regulations. As discussed above, the purpose of the business cycle adjustment is to ensure that the historical data on which the PD values are based is representative of a business cycle. When Handelsbanken developed the IRB methods in the mid-2000s, the time period with detailed data for PD measurements contained only relatively favourable economic conditions. The economic climate had been good for a long period and the downturn in was relatively mild both in terms of the defaults measured at Handelsbanken and the number of bankruptcies in the economy in general. Since the conditions were favourable, relatively large increases were made in the PD values, so that they would not be lower than could be expected over an average business cycle. The method for business cycle adjustment took account of a far longer history of losses, including the extremely serious bank crisis in Sweden in the 1990s. The fact that a very serious crisis was taken into account when calculating the PD values ensured that they were conservatively estimated. During the recent financial crisis, most developed economies suffered a serious economic downturn and increased default rates. As a result of this, there is less need for business cycle adjustments of the PD values due to an over-positive default history. Instead the discussion has come to deal with business cycle adjustments in the other direction, in other words that the PD values should be lowered because the data period on which the PD values are based is worse than can be expected during a normal business cycle. Although Handelsbanken experienced far higher defaults and losses during than in the previous downturn in , in order to ensure that the PD values are always cautious, the Bank has decided not to adjust them down temporarily. Banks normally provide no or very little information about their business cycle adjustments, so it is difficult for external stakeholders to assess to what extent this affects the conservatism of the internally calculated risk weights. This is probably 15

18 an area where there is scope for increased transparency on the part of the banks and more harmonised application on the part of the supervisory authorities. In one public case, however, the business cycle adjustment, combined with other circumstances, has led to one bank s risk weights for corporate exposures being questioned by the supervisory authority. In its decision memorandum regarding the assessment of the bank s risk weights in its corporate portfolio, the financial supervisory authority in question stated that the use of the point-in-time method compared to the through-the-cycle method used by the bank would raise the risk weights by eight percentage points. This was a major difference, since the average risk weight in the corporate portfolio in question was 26 per cent (if defaults were not included in the average risk weight). Although it can be discussed how appropriate it is to use an outright point-in-time measurement, it can be seen that the application of the business cycle adjustment is of major importance. An increase of eight percentage points corresponds to an increased capital requirement of 31 per cent. By way of comparison, Handelsbanken s average risk weight for corporate exposures is also 26 per cent. Over the past 16 years, Handelsbanken has had an average loan loss ratio of 0.05 per cent, as compared with the above-mentioned bank s 0.31 per cent during the same period. 5. Using the final PD values to calculate the capital requirement. Of course, this step does not contain any degree of freedom in terms of calculating the PD, since the parameters produced are used directly in the Basel formula. Approval The Swedish Financial Supervisory Authority s process for approval of the use of the IRB models is very extensive. It contains not only the risk classification itself, measurement of the historical data and how the data is used to obtain the risk parameters used in the capital requirement calculation, but also how the bank works with its credit assessments and the internal rating of counterparties, how the bank ensures that the rating is set consistently throughout the organisation and the various internal controls to ensure a well-functioning IRB system. The latter comprises extensive requirements regarding the internal audit s examination of the IRB process and how the annual validation is carried out. We will not go into the details of all the steps in this extensive process here. But it is clear that the resources used are very large, both on the part of the Swedish Financial Supervisory Authority and the banks, comprising thousands of hours. It is far from a process where the bank presents a proposal which the supervisory authority then approves. Instead, it comprises a detailed, continuous examination where the bank must prove that its approach is correct and that the choices it makes ensure as cautious an application of the IRB system as possible. 16

19 Validation Validation is one of the most important components in ensuring that the IRB system gives fair risk weights for calculating the capital requirement. As shown in the above description of how PD values are produced, with respect to the Swedish supervision, banks have extremely limited opportunity to optimise measurement of the IRB-based capital requirements. The options that nevertheless exist, arise above all at an early stage before the methods are approved, for example when the risk classification and the internal rating are designed and when statistical methods are determined for the business cycle adjustment. As is the case for all usage of historical data and statistical measurements, it is easier to find fair models in sample when designing the model on the basis of existing data. It is always vital to test these models out of sample, since it is only then that you actually see that the model works as a reliable prediction of the future. The validation is the process during which the bank continually tests whether the risk parameters used are relevant estimates of the actual risk. This process allows the supervisory authority to properly check that the IRB system actually provides a true and fair view of the bank s capital requirement. The validation is also a comprehensive process containing many steps and statistical tests which cannot be described in detail here. Based on the example with the PD values above, one can nevertheless show that it is basically a simple process which measures the proportion of counterparties which have defaulted in the past year, for each group of exposures and for each risk class. Then it is checked that the default rates measured correspond to the anticipated default rates as expressed in the PD values. In the table below, the actual default rates are shown as measured for each risk class for all counterparties in the Corporate exposures category at Handelsbanken. The table shows the outcome of the latest two validation years 2012 and 2013, which were relatively weak in business cycle terms in most of Handelsbanken s home markets. The outcome for 2009 is also shown, this being the worst year for loan losses at Handelsbanken since the bank crisis in the early 1990s was a weak year in terms of the economic climate in all of Handelsbanken s home markets. Sweden, for instance, experienced a five per cent fall in GDP. 17

