Quantifying the Impact of Non-Modeled Catastrophes David Chernick, FCAS, MAAA Michael Devine, FCAS, MAAA Eric Huls, FCAS CAS Ratemaking Seminar March, 2005 1 7/26/2016 Agenda Introduction History of Methods – Terminology – Exposure Base – Capping Discussion of Several Alternative Approaches – – – – Traditional Methods Using Cat/AIY – State Based Methods Using Cat/AIY – Countrywide Based Total Weather Method Wrap-up 2 7/26/2016 Introduction The Panelists The Data The Issue 3 7/26/2016 Introduction The Panelists Introductions Eric Huls Michael Devine David Chernick 4 7/26/2016 Introduction The Data The base data we are using in this presentation is included in the handouts. Real data Large Catastrophe in 1998 343.2% loss ratio 9.67 Ratio of Cat/AIY 5 7/26/2016 Introduction The Issue – Operating results $19.1 Million profit prior to 1998 $41.4 Million loss in 1998 – Statistics Mean: 1998 was 4.3 Standard Deviations from mean. 6 7/26/2016 Introduction The Issue – A rate should include all costs associated with the transfer of risk. – 20 or 30 or even 40 years of data is not sufficient to properly quantify the tail of the distribution – What is the true prospective average (mean) catastrophe provision? 7 7/26/2016 Introduction The Issue – Perspective of this presentation is from large insurers without reinsurance coverage. – Reinsurance covering some portion of the catastrophe exposure would most likely be an upper bound of the true mean. – What is the true prospective average (mean) catastrophe provision? 8 7/26/2016 Defining a “Catastrophe”: Dollar/Claim Count Thresholds: Easy to determine when cat has occurred; Standard method; Static dollar threshold erodes over time due to inflation; Not responsive to different exposure concentrations or growth/decline in exposures; Percentage of Policyholders Threshold: Avoids inflation issue of static dollar threshold; Requires additional data (exposures) to categorize events; Ignores severity; Percentage of Days with Highest Frequency Also avoids inflation issue of static dollar threshold; Requires additional data (exposures) to categorize events; Categorization can change over time; Ignores severity; 9 7/26/2016 Additional Terminology AIY – Amount of Insurance years 1AIY=$1,000 of dwelling coverage Losses/AIY – Damage Ratios or Cat/AIY 10 7/26/2016 History of Methods Data Category Category # 1 ISO Excess Wind & Variations A. Ratio of wind losses to ex-wind losses Averaging Methods Base: Non-wind incurred losses Cat Data: Annual wind incurred losses over threshold B. Ratio of cat losses to ex-cat losses Base: Non-cat incurred losses Cat Data: Incurred event losses over fixed external threshold Category #2 Cat/AIY Methods Base: AIY Cat Data: Incurred event losses over fixed external threshold A. Utilize state Cat/AIY with no direct use of countrywide cat experience Adjustment for Extreme Cats Arithmetic average of excess wind factors and average wind to ex-wind ratios. For each additional year, give 95% weight for long-term average, 5% weight for additional year. Can give less weight to unusual years Can adjust catastrophes for unusual years Straight long-term average Confidence interval approach Average adjusted catastrophes and add increment for extreme Trended exponential smoothing Balance state provisions to countrywide expectation May or may not include adjustments for extreme catastrophes Average annual losses are based on the stochastic event set Extreme events are given their appropriate statistical weight B. Utilize state Cat/AIY with direct use of countrywide cat experience Category #3 Modeled Cat Losses Base: AIY Cat Data: Computer application generated losses Category #4 Reinsurance Cost Based Cat cover reinsurance cost passed through primary pricing 11 7/26/2016 What Base to Relate Catastrophes To? Premium: Cat provisions impacted by rate changes Trends in non-catastrophe loss & expense dictate cat provision 12 7/26/2016 What Base to Relate Catastrophes To? Premium: Cat provisions impacted by rate changes Trends in non-catastrophe loss & expense dictate cat provision Non-Cat Loss/Ex-Wind Loss: Still heavily dictated by trends in Crime, Liability, etc. loss Ex-wind losses can include catastrophic losses 13 7/26/2016 What Base to Relate Catastrophes To? Premium: Cat provisions impacted by rate changes Trends in non-catastrophe loss & expense dictate cat provision Non-Cat Loss/Ex-Wind Loss: Still heavily dictated by trends in Crime, Liability, etc. loss Ex-wind losses can include catastrophic losses AIY or Amount of Insurance Years Definition: $1000 of Building Coverage in force for one year Inflation sensitive Direct measurement of exposure – incorporates policy growth and changes in building costs 14 7/26/2016 Should Individual Catastrophes Be Capped? 15 7/26/2016 Should Individual Catastrophes Be Capped? Stabilizes provision Can serve to more appropriately match experience period used with event return periods Potentially more accurate estimate of expected value results 16 7/26/2016 What Are Some Problems With Capping Individual Catastrophes? 17 7/26/2016 What Are Some Problems With Capping Individual Catastrophes? What criteria should be used? The “unthinkable” is happening every year somewhere. Is the result systematic underestimation of loss costs? How do we really know appropriate event return periods? 18 7/26/2016 Insurance Services Office (ISO) Excess Wind Procedure Basic Approach Separate wind & non-wind losses Examine wind/non-wind ratios Years where wind/non-wind exceed 1.5 times median are “excess” Average factor for excess wind Factor developed for excess wind applied to non-wind, non-excess losses 19 7/26/2016 Insurance Services Office (ISO) Excess Wind Procedure Basic Approach Separate wind & non-wind losses Examine wind/non-wind ratios Years where wind/non-wind exceed 1.5 times median are “excess” Average factor for excess wind Factor developed for excess wind applied to non-wind, non-excess losses Characteristics Straightforward application Definition of “excess wind” can change as median changes Assumes stable relationship between wind & non-wind losses Doesn’t consider variability of wind losses 20 7/26/2016 Doesn’t consider non-wind catastrophes The Fix ‘Em Up Insurance Group Homeowners The State of Mich-con-ota 20-Year Average Approach Year Amount of Insurance Years (AIY) 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 4,276,135 4,306,815 4,540,913 4,774,783 5,001,164 5,193,190 5,367,566 5,574,506 5,745,344 6,223,199 Cat Incurred Loss $ CAT/AIY 307,946 0.0720 259,784 0.0603 4,378 0.0010 1,529,513 0.3203 2,736,486 0.5472 50,241,886 9.6746 8,141,594 1.5168 6,676,296 1.1976 12,086,512 2.1037 6,091,053 0.9788 Provision (20-Year Average ) Running Provision 0.2148 0.2102 0.1719 0.1862 0.2128 0.6948 0.7114 0.7566 0.8465 0.8929 0.8929 21 7/26/2016 Confidence Interval Approach Step 1 – Establish Company Objective: Factors include risk tolerance, surplus position/ availability of capital, reinsurance Determine confidence demands for long-term companywide cat provision Calculate companywide mean cat/aiy Calculate standard deviation of mean cat/aiy 22 7/26/2016 Confidence Interval Approach Step 1 – Establish Company Objective (Cont.): Company has established that it would like to be 90% certain it has an adequate catastrophe provision over the long-term The following have been calculated from the companywide catastrophe history: Mean Cat/AIY = .3151 Standard Deviation of Mean Cat/AIY = .0372 23 7/26/2016 Confidence Interval Approach Step 1 – Establish Company Objective (Cont.): The long-run companywide benchmark cat provision is established as follows: Provision (Cat/AIY) = Mean + (t) x (Standard Deviation) = .3151 + (1.323) x (.0372) = .3643 Where : Mean = average cat/aiy companywide 