Weathering the Storm - Aon Benfield Events Center
Transcription
Weathering the Storm - Aon Benfield Events Center
Weathering the Storm Steve Drews, Impact Forecasting & Paul Eaton, Aon Benfield Analytics 2014 Analytics Insights Conference July 22 - 24 Agenda Section 1 IF’s New Historical Severe Thunderstorm Model: STS RePlay Section 2 Experience Analysis Section 3 Enhanced Tornado Viewing Guide Aon Benfield | Analytics Proprietary & Confidential | July 2014 2 Agenda Tracker Section 1 IF’s New Historical Severe Thunderstorm Model: STS RePlay Section 2 Experience Analysis Section 3 Enhanced Tornado Viewing Guide Aon Benfield | Analytics Proprietary & Confidential | July 2014 3 Why… the current severe Not…are enough emphasis thunderstorm stochastic on or a lack of recent catastrophe models so far off historical hazard data with their AAL values???? Aon Benfield | Analytics Proprietary & Confidential | July 2014 4 Annual Severe Weather Frequency By Different Periods Average Annual SPC Reports: All Tornadoes Average Annual SPC Reports: All STS Perils 35,000 30,000 26,931 28,642 29,638 1,283 1,336 924 500 0 15,000 20,000 Reports / Year Reports / Year 1,000 1,284 Average Annual SPC Reports: All Hail 25,000 15,000 Reports / Year 1,500 11,271 10,000 12,588 13,461 13,858 10,000 5,000 5,206 0 5,000 Average Annual SPC Reports: All Convective Wind 0 1996-2013 Aon Benfield | Analytics Proprietary & Confidential | July 2014 2000-2013 2004-2013 Reports / Year 1950-2013 20,000 13,059 10,000 5,941 0 5 13,898 14,444 Annual Severe Weather Frequency: Tornadoes The vast majority of tornado frequency increase stems from weaker (and not as easily detectable) tornadoes Note: 2013 lowest tornado total since 1989 Aon Benfield | Analytics Proprietary & Confidential | July 2014 6 Annual Severe Weather Frequency: Hail The vast majority of hail frequency increase stems from smaller hail occurrences Aon Benfield | Analytics Proprietary & Confidential | July 2014 7 Annual Severe Weather Frequency: Wind Both weaker AND stronger thunderstorm-produced winds continue to show frequency increases Aon Benfield | Analytics Proprietary & Confidential | July 2014 8 What if… …there was a severe thunderstorm catastrophe model that would more accurately predict my AAL? Well, now there is… Aon Benfield | Analytics Proprietary & Confidential | July 2014 9 New IF ELEMENTS Severe Thunderstorm Tool: STS RePlay Based on recent historical data obtained from the Storm Prediction Center – Contains last 10 years of severe thunderstorm activity – Contains tornado, hail and convective wind occurrences – Recent historic experience contains a richer catalog of activity based on better data capture using Doppler radar and other technological advancements ELEMENTS STS RePlay event set – Can be used for a replay of a large catalog of historical daily outbreaks to better understand loss behavior – A replay of historical event catalog against current exposures allows a recast to current loss expectations – RePlay catalog can be used for obtaining a view of average yearly losses as an alternative to Average Annual Losses produced by stochastic models – Note: the scenario set is not a replacement for stochastic models, but is an alternative recast view using actual historical experience Aon Benfield | Analytics Proprietary & Confidential | July 2014 10 STS RePlay Source Data Source data: Storm Prediction Center – www.spc.noaa.gov Local Storm Reports (LSRs) – Gathered from trained storm spotters, emergency managers, law enforcement, automated weather stations and Doppler Radar – Three types of severe weather data gathered • All tornadoes • Hail 1” or greater • Convectively-induced (thunderstorm-produced) winds of 58 mph or greater – Reports characterize severe weather occurrences • Date and time, location (lat/long), etc. • Intensity, damage, etc. Data used – Total of 260,610 LSRs analyzed and converted into ELEMENTS