Chief Risk Officer Forum Principles for Regulatory Admissibility of Internal Models

Transcription

1 10 June 2005 Chief Risk Officer Forum Principles for Regulatory Admissibility of Internal Models Benchmarking Study of Internal Models

2 This presentation is based on research (Internal Models Benchmarking Study) carried out on behalf of the Chief Risk Officer Forum by Damir Filipovic and Daniel Rost of the University of Munich, with additional support from Mercer Oliver Wyman in the presentation of the results and conclusions. The internal models benchmarking study is based on research carried out on behalf of the Chief Risk Officer Forum by Damir Filipovic and Daniel Rost of the University of Munich Copyright 2005 Damir Filipovic and Daniel Rost, University of Munich Chief Risk Officers Forum Contact Details: Giselle Lim, KPMG, phone number , fax number Benchmarking Study of Internal Models

3 Contents 1. Introduction and objectives 2. Risk Modelling in the context of solvency regulation 3. Evaluation of current frameworks - Principles - Methodologies - Application 4. Summary of Chief Risk Officer Forum principles and modelling standards Benchmarking Study of Internal Models 2

4 Section 1 Introduction and Objectives Benchmarking Study of Internal Models 3

5 The Chief Risk Officer Forum The Chief Risk Officer Forum is comprised of risk officers of the major European insurance companies and financial conglomerates, and was formed to work on key relevant risk issues for advanced practitioners It is a technical group focused on developing and promoting industry best practices in risk management. The membership comprises: Aegon NV, Allianz AG, Aviva PLC, AXA Group, Converium, Fortis, Generali, ING Group, Munich Re, Prudential PLC, Swiss Re, Winterthur and Zurich Financial Services. Benchmarking Study of Internal Models 4

6 Background and Rationale for this Study The discussions at the European level around the new prudential regime for insurers defined under the banner of Solvency II have gathered pace in recent months 3 pillar approach is clearly going to be adopted Pillar 1 (Capital requirements), Pillar 2 (Regulatory supervision) and Pillar 3 (Disclosure), but the respective roles of these three pillars still unclear European Commission sent out the first request for advice in July 2004 CEIOPS provided a progress report and plan on how to proceed in October 2004 Draft answer to 2 nd wave call for advice expected at the end of June 2005; Comments from external stakeholders to CEIOPS expected to be accepted until end of September; CEIOPS due to submit report on CEIOPS Working Groups have started work on the next calls for advice and as such require stakeholders input notably the insurance industry s by the end of Jan 2006 Leading industry players strongly in favour of admissibility of internal models Reflect the true risk profile and solvency position Provide a real incentive for improved risk measurement and management but the admissibility of internal models within Solvency II still as yet unclear High-level principles for internal models have been defined by the International Actuarial Association (IAA) and endorsed by the International Association of Insurance Supervisors (IAIS), but regulators still unclear as to what minimum standards should be for determining whether internal models could be used instead of standardised regulatory models for Pillar 1 Capital requirements Disclosure requirements relating to internal models undefined (a) (b) (c) Rationale for this Study There is at present no clear framework for evaluating the use of internal models within a Pillar 1 solvency capital framework To provide a practical framework, building on current practices within the industry, for regulators to assess whether an individual company s internal model meets minimum admissibility standards for Pillar 1 To provide a practical framework, building on current practices within the industry, for determining which aspects of current internal risk models should be considered within Pillar 2, recognising that Pillar 2 encompasses a wide range of risk management practices beyond those related to internal risk models Benchmarking Study of Internal Models 5

7 Objectives of this study Take inventory of the risk measurement frameworks used by Chief Risk Officer Forum member companies Evaluate strengths and weaknesses of various frameworks and compare them with the standard solvency models developed by the Swiss and Dutch insurance regulators Provide a set of common standards based on analysis of the internal risk models Propose a summary of principles for regulatory admissibility of internal models supported by the Chief Risk Officer Forum member companies Develop a glossary of common terminology Covered in Section 2 Covered in Section 3 Covered in separate portrait document Benchmarking Study of Internal Models 6

8 Process followed Key participants Chief Risk Officer Forum steering team Allianz, AXA, Swiss Re Core benchmarking team Damir Filipovic and Daniel Rost, University of Munich Supplementary support for final executive summary / presentation Mercer Oliver Wyman Process followed Detailed questionnaire completed by all participants and 3 regulatory agencies Interviews with risk management groups of all 13 member companies in Chief Risk Officer Forum* Responses from BPV (Switzerland), DNB (Netherlands) and BaFin (Germany)** Observations on the frameworks of the IAA and IAIS, and on the approaches adopted by other national regulators chiefly FSA (UK), OSFI (Canada) and APRA (Australia) where relevant*** Chief Risk Officer Forum Principles are largely aligned with the overall framework laid out by the IAA and endorsed by the IAIS * For a more comprehensive overview of public-domain solvency models already in use, see Solvency Assessment Models Compared, CEA and Mercer Oliver Wyman, March 2005 ** BPV (Bundesamtes für Privatversicherungen) is the insurance regulator in Switzerland, DNB (de Nederlandsche Bank) is the integrated financial services regulator in the Netherlands, and BaFin (Bundesanstalt für Finanzdienstleistungsaufsicht) is the integrated financial services regulator in Germany *** FSA (Financial Services Authority) is the UK integrated financial services regulator, OSFI (Office of the Superintendent of Financial Institutions) is the integrated financial services regulator in Canada, and APRA (Australian Prudential Regulatory Authority) is the integrated financial services regulator in Australia Benchmarking Study of Internal Models 7

