Michael Eves
Overview and context
Why Are We Talking About This Now? One facet of a long-term reaction to the financial crisis by many stakeholders: Increasing knowledge of models Decreasing confidence in models Modelers Risk Managers Finance/Actuarial Senior Executives Board Members Need to fix models and improve tools Need to understand risks and framework Line Management Regulators Legislators Need more transparency Public General perception that models didn t work
IAIS: Insurance Core Principle 16.1: The solvency regime requires the insurer s enterprise risk management framework to provide for the identification and quantification of risk under a sufficiently wide range of outcomes using techniques which are appropriate to the nature, scale and complexity of the risks the insurer bears and adequate for risk and capital management and for solvency purposes. How is this implemented?
Mandatory scenarios: Europe Solvency II stress testing U.S. NAIC stress testing sub-group in early stages Fed stress tests for subject insurers (SIFIs and Bank Holding Companies) Internal scenarios: ORSA requirement: the ORSA Summary Report should provide a high level summary of the quantitative and/or qualitative assessments of risk exposure in both normal and stressed environments for each material risk category Fed/OCC Guidance for subject insurers State regulatory requirements and expectations Rating agency expectation
Generally recognized as an important element of Enterprise Risk Management and solvency analysis Integrated with other risk management tools: Economic Capital and stochastic modeling Crisis simulation and continuity planning Emerging risk analysis Strategic planning Best practices for insurers are being defined by a number of professional organizations: North American CRO Council European CRO Forum International Actuarial Association SOA/CAS/CIA Board of Directors expectations
Formulation: A convincing narrative is critical to achieving buy-in from stakeholders Include global events affecting the firm and other financial institutions Emphasize the systematic risk linkage between different economic sectors Challenge of Risk Dependencies: Dependencies in many situations are different than expectations under normal situations Stress scenarios are rare and historical data is not relevant in estimating stressed dependencies Interdependencies evolve with an expanding web of relationships among risk factors Resources: Proper experienced resources important for adequate oversight Unbiased models challenge management
As actuaries do we just produce numbers or can we provide a context for managing what we don t know? How do we communicate uncertainty? Can we use more formalized stress testing to manage Black Swans and the unknown? 8
Severe Stress Stress Tests Stress tests/scenarios relevant for risk management and regulation Severity Scenarios Low Stress Complexity Single risk, single time period Multiple risks, interactions, time period
Picture of a Stress Test
With Stress Testing only one thing is different
With Scenario Testing, you imagine an entire world
Reverse Scenarios Historical Scenarios Synthetic Scenarios Company-Specific Scenarios Single-Event Scenarios Multi-Event Scenarios Global Scenarios Top Down Stress (Balance Sheet) vs. Bottom Up (Transaction Level)
Emerging Risk Evaluation Primary Risk Metric Communicating Risk Supervisors and Rating Agencies Risk Capital Determination Validation of Internal Models
Economic Environmental Geopolitical Societal Technological Source: World Economic Forum, Global Risks 2014
Economic Fiscal crises in key economies Failure of a major financial mechanism or institution Liquidity crises Structurally high unemployment/underemployment Oil-price shock to the global economy Failure/shortfall of critical infrastructure Decline of importance of the US dollar as a major currency Technological Breakdown of critical information infrastructure and networks Escalation in large-scale cyber attacks Massive incident of data fraud/theft Source: World Economic Forum, Global Risks 2014
Many people are not conversant with statistics The average loss for stochastic scenarios with likelihood less than 1% is just not something that resonates Stress Tests and especially Scenario Tests are stories People all relate to stories
Might include: Reason for the choice of scenario Relevance of Scenario to the company Back story on the scenario Description of the scenario Description of post event happenings Comparison against actual experience following similar events Story must be believable!
Asset type Historical scenarios Interest rates 1994 bond market sell-off 1997 Asian financial crisis 1998 Russian debt default and LTCM failure Asset type Historical scenarios Equities 2001 9/11 terrorist attacks in the U.S. 2003 bond market sell-off 2007-8 liquidity crisis 2009 Eurozone crisis 1987 October Black Monday 1997 Asian financial crisis 2000 bursting of the IT bubble 2001 9/11 terrorist attacks in the U.S. 2007-8 global financial crisis F/X Commodit ies Credit 1992 European Monetary System) crisis 1997 Asian financial crisis 1998 Russian debt default 2009 Eurozone crisis 1973-4 Oil crisis 1997 Asian financial crisis 1998 Russian debt default and LTCM failure 2001 9/11 terrorist attacks in the U.S. 2007 credit crisis (from housing market slowdown) 2009 Eurozone crisis
Scenario Analysis Example: Oil Crisis Scenario Analysis Ongoing fears of supply restrictions Extreme weather conditions trigger a 40% increase in Crude prices (with a high of 70% around day 40) US releases oil from the strategic reserve Constrained capacity at refineries mitigates the relief somewhat Fed raises rates to respond to potential inflation Towards the end of the quarter, safe-haven buying of short-term Treasury securities reverses some of the impact Credit Spreads widen sharply as the US economy deteriorates and declines by 1% (B by 400 bps, BBB by 113bps) US Dollar gains moderately (4% vs. JPY, 8% vs. EUR, 6% vs. CAD) Dollar strengthening is limited by the possibility that oil producers will seek to price oil in Euros rather than Dollars Rates rise modestly with a flattening bias (40bps in the long end, 100 bps in the short end) Initial higher increases being offset by safe-haven buying in the later stage of the quarter
