1、I1:Intermediate Topics on Experience and Exposure RatingCARe Seminar,June 5-6,2023 Philadelphia,PAYinglu Fan,FCAS,AVP Treaty Underwriter(Moderator)John Maher,FCAS,Senior Vice PresidentRalph Dweck,FCAS,Actuarial Director,Verisk/ISO2Antitrust Notice The Casualty Actuarial Society is committed to adher

2、ing strictly to the letter and spirit of the antitrust laws.Seminars conducted under the auspices of the CAS are designed solely to provide a forum for the expression of various points of view on topics described in the programs or agendas for such meetings.Under no circumstances shall CAS seminars

3、be used as a means for competing companies or firms to reach any understanding expressed or implied that restricts competition or in any way impairs the ability of members to exercise independent business judgment regarding matters affecting competition.It is the responsibility of all seminar partic

4、ipants to be aware of antitrust regulations,to prevent any written or verbal discussions that appear to violate these laws,and to adhere in every respect to the CAS antitrust compliance policy.3Intermediate Track Pre-RequisitesThese presentations are considered intermediate level and assume you alre

5、ady have a basic understanding of the following concepts:General purpose of exposure rating vs experience rating Losses occurring vs risks attaching Treaty vs facultative Excess of loss reinsurance Primary vs excess policies Claims development and trending/on-leveling:purpose and methodologies ALAE,

6、rate change,ILFs,credibility 4This session will build upon basic CARe track and prior boot camp materials and will presuppose familiarity with the basics of experience and exposure rating methodologies.This session will include the usage of more advanced techniques to identify and address common exc

7、ess rating challenges.These challenges have been exacerbated by the significant but waning pandemic disruptions and resulting supply constrained inflation impacts over the last 3 years.These additional techniques and distortions include:Rating methods:Shifting policy limits,credibility and blending

8、of loss development factors,and combining experience&exposure ratings Measuring benchmark distortions:LDFs,severities,frequencies,closure ratios,recent adverse development and resulting loss ratiosAccurately assessing these impacts holistically,and avoiding overconfidence,will lead to more refined p

9、ricing/reserving benchmarking and individual account analysis.Moderator:Yinglu Fan,FCAS,AVP Treaty Underwriter,QBE Re 5 mins(1 intro slide+potential polls updates 2022)Panelists:John Maher,FCAS,Senior Vice President,QBE Re 25-30 mins(27 slides)Ralph Dweck,FCAS,Actuarial Director,Verisk/ISO 25-30 min

10、s(26 slides)Q&A 10 minsI1:Intermediate Topics on Experience and Exposure Rating5Measuring Confidence Covid/Inflation Trends-6/2022We asked 12 Qs(10 US,2 UK)via Survey Monkey that was presented at the 2022 CARe Conference in a pair of linked sessions.The poll was left up during the course of the Mond

11、ay June 13 CS10 2pm presentation.To answer the questions:If you feel 90%of the time the answer will be between-15%to-5%then enter-15 and-5 in the 2 boxes.Should carefully read the question being asked,such as LOB,frequency or severity,and time period.You can answer either anonymously,or provide your

12、 nameat the end.You dont need to answer all the Qs leaving certain ones blank or just providing a wider range on those.“Answers”will be presented during CS10s Covid/Inflation section.Measuring Confidence answers,comparing aggregated confidence interval ranges to the“Answers”,was provided in the 2022

13、 Tuesday linked session CS23“Overinflated Wheels”.That session will also go deeper into the Covid/Inflation impacts in the Commercial and Personal Auto poll Q results.6ISO CARe 6/2022 Survey of Covid/Inflation TrendsMetrics for Pre Covid,1st Covid 90%CI(Responses)ActualResponses in Rangeand 2nd Covi

14、d yearLowerUpper1.Total GL Frequency Change 2015-2019-10%7%-4.0%33.0%2.Total GL Frequency Change 2019-2020-20%5%-29.5%0.0%3.Total GL Frequency Change 2020-2021-10%15%-2.0%33.0%4.Total GL Severity Change 2015-20190%15%5.2%82.0%5.Total GL Severity Change 2019-20200%20%10.7%27.0%6.Total GL Severity Cha