9 Section 2 Risk Modelling in the context of solvency regulation Benchmarking Study of Internal Models 8

10 Trade-offs in formulating Chief Risk Officer Forum Principles for admissibility of internal models but to get credit for internal models in Pillar 1 highly standardised admissibility criteria are needed Chief Risk Officer Forum position on admissibility of internal models Increasing flexibility of standards Standardised model A single model (e.g. DNB s standardised approach, or FSA s standard risk capital model) (+) Simple, standardised and low implementation costs (-) Limited incentives for improved risk management (-) Poor representation of many risks (-) Fails to take advantage of industry s advances in risk measurement Minimum Standardised framework Clear set of principles for risk / capital measurement available capital definition, risk measure calibration, valuation, risk coverage and usage (+) Standardised, allowing regulators to ensure consistency and comparability across companies (+) Incentives for improved risk management (+) Coverage of all material risks (+) Takes advantage of industry s advances in risk measurement Standardised philosophy A high-level standard, similar to the IAA modelling standards (+) Allows risk modelling to be tailored to institution, aligning regulatory framework with internal perspective and management of risk (-) Not specific enough to ensure consistency / comparison of results across insurers Too prescriptive to allow internal models not acceptable to industry Most likely to be how a Pillar 1 standard for internal models can be defined in an acceptable way for both regulators and industry Standards not specific enough to form the basis of admissibility criteria of internal models for Pillar 1 Benchmarking Study of Internal Models 9

11 Pillar 1 needs to recognise the different role of different solvency control levels the MCR and SCR and ensure that these are both linked to the risk profile of the insurer Objectives of SCR and MCR SCR is a target level of solvency not a minimum MCR is a strict minimum level of solvency, below which regulatory intervention should occur: The MCR should define a legal intervention point so that there can be an orderly wind-up of liabilities The level of the MCR should ensure that there is prudence in the resources available to meet policyholders claims in the event of a wind-up Key principles The SCR should be set in such a way as to ensure the likelihood of economic loss for policyholders is no higher than a standardised specified level The level at which the MCR is set should not interfere with the operation of the SCR, and should be based on the same framework as the SCR The MCR needs to strike a balance between being linked to the economic value of liabilities and their risk in a transparent manner, and allowing for continuous monitoring and the need for a legally certain trigger for intervention Benchmarking Study of Internal Models 10

12 Chief Risk Officer Forum Principles therefore need to recognise that solvency requirements need to be evaluated separately from the accounting treatment of liabilities Likely Structure of Pillar 1 capital requirements in Solvency II 3. Solvency capital requirement (SCR) Role of Internal Models in Pillar 1 should be in determining the aggregate required assets (and therefore the SCR) Required assets to reach a specific solvency standard 2. Minimum capital requirement (MCR) Liabilities for solvency purposes need to be set on an economic basis 1. Economic liabilities Accounting liabilities Accounting liabilities likely to be set incorporating some margin for prudence Benchmarking Study of Internal Models 11

13 The only role for accounting liabilities in a solvency framework is to allow minimum solvency requirements to be expressed in a manner which allows objective, external monitoring ILLUSTRATION: Possible mechanism for how MCR might operate consistently in relation to SCR Required assets 100m SCR 20m Economic MCR 6m Liabilities for solvency purposes need to be set on an economic basis Economic Liabilities 80m Accounting Liabilities 84m Accounting MCR 2m In this example, for Intra-year continuous solvency testing, Accounting MCR = 2.4% x Accounting Liabilities Further technical work regarding calibration is needed on defining the relationship between SCR, Economic MCR and Accounting MCR Benchmarking Study of Internal Models 12

14 Chief Risk Officer Forum Principle 1 SCR, MCR and Reserves The SCR should be able to be calculated using the output from internal models The SCR should be set in order to ensure a target standard likelihood of economic loss to policyholders SCR should be based on the economic value of liabilities and the insurer s risk profile, and should be independent of the accounting liabilities The level at which the MCR is set should not interfere with the operation of the SCR, and should strike a balance between being linked to the economic value of liabilities and the risk profile of the insurer risk in a transparent manner, and allowing for continuous monitoring and the need for a legally certain trigger for intervention Benchmarking Study of Internal Models 13