1.323 = t – stat for 90% and (N-1) degrees of freedom .0372 = standard deviation of the mean 24 7/26/2016 Confidence Interval Approach Step 2 – Establish State Level Objective: Goal period becomes interval rates are in effect Need to be reasonably certain provision is adequate Desire to use cap on individual cats to limit volatility Largest 5% of companywide cats exceeded .65/AIY Establish required confidence for state capped average 25 7/26/2016 Non-Hurricane Catastrophe Provisions Confidence Sum of States Intervals Uncapped Capped 50% 55 60 65 70 75 80 85 90 95 0.3334 0.3825 0.4328 0.4848 0.5392 0.5988 0.6657 0.7447 0.8452 0.9999 0.2682 0.2998 0.3323 0.3657 0.4008 0.4392 0.4823 0.5332 0.5981 0.6973 Companywide Uncapped 0.3151 0.3198 0.3247 0.3297 0.3349 0.3406 0.3471 0.3546 0.3643 0.3791 26 7/26/2016 Confidence Interval Approach Step 2 – Establish State Level Objective (Cont.): It’s determined that 65% confidence is required Calculate state mean cat/aiy (capped) Calculate state standard deviation of cat/aiy (capped) 27 7/26/2016 Confidence Interval Approach Step 3 – Calculate State Provision: The following was calculated from the capped state level catastrophe history: Mean Cat/AIY = .3912 Standard Deviation of Cat/AIY = .5450 The short-run state cat provision is established as follows: Provision (Cat/AIY) = Mean + (t) x (standard deviation) = (.3912) + (.389) x (.5450) = .6032 Where: Mean = average capped cat/aiy for Mich-con-ota .389 = t – stat for 65% and (N-1) degrees of freedom .5450 = standard deviation of the annual capped 28 7/26/2016 cat/aiy The Fix ‘Em Up Insurance Group Homeowners The State of Mich-con-ota Confidence Interval Approach Year Amount of Ins. Years (AIY) 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 4,276,135 4,306,815 4,540,913 4,774,783 5,001,164 5,193,190 5,367,566 5,574,506 5,745,344 6,223,199 Cat Incurred Loss $ 307,946 259,784 4,378 1,529,513 2,736,486 50,241,886 8,141,594 6,676,296 12,086,512 6,091,053 CAT/AIY Capped CAT/AIY Running Provision 0.0720 0.0603 0.0010 0.3203 0.5472 9.6746 1.5168 1.1976 2.1037 0.9788 0.0720 0.0603 0.0010 0.3203 0.5472 1.6175 1.5168 1.1976 1.9537 0.9788 0. 2983 0.2911 0.2825 0.2865 0.3016 0.3934 0.4612 0.5004 0.5835 0.6032 Provision (Confidence Interval Approach) 29 7/26/2016 0.6032 Issues With Confidence Interval Approach Pluses Recognizes individual state variability Stable provision Provides means to assure companywide sufficiency 30 7/26/2016 Issues With Confidence Interval Approach Pluses Recognizes individual state variability Stable provision Provides means to assure companywide sufficiency Drawbacks Not particularly responsive to distributional changes, coverage changes, etc. (data back to 1971) Capping can result in less responsiveness Recognition of variability interpreted as risk margin 31 7/26/2016 The Fix ‘Em Up Insurance Group Homeowners The State of Mich-con-ota Extreme Events Adjustment Year Amount of Ins. Years (AIY) 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 4,276,135 4,306,815 4,540,913 4,774,783 5,001,164 5,193,190 5,367,566 5,574,506 5,745,344 6,223,199 Cat Inc. Loss $ Non Extreme Cat Inc. Loss Extreme Cat Inc. Loss 307,946 $ 259,784 4,378 1,529,513 2,736,486 50,241,886 45,217,697 8,141,594 6,676,296 12,086,512 6,091,053 307,946 259,784 4,378 1,529,513 2,736,486 5,024,189 8,141,594 6,676,296 12,086,512 6,091,053 Extreme CAT/AIY Contrib. From Extreme $ 20 Year Average 8.7071 1.7414 0.0871 Provision (Extreme plus Non-Extreme) Contrib. From Non Extreme Running Provision 0.0720 0.0603 0.0010 0.3203 0.5472 0.9675 1.5168 1.1976 2.1037 0.9788 0.2148 0.2102 0.1719 0.1862 0.2128 0.3465 0.3631 0.4083 0.4982 0.5446 0.4576 0.5446 32 7/26/2016 Extreme Events Adjustment Pluses Relatively stable As opposed to censoring, reflects events fully Drawbacks Accurate determination of event return period difficult Can be viewed as arbitrary and