format after initial filtering (hail below 1”, winds below 58 mph, reports with incomplete data) Aon Benfield | Analytics Proprietary & Confidential | July 2014 11 STS RePlay Input Data Occurrence Counts – Tornadoes: 13,764 (average 1,376/year) – Hail: 125,924 (average 12,592/year) – Wind: 120,922 (average 12,092/year) Aon Benfield | Analytics Proprietary & Confidential | July 2014 12 STS RePlay Event Set Generation Hail and Wind Data usage within SPC LSR database files – Start coordinates – Hail Size – Wind (knots) Directionality calculated as a random number due to the uncertainty of the direction (i.e. no end coordinates) Path Length and Path Width calculated as a random number based on average, lower bound, upper bound, and standard deviation values given within the ELEMENTS Severe Thunderstorm Model stochastic event set for each hail size bin (0-5) and wind intensity bin (0-3) Hail Size calculated using hail size provided by SPC and converted from inches to millimeters – If hail size is less than 1” (25.4 mm), data is not used Wind intensity calculated using wind speed provided by SPC and assigned to Fujita Scale – If wind intensity is less than 50 knots (58 mph) data is not used Aon Benfield | Analytics Proprietary & Confidential | July 2014 13 Special Considerations For Wind and Hail SPC Reports Limited to point estimates – No end coordinate reported in SPC data – No data on swath shape (length, width) – SOLUTION: Implement current inplace Impact Forecasting research on characterization of hail and wind swaths Multiple Hail Reports Multiple reports could represent portions of same event swath – Overlapping swaths over-count losses – SOLUTION: Bin events by report time (5 minute increment) and spatial proximity (0.6 mile increment) and treat similar reports as duplicates of the previous reports Single Hail Swath Aon Benfield | Analytics Proprietary & Confidential | July 2014 14 STS RePlay Event Set Generation Tornadoes Data usage within SPC LSR database files – Start and end coordinates – Intensity (Fujita Scale/Enhanced Fujita Scale) – Path Length (verification only) – Path Width Directionality calculated by utilizing the start and end coordinates given by SPC data Path Length calculated by utilizing the start and end coordinates given by SPC data and converted into kilometers – Output verified by SPC data values Path Width (maximum) taken directly from SPC data and converted into meters (given in yards) Aon Benfield | Analytics Proprietary & Confidential | July 2014 15 Special Considerations For EF3+ Tornado SPC Reports Strong to violent tornadoes (EF3+) often long lived with complex track shape and varying intensity – Example: Joplin, MO 5/22/2011 EF5 tornado SPC LSR data characterizes start and end coordinates and single intensity rating – Example: Joplin EF5 as straight line between SPC start and end coordinates – 490 EF3+ events in 2003-2012 data set – All 490 EF3+ events researched – Majority didn’t have detailed information 28 EF3+ tornado tracks modified – Ad hoc research for event shape and intensity (NWS-sourced official track data) – Re-parameterized events as necessary for proper representation within event set Aon Benfield | Analytics Proprietary & Confidential | July 2014 16 Hazard Uncertainty Two mile tolerance for coordinate accuracy creates uncertainty in LSR event location – Street-level addressing not available – cross-street coordinate assignments most likely – SOLUTION: Sample every LSR 25 times, adding position uncertainty • 5% position uncertainty for tornadoes (more accurate reporting) • 20% position uncertainty for hail and wind • 6.5 million event set generated in ELEMENTS from 260,610 individual LSRs Aon Benfield | Analytics Proprietary & Confidential | July 2014 17 STS RePlay Results Verification Comparison of Severe Convective Storm Models To Actual Client Historical AALs Client Client Size* IF STS RePlay to Actual Stochastic Model A to Actual Stochastic Model B to Actual Stochastic Model C to Actual A Large 0.90 0.51 0.54 0.40 B Large 1.00 0.49 0.58 0.40 C Small 1.78 0.87 2.30 0.95 D Small 1.33 2.06 1.54 1.16 E Large 1.04 0.85 0.84 n/a F Large 0.90 0.65 0.61 n/a 0.93 0.54 0.58 0.41 Weighted Average * Large client actual loss approx. 