15 In developing standards for the admissibility of Internal Models, four sets of criteria are needed Chief Risk Officer Forum Principles currently address three of these criteria Framework definition ( Output Criteria ) Choice and calibration of risk measure Time horizon Definition of available capital / insolvency Modeling Methodologies ( Design Criteria ) Market risk Credit risk Insurance risk Operational risk Risk aggregation Implementation ( Usage Criteria ) Frequency of calculation and assessment of risk model Documentation, Sign-off and Review of methodologies and tools Use for decision-making Integrity of data and systems environment Input Criteria not addressed in Chief Risk Officer Forum Principles Appropriate input criteria will depend on the adopted Framework and Methodologies Need to be take into account the heterogeneity of insurance risks and modeling techniques Requires further technical work before appropriate standards can be defined Draft Benchmarking Study of Internal Models 14

16 Section 3 Evaluation of current frameworks Framework definition Benchmarking Study of Internal Models 15

17 Chief Risk Officer Forum Principles relating to the general framework for internal models cover 3 main areas #3. Choice and calibration of risk measure Covered in this section Internal risk modelbased capital requirement Overall risk profile #1. Time horizon #2. Definition of available capital Risk aggregation Covered in next section Market Risk Credit Risk Insurance Risk Operational Risk Interest rates Equities Real estate FX Default risk Migration risk Spread risk* Reinsurer credit risk Mortality risk Non-Life Premium risk Non-Life Reserve risk Compliance Administration Systems * This is also treated by many insurers as a market risk Benchmarking Study of Internal Models 16

18 Chief Risk Officer Forum Principles relating to the general framework for internal models cover 3 main areas Time horizon One year versus Multi-year Risk modelling time horizon and valuation time horizon Definition of available capital / insolvency Definition of insolvency economic, statutory or other Inclusion of future new business Choice and calibration of risk measure VaR versus TailVaR / expected shortfall Calibration approaches Benchmarking Study of Internal Models 17

19 #1 Time horizon there are good theoretical and practical rationales for both multi-year and one-year time horizons Description Pros Simple and transparent One year risk horizon Risks modelled over one year, then economic value calculated by projecting all subsequent cash flows over the remaining runoff and discounting risk measure typically VaR-style loss in economic value Consistent and comparable with most regulatory risk-based standard models Multi-year risk horizon Risks modelled over multiple years, with risk measured over the whole multi-year time horizon (e.g. probability of default over 20 years, expected economic loss to policyholders over entire run-off) Provides deeper understanding of dynamic, path-dependent risk exposures Allows for analysis of inter-temporal aspects, including business cycles, regime changes and management actions Cons Fails to accurately reflect time-dependent or path-dependent risks Can be hard to model risks which in practice take several years to emerge (e.g. modelling longevity changes requires one to estimate the adverse change in expectations that might occur over 1 year) High sensitivity to assumptions and prone to error propagation Larger number of parameters to be estimated and risk factors to be modelled Computationally complex when checking for economic solvency every year during the simulation Usage by insurers Regulatory perspective 9 participants 4 participants (1 with a hybrid view) DNB, BPV, FSA, APRA OSFI (Canada) Proposed approach for defining calibration of internal models Multi-year models allowed where these are consistent with the 1- year risk horizon calibration Benchmarking Study of Internal Models 18

20 #2a A Strong view that Economic, not accounting or cash flow, is the appropriate way to define insolvency and available capital Economic Statutory / Accounting Cash flow Definition Economic value of assets < Economic value of liabilities Statutory surplus or accounting net asset value < 0 Assets are not available to support cash outflow Comments Measures true ability of insurers to finance obligations at current point in time Arbitrary leads to capital requirements changing purely due to changes in statutory rules or accounting treatment Confuses liquidity with solvency, unless modelled over the entire run-off of the business Usage by insurers 10 participants 3 participants (largely due to current regulatory constraints and all support moving towards an economic view of solvency) Regulatory perspective DNB, BPV, FSA, APRA, OSFI Proposed approach Benchmarking Study of Internal Models 19

21 #2a (Ctd.) Economic Valuation of Assets and Liabilities Where economic value is used, it is defined by all insurers as being The present value of future cash flows, valued in such a way as to be consistent with current market prices where these are available, with several implications: All assets should be valued at market value, where market prices are available All liabilities that depend on market returns should be valued based on the arbitrage-free principles of derivative pricing theory All fixed cash flows should be valued using the current term structure of interest rates For risks which are hedgeable no market value margin should be applied For unhedgeable risks that cannot be fully diversified (such as certain large-loss insurance risks, or major parameter risks), a market value margin should be applied to best-estimate cash flows in order to ensure that their discounted value is consistent with the price at which the liabilities could be transferred to a willing, rational, diversified counterparty market value margins are currently calculated by insurers in three ways: By using external pricing benchmarks where such benchmarks exist (e.g. quotes from reinsurers) By projecting the capital required to be held by a well-diversified insurer to support such risks, and discounting back the cost of holding such capital By adopting a confidence interval in excess of the 50 th percentile for such risks in projecting cash flows (at an appropriate level of aggregation, taking into account natural netting / offsets that may occur within the portfolio) Benchmarking Study of Internal Models 20

