FUTURE FLOODS: An exploration of a cross-disciplinary approach to flood risk forecasting. Brad Weir: Catastrophe Models

Transcription

1 FUTURE FLOODS: An exploration of a cross-disciplinary approach to flood risk forecasting Brad Weir: Catastrophe Models

2 Agenda Reinsurance & Catastrophes Catastrophe Modelling Models in Asia Limitations and Challenges Consideration

3 What is reinsurance? Insurance of insurance companies Spreading the risk to reinsurance companies Reinsurer Reinsurer many portfolios Contingent capital an important aspect of an Insurer s capital management strategy Aon Benfield Benefits Reinsurer absorbs large losses and hence protects: insured against fluctuations in cost insurer s shareholders staff of insurer against losing their jobs and social benefits government protection Insurance Brokers (Aon Risk Services) Insurance Company portfolio of risk Insurable Property single risk 3

4 What are Catastrophes? Infrequent large events that have the potential to result in large economic/insured losses and/or loss of life. Commonly the term is associated with Natural Disasters such as Earthquakes, Typhoons and Flood. Can be natural or man-made (accident, terrorism) Catastrophic insured loss characteristics Infrequent Difficult to assess probability of events Difficult to project historical losses forward Repeat of the same/similar event may not produce the same loss How can we understand these better? 4

5 Application of Catastrophe Modelling in insurance and reinsurance As an insurer; what am I exposed to, how bad can my performance be, how much capital do I need to have access to following an event can I answer these questions with my existing loss history? Thailand flood Japan Earthquake and Tsunami New Zealand Earthquake This is the principal role of catastrophe modelling Necessary for insurance / reinsurance CAT risk pricing, portfolio management and the structuring of reinsurance treaties. Insurers and risk managers use cat modeling to assess and the risk in a portfolio of exposures. This might help guide an insurer's underwriting / pricing strategy or help them decide how much reinsurance to purchase It also provides an indication as to what such reinsurance purchasing should cost Insurance rating agencies use cat modelling to assess the financial strength of insurers. Necessary for capital markets and other innovative risk transfer mechanisms. 5

6 RMS 1996 NZ Release (EQ) 2007 NZ Upgrade AIR The History of Catastrophe Modelling: reactive EQECAT 1996 AU Release (EQ/CYC) 1998 NZ Release (CYC) 2002 AU Release 2004 NZ Release (EQ) 2012 AU Upgrade 1994: EQECAT Founded 1999: IF Founded 2003: RMS Terrorism Model Released 2012: AIR Bushfire Release 2010: RMS Sever Convective Storm 1988: RMS Founded 1994: Northbridge EQ 1999: Sydney Hailstorm 1999: European Winter Storm 2010: IF Asian Typhoon Model Released 2007: AIR Californian Wildfire Model Released 1987: First Commercial Cat Model AIR 1999: Chi Chi EQ 2003: RF Hail Model 2006: RMS Pandemic Model Released 5E+10 8E+10 1E+11 2E+11 3E+11 4E+11 7E+11 1E+12 2E+12 3E+12 4E+12 6E+12 1E+13 2E+13 2E+13 4E+13 6E+13 9E+13 1E+14 2E+14 3E+14 5E+14 8E+14 1E+15 2E+15 3E+15 5E+15 7E+15 1E+16 Koomey's Law : Computations per kwh (ie computing power through time) : Hurricane Hugo 1995: Kobe EQ 2001: September 11 th Attacks 2005: KRW 1989: Loma Prieta EQ 2002: SARS Outbreak 1989: Newcastle EQ 2010: Darfield EQ 2011: Lyttelton EQ 2011: Tohoku EQ 2011: Thailand Floods : Hurricane Andrew 2004: Asian Tsunami 2011: Brisbane Floods 6

7 Inside a Catastrophe Model Hazard (Science) Module Generates the event information including the event impact or hazard measure, such as ground motion for earthquakes and wind speed for typhoons. Vulnerability Module Defines the potential damage vulnerability to a particular type of structure caused by a specific event. Financial Module Determines the loss to various financial perspectives, such as gross loss, client loss (deductibles), reinsurance treaty losses, etc. Hazard 灾害 Vulnerability 损毁 Financial 经济价值 EP Curve 7

8 The Hazard Module Stochastic event generation with associated hazard by geographical region Earthquake Ground motion (or shaking) depends on the distance to fault and the soil type (soft soil, bedrock) under a building Windstorm Peak gust wind speed depends on distance to coast and surface roughness (obstacles such as buildings) upwind Floods Water depth depends on terrain elevation and flood zone Driven by scientific understanding but challenged by: Data availability / accuracy Historic experience or time series Man-made impacts Supposed representation of all physically possible scenarios 8

9 Vulnerability - what will the damage be? Given a local hazard intensity - what will the damage be? damage varies by building construction, age, height, local conditions, etc coverage: damage to structure only? contents? business interruption? vulnerability curves constructed using engineering principles and/or claims data Loss Ratio Flood Vulnerability Data Loss Data: Buildings Log. (Loss Data: Buildings) Depth (cm) local experience? often contain a large amount of uncertainty (i.e. flood) Typically the largest source of model differentiation in APAC 9