difficult to explain 33 7/26/2016 95% / 5% Trended Approach: Methodology: All years used Exponential smoothing Trend factor applied – recognizes static cat definition 10% annual cap to change in provision 34 7/26/2016 The Fix ‘Em Up Insurance Group Homeowners The State of Mich-con-ota 95/5 Trended Year Amount of Insurance Years (AIY) 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 4,276,135 4,306,815 4,540,913 4,774,783 5,001,164 5,193,190 5,367,566 5,574,506 5,745,344 6,223,199 Cat Incurred Loss $ 307,946 259,784 4,378 1,529,513 2,736,486 50,241,886 8,141,594 6,676,296 12,086,512 6,091,053 Provision (95/5 Trended) CAT/AIY 0.0720 0.0603 0.0010 0.3203 0.5472 9.6746 1.5168 1.1976 2.1037 0.9788 Running Provision 0. 1972 0.1930 0.2054 0.2259 0.2485 0.2733 0.3007 0.3307 0.3638 0.4002 0.4002 35 7/26/2016 95% / 5% Trended Approach: Advantages: Not volatile, yet responsive for non-extreme events Simple to understand Trend factor to compensate for static definition of cats Reduced data complications 36 7/26/2016 Summary of Results So Far: Approach Mich-con-ota Cat/AIY Stability Rank Cat/Ex-Cat 20-Year Average Confidence Interval Approach Extreme Events 95% / 5% Trended All Years Weighted Average .7292 .8929 .6032 .5446 .4002 .9940 4 5 2 3 1 6 37 7/26/2016 Pivotal Question: Can Countrywide or Regional Data Help Quantify the True Prospective Mean Catastrophe Loss in a Given State? 38 7/26/2016 Pivotal Question: Can Countrywide or Regional Data Help Quantify the True Prospective Mean Catastrophe Loss in a Given State? Issues: Provisions need to reflect adequacy and stability All company surplus is generally available and at risk Are regional or sub state provisions appropriate? Perceived cost sharing will be scrutinized 39 7/26/2016 Goals of Relativity Method Develop accurate, stable results by state that results in an appropriate provision on a countrywide basis Systematic approach to handle extreme events so a single outlying year does not drive the cat provision for a state Appropriate application of credibility procedure Provide result that is responsive to recent demographic and cat definition shifts 40 7/26/2016 Issues Addressed How to be responsive to changes in risk due to population shifts or cat definition changes while still including an appropriate number of years How does one define an outlying event – Individual state vs. countrywide How to incorporate credibility 41 7/26/2016 State Relativity Weighted with Countrywide Complement – General Outline I. II. III. IV. V. VI. VII. VIII. Develop State Damage Ratios Calculate Countywide Damage Ratios Calculate State Relativities Cap State Relativities Average Capped Relativities Credibility Weight with CW Average of 1.000 Balance Back to CW Average of 1.000 Calculate Statewide Catastrophe Provision 42 7/26/2016 State Relativity Weighted with Countrywide Complement I. II. III. IV. V. VI. Develop State Damage Ratios Calculate Countrywide Damage Ratios Calculate State Relativities Cap State Relativities Average Capped Relativities Credibility Weight with CW Average of 1.000 VII. Balance Back to CW Average of 1.000 VIII. Calculate Statewide Catastrophe Provision Develop each state’s damage ratios for years 1981-2000 – State Damage Ratios – Losses/AIY – Only use years 1981 forward. Data for years 1971 through 1980 is sparse as evidenced by yearly variance. 