15、nge 2020-20210%20%9.1%91.0%7.Total CAu Frequency Change 2019-2020-40%20%-26.3%17.0%8.Total CAu Severity Change 2020-20212%20%10.7%45.0%9.Total PAu Frequency Change 2019-2020-50%2%-22.5%33.0%10.Total PAu Severity Change 2020-20212%30%7.5%55.0%11.UK Personal Motor Frequency Change 2019-2020-50%10%12.U

16、K Personal Motor Severity Change 2020-20210%20%NB:Above frequency indications are Nominal,before rate change impacts7Shifting Limits in Excess of Loss Ratings8Shifting Limits in Excess of Loss Rating Changing Policy Limits Distribution Suppose we are pricing a 500,000 excess of 500,000 layer,but the

17、 ceding company has recently begun writing higher limit policies that result in more exposure to the layer.Can we still use the historical experience rating?If so,what adjustments can be made?9Shifting Limits in Excess of Loss RatingThere are many possible approaches to overlay an adjustment to the

18、experience rating.One approach:Adjust historical experience period burn cost based on the relative exposure rating of each historical period(i.e.limits drift factor)Advantage:This is one of the most accurate of possible methods.Disadvantage(s):Requires full policy limit profile for each historical p

19、eriod Potential difficulty in explaining adjustment factorsExample on the next slide10Shifting Limits in Excess of Loss RatingAdjust historical experience period burn cost based on the relative exposure rating of each historical period(i.e.limits drift factor)The exposure rates from this table are u

20、sed to adjust the experience rated loss costs.The change in exposure rate combines the impact of the changing layered loss and the change in premium that results from the shift in the limits profile.Mata&Verheyen“An Improved Method for Experience Rating Reinsurance Treaties using Exposure Rating Tec

21、hniques”(2005)http:/www.casact.org/pubs/forum/05spforum/05spf171.pdf11Shifting Limits in Excess of Loss RatingAdjust historical experience period burn cost based on the relative exposure rating of each historical period(i.e.limits drift factor)Limits drift factor for 2011=Expected Loss for 2020/Expe

22、cted Loss for 20114.17%/2.22%=1.88The experience rated loss cost indication for 2011 would then be adjusted by a factor of 1.88 to account for the fact that the ceding company is now writing more high limit policies than they have in the past.This adjustment factor would be calculated for each year

23、in the experience period.IMPORTANT this methodology can be used for an increasing shift in limits or decreasing shift in limits12AYUltimate Loss RatioExposure Indication$500k Xs$500kLimits Drift FactorAdjusted Ultimate Loss Ratio20112.8%2.22%1.885.2%20122.0%2.22%1.883.8%20131.4%2.22%1.882.6%20143.3%

24、2.22%1.886.2%20154.0%2.22%1.887.5%20162.8%2.87%1.454.0%20173.4%3.52%1.184.0%20183.0%4.17%1.003.0%20192.7%4.17%1.002.7%20203.1%4.17%1.003.1%20214.1%4.17%1.004.1%Straight Avg=3.0%4.2%13Credibility in Loss Development14The Issue The client data we get is usually not 100%credible,due to volume and insuf

25、ficient time frame.We have some prior knowledge of what the development pattern should look like,either from external data or wider samples of similar business.How do we blend our prior knowledge with the new observation in a systematic way?15Brief Introduction to Bayesian Credibility“Probability is

26、 orderly opinion,and inference from data is nothing other than the revision of such opinion in the light of relevant new information.”Edwards,Lindman and Savage Bayesian Theory16Bayesian Made SimpleTwo coins are in a box:one with both sides heads and one fair coin.Select one coin at random and flip

27、it,the odds of a heads are:=12 1+12 0.5=0.75(one-half chance selecting the sure heads coin and one-half chance selecting the fair coin)The first result was heads.Now use the same coin and flip it a second time.The odds of a second heads are:We need to first calculate the odds that each of the coins