10 Summary of Catastrophe Requirements Capital Model Return Period / Peril Basis Australia Greater of: Natural Perils Vertical Requirement (NP VR)# Natural Perils Horizontal Requirement (NP HR)# Other Accumulations Vertical Requirement (OA VR) NP VR & OA VR Occurrence NP HR -- Aggregate Bermuda 1:100 TVaR All Perils Aggregate Canada 1: Earthquake Occurrence Japan Greater of: Return of Kanto Equivalent Earthquake* Return of Equivalent Typhoon as Isewan Typhoon* Occurrence Lloyds Realistic disaster scenario of 1 in 200 year all risk estimate within the ICA Aggregate Solvency II 1:200 All Perils Aggregate UK 1:200 years All Risk Estimate within ICA Aggregate AM Best Greater of : 1: Wind 1: Earthquake Occurrence S&P 1: All Perils Aggregate # NP VR: 1 in 200 year return period loss after allowing for all classes of business, non-modelled perils and potential growth in the insurer s portfolio # NP HR: Three 1 in 10 year losses or four 1 in 6 year losses less an allowance for the net premium liability provision which relates to catastrophic losses * In practice usually modelled as 1:200 earthquake and 1:70 typhoon respectively 10

11 Model Coverage in APAC Country RMS EQECAT AIR Impact Forecasting EQ WS EQ WS EQ WS EQ WS FL Australia * China * Guam Hong Kong India * Indonesia * Japan Malaysia /2015 New Zealand Pakistan * Philippines * 2010 Singapore * South Korea * 2010 Taiwan * 2010 Thailand * Vietnam 2015* * *Denotes RDS or Scenario Event Italics is WIP 11

12 Model Coverage in APAC Country RMS EQECAT AIR Impact Forecasting Australia EQ WS EQ WS EQ WS EQ WS FL China 2014* Guam Hong Kong India Indonesia 2015* Japan Malaysia 2011/2015 New Zealand Pakistan Philippines Singapore South Korea Taiwan Thailand 2012 Vietnam 2014* *Denotes RDS or Scenario Event Italics is WIP 12

13 Challenges in Catastrophe Modelling (Asia) Nature of typical insured portfolio In some cases, smaller portfolios of high valued risks higher potential for high valued accumulations Aftermath of Hurricane Andrew 1992 note the flattened residential neighbourhoods Access to and lack of loss experience Low insurance penetration, specialist portfolios, historically non technical environment. Typhoon Haiyan is typical example Access to development data Difficult to access required data (not available or not accessible) 2010 FEMA Historically US centric development with catastrophe modelling Lack of local understanding around risk and perils Recently changing with recognition of local needs Unmodelled perils can give rise to large losses Flood, surge, fire following, tsunami etc. Exasperated by all points above Red areas: industrial estates 13

14 Challenges in loss estimation Tohoku Japan Earthquake example Insurance policy might cover all these elements Nuclear Power Meltdown Fire Source: Digitalglobe Source: Alertnet Earthquake Flooding Liquefaction Landslides Source: earthquakejapan2011 Source: earthquakejapan2011 Tsunami Source: earthquakejapan2011 Source: earthquakejapan

15 Non modelled hazards before Christchurch Awareness of the risk exists scientifically does it within the industry, or within the modeling? surface faulting fire following earthquake (FFE) liquefaction landslide 15

16 Various components of model miss in catastrophe models Model miss - Difference between actual and modelled loss where non modelled loss is a potentially significant contributing component Model miss is the uncertainty in the modelling results not underestimation Examples include: Actual Loss Model Miss Model Miss Model Uncertainty Hazard Vulnerability Modelling Assumptions Uncertainty in Exposures Non Modelled Loss components Coding of Policy Condition Impact of model miss (uncertainty) can be reduced or better quantified through improved understanding, addressing data concerns and reviewing non modelled loss potential 16

17 Accounting for non modelled components Awareness must exist first Consider if the model already accounts for some level of non modelled elements and what are the limitations: Scientific basis of the modelling and its limitations What experience forms the basis of vulnerability formation or model calibration? Events used in calibration Engage the modelling company to understand more Use scenarios to understand and stress test the potential impact of non modelled elements Use experience from other regions to apply relevant loads to outputs Attempt to address individual elements through expert solicitation (in and out of the industry) 17

18 Inside a Catastrophe Model - recap Hazard (Science) Module Generates the event information including the event impact or hazard measure, such as ground motion for earthquakes and wind speed for typhoons. Vulnerability Module Defines the potential damage vulnerability to a particular type of structure caused by a specific event. Financial Module Determines the loss to various financial perspectives, such as gross loss, client loss (deductibles), reinsurance treaty losses, etc. Hazard 灾害 Vulnerability 损毁 Financial 经济价值 EP Curve 18

19 Hazard: some of the challenges we need to overcome Understanding Focussed and timely research Acquisition & Cost Education and leverage local/regional relationships Completeness How can we fill in the gaps (instrumental vs. other) Fit for Purpose Education 19

20 Vulnerability: unique environment and lack of experience Urbanisation Relatively new building stock Non-homogenous risks Low insurance penetration Lower level of regulation and sophistication New risks (insured) from a modelling perspective Satellite images by the National Aeronautics and Space Administration (NASA) 20

21 Utilising local knowledge to our advantage Local research and experience is a necessity in correctly understanding the challenges in Asia in particular if implementing approaches from elsewhere before after Industrial Estate - Thailand Industrial Estate Vietnam drains and defences Source: Philibosian et al,

22 Collaboration for Innovative solutions Academia Potentially looking for application Long research timelines Access to expertise Network of partners Linkage into government Awareness Continues engagement Focussed/collaborative research Applying research Leverage regional relationships Industry Driven by immediate needs Commercial constraints Data accessibility Short horizon Forefront of education Link to commercial partners 22