10x small client actual losses Aon Benfield | Analytics Proprietary & Confidential | July 2014 18 IF ELEMENTS STS RePlay Summary Designed only as an alternative AAL view using rich data due to technological improvements in data capture and an overall active severe convective weather period Contains historical SPC occurrences of tornadoes, hail and convective winds from active 10-year period Each severe weather occurrence is sampled 25 times to account for hazard position uncertainty Reports were analyzed spatially and temporally to eliminate duplicate reports Significant tornadoes were analyzed and modified (where information existed) to better represent the overall track and intensity fluctuations that EF3+ tornadoes often exhibit Over 6.5 million severe weather occurrences are contained within the STS RePlay event set Official Release: September 2014 (in coordination with new IF STS Stochastic Model) Aon Benfield | Analytics Proprietary & Confidential | July 2014 19 Agenda Tracker Section 1 IF’s New Historical Severe Thunderstorm Model: STS RePlay Section 2 Experience Analysis Section 3 Enhanced Tornado Viewing Guide Aon Benfield | Analytics Proprietary & Confidential | July 2014 20 What if I… Instant Have Need Want Don’t Don’t noisy ato ahave realistic quantifiable have data experience credible that assessment makes basis loss that for trend of Want see Derrick Rose fast experience selections my comparing represents average and because stochastic annual on-leveling mythe current I’ve loss model grown from very break through entire underwriting severe output into new thunderstorms? difficult? todefense experience? regions? paradigm? Mavericks again? Aon Benfield | Analytics Proprietary & Confidential | July 2014 21 Severe Weather Experience Templates Superior Trend Diagnostics Daily Incurred Losses Log Scale 100,000 Loss Thousands 10,000 1,000 100 10 1 Jan-98 Trend: 8.2% Jan-00 Jan-02 Jan-04 Jan-06 Jan-08 Jan-10 Jan-12 Date Daily Average Severity Log Scale Trend: 9.1% Daily Claim Counts Log Scale 100,000 Trend: 0.0% 10,000 Claims Severity 10,000 1,000 1,000 100 10 Trend: 0.2% Trend: 8.1% 100 Jan-98 Jan-00 Jan-02 Jan-04 Jan-06 Jan-08 Jan-10 Jan-12 1 Jan-98 Jan-00 Jan-02 Date Aon Benfield | Analytics Proprietary & Confidential | July 2014 Jan-04 Jan-06 Date 22 Jan-08 Jan-10 Jan-12 Severe Weather Experience Templates Analysis Of Daily Loss Data Automated Aggregator of Events Per Occurrence XOL & Aggregate Structure Analysis Cat XOL Co-Part 5% Limit 210,000,000 Retention 20,000,000 Trended IF RePlay Selected Incurred AccidentDate 31,438 07-Jun-92 18,838 08-Jun-92 86,403 09-Jun-92 588,810 10-Jun-92 169,952 11-Jun-92 32,875 12-Jun-92 33,038 13-Jun-92 7,286,350 14-Jun-92 104,998 15-Jun-92 67,508 16-Jun-92 62,157 17-Jun-92 61,605 18-Jun-92 2,402,488 19-Jun-92 23,295 20-Jun-92 22,837 21-Jun-92 54,423 22-Jun-92 40,066 23-Jun-92 52,521 24-Jun-92 524,827 25-Jun-92 37,133 26-Jun-92 29,800 27-Jun-92 15,283 28-Jun-92 60,824 29-Jun-92 DOY 96 Hour Total 158 152,342 159 164,301 160 207,116 161 725,489 162 864,003 163 878,039 164 824,675 165 7,522,215 166 7,457,260 167 7,491,894 168 7,521,013 169 296,268 170 2,593,758 171 2,549,545 172 2,510,226 173 2,503,044 174 140,622 175 169,848 176 671,837 177 654,547 178 644,281 179 607,043 180 143,040 Losses by day tracking Aon Benfield | Analytics Proprietary & Confidential | July 2014 Regular XOL - Occ Ceded Loss 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 Aggregate Per Event Deductible 2,000,000 Deductible Type Standard Per Event Limit 18,000,000 Regular XOL Franchise XOL Ceded Loss 0 0 0 0 0 0 