43 7/26/2016 State Relativity Weighted with Countrywide Complement I. II. III. IV. V. VI. Develop State Damage Ratios Calculate Countrywide Damage Ratios Calculate State Relativities Cap State Relativities Average Capped Relativities Credibility Weight with CW Average of 1.000 VII. Balance Back to CW Average of 1.000 VIII. Calculate Statewide Catastrophe Provision Each year’s Countrywide damage ratio is calculated as the weighted average of state damage ratios using latest year AIYs as weights – Eliminates distortion of state distributional shifts over time Countrywide catastrophe provision is the arithmetic average of the most recent 10 years of damage ratios 44 7/26/2016 Figure 1 CW Cat damage ratios including 2000 0.80 0.70 Damage ratio 0.60 0.50 0.40 0.30 0.20 0.10 0.00 1970 1975 1980 1985 1990 1995 2000 2005 Year Years Linear trend 1971-1978 0.006 1979-1989 0.000 1990-1999 -0.019 1990-2000 -0.010 45 7/26/2016 State Relativity Weighted with Countrywide Complement I. II. III. IV. V. VI. Develop State Damage Ratios Calculate Countrywide Damage Ratios Calculate State Relativities Cap State Relativities Average Capped Relativities Credibility Weight with CW Average of 1.000 VII. Balance Back to CW Average of 1.000 VIII. Calculate Statewide Catastrophe Provision Calculate state relativities as the ratio of state damage ratios to countrywide damage ratios Relativities should be more stable than damage ratios Trend should not be a problem so we can use more years of data than the Countrywide Catastrophe Provision 46 7/26/2016 State Relativity Weighted with Countrywide Complement I. II. III. IV. V. VI. Develop State Damage Ratios Calculate Countrywide Damage Ratios Calculate State Relativities Cap State Relativities Average Capped Relativities Credibility Weight with CW Average of 1.000 VII. Balance Back to CW Average of 1.000 VIII. Calculate Statewide Catastrophe Provision Any relativity greater than the mean plus three standard deviations is capped to the next lowest relativity (not the cap number) – Intuitively we are replacing a once in a hundred year event with a once in 20 Benefit of capping process – Represents a systematic approach to dealing with extreme events – Cap is dynamic and is allowed to shift if exposure in a state is changing over time – Censoring at the cap would not have much impact and therefore would not result in increased stability 47 7/26/2016 State Relativity Weighted with Countrywide Complement I. II. III. IV. V. VI. Develop State Damage Ratios Calculate Countrywide Damage Ratios Calculate State Relativities Cap State Relativities Average Capped Relativities Credibility Weight with CW Average of 1.000 VII. Balance Back to CW Average of 1.000 VIII. Calculate Statewide Catastrophe Provision Calculate arithmetic average of 1981-2000 capped relativities – Using a arithmetic average is simple – No benefit of weighting relativities has been shown since relationship of variability to exposure level is unclear – Arithmetic average relativity does not differ significantly from an AIY weighted average 48 7/26/2016 State Relativity Weighted with Countrywide Complement I. II. III. IV. V. VI. Develop State Damage Ratios Calculate Countrywide Damage Ratios Calculate State Relativities Cap State Relativities Average Capped Relativities Credibility Weight with CW Average of 1.000 VII. Balance Back to CW Average of 1.000 VIII. Calculate Statewide Catastrophe Provision Uses Buhlmann credibility factor: n/(n+k) – n = number of years of relativities in average We use number of years rather than exposures because exposures not independent, especially past a certain threshold where exposure concentration increases – k = average process variance/variance of hypothetical means The process variance and variance of hypothetical means are calculated using all available years of capped relativities across all states. Complement of credibility of 1.000 is not appropriate when there is a wide spread of average relativities – Solution lies in balancing process described on next slide 49 7/26/2016 State Relativity Weighted with Countrywide Complement I. II. III. IV. V. VI. Develop State Damage Ratios Calculate Countrywide Damage Ratios Calculate State Relativities Cap State Relativities Average Capped Relativities Credibility Weight with CW Average of 1.000 VII. Balance Back to CW Average of 1.000 VIII. Calculate Statewide Catastrophe Provision At this point, the