CNSF XXIV International Seminar on Insurance and Surety Internal models 20 November 2014 Mehmet Ogut
Internal models Agenda (1) SST overview (2) Current market practice (3) Learnings from validation of internal models 2
SST overview 3
SST overview Aims to protect policyholders In-force since 2011 1973 Compulsory reporting for large insurers Compulsory reporting for all insurers Deemed largely equivalent with SII 2011 Temporary adjustments Solvency I 2006 2008 SST in force 2013 Largely equivalent with Solvency II SST very strong in Pillar I FINMA may increase Pillar II (e.g. ORSA) and Pillar III (reporting) requirements 4
SST overview The risk measure is the 99% Tail Value at Risk (aka Expected Shortfall) Requirement to model each legal entity explicitly Standard model consists of separate modules for market risk, life insurance risk, non-life insurance risk, credit risk, scenarios (insurance and financial scenarios, prescribed and company-specific), aggregation and the Market Value Margin (risk margin) 5
SST overview Internal model requirements are principle based Documentation requirements covering methodology, parameterisation, data quality, limitations and weaknesses Decision Model accepted Conditional accepted Model declined To be used until essential changes occur Condition Legal requirements Fall-back solutions Condition subsequent Condition precedent Standard Model Capital addon / scenario Validation by FINMA is not a one-off exercise 69 companies (just over half of total number of SST participants) in Switzerland have applied for internal models. 6 were accepted, 34 conditionally accepted and 29 declined 6
Current market practice 7
Current market practice All reinsurers are required to use internal models Most P&C insurers use internal models at least for insurance risk, most life insurers use internal models for valuation, market and credit risk Modular approach Market risk Insurance risk (separate models for life and non-life) Sub-modules: reserve risk, premium risk, nat cat risk for non-life for example Scenarios Credit risk Outward reinsurance (including intra-group transactions) Aggregation MVM Expected result (technical plus financial) 8
Current market practice materiality of different risk types Most material risk is insurance risk for P&C insurers and reinsurers, and market risk for life and health insurers Composition of required capital P&C Composition of required capital Reinsurers Composition of required capital - Life Composition of required capital - Health Source: FINMA website, SST 2014 results 9
Current market practice Example: P&C Reserve Risk Definition: Risk relating to business from prior accident years Different lines of business modelled separately Approaches vary in terms of: Analytical (e.g. Mack) or Stochastic (e.g. bootstrap) Modelling ultimate risk and transforming it to one-year risk, or modelling one-year risk directly Modelling claims together or separate modelling of large claims All approaches involve significant assumptions, increasing the importance of validation 10
Current market practice Example: Natural Catastrophes Scenario 1, e.g. WS Europe Portfolio information Hazard model Vulnerability model Event loss table Adjustments, Scenario 2, e.g. EQ California Portfolio information Hazard model Vulnerability model Event loss table Aggregation with Other Risks in a Monte Carlo Scenario 3, e.g. TC North America Simulation, Portfolio information Hazard model Vulnerability model Event loss table Modelling of Outward Reinsurance Conditions Scenario 4, Scenario 5,.. 11
Current market practice Example: Natural Catastrophes (Hazard Model) List of probabilistic events, frequency and intensity footprint of each Historic events are insufficient to capture all possibilities, so probabilistic events are created using scientific models, with sub-models for: Generating the origin of the event (e.g. epicentre for earthquakes, or the tracks for windstorms) Attenuation functions to estimate the intensity (e.g. ground shake, wind speed, water depth) at all locations affected by the event Allowance for soil characteristics, topographic factors, climatology, etc etc Sub-perils, e.g. storm surge, may be modelled as an extension of the hazard model for the primary peril, e.g. wind, or separately Global Tropical Cyclone Tracks, Wikipedia Sample output from the SLOSH storm surge model, US National Hurricane Center 12 Earthquake zoning map for Turkey Image from website of Kandilli Observatory and Earthquake Research Institute