28、was initially selected,given the result of heads.These are called Conditional Probabilities.1.=0.5 10.75=232.=0.5 0.50.75=13Finally,we use theses conditional probabilities as weights and multiply them by the odds of a heads on those respective coins:23 1+13 0.5=0.8317Application to Loss Development

29、Organize the prior beliefs into an explicit distribution By staying in the context of conjugate(posterior distribution follows the same parametric form as the prior distribution)models,the blending of prior knowledge with new data can be done with very simple calculations.ZA+(1-Z)B Can be derived fr

30、om Bayes Theorem either by assuming that the number of claims follow a Bernoulli process,with a Beta prior distribution on the unknown parameter p,or a Poisson process,with a Gamma prior distribution on the unknown parameter m.Allen L.Mayerson A BAYESIAN VIEW OF CREDIBILITY(casact.org)18Generalized

31、Dirichlet Distribution First introduced in the context of biological science.Parameter set with alphas and betas Alphas proportional to incremental loss and betas proportional to cumulative loss.Different weights for each cumulative development age,making it a natural for the development triangle fo

32、rmat.1224=+=+=1,+1 +=1,19 Bayesian theory assumes that an analyst working with a loss development triangle does not start as a“blank slate”with no idea of what a development pattern looks like.Instead,it assumes that the analyst comes with some“prior”expectation and is willing to change that prior b

33、elief on what is observed in the new data.(Clark 2016)Our prior knowledge,in this case of the industry or market development patterns,is used as though it had been previously observed data.There are two main sources of uncertainty in prior information(Parodi and Bonche 2010)Market heterogeneity the

34、spread of different risks around some industry average Estimation uncertainty the industry average,though large,may still be of limited size As a result,we may choose to give the prior distribution more or less variance(and ultimately credibility)depending on how we view these sources of uncertainty

35、.2028496Prior Pattern LDFs=21.9507.7873.9462.5121.8421.5581.4151.315%Reported=4.6%12.8%25.3%39.8%54.3%64.2%70.7%76.0%ATA=2.8191.9731.5711.3641.1821.1011.0761.315Alpha2.62.01.51.10.60.40.31.0Beta1.41902.02.52.93.43.63.73.0Alpha+Beta(+)4.04.04.04.04.04.04.04.0Variance/Mean Ratio()1,000Col.1

36、1,419 2,027 2,546 2,933 3,383 3,633 3,717 3,042 Col.24,000 4,000 4,000 4,000 4,000 4,000 4,000 4,000 User judgmentally selects+and,the variance to mean ratio.is (1-1/ATA)is(+)Col.1 is()Col.2 is(+)Prior Information21Client Data(new observation)28496199073 262 469 528 536 591 604 606 199114

37、8 346 391 502 522 514 567 199299 198 219 394 408 430 1993118 255 352 412 581 1994275 415 645 803 1995261 446 637 1996130 471 1997148 Col.11,104 1,922 2,076 1,836 1,466 1,105 604 Col.22,393 2,713 2,639 2,047 1,535 1,171 606 Avg ATA2.1681.4121.2711.1151.0471.0601.00322Combine the two for Credibility W

38、eightingPrior KnowledgeCol.11,419 2,027 2,546 2,933 3,383 3,633 3,717 3,042 Col.24,000 4,000 4,000 4,000 4,000 4,000 4,000 4,000 ATA2.8191.9731.5711.3641.1821.1011.0761.315New ObservationCol.11,104 1,922 2,076 1,836 1,466 1,105 604 Col.22,393 2,713 2,639 2,047 1,535 1,171 606 ATA2.1681.4121.2711.115

39、1.0471.0601.003Credibility WeightedCol.12,523 3,949 4,622 4,769 4,849 4,738 4,321 3,042 Col.26,393 6,713 6,639 6,047 5,535 5,171 4,606 4,000 New ATA2.5341.7001.4361.2681.1411.0911.0661.31523The higher selection for the parameters(+),)result in more weight being given to the prior knowledge.Prior Inf

40、ormation(more weight to prior)28496Prior Pattern LDFs=21.9507.7873.9462.5121.8421.5581.4151.315%Reported=4.6%12.8%25.3%39.8%54.3%64.2%70.7%76.0%ATA=2.8191.9731.5711.3641.1821.1011.0761.315Alpha3.93.02.21.60.90.60.41.4Beta2.12863.03.84.45.15.45.64.6Alpha+Beta(+)6.06.06.06.06.06.06.06.0Vari