22 #2a (Ctd.) Choice of risk free rate for discounting Government bonds Swap rates Corporate bond yield consistent with credit rating of well-capitalised counterparty Comments Can be illiquid and prone to discontinuities, especially in markets with limited government bond issuance and excess demand (e.g. from pension funds) Highly liquid in most currencies Incorporates a (very) small additional premium to account for credit risk in Interbank market and credit risk on swap, but still lower than strong AAA corporate credit Relatively liquid in most currencies Incorporates sizeable spread to account for the value (to shareholders) of the option to default on liabilities Additional value created by this valuation approach only has value to protect policyholders if liabilities are actively traded, which is not generally the case with insurance liabilities Usage by insurers 5 participants 5 participants Proposed compromise approach Benchmarking Study of Internal Models 21

23 #2b Inclusion of future new business Definition Comments Usage by insurers Regulatory perspective Inclusion of no more than one year of new business Balance sheet cut-off at the measurement date All future cash flows from existing business projected and included in the economic valuation Where relevant (e.g. short-tailed P&C risks that renew over the course of a year), risks arising from one year s new business accounted for in internal risk model Many insurers typically have a hybrid approach with cash flows from existing business applied to life businesses, and one-year of risk but no renewals applied to non-life companies Simple and requires no explicit assumptions about the economics / risk profile of new business 11 participants FSA (Standard model), DNB (Standard model), APRA, BPV, OSFI Multiple years new business Expected new business projected for more than one year (typically up to 4 years), and associated risks modelled A realistic approach, mostly for P&C business when considering going concern as renewals are material Requires assumptions to be made about the economics / risk profile of new business which may be hard to compare across insurers 2 participants DNB (Continuity test) Proposed approach Should be a Pillar 2 requirement Benchmarking Study of Internal Models 22

24 #3 Choice and calibration of risk measure VaR more widely used, but both VaR and TailVaR approaches should be allowed VaR / Confidence Interval TailVaR / CTE Pros Cons Widely accepted, especially in banking industry Straightforward to calibrate to solvency standard defined by historical data on frequency of default Can lead to inconsistent results when aggregating across skewed loss distributions (not a coherent risk measure) Does not take account of the severity of insolvency Sensitivity analysis necessary in order to identify possible stability issues for certain distributions Consistent in aggregation (a coherent risk measure) Accounts for the severity of insolvency, not just the probability of insolvency Less widely known outside of industry, a newer approach More complex to calibrate to solvency standard defined by historical data Usage by insurers Regulatory perspective 11 participants (9 use VaR, 2 use multi-year probability of default) 2 participants (also calculate VaR / probability of default as well) DNB (Standard Model), FSA, APRA * BPV, OSFI * Proposed primary approach * Source: Solvency Assessment Models Compared, CEA and Mercer Oliver Wyman, March 2005 Alternative approach, provided calibration is consistent with the VaR approach Benchmarking Study of Internal Models 23

25 #3 (Ctd.) Regulatory consensus appears to be converging on a target 99.5% annual probability of solvency and the CROs support this level for the SCR # participants 4 3 Calibration of risk measures to confidence intervals All participants that use confidence intervals calibrate to at least 99.5% annual solvency standard Industry practices 11 participants that use VaR calibrate to confidence intervals that range between 99.6% and 99.99% Majority of regulators have so far adopted 99.5% confidence interval 99.60% 99.75% 99.80% 99.90% 99.95% 99.97% 99.98% >99.99% Regulatory perspective* BPV calibrates to 99% expected shortfall DNB and FSA calibrate standard models to 99.5% confidence intervals APRA each insurer has to set an entity-specific specific confidence interval, of at least 99.5% OSFI For segregated funds, capital required to be held up to CTE(95%) * Solvency Assessment Models Compared, CEA and Mercer Oliver Wyman, March 2005, Page 20 Proposed approach = Annual confidence interval of 99.5% Further work is needed concerning regulatory action around the target SCR, particularly bearing in mind the need to avoid amplifying cyclical effects Benchmarking Study of Internal Models 24

26 Chief Risk Officer Forum Principle 2 Risk Modelling Framework Internal models need to be based on the adverse movement in the Economic Value of (Assets-Liabilities), calibrated to an annualised 99.5% probability of solvency Modelling approaches based on longer time horizons or alternative risk measures (e.g. TailVaR) are permissible, provided the calibration approach used can be shown to be consistent with an annualised 0.5% probability of economic insolvency One year s new business should be explicitly modelled, based on assumptions that are consistent with business plans, where this has a material impact on the risk profile of the Group Assets which are not likely to be available in the event of insolvency (for example, profits from future new business, the component of deferred tax assets arising from losses carried forward), should not be included as available capital in the internal model Best estimate liability cash flows should be discounted at swap rates, as they are typically the most liquid, complete and reliable such risk-free rates available this is more conservative than using a truly economic discount rate that would include an allowance for the credit spread of the insurer itself (or of the counterparty to whom the liabilities would be transferred in the event of insolvency) Internal model features to be covered by Pillar 2 Insurers need to have a stated risk tolerance, which should be at least as conservative as a 99.5% probability of economic insolvency, and which is used for internal capital allocation and risk management Insurers should also model risk and solvency levels over multiple years taking into account the effects of new business, over at least their business planning horizon, in the form of a risk-based continuity test Further work is needed concerning regulatory action around the target SCR, particularly bearing in mind the need to avoid amplifying cyclical effects Benchmarking Study of Internal Models 25

