How to manage the risk of Economic Capital

1. How to manage the risk of Economic Capital Pricing Model Validation9th - 11th September 2013London, UK This document presents the views of its author that are not necessarily those of its employer.

2. Agenda 2 • The Risks of Risk Models • How to Validate Economic Capital

3. Background . 3 • Luc Léonard • Belfius Bank (April 12 – to date)Head of Model Validation & Quality Control • Dexia SA (April 10 – March 12)Head of Model Validation & Quality Control • Dexia SA (July 02 – March 10)Head of Credit Modelling

4. Did something go wrong? 4

5. Is it the fault of model(ler)s? 5

6. Structured Products Annual Impairment Rate Well … perhaps a little bit … 6

7. So, what? • The issue is not about making better risk models • Not much room for improvement • It is about better understanding their limits / capacitiesand about taking those into account • Mitigate model risk with critical eye, expertise • This requires efforts • From modelers and validation • Be transparent and down-to-earth when communicating about the models • Don’t oversell what you (and your models) can do • From models users (inc. top management) • Risk models are not that complex • Be more realistic in your expectations 7

8. Realistic expectations Here, we need a modelthat will do this and that 8 • Associate Modeling teams as early as possible in the definition of the needs • They must not be seen as internal service provider who deliver models on demand without being associated to the strategy.

9. Transparent, down-to-earth look at risk models Central role for Validation (if not done by modeling) Main responsibility: Enlighten what the model actually does … … then check how it fits with what it is supposed to do If this is not well defined, explain what the model can be used at Keeping in mind the intrinsic limitations of risk models which are … 9

10. What does the model deliver? 10 • Not always so simple to define • Interesting exercise … even for modelers • Potential risk: some terms are misleading • Ex: Probability of Default

11. What does the model deliver? 11 • Rating model • Delivers: Ratings with associated Probabilities of Default • Belfius: we wanted stability  Through the Cycle Models

12. What does the model deliver? 12 • The PD is an estimate of … • the default probability of an issuer over the coming year • the average default probability of an issuer over a cycle • the average default rate of a rating class over a cycle • OK to meet regulatory requirements but … • … it should not be used to make risk/return estimates • The rating includes state support (important impact for banks). Hence … • … it should not be used to assess the risk on equities

13. What are the limits of Risk Models? 13 • A risk model extracts information from historical data • It is merely an explanation of a certain (reference) period • Hence … • It is predictive only to the extent that the future will be comparable to that reference period • It will not anticipate risk factors that did not show their effects during the reference period • Estimating “exceptional” risks remains a step in the dark: the higher (hence, the more unusual) the risk level, the more arbitrary the model • It could even be rather slow to adapt to new circumstances • You should know • What the reference period is • How reliable the explanation is

14. What is the reference period? 14 • What is the reference period? • How much data is available? • Was available data reliable? Did one have to “correct” it? • What were the explanatory variables? • Was the data coherent with the modeled sector? • Was the available data up-to-date? • How reliable is the model? • Performance indicators (R2, Gini coefficient) • Has it been backtested? • Are there restrictions of use? • Are there margins of error?

15. Model Risk Indicators 15 • What is the Model Risk? • Three types of Model Risk • Unavoidable: The future will not be like the past • Something unprecedented occurs • Very difficult to mitigate  live with it • Structural • Lack of good quality data • Numerical limitations … • Difficult to overcome but can be quantified • Avoidable: The model could be better • It does not exploit available data to the full • It is not properly used • It is not properly implemented • That risk can be reduced (sometimes at a cost)

16. Model Risk Indicators 16 • What is the level of Model Risk? • Four indicators with three levels of severity (low, medium, high) • Avoidable risk • Methodological: In line with best practices? Improvements possible? Sufficiently conservative? • Operational: Correctly implemented? Adequately operated? • Structural risk • Reliability: length of reference period, statistical significance of the results • Unavoidable risk • Stress Test:: can one conceive a situation where model problems would cost x mios, xx mios, xxx mios? • Ex: you would have to recalibrate your PD model and the impact would be up to 1 bios Eur in RWA • Focus on the impact in absolute terms