0 5,522,215 7,522,215 5,522,215 5,457,260 7,457,260 5,457,260 5,491,894 7,491,894 5,491,894 5,521,013 7,521,013 5,521,013 0 593,758 2,593,758 593,758 549,545 2,549,545 549,545 510,226 2,510,226 510,226 503,044 2,503,044 503,044 0 0 0 0 0 0 0 Optimization of Program Total Ceded 0 0 0 0 0 0 0 5,522,215 5,457,260 5,491,894 5,521,013 0 593,758 549,545 510,226 503,044 0 0 0 0 0 0 0 Initial step to aggregate into 3 day step cumulative totals before and after an aggregate deductible Max ceded through day n assuming an event ends on Max ceded Events, Gross day n through day n of Reinsurance 3,128,895 3,128,895 0 3,128,895 3,128,895 0 3,128,895 3,128,895 0 3,128,895 3,128,895 0 3,128,895 3,128,895 0 3,128,895 3,128,895 0 3,128,895 3,128,895 0 8,651,110 8,651,110 7,522,215 8,586,156 8,651,110 0 8,620,789 8,651,110 0 8,649,908 8,651,110 0 8,651,110 8,651,110 0 9,244,868 9,244,868 2,593,758 9,200,655 9,244,868 0 9,161,336 9,244,868 0 9,154,154 9,244,868 0 9,244,868 9,244,868 0 9,244,868 9,244,868 0 9,244,868 9,244,868 0 9,244,868 9,244,868 0 9,244,868 9,244,868 0 9,244,868 9,244,868 0 9,244,868 9,244,868 0 Event Loss to Per Occ Layer 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 Event Loss to Agg Layer 0 0 0 0 0 0 0 5,522,215 0 0 0 0 593,758 0 0 0 0 0 0 0 0 0 0 Aggregation of daily losses via local optimization rule 23 Agenda Tracker Section 1 IF’s New Historical Severe Thunderstorm Model: STS RePlay Section 2 Experience Analysis Section 3 Enhanced Tornado Viewing Guide Aon Benfield | Analytics Proprietary & Confidential | July 2014 24 TVG Concentrations Are Tornado-Track Based TVG uses scenario or “What if” analysis Example: what if the Moore, OK tornado from 2013 hit other exposure concentrations… St. Joseph Columbia Kansas City Sedalia St. Louis 2013 Moore, OK EF-5 Track Cape Girardeau Springfield Poplar Bluff Aon Benfield | Analytics Proprietary & Confidential | July 2014 25 Accumulations By Census Statistical Area Kansas City, MO‐KS TIV No. Events 39,466,872 1 17,585,432 119 5,412,883 1,191 3,771,730 2,381 1,475,490 5,954 2,466,163 11,908 St. Joseph, MO‐KS TIV No. Events 37,366,568 1 29,453,144 32 7,266,844 317 3,195,987 633 832,917 1,584 2,836,071 3,168 Cape Girardeau‐Jackson, MO‐IL TIV No. Events 24,297,988 1 20,470,170 25 6,639,354 247 2,609,244 495 512,595 1,237 2,137,388 2,474 Springfield, MO TIV No. Events 23,878,528 1 17,553,980 81 7,988,126 809 4,723,276 1,617 1,415,083 4,043 2,872,752 8,086 Fayetteville‐Springdale‐Rogers, AR‐MO TIV No. Events 23,048,812 1 13,348,868 58 4,483,250 577 2,267,785 1,155 553,076 2,886 1,582,815 5,772 Micropolitan Conditional Percentile Max 99.0% 90.0% 80.0% 50.0% MEAN Total Events Poplar Bluff, MO TIV No. Events 63,045,104 1 50,838,456 20 12,989,744 199 5,914,346 397 1,541,905 994 4,923,481 1,988 Sedalia, MO TIV No. Events 49,566,616 1 41,498,012 18 9,920,330 180 3,409,520 361 1,268,927 902 3,920,615 1,804 Sikeston, MO TIV No. Events 34,808,164 1 30,377,070 12 9,320,646 119 3,991,476 238 1,499,919 596 3,515,049 1,192 Paragould, AR TIV No. Events 34,479,808 1 27,025,388 12 6,414,103 123 3,442,260 245 573,286 613 2,743,152 1,226 Hannibal, MO TIV No. Events 27,629,492 1 21,423,680 18 3,066,074 177 1,360,789 354 219,738 884 1,351,080 1,768 Conditional Percentile Max 99.0% 90.0% 80.0% 50.0% MEAN Total Events Macon, MO TIV No. Events 42,836,672 1 42,046,328 3 23,580,500 35 9,327,144 70 2,887,726 174 6,871,468 348 Anabel, MO TIV No. Events 38,361,144 1 38,361,144 1 12,649,454 11 6,150,253 23 1,319,952 56 4,355,448 112 Bertrand, MO TIV No. Events 33,734,388 1 33,734,388 1 26,275,186 13 17,446,704 26 1,635,025 64 7,894,294 128 Matthews, MO TIV No. Events 32,353,402 1 31,614,578 3 6,482,090 30 1,986,192 60 436,107 150 2,514,532 300 Bevier, MO TIV No. Events 31,226,224 1 31,226,224 1 16,643,744 14 8,326,898 27 2,186,144 68 5,466,836 136 Metropolitan Metropolitan Conditional Percentile Max 99.0% 90.0% 80.0% 50.0% MEAN Total