27 Section 3 Evaluation of current frameworks Modelling Methodologies Benchmarking Study of Internal Models 26

28 Chief Risk Officer Forum principles issues relating to methodologies for risk modelling cover 5 main areas Covered in previous section Internal risk modelbased capital requirement Overall risk profile #5. Risk aggregation Covered in this section #1. Market Risk #2. Credit Risk #3. Insurance Risk #4. Operational Risk Interest rates Equities Real estate FX Default risk Migration risk Spread risk Reinsurer credit risk Mortality risk Non-Life Premium risk Non-Life Reserve risk Benchmarking Study of Internal Models 27

29 #1 Market Risks All insurers model and measure market risk, typically including risks from all financial instruments / indices: Interest rates (the entire yield curve) where cash flow matching is carried out insurers typically model the entire yield curve (10 participants) e.g. through key rate interest rates Equities Real estate indices FX rates etc. However approaches for modelling market risks (scenarios, analytical approaches, simulation) vary across companies and lines of business For markets / businesses where optionality is significant, simulation approaches typically used (occasionally scenario or analytical approaches), otherwise analytical approaches more prevalent Liquidity risk Liquidity risk only measured quantitatively within the risk model by 3 participants, others use qualitative approaches within a broader liquidity management framework Dependencies between market risks modelled explicitly Where simulation modelling is widely used, this is increasingly through consistent, global Economic Scenario Generators (ESGs) But variance / covariance approaches are still used for some businesses where ESGs not available FX risk Most institutions distinguish between FX mismatch risk, where there are differences in the currencies of assets and liabilities / supporting capital, and FX translation risk, which arises in Groups where the currency of both assets and liabilities / supporting capital in a local entity is different to the base reporting currency of the Group FX mismatch risk is modelled for risk-based capital purposes for Groups, this modelling needs to be at a Group level if there is excess capital held in one currency in one part of a group that is effectively supporting risks in another currency taken elsewhere within the same Group FX translation risk is typically not modelled for risk-based capital purposes, as the solvency of an insurer is independent of the currency in which it reports its financial results (this is a pure shareholder risk, not a solvency risk) Benchmarking Study of Internal Models 28

30 #1 (Ctd.) Where liability cash flows are dependent on market risk factors, more advanced methodologies are used Industry practices Embedded options and guarantees 9 participants assess these risks explicitly, typically through simulation modelling / scenario modelling (similar to models used by banks), and the other participants developing models towards capturing these risks Participating business modelled explicitly, with profit sharing rules either linked to internal management rules, or to external indices Management behaviour 8 participants explicitly model management actions, for example by linking bonus rates and asset mix to level of solvency within the simulation 5 participants do not currently have quantitative rules for management actions, but 3 of them are investigating this Policyholder behaviour 7 participants explicitly link policyholder behaviour (e.g. lapses) to market movements (e.g. interest rates) 5 participants use static best estimate assumptions for policyholder behaviour Regulatory perspective* supports the use of more advanced methodologies where needed FSA parallel yield curve shifts in standard approach, but embedded options, management and policyholder behaviour expected to be dynamically modelled in simulation-based valuation of liabilities for participating life products BPV use of 23 market risk factors, including granular term structure, and dynamic modelling of lapses NAIC impact of specific interest scenarios (e.g. New York 7 ) required to be tested in certain states, with additional scenarios required if interest rate risk is significant (>40% of RBC) * Source: Solvency Assessment Models Compared, CEA and Mercer Oliver Wyman, March 2005, Page 31 Benchmarking Study of Internal Models 29

31 Chief Risk Officer Forum Principle 3.1 Market Risk All sources of market risk need to be modelled probabilistically with inter-factor dependencies explicitly modelled Choice of modelling approach (simulation-based or analytical) and granularity of modelling needs to be proportionate to the risks / businesses being modelled. For example: Interest rates Cash flow matching taken account of by modelling of the whole yield curve FX mismatch risk Currency mismatches between assets and liabilities / supporting capital explicitly modelled Equity risk Equity risk modelled based on analysis of the relevant market index where concentration in individual sectors / individual stocks differs from the index, such concentrations should be explicitly modelled Real estate risk Real Estate risk modelled based on analysis of the relevant property market index, or reasonable proxies if such an index is unavailable where concentration in individual sectors / individual stocks differs from the index, such concentrations should be explicitly modelled Derivatives / market risk mitigation Explicit modelling through simulation / scenarios, with counterparty credit risk also being measured Embedded options and guarantees explicitly modelled through simulation modelling: Management actions (e.g. bonus rates on participating business, dynamic asset allocation policies), where material, should be explicitly and realistically modelled, with modelled management actions codified as policy and disclosed to the supervisor Policyholder behaviour, where material, should be explicitly and dynamically modelled, with key assumptions (which could be either expert-opinion-based or empirically-based) being disclosed to the supervisor Parameterisation of volatility and dependencies between market risk factors should be derived from an appropriate and reliably time series of market data, and should be estimated accounting for tail dependencies (e.g. under stressed conditions) Benchmarking Study of Internal Models 30