17. How to communicate about models 17 • Example: LGD Model for Countries • Perimeter: central states, central banks, export credit agencies, embassies and the debtors whose obligations are guaranteed by a central state. • 386 issuers EAD: 7G€ RWA: 2,5G £ Avg PD:8bp Avg LGD:40% • Estimate of the LGD in case of default during the coming year. • LGD are grouped by multiples of 5% • Reference Period: 1998 – 2006 8 default cases • 2 more defaults since then • Main explanatory factors: GDP (33%), External Balance (27%), Rating (11%)

18. How to communicate about models 18 • Example: LGD Model for Countries • Validation history: • Methodology: June 2007 Implementation: Jan 2008 • Annual BT since 2008 (last: 2011) Annual use test since 2008 (last 2012) • Model Risk • Methodological: Low Market standard method, conservative calibration (45% on average vs 30% observed) • Operational: Low • Reliability: High Mostly expert model given the lack of data. • Stress Test: Medium Given the lack of data, regulators could enforce a standard value. 60% would cause an increase of RWA of 1,25G£ • Comments: • Expert model whose validity cannot be tested because of a lack of data. • Data is in general of Year-2 • Model aims to be good “on average” but is not very good on a case by case basis

19. By the way: Were Model(er)s responsible for the crisis? 19 • BIS published in 2008 a report1 analyzing the reasons why Rating Agencies failed to estimate the risks of Structured Products correctly. They presented three main conclusions regarding modeling issues: • Agencies had largely underestimated what could be an extreme decrease of the housing prices • Agencies had badly estimated the impact of a decrease of the housing prices on the default rates, mainly because they only had data covering growth periods • Agencies did not anticipate some new risk elements that had never been observed before (like frauds committed by the originators) • In other words: models were built on a certain reference period. They completely missed what was not present in their data. • But: the last word at the Agencies was to an Expert Panel. Alas, … Experts did not help mitigate the models’ limitations. 1Ratings in structured finance: what went wrong and what can be done to address shortcomings?, CGFS Papers n°32, Bank for International Settlements, June 2008

20. Agenda 20 • The Risks of Risk Models • How to Validate Economic Capital

21. What is Economic Capital? 21 • No “official” definition • Independent (i.e. not defined by regulation) estimation by the bank of its capital needs • A way to answer the regulatory Pillar 2 requirements • “Bank management clearly bears primary responsibility for ensuring that the bank has adequate capital to support its risks.” • Covers “all” the risks • IR in the banking book, spread, funding, pension, strategic … • Goes beyond regulation’s known limits • Concentration, single systemic risk …

22. What is Economic Capital? Credit Diversification Operational Market Pension Funding Strategic Sum with diversification Risks computed independently 22 • Most common implementation • X% - 1 year Value at Risk • Other uses • RAROC • “Equal level playing field” between all risks and business lines • Strategic decisions  capital allocation, risk / return comparisons • A communication tool on the solidity of the bank

23. What is (also) Economic Capital? 23 • A still immature concept • No general acceptance (especially after the crisis) • No common standard • Debatable assumptions • A challenge for Validation • Lack of clear definitions and objectives • Almost impossible to backtest • Requires wide banking knowledge • All risks • Accounting • Finance process

24. No general acceptance ... 24 • 2011 KPMG survey: “ICAAP in Europe … moving in opposite directions” • “The approaches used by banks and regulators to determine sufficient capitalisation of large banks differ widely, ranging from score card models with an apparently high weighting of qualitative factors (e.g. France), to fully quantitative economic capital-based models (e.g. Germany and Belgium) and approaches requiring a flat, significantly higher capitalisation for large and for all banks (e.g. Spain and Switzerland). • The (required) coverage of risk types under Pillar 2 apparently places different weights on risks such as credit spread risk (in the banking book), funding spread risk, business risk and pension risk, while other types of risk such as interest rate risk in the banking book and concentration risk always seem to be in focus.