Events Rural Joplin, MO ‐ 2011 Historical Tornado Track Aon Benfield | Analytics Proprietary & Confidential | July 2014 Conditional Kansas City, MO‐KS Percentile TIV No. Events MaxJoplin, MO ‐ 2011 Historical Tornado Track 39,466,872 1 99.0% 17,585,432 119 90.0% 5,412,883 1,191 80.0% 3,771,730 2,381 50.0%Joplin, MO ‐ 2011 Historical Tornado Track 1,475,490 5,954 MEAN 2,466,163 Total Events 11,908 26 Tornado Viewing Guide – Top Accumulations Loss Scenarios Mayflower, AR ‐ 2014 Historical Tornado Track Conditional Percentile Max 99.0% 90.0% 80.0% 50.0% MEAN Total Events Sikeston, MO TIV No. Events 70,566,784 1 59,718,096 12 28,595,964 119 19,031,244 239 8,337,714 598 12,255,941 1,196 Poplar Bluff, MO TIV No. Events 66,802,704 1 56,327,168 22 23,891,564 216 16,505,950 433 6,080,895 1,082 10,235,657 2,164 Sedalia, MO TIV No. Events 51,827,928 1 43,565,912 18 25,989,804 181 9,558,846 362 3,506,265 905 8,144,893 1,810 Lebanon, MO TIV No. Events 41,964,000 1 39,725,136 20 29,740,616 203 23,858,452 406 12,674,278 1,015 15,490,019 2,030 US $ in Ones Mayflower, AR ‐ 2014 Historical Tornado Track Conditional Percentile Max 99.0% 90.0% 80.0% 50.0% MEAN Total Events TIV 70,566,784 59,718,096 28,595,964 19,031,244 8,337,714 12,255,941 1,196 Sikeston, MO Loss Potential Inside Track Low Intensity (20%) High Intensity (50%) 14,113,357 35,283,392 11,943,619 29,859,048 5,719,193 14,297,982 3,806,249 9,515,622 1,667,543 4,168,857 2,451,188 6,127,971 US $ in Ones Aon Benfield | Analytics Proprietary & Confidential | July 2014 27 Kennett, MO TIV No. Events 40,750,440 1 36,633,496 14 16,024,542 142 10,517,483 284 3,058,194 711 6,411,309 1,422 Major (EF4 to EF5) Tornado Days Per Century A “Tornado Day” is any day in which you see at least one EF4+ tornado within 80km Aon Benfield | Analytics Proprietary & Confidential | July 2014 28 TVG Loss Ranges Using Historical EF4 & EF5 Event Frequency Exceedance Probabilities Modeled Other Wind Loss Potential Inside Track Return Period 1,000 500 250 100 50 RMS TIV 100% 58,500 47,008 36,813 25,456 18,656 Low Intensity 20% 11,700 9,402 7,363 5,091 3,731 High Intensity 50% 29,250 23,504 18,406 12,728 9,328 12,830 2,566 6,415 25 675 1,687 Annual avg Std dev 25 Annualized Loss from EF4s and EF5s in TVG US $ in Thousands Assumptions: Tornado frequency uses NOAA Storm Prediction Center data 1950-2012 TVG event frequency by smoothed historical average of EF4+ tornadoes is SPC data using 80km Gaussian filter Excess of 1,000 Year Return Period RiskLink v13.1 46,842 44,035 29,864 17,335 10,786 Return Period 1,000 500 250 100 50 7,237 12,338 21,658 5,316 22,996 7,543 US $ in Thousands Excess of 100 Year Return Period Memphis Memphis Denver Other Denver Monroe Tulsa Monroe Other Clovis Clovis Aon Benfield | Analytics Proprietary & Confidential | July 2014 Tulsa 29 AIR Touchstone v1.5 45,405 39,339 30,260 20,955 16,684 TVG Loss Ranges Using Historical EF4 & EF5 Event Frequency Exceedance Probabilities 2013 Other Wind 2011 Return Period 1,000 500 250 100 50 25 AAL from EF4s and EF5s in TVG 2013 Change RMS TIV 100% 17,191 14,590 11,360 7,825 5,888 Loss 50% 8,596 7,295 5,680 3,912 2,944 TIV 100% 15,423 13,045 10,042 6,792 4,996 Loss 50% 7,712 6,523 5,021 3,396 2,498 2011 to 2013 -10% -11% -12% -13% -15% 4,306 2,153 3,664 1,832 -15% 711 -9% 781 US $ in Thousands Assumptions: Tornado frequency uses NOAA Storm Prediction Center data 1950-2012 TVG event frequency by smoothed historical average of EF4+ tornadoes is SPC data using 80km Gaussian filter Other Louisville Cincinnati Cleveland Aon Benfield | Analytics Proprietary & Confidential | July 2014 25 3,821 3,417 Annual avg Std dev 6,585 1,435 3,351 1,667 Return Period 1,000 500 250 100 50 Key Metropolitan Risks 2013 Excess of 1,000 Year RP Cincinnati Cleveland Touchstone v1.5 13,234 10,419 8,767 6,714 4,982 US $ in Thousands Key Metropolitan Risks 2011 Excess of 100 Year RP AIR RiskLink v13.1 12,575 10,318 