32 Internal model features to be covered by Pillar 2 Market risks not included in Pillar 1 need to be actively measured and controlled Liquidity risk is not part of Pillar 1, but should be measured and controlled as part of Pillar 2 supervisory review FX risks arising from translation to a base currency are not part of Pillar 1 since it is not a solvency risk, and should be measured and managed as part of Pillar 2 insofar as they represent a risk to shareholders (including dividend payments for example) Management and control of market / ALM risk needs to be consistent with the assumptions and philosophy behind the internal risk model The cost of guarantees and options needs to be evaluated using internal risk models during product development and pricing, and subsequently over the lifetime of the contracts Market / ALM risk reports need to be produced using the internal risk model on a regular basis, of at least a frequency that enables risk mitigation action to be taken Actual ex-post management actions need to be consistent with the codified management actions that are assumed to take place in the internal risk model Regular stress-testing and back-testing of internal models and their calibration should be carried out Benchmarking Study of Internal Models 31

33 #2 Credit Risks All insurers model and measure credit risk, typically distinguishing between two sources: Investment credit risk Counterparty credit risk (primarily reinsurer and derivative counterparties, but other debtors also considered if they are lead to material exposure) In addition, credit insurers take on credit risk, but typically measure it using methodologies that reflect the specific exposure characteristics and risk mitigation options inherent in the business Investment credit risk modelled quantitatively, with some convergence on the modelling approaches in general, insurers practices mirror those of banks, and explicitly take a portfolio view of credit risk* Three manifestations of credit risk considered Default risk, migration risk and spread risk** Simulation-based industry standard models often in use, especially where it is important to capture concentration effects e.g. KMV, CreditMetrics Simulation-based integrated Economic Scenario Generator models also increasingly in use Reinsurance credit risk modelled quantitatively where it is a significant risk (10 participants) Where modelled, (typically insurers with major non-life businesses), in-house stochastic modelling is typically used to account for dependencies between insurance risks, market risks and reinsurance credit risk A few exceptions for companies where this risk is not material (typically life insurers) * Note that this is more sophisticated than the approach adopted by Basel II, which uses internal models only for determining ratings, but which does not link ratings to capital requirements taking into account the actual risk portfolio of the bank ** Note that spread risk is classified and measured by many insurers as a market risk Benchmarking Study of Internal Models 32

34 Chief Risk Officer Forum Principle 3.2 Credit Risk All sources of credit risk need to be modelled, or demonstrated to be insignificant Investments Reinsurance / derivative counterparty failure Credit insurance Trade creditors, debtors All different manifestations of credit risk should be modelled Default risk Migration risk Spread risk* Credit insurance should be modelled using methodologies that reflect the specific exposure characteristics and risk mitigation options inherent in the business If credit exposures can be accurately represented by external credit indices (e.g. Euro A corporate bond index) and credit concentrations are not material relative to the relevant index, then default risk, migration risk and spread risk can be modelled on integrated basis through direct modelling of the index (e.g. through an Economic Scenario Generator) If representative credit indices are not available, or credit concentrations are material, then default and migration risk need to be modelled explicitly in a manner aligned with the principles of Basel II** Individual credit exposures rated, with ratings calibrated Each exposure assigned an expected probability of default, an exposure and an expected Loss Severity Credit risk concentrations captured through the use of portfolio models such as Moody s KMV or CreditMetrics Spread risk captured separately based on historical volatility in spreads of similarly-rated instruments (e.g. bonds) Note that spread risk is classified and measured by many insurers as a market risk in Basel II, those credit instruments that are held at mark-to-market (primarily bonds) capitalise for spread risk under the Market Risk capital component. Banks do not capitalise for spread risk in their lending portfolio, which is held at book value ** By taking both a mark-to-market value approach for all forms of credit risk and an explicitly portfolio-based approach, we are proposing an approach that is more sophisticated and closer to economic reality than that of Basel II Benchmarking Study of Internal Models 33