25. Why is it difficult? 25 • Very large and technical scope • Covers “all” the risks of a bank • Mix of issues from risk, finance, regulation, market views • Not all the goals are compatible • What you do depends on the priorities you set • “Extreme risks” cannot be appropriately modeled • Strong assumptions are required … and can thus vary

26. How to Validate then?1- Ideal Version 26 • Economic Capital is a framework • Its purpose must be clearly defined and in particular, an adequate answer must be provided to the fundamental question: Why do you want to hold Capital? • This determines all the other principles • How will you measure your risks? • What risks will you include or not? • How will you define your capital? • Validate first whether the framework is adequately and coherently defined • Check whether each risk models meets the framework’s objectives and guidelines

27. How to Validate then?2- Realistic Version 27 • … don’t expect your EC to solve all the problems ahead of the market or the regulators • EC framework is likely to be imperfect • Main added-value of Validation is here to give a perspective and to make clear what makes sense or not • What are the goals (if any) your approach is suitable for

28. What is your EC supposed to do? Assets Equity Assets Debt 28 • Standard answer Estimate a buffer of capital against “extreme” losses in order to protect your creditors • “Extreme” meaning in general: very unlikely losses over 1Y

29. What is your EC supposed to do? 29 • Such definition does not tell you much about your risk of bankruptcy (likely to occur long before you have lost all your capital) • Interesting question: how much can I loose before I need a recapitalisation?  going concern view asked by German regulator

30. Is it what your EC actually does? 30 • Some topics to consider • Resolution losses • Do I foresee also “resolution” losses (German regulator says yes) • Complicated to estimate • Or do I consider a “going concern” perspective • Not credible • Or do I simply not care? The risk estimate is rather arbitrary anyway (uncertain measures, 1 year period …)

31. Is it what your EC actually does? 31 • Which losses are taken into account? • Do I compute my losses from an economical or from an accounting perspective? • i.e. do I take all the losses into account or only those with an impact on my results / available capital? • Strong impact on the decision to include some risks or not • Example: spread & transition risk on the loans • If the goal is to assess the (remaining) value of my assets: economic value would be normal • If it is to prevent from bankruptcy: more difficult to say • Risks that impact regulatory capital • What is the market perception

32. Is it what your EC actually does? 32 • NB: no clear answer from the regulation • Maturity adjustments in the Basel formula can be seen as an estimate of market value losses due to credit quality degradation • They are imposed to all credits  economic view • But they are too light to account for spread volatility • Basel III foresees adjustments for AFS reserves losses, not for loans

33. Is it what your EC actually does? 33 • How do you define your available capital? • Regulatory capital? • plus or less some adjustments? • Available Financial Resources? • Coherence with the risks • e.g. value of your assets

34. Is it what your EC actually does? 34 • How do you measure your risks? • Most problematic issue actually • How to “model” extreme losses … without enough data? • Best cases: 30 years? of reasonably relevant data • Often, almost no meaningful data (e.g. operational risk) • And you aim at losses that would occur with a probability of 1/1000? 1/2000? 1/3000?

35. How do you measure your risks? 35

36. How do you measure your risks? 36 • In practice: no other solution than to “extrapolate” with strong assumptions • As a consequence “statistics” do not make much sense and should be interpreted cautiously • Risk of being misleading otherwise

37. What makes sense then? 37 • Very useful exercise to go through all your risks and assess their materiality • Good basis for management to decide • Provided the assumptions are presented in a transparent manner • Can be used for strategic decisions • But again in a transparent manner • NB: it is not mandatory to: • Compute just one figure • Use VaR methods. Stress Tests (deterministic scenarios) can be more explicit.

38. Summary: what to validate? 38 • Require a clear definition of the purpose(s) • Check that the framework(s) is (are) consistent with these purposes • Risks retained • Capital definition • Risk measures • Check that the list of risks is complete • Check that all the risk models are in line with the framework • Check that the communication around the EC framework is adequate • Otherwise, do it yourself

39. Summary: what to validate? 39 • How to decide that an “extreme” figure is adequate • Precise stats do not make much sense • Challenge the level with stress tests: it must be at least significantly above “past” events. • Challenge also with imagination • Beware of over-reliance on past data (no catastrophe ever occurred then … no catastrophe will ever occur?). • It is not necessarily up to you to decide. • Provide management with a clear view on the strengths and weaknesses of the approach. • It is mainly an “expert” management decision