individual state relativities result in a countrywide relativity of less than 1.000. Relativities are adjusted to achieve an overall adequate level as follows: – Determined on a countrywide basis what our expected losses would be based on the countrywide selected catastrophe factor – Sum the pre-balanced expected losses across all states – We distribute the difference between 1 and 2 in proportion to each state’s standard deviation measured in latest year expected losses. Using this approach has several benefits: – Results in an appropriate provision countrywide – It compensates for high (low) relativity states being underestimated (overestimated) by the use of a 1.000 complement of credibility. – Each state’s resulting cat load is a function of its own size and variability 50 7/26/2016 State Relativity Weighted with Countrywide Complement I. II. III. IV. V. VI. Develop State Damage Ratios Calculate Countrywide Damage Ratios Calculate State Relativities Cap State Relativities Average Capped Relativities Credibility Weight with CW Average of 1.000 VII. Balance Back to CW Average of 1.000 VIII. Calculate Statewide Catastrophe Provision Statewide catastrophe provision is calculated by multiplying capped, credibility weighted, balanced relativity by the countrywide catastrophe provision Benefits of method: – Allows use of long term data to determine relativity while using more responsive data for countrywide provision – Adjustments to data are determined objectively with each state’s characteristics used to determine both capping and balancing 51 7/26/2016 Issues Addressed How to be responsive to changes in risk due to population shifts or cat definition changes while still including an appropriate number of years – We use as many years of relativities as possible while using only the latest 10 years of Countrywide to determine the Countrywide load. How do you define an outlying event – Greater than the mean plus 3 standard deviation in any given state. On a countrywide basis these “outlying” events occur fairly regularly (about 2% of relativities have been capped) How to incorporate credibility – Uses Buhlmann credibility to account for variability in relativities. Used a relativity of 1.000 as complement. Balancing method adjusts for bias in complement for when a 1.000 may not be an appropriate complement. 52 7/26/2016 Results of Damage Ratiobased Methods: Approach Cat/Ex-Cat 20-Year Average Confidence Interval Approach Extreme Events 95% / 5% Trended All Years Weighted Average Relativity Method Mich-con-ota Cat/AIY .7292 .8929 .6032 .5446 .4002 .9940 .6270 53 7/26/2016 Average Frequency / Trended Severity Approach: Methodology: Calculates total weather provision, as opposed to cat-only provision All available years used Projected frequency equals arithmetic average frequency Projected severity equals arithmetic average trended severity Projected pure premium = Proj. Freq x Proj. Sev 54 7/26/2016 Average Frequency / Trended Severity Approach: Pluses Simple to understand Avoids problems associated with static definition of cats by considering all weather events together Avoids identical events being classified differently as cats and non-cats, depending on where they occur Less sensitive than AIY methods to changes in insurance to value 55 7/26/2016 Average Frequency / Trended Severity Approach: Drawbacks Ignores changes in frequency over time due to deductible and geographical distribution shifts, claiming behavior Must select frequency and severity trends Still must account for non-weather related catastrophes (like wildfires) 56 7/26/2016 Conclusion We have made significant progress as a profession in quantifying the catastrophe exposure. Do our current methods capture the true mean? My purpose will be achieved if this session helps to keep focus on this issue 57 7/26/2016 Conclusion Comments and Discussion 58 7/26/2016