For communicating results of Reverse Stress Tests 1 Black Swan (BLS) = Worst Historical Scenario So a company could report that their failure point was: 1.5 BLS for Equities; 2.5 BLS for Hurricane 3.0 BLS for Credit; 3.5 BLS for Pandemic
Recently published papers
Two recent papers have been published by the actuarial community and by the North American CRO Council: Stress Testing and Scenario Analysis Published by the International Actuarial Association in July 2013 Expanded on previous model risk management guidance Defines terms and establishes standards for stress testing, sensitivity analysis, and scenario analysis Scenario Analysis Principles and Practices in the Insurance Industry Published by the North American CRO Council in December 2013 Strives to promote sound practices related to stress testing and scenario analysis Lays out various principles, drawing on actual and planned practices of CRO Council members
Risk appetite and limits Links to business strategy Identify and assess risks Stress testing is an essential component of the North American CRO Council s view of the ERM framework Capital Management Risk Culture and Governance Risk measurement Stress and Scenario Testing Monitoring and reporting
Regulatory and rating agency considerations
IAIS US and European Bank Stress Tests European Insurer Stress Tests IMF A.M. Best SRQ Federal Reserve NAIC State of New York
2014 = Propose a simple Basic Capital Requirement for G-SII s 2015 = Propose a High Loss Absorbency Requirement for G-SII s 2016 = Propose an International Capital Standard (likely to be based off stress testing concepts)
Educational Phase-by April 30 Consider Straw Man Proposal-by May 31 Recruit Companies for Field Test-by June 30 Run Field Test-July through December Review Field Test Results-Early 2015 Refined Stress Testing Proposal-Spring, 2015 Updated Field Test-Summer, 2015 NAIC Adoption-Fall, 2015 with effective date of yearend 2016 or 2017
Move towards publicly disclosed stress tests They become bright line May overwhelm all other uses of stress tests At least under that name Over use by regulators (but likely is better than over reliance on capital)
Michael Eves
GROUPS Implications of Capital Being Managed at Group Level vs. Legal Entity Fungibility of Capital Impact of Reputational Risk Across Group REINSURANCE Intra-Group Non-Proportional Diversification vs. Concentration Risk Value for & Impact on Ceding & Assuming Entities
Michael Eves
What s the real story? A Process of applying quantitative estimates to future unknowns that creates a sustainable pooling of risk for all stakeholders Those quantitative estimates are always based on a model. 35
May be the intuition of a Smart person May use historical data as basis to estimate the future May use today s market data to estimate the future 36
Done correctly it is an investigative process using a series of questions. How to reliably estimate or quantify through a model what is being managed? Errors become stepping stones for improved understanding of what was missing. All models are wrong, even accounting ones Models produce a number(s), but it has a cloud of uncertainty around it. How big is that cloud? Learning comes from creating accountability. Weather prediction gets better through daily accountability. 37
Morality data collection 1700 s through 20 th century Next Step was, how to address volatile interest rates? Unbundled UL PRODUCTS in the US Rise of ALM (Asset Liability Matching) Duration, option adjusted & convexity Rise of seeking hedging value from equity risk 38
Earthquake & Other Cat models Managing Delta & Basis risk to a bounded level There is always a Gap - Mind the Gap In 5 years model will be thrown away. They are disposable, but they taught us the next question to ask. Define unknown - Then scope it out 39
NOT the real world, but a simplification used to better understand real world systems. There are always gaps because: 1. We cannot capture all the variables that impact the results. 2. Key assumptions about the future are uncertain when based on past history and/or current market conditions. 4 0
Examples of models: Financial reporting (IFRS, GAAP, Statutory, Tax, MCEV, CFT) Pricing Capital & Inforce management New business profitability reporting (IRR, MCEV) Hedging Basic Fundamental for managing model risk: Mind the Gap Models can be flawed, leading to incorrect management decisions or inaccurate financial reporting. But are the best basis for getting better.
How to regulate something that is always wrong?
1. Documentation: Key assumptions Simplifications Controls (More Later) Limits of the Model 2. Sensitivity of output to changes in parameters Gives an estimate of the uncertainty within the model
3. Calibration of assumptions/parameters Understand that real world means average the past Market consistent means average all of the information of a single point in time Is there a committee that documents and reviews the basis for the assumptions via a controlled process? 4. Validation of output Use of Actual to Expected Risk Management independent tracking of hedging results Projected policyholder behavior to actually observed Build an Independent Model Use of Three Pillar Approach (More later)
Controlled approval for model changes Documentation of model and reasons for change Locked down IT controlled production models Formal governance to approve changes to methods and assumptions Exception escalation process Audit Periodic internal audit External audit (IFRS, GAAP, Statutory, MCEV) Continuous Improvement via Bread Crumbs Always document why changes were made 4 5
1. Users build the models 2. Independent Checkers sign off on models 3. Use of Internal and External Audit to Validate Checkers 4 6
1. Builder uses control process and gap analysis to own its validity 2. Checkers confirm operational and ERM (Enterprise Risk Management) Controls 3. Who owns the model in a Three Pillar Defense? 4 7
What went wrong at JP Morgan? 1. People running model changed key assumptions so they did not break a risk limit. Had authority to both use and change the model. 2. Change in assumptions had no governance/escalation process. 4 8
Companies using these best practices had internal models and hedging programs that worked exactly as expected throughout 2008 and 2009. Want to ensure that models are the backbone for managing risk instead of taking risk. 4 9