35 #3 Insurance Risks Insurance risks are largely heterogeneous, varying in their time profile (from short-tailed to long-tailed), in their proximate causes (meteorological, socio-economic, geological, medical, etc.), and in the extent to which they are amenable to traditional actuarial approaches As a result, liability modelling approaches are highly tailored depending on the type of business and types of insurance risks stress scenarios, and probabilistic approaches (both analytical and simulation-based) are all used Life insurance risk (which is not the major risk for most life insurers) exhibits significant variety in modelling approach Simulation-based modelling in use by half the participants (6 out of 12) The remainder use a mixture of stress scenario and analytical Var/Covar approaches Non-Life insurance Premium and reserving risk often separated out for modelling purposes, although some insurers model them in an integrated manner Premium risk Catastrophe risk modelled probabilistically, based on dedicated scientific models Non-catastrophe risk modelled by a mixture of simulation, analytical and scenario-based approaches depending on the loss distribution of the line of business (more asymmetric loss distributions tend to be modelled through simulation) volatility parameters driven by historical loss triangles Pricing cycle typically not explicitly included, pricing in current year assumed to be adequate, and risk factor volatility assumptions often incorporate historical volatility due to pricing cycle Reserve risk Modelled by a mixture of simulation, analytical and scenario-based approaches depending on the size / materiality of the risk Volatility parameters driven by historical loss triangles, overlaid with expert opinion Reinsurance 11 participants take quantitative account of reinsurance, with large programs being modelled explicitly, using simulation approaches for significant non-proportional programs Benchmarking Study of Internal Models 34

36 Chief Risk Officer Forum Principle 3.3 Insurance Risk For Life / Health insurance, mortality, morbidity and persistency risk should all be measured, ensuring that parameter, process and calamity risks are all covered by the modelling For Non-Life insurance, the risk associated with current year underwriting (premium risk) and prior years underwriting (reserve risk) should both be measured (either in an integrated model, or separately), again ensuring that parameter, process and calamity / catastrophe risks are all covered by the modelling For both Life / Health and Non-Life insurance, process, catastrophe / calamity and parameter risk should be measured using either scenario or probabilistic approaches Process (or volatility) risk, the risk associated with the anticipated year-to-year volatility in insurance result, should be measured probabilistically, supported by scenario analysis where appropriate Separate estimation of calamity / catastrophe risk should be carried out using scenarios / probability distributions based on scientific analysis and expert opinion Parameter risk if significant, level and trend risk should be measured separately based on a combination of scientific analysis, expert opinion and analysis of historical experience Reinsurance / risk transfer Proportional reinsurance can be modelled consistently with the approach used for modelling the gross losses For additional credit to be given for non-proportional reinsurance, scenario or probabilistic approaches must be used Capital must be held to cover the risk of counterparty failure, taking into account possible dependencies between the size of gross losses occurring and counterparty failure Benchmarking Study of Internal Models 35

37 #4 Approach for modelling risk Operational Risks Quantitative measurement of operational risk still in its infancy: 7 participants apply flat percentage add-ons (e.g % of capital) for operational risk 2 participants are developing stochastic operational risk models 3 participants use other approaches (e.g. comparatives0 to estimate the aggregate cost of op risk For those using quantitative approaches, the same general methodology is used External / internal operational event databases (e.g. F1RST, ORX) used to calibrate quantitative loss distributions for each risk category Scenario analysis discussions with business unit management, and simple scorecards / templates used to subjectively assess and rank operational risks and controls, and to adjust the outputs of quantitative approaches but details differ considerably from company to company Taxonomy for defining operational risk categories Choice of external operational event databases Choice of loss distribution for different operational risk types Relative weight of internal scenarios versus external / internal event databases Regulatory perspective* Most insurance regulators do not currently explicitly require capitalisation for operational risk, and implicitly assume that operational risk is covered by the capital requirements imposed for other risks A few insurance regulators have proposed factor-based capital requirements e.g. FSA Standard Model 1% of reserves for non-participating life products NAIC RBC 3% of premiums (pre-tax, so ~2% of premiums after tax) for certain products * Source: Solvency Assessment Models Compared, CEA and Mercer Oliver Wyman, March 2005, Page 31 Benchmarking Study of Internal Models 36

38 Chief Risk Officer Forum Principle 3.4 Operational Risk Operational risk needs to be explicitly accounted for under Pillar 1, in a manner aligned with the principles of the Basel II approach A standardised charge for operational risk is acceptable, similar to the Basic Indicator Approach and Standardised Approach of Basel II however, further research needs to be done to establish the level of the charge and the metric that it should apply to (e.g. net income, assets, SCR excluding Operational Risk, premiums, economic value of liabilities, etc.), in order to ensure that it is consistent with a annualised 0.5% probability of insolvency Advanced Measurement Approaches (AMA) based on insurers own internal models should be allowable as substitutes for standardised charges, subject to criteria consistent with those adopted by Basel II for determining the acceptability of AMA for determining operational risk capital requirements for banks Internal model features to be covered by Pillar 2 Operational risks need to be identified, classified, quantified, reported and controlled, using both qualitative and quantitative approaches, in a manner consistent with the principles of the Basel II Pillar 2 requirements Significant further work is needed in order to establish how Operational Risk should fit within the Pillar 1 framework Benchmarking Study of Internal Models 37