20 PD and default for counterparties in the Corporate exposure category (foundation and advanced approach), number-weighted average Risk No of counterparties class 2013 Average PD 2013 Expected no of defaults No of defaults 2013 No of defaults 2012 No of defaults By multiplying the number of counterparties in each risk class by the PD value used for each risk class, it is possible to see, based on the IRB system, how many counterparties can be expected to default during one year in each risk class. This is shown in the fourth column of the table. We then show how many counterparties actually defaulted in 2013 and 2012, and in the very negative year The number of counterparties per risk class and the PD levels per risk class have not changed dramatically over the years, so only the latest year 2013 is shown for these variables. The table shows that during 2013, the number of defaulting counterparties was lower than expected in all risk classes except risk class 7. In 2012, the number of defaulting counterparties was larger than expected only in the lower risk classes 6 to 9, and it is in fact only in risk class 7 that the outcome deviates more substantially from the expected value. During the crisis year 2009, all the poorer risk classes and risk class 5 deviated negatively from the expected value. The fact that the outcome is worse in weak years in business cycle terms than indicated by the PD values is not unexpected, since the PD values are supposed to correspond to the expected default frequencies through a full business cycle. What is particularly positive for Handelsbanken is that the outcomes confirm that Handelsbanken s rating system works well in terms of sorting counterparties with a high risk of default. In the high risk classes 1 4, the outcome is better than expected, even for the worst crisis year of And it is here that most counterparties are located and where it is particularly important that the risk classification works. Only the PD values are presented here, since they are of greatest importance for the risk weights. The LGD and CCF factors are validated in a similar way to the PD values, where the outcome for the latest period s defaults is compared with the values used to calculate the risk weights. One difference for the LGD values is that they must correspond to what can be expected in an economic downturn and not the average over a business cycle. This means that it is primarily in a recession situation that the 18

21 actual outcomes are crucial and that the annual follow-up of LGD often does not have the same importance, provided that it does not refer to an actual recession situation. So far, the validation results are only shown to a small extent in banks Pillar 3 reports and the information that exists is presented on a relatively general level. Based on the existing information, however, comparisons can be made between the Nordic banks. The figures below show some of these comparisons. Comparison between expected loss and actual loan losses for Corporate exposures. Proportion of total exposure in the category. 0.40% Expected losses (EL) 0.35% 0.30% 0.25% 0.20% 0.15% 0.10% DNB 2012 SHB 2010 SHB 2008 SHB 2013 SHB 2011 SHB 2012 DNB 2011 DNB 2013 Nordea 2010 Nordea 2008 Nordea 2009 SHB 2009 Nordea 2011 Nordea 2012 Nordea % 0.00% 0.00% 0.10% 0.20% 0.30% 0.40% 0.50% 0.60% 0.70% 0.80% 0.90% Net loss The figure shows how the expected losses (EL) measured on the basis of the values used in the IRB approach compare to the actual losses reported by the banks. EL is calculated by multiplying the probability of default (PD) by the loss in the case of default for each exposure value (EAD). The expected loss for the whole credit portfolio is aggregated by adding the EL numbers for each credit agreement in the portfolio. When the observations in the figure are to the left of the diagonal line, the expected loss for the portfolio based on the parameters in the capital adequacy calculation are higher than the actual loan loss, which indicates that the risk categorisation is conservatively performed. Of course, the actual losses may be expected to deviate from EL in certain years, particularly in years of economic downturn when loan losses are higher than normal. But over a long period, EL should be close to the average loan losses 5. The selection of banks is determined by which banks have chosen to show 5 Theoretically, EL should actually be slightly higher than the average loan losses over time, since the LGD is calculated as an estimate pertaining to a cyclical downturn, while PD s should reflect an estimated average over the business cycle. The LGD in a downturn should be slightly higher than a long term average, meaning that the EL will be higher when calculated for regulatory purposes, compared to if a best estimate of LGD would be used. 19