8,379 6,274 4,941 Excess of 100 Year RP Other Excess of 1,000 Year RP Louisville Cincinnati Cleveland Louisville Cincinnati Cleveland Louisville 30 TVG Historical Tornado Shapes Moore, OK Tornado Joplin, MO Tornado Date: May 22, 2011 Rating: EF‐5 Date: Rating: Mayflower, AR Tornado May 20, 2013 Date: EF‐5 Rating: April 27, 2014 EF‐4 Est. Max. Wind: 200+ mph Est. Max. Wind: 210 mph Est. Max. Wind: Path Length: 22.1 Miles Path Length: 17 Miles Path Length: 41.3 Miles Path Width: 0.75 Mile Path Width: 1.3 Miles Path Width: 0.75 Miles Fatalities Aon Benfield | Analytics Proprietary & Confidential | July 2014 162 Fatalities 24 Fatalities 31 190 mph 16 What if you could run every EF4 and EF5 tornado? Aon Benfield | Analytics Proprietary & Confidential | July 2014 32 Stop Light Exposure Management Aon Benfield | Analytics Proprietary & Confidential | July 2014 33 Paul’s Everyone’s Favorite Quotable Statistician Essentially, all models are wrong, but some are useful George E. P. Box Aon Benfield | Analytics Proprietary & Confidential | July 2014 34 Biographies Steve Drews, Associate Director & Lead Meteorologist Steve Drews is Associate Director and Lead Meteorologist for Impact Forecasting. He currently heads the development of model customization for Impact Forecasting’s catastrophe modeling suite, ELEMENTS. Over his 12 years with Aon, Steve has provided expert analysis on hurricane and tornado damage and has been a featured speaker at client, meteorological and government presentations about coastal urbanization, global climate change, tropical cyclone frequency, and severe convective weather frequency as well as provided scientific testimony on behalf of insurance and reinsurance companies in legal matters. Paul Eaton, FCAS, Director Paul Eaton has worked in the Aon Benfield Analytics Chicago office for seven years. Paul’s current role focuses on catastrophe management including risk adjusted pricing and capacity allocation. Paul dabbled in telephony consulting, home improvement sales, and even driving an ambulance prior to joining Aon Benfield. Aon Benfield | Analytics Proprietary & Confidential | July 2014 35 Contacts Steve Drews Associate Director & Lead Meteorologist Impact Forecasting +1.312.381.5888 steven.drews@aon.com Paul Eaton, FCAS Director Aon Benfield Analytics +1.312.381.5553 paul.eaton@aon.com Aon Benfield | Analytics Proprietary & Confidential | July 2014 36 Legal disclaimer: Impact Forecasting LLC The results listed in this report are based on engineering / scientific analysis and data, information provided by the client, and mathematical and empirical models. The accuracy of the results depends on the uncertainty associated with each of these areas. In particular, as with any model, actual losses may differ from the results of simulations It is only possible to provide plausible results based on complete and accurate information provided by the client and other reputable data sources. Furthermore, this information may only be used for the business application specified by Impact Forecasting, LLC and for no other purpose. It may not be used to support development of or calibration of a product or service offering that competes with Impact Forecasting, LLC. The information in this report may not be used as a part of or as a source for any insurance rate filing documentation. THIS INFORMATION IS PROVIDED "AS IS" AND IMPACT FORECASTING, LLC HAS NOT MADE AND DOES NOT MAKE ANY WARRANTY OF ANY KIND WHATSOEVER, EXPRESSED OR IMPLIED, WITH RESPECT TO THIS REPORT; AND ALL WARRANTIES INCLUDING WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE HEREBY DISCLAIMED BY IMPACT FORECASTING, LLC. IMPACT FORECASTING, LLC WILL NOT BE LIABLE TO ANYONE WITH RESPECT TO ANY DAMAGES, LOSS OR CLAIM WHATSOEVER, NO MATTER HOW OCCASIONED, IN CONNECTION WITH THE PREPARATION OR USE OF THIS REPORT. Aon Benfield | Analytics Proprietary & Confidential | July 2014 37