39 #5 Four issues relevant in assessing approaches for risk aggregation Aggregation methodology Industry practices a variety of tools used depending on the level of granularity and the risk Dependency between financial risks typically measured explicitly using either integrated stochastic economic scenario generators, or through using explicit correlation matrices Dependency between specific catastrophe events typically measured through stochastic modelling taking account of geographical reach of such events, with convolution between independent events Across different risks a mixture of approaches 6 participants use simulation to generate an aggregate distribution at group level, 6 participants use Var/Covar approaches to aggregate standalone risk measures Regulatory perspective* BPV Var/Covar approach used to aggregate standalone risk positions, supplemented with tail scenarios DNB Standard model implicitly uses Var/Covar approach with standard correlation assumptions FSA inter-risk correlation implicitly accounted for by choice of scenarios Setting of risk dependency assumptions mixture of stressed correlations and copulas in use, with tail dependencies set by a combination of expert opinion and empirical analysis Correlations in partial use by all participants Copulas in partial use by 6 participants Allocation / attribution methodology 8 participants explicitly allocate out diversification benefits either on marginal or proportional basis 5 participants do not explicitly allocate diversification benefits to individual units, but implicitly incorporate it into target setting and performance measurement Treatment of Intra-Group issues addressed through the separate report from the Chief Risk Officer Forum A Framework for Incorporating Diversification in the Solvency Assessment of Insurers ** * Source: Solvency Assessment Models Compared, CEA and Mercer Oliver Wyman, March 2005, Page 48 ** To be published by Chief Risk Officer Forum, forthcoming Benchmarking Study of Internal Models 38

40 Chief Risk Officer Forum Principle 3.5 Risk Aggregation Pillar 1 admissibility requirements Risk dependencies and concentrations should be explicitly measured, with dependency parameterisation based on empirical evidence, scientific modelling or expert opinion Dependency parameters should be estimated based on tail dependencies (e.g. through stressed correlations) Probabilistic risk aggregation approaches should be used for highly skewed risks or for non-linear risk dependencies Benchmarking Study of Internal Models 39

41 Section 3 Evaluation of current frameworks Implementation Benchmarking Study of Internal Models 40

42 Four key sets of Pillar 1 requirement regarding the use / implementation status of internal models Frequency of calculation of internal model results, and frequency of re-evaluation of internal model Documentation, sign-off and review of model Use of internal model in management decisions Integrity of data and systems environment Benchmarking Study of Internal Models 41

43 #1 Frequency of calculation and assessment / parameterisation most participants moving to more frequent calculations and reassessment of risk modelling Frequency of risk calculations Frequency with which internal risk model is evaluated, re-parameterised and refined # participants 5 # participants Monthly Quarterly Half yearly Yearly 0 Continuously Quarterly Half yearly Yearly When necessary Proposed frequency for update* calculations Proposed frequency for full calculations Proposed approach * Permissible to base quarterly update calculations on approximation if these are robust Benchmarking Study of Internal Models 42

44 #2 Documentation, Sign-off and Review Documentation Internal Sign-off Independent Review # participants 10 # participants # participants Detailed documentation exists Some documentation exists, but w ith some inadequacies* 0 Formalised sign-off process for internal model Partially formalised signoff process No formal signoff process 0 Had external review Planning to have external review No plans as yet for external review Proposed approach Proposed approach Proposed approach requires a regulatory review of internal models** * Not suitable for distribution due to for example being incomplete, only available for individual models, not unified / integrated, or inconsistent ** Regulatory review can be outsourced to external 3 rd party at regulator s discretion (e.g. if regulator does not have the resources to support such a review), and for multinational insurers needs to be coordinated with other relevant regulators in line with the Helsinki protocol Benchmarking Study of Internal Models 43

45 #3 Primary applications to date are in risk / capital management, performance measurement and pricing links to management compensation still in their infancy, but critical for gaining senior management commitment # participants Capital Allocation A-L Management Risk Steering Pricing Performance Measurement Risk-Taking Activities Regulatory Purposes Underw riting Management Compensation Reserving Others In use In partial use Intended for use Intended for partial use Benchmarking Study of Internal Models 44

46 #4 Integrity of data and systems environment an issue, but one that is best addressed through structured internal reconciliation processes and audit than through enforced harmonisation if IT platforms / systems A variety of IT platforms and software tools currently in use Some off-the-shelf vended systems for specific risk modelling issues Asset modelling ESG (Barrie and Hibbert), KMV, CreditMetrics Life liability modelling Moses, Prophet, Atlas, VIP, Alpha Non-Life insurance risk modelling Igloo, RMS / Eqecat / AIR, Remetrica Some use of standard tools, especially for general purpose modelling and risk aggregation Technical software MatLab, Mathematica PC tools Excel, Access Some use of proprietary systems (for example, some liability cash flow projection systems) Harmonisation and integration of IT platforms not a major priority for most insurers Horses for courses no one tool / platform is likely to have the functionality or flexibility to be able to handle all aspects of an internal model Rapid evolution of methodologies means that flexible software environments are needed Clear and precise definitions of inputs / outputs reduces the need for harmonisation of platforms Data and process management primarily manual Data transfer between systems primarily through manual feeds (copy-paste) Data validation and reconciliation processes in place for most insurers Group Finance / Risk functions typically coordinate process in order to ensure integrity of assumptions / approaches and data Benchmarking Study of Internal Models 45