Expecting the Unexpected: Insurance and Cat Modeling in a World of Multi-colored Swans, Ducks and Other Animals Gero Michel Montpelier Re, Bermuda
Catastrophe Global Losses losses 2013 (AON YTD Benfield, vs. Prior Baden-Baden Years 2013) USD BN ~45bn as of today AON, BB, 2013 2
Catastrophe Black Swan losses Years 20, 6013 YTD vs. Prior Years 1992 (Hurricane Andrew) Modeling provided a solution 2001 Sept. 11, terror/clash Terror modeling changed policies 2004 Clash of hurricanes Change in models 2005 KRW/flooding Increase in model penetration 2010 RMS v.11 model change Twice the risk 2011 Clash of events/indirect losses/flood In-house modeling Black Swan Years were dominated by Clash, Hurricanes and Flood, Man-Made Perils (including model change) and Indirect Losses were of second order Future Black/Grey Swans years might include: I.USEQ/clash of EQ events, II.Indirect losses/cyber Risk, III.Accelerated change in reinsurance landscape/capital market influx, IV.Social unrest/political risk, V.Solar flares among others. 3
Catastrophe Evolution of losses Models20, 6013 YTD vs. Prior Years From: After 2005: True Exposure + True Models = True Risk Results (Gold in Gold out) To After 2011: True Exposure + Corrupted Models = Uncertain Risk results To Today: True Exposure + Calibrated Models = More Certain Risk Results To Tomorrow: Stochastic Exposure + In-house/Simulation/Ensemble Model = Realization of Uncertain Truth Decision Making Under Uncertainty 4
Number of Insured USD BN Events around the Globe average: 10+/-4 events/yr >= USD 1 BN/yr, (no obvious trend!) US=United States AM=Americas non-us EMEA=Europe, Middle East, Africa APAC=Asia Pacific/Oceania 6/yr 0.4/yr 1.7/yr 2/yr Actual loss data based on Impact forecasting 2014 5
Catastrophe How Much has losses Been 20, Modelled? 6013 YTD vs. Prior Years Numbers are estimates taking into account percentages of deemed not (Vendor) modelled portions of Tornadoes losses (US), hail, earthquake (e.g. NZ), flood, among others. Non Modelled Losses FL=flood TOHA=tornado/hail TC=tropical cyclone WW= winter weather WF=wildfire EQ=earthquake DR=drought Models have been improving but % nonmodelled has not been decreasing! In-house and Broker models tend to fill gaps 6
United States Perils: Actual vs. Modelled Losses Average Annual Losses since 2003 Modelled 22bn 4.5bn 1.8bn 2.6bn 1.3bn 1bn 2.1bn 0.5bn Actual 17.4bn 10bn 2.1bn 0.1bn 1.8bn 0.4bn 2.7bn 0.3bn? black: vendor model, red: needs improvement! blue: in-house FL=flood HA=hail TC=tropical cyclone WW= winter weather WF=wildfire EQ=earthquake DR=drought USAP (USDbn) OCC AGG 1 in 100 195 230 1 in 25 92 120 1 in 10 46 70 7
Long-term Gains and Losses Hard Market Soft Market Soft Market Soft Market Hard Market Hard Market Hard Market Hard Market Hard Market Large Loss Years Large Loss Years Long-Term Gains Long-Term Losses 8
Does the Average Annual Loss Even Matter If Events/Loss Years Cluster? Over-dispersed and clustered HU events: Last 8 years without a landfalling HU?? This did not happen once - in 150yrs! 8 years w/o FL loss ~ -90bn of insured AAL (x2 for economy) 1.3bn 4bn 2.5bn 11bn AALs AAL, average annual loss 9
Can We Forecast Highly Active Years? AMO: Atlantic Multidecadal Oscillation (Variability) vs. ACE A Virtual Risk Institute??? ACE: Accumulated Cyclone Energy, Atlantic HU Regimes? ENSO (MEI)? Atlantic, 2013 NW Pacific Prediction Failure 2013 Atlantic 2013: Forecast: Above average Result: NOAA Hurdat reanalysis data Among lowest severity in last 100 years! 10
Forecasting with a Trend: Climate Change - increasing HU Frequency and Severity? SST A simple linear trend might not work for forecasting HU losses! Tom Knutson 11
Reproducing History has been (relatively) easy - But the Future might be (very) different We are Good at Hint-cast A Virtual Risk Institute Last 11 years: USD 635bn/11yrs (as of today) 58bn+/- 38bn per annum Actual Modelled AAL Modelled Variability 12
More Non-(Partially)Modelled Perils US Terror, several 10s to 100s of bn Non-US Terror, (1/100 property losses of USD several 100mn to several bn per country) Cyber Risk, Cyber Terror, indirect and intangible losses (USD several 10s of bn or more?) Pandemic losses (see graph) And more 13
Non-Modelled Extreme Tail Events Solar flux has been on the rise! The solar storm of 1859, Carrington Event, a solar flare or coronal mass ejection hit earth's magnetosphere and induced the largest known solar storm, recorded by Richard C. Carrington. Assumed economic loss today: USD 1 - >2.5 TN Effects on: 1. Communications (Satellite) 2. Communications (Wireline) 3. Energy (Electric Power) 4. Information Technology 5. Transportation (Aviation) 6. Transportation (Mass Transit) 7. Transportation (Pipeline) 8. Transportation (Rail) Not excluded from most property policies 5 insured loss realizations of a Carrington Event Affected Territories: N-Am, N-EU, S- Africa Our studies suggest that the Carrington event might (in a solar maximum) have a RP of 100 years 14
Extreme Tail Events 15
Conclusion Amount of Non-Modelled Cat. Losses has been ranging around 40% Dominated by Flood and Tornado/hail This number has been diminishing due to in-house models various new third-party data/model providers Most obvious future Grey Swan Years might include USEQ/clash, Solar Flare events, Political Risk losses, among others There is no general cure to the unexpected and modeling (or loading) alone will not make us survive in a competitive market A risk/uncertainty aware culture/strategy = Network/high penetration of risk-conscious underwriting and technical activities throughout your company, awareness of uncertain outcome! 16