Current market practice Example: Natural Catastrophes (Vulnerability Model) Given the intensity of an event at each location, and characteristics of the insured item(s), what will the insured loss be? Characteristics of the insured item: Location Policy conditions (e.g. deductibles, limits) Coverage (buildings, contents, business interruption, etc) Occupancy types (residential, commercial, industrial, etc) Construction type (brick, timber, etc) etc Vulnerability curves are often based on: Historical data Post-disaster surveys Engineering analyses, building code information PPA (Probability of Property to be Affected) and MDD (Mean Damage Degree) may be modelled separately, or MDR (Mean Damage Ratio, PPA*MDD) modelled directly It is common for secondary uncertainty to be allowed for in the model 13
Current market practice Example: Natural Catastrophes Modelling NatCats is a complex task, parameter uncertainty and model uncertainty can be significant There is no perfect model Significant differences between the results from different vendors FINMA expects that the companies are able to explain and justify their selection between vendor models and own models, and the adjustments to these by considering: Assumptions and limitations of the different models Their own risk profile The effort needed to build up this understanding, if not already there, is a big task 14
Learnings from validation 15
Learnings from validation of internal models 1) Interviews & workshops in addition to reading documentation is often essential Good documentation is essential but it often comes with gaps and ambiguity and can be aspirational 2) Some insurers criticise the speed of the regulatory reviews but in my opinion proper and valuable reviews need time, A lot of terms in requirements are ambiguous, e.g. appropriate, reasonable Quick checklist approach may lead to misleading conclusions IMs contains many parameters, takes time to understand them And changes in economic environment and modelling literature and market practice may lead to new request/questions from the regulator 3) Benchmarking requires care 4) Assigning materiality to findings can be challenging 16
Learnings from validation of internal models 5) Some commonly not well-understood model elements, e.g. One-year risk Risk margin Model risk, parameter risk Intra-group guarantees (consolidated models do not give policyholder view) 6) Two types of validation 1.Checking whether the model works as documented 2.Checking whether the methodology makes sense and fit for the company using it Both are important, both require different types of skill set. FINMA sometimes makes use of consultants. Specialist expertise and creativity required for the second type. 7) Despite challenges in making conclusions on materiality, appropriateness, etc, in most cases the discussions involved in design and validation of internal models improve the company s and the regulators understanding of risk profile and lead to better decisions 17
Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited ( DTTL ), a UK private company limited by guarantee, and its network of member firms, each Deloitte of which refers is a to legally one or separate more of and Deloitte independent Touche Tohmatsu entity. Please Limited see ( DTTL ), www.deloitte.com/ch/about a UK private company for a limited detailed by description guarantee, of and the its legal network structure of member of DTTL firms, and each its member of which firms. is a legally separate and independent entity. Please see www.deloitte.com/ch/about for a detailed description of the legal structure of DTTL and its member firms. Deloitte Consulting AG is a subsidiary of Deloitte LLP, the United Kingdom member firm of DTTL. Deloitte Consulting AG is a subsidiary of Deloitte LLP, the United Kingdom member firm of DTTL. This publication has been written in general terms and therefore cannot be relied on to cover specific situations; application of the principles set out will depend This upon publication the particular has been circumstances written in general involved terms and and we therefore recommend cannot that be you relied obtain on professional to cover specific advice situations; before acting application or refraining of the principles from acting set on out any will of the depend contents upon of this the publication. particular circumstances Deloitte Consulting involved AG and would we be recommend pleased to that advise you readers obtain professional on how to apply advice the before principles acting set or out refraining this publication from acting to on their any of specific the contents circumstances. of this publication. Deloitte Consulting Deloitte AG Consulting accepts AG no would duty of be care pleased or liability to advise for any readers loss occasioned on how to apply to any the person principles acting set or out refraining this publication from action to as their a result specific of any circumstances. material in this Deloitte publication. Consulting AG accepts no duty of care or liability for any loss occasioned to any person acting or refraining from action as a result of any material in this publication. 2014 Deloitte Consulting AG. All rights reserved. 2014 Deloitte Consulting AG. All rights reserved.