41、ance/Mean Ratio()5,000Col.110,643 15,202 19,098 21,998 25,375 27,246 27,880 22,814 Col.230,000 30,000 30,000 30,000 30,000 30,000 30,000 30,000 Credibility WeghtedCol.111,747 17,124 21,174 23,834 26,841 28,351 28,484 22,814 Col.232,393 32,713 32,639 32,047 31,535 31,171 30,606 30,000 Avg ATA2.7581.9

42、101.5411.3451.1751.0991.0751.31524Using a Library of Benchmark PatternsBenchmark Loss Development Factors(LDF to Ultimate)28496Fast14.0144.932.6071.7591.4061.2631.1911.155Medium21.957.7873.9462.5121.8421.5581.4151.315Slow49.2415.867.4074.1632.7062.0571.751.567In this case,we have not just

43、 one,but three benchmark patterns.These may be based on reporting lag,settlement strategies,case reserving practices,etc.If we have no knowledge of our clients practices,we can start with giving each benchmark pattern equal weights.We perform the credibility weighting of our clients data with each o

44、f these three benchmarks.Then use their likelihood functions to update the weights.25Example(Fast Pattern)Fast Pattern28496LDF14.0144.9302.6071.7591.4061.2631.1911.155Pattern7.14%20.28%38.36%56.85%71.12%79.18%83.96%86.58%ATA2.8431.8911.4821.2511.1131.0601.0311.155Alpha6.54.73.32.01.00.60.

45、31.3Beta3.55.36.78.09.09.49.78.7Alpha+Beta10.010.010.010.010.010.010.010.0Variance/Mean Ratio1,000Col.13,518 5,288 6,747 7,993 8,983 9,430 9,698 8,658 Col.210,000 10,000 10,000 10,000 10,000 10,000 10,000 10,000 Blended ATA2.6811.7631.4321.2261.1041.0601.0301.155LDF11.5544.3092.4441.7061.3921.2611.1

46、891.155Loglikelihood-0.9363-1.0052-0.8252-0.5260-0.2687-0.2535-0.02900.0000-3.844126Posterior WeightsBayesian Updating of ProbabilitiesLog-LikelihoodDifference in LLRelative LikelihoodOriginal Weights Revised WeightsAB=B-max(A)C=exp(B)DE=C*D/Avg(C)Fast-3.8401.0033.33%43.98%Medium-4.06-0.210.8133.33%

47、35.61%Slow-4.61-0.770.46433.33%20.41%27Credibility In Loss Development1.Sample Company Data First step would be to check for stability in the profiles and policy limit drift.Triangle observations:The lower attaching 400K xs 100K layer has a far more credible triangle than the 500K xs 500K layer.The

48、empirical tail factor generated by the 400K xs 100K layer also significantly longer than the empirical tail factor in the 500K xs 500K triangle.28Credibility In Loss Development400K xs 100K graphThe blue lines represent an approximate 90%confidence interval around the industry pattern.Similarly,we c

49、an fit the client data to a curve to see a similarly calculated 90%confidence interval in orange above.The client data has a slower development pattern than the industry data.500K xs 500K graphThe client data has a faster development pattern than the industry data.29ApplicationThe credibility weight

50、ed patterns are simply the dollar weighted average(utilizing the column 1 and column 2 figures)of the client/benchmark sections.30ApplicationThe same procedure is performed with the Slow and Fast benchmark patterns(Slow shown below).31ApplicationOur prior weights(33.33%)are adjusted to posterior wei

51、ghts to reflect the fact that the client data is most representative of the slow curve.The final pattern is a credibility-weighted average of the individual benchmark patterns weighted with the client data.32ApplicationSame process is followed for the 500K xs 500K layer.However,now we can use what w

52、e learned on the 400K xs 100K layer and begin with our apriori weights equal to the posterior weights from the previous slide.Since the 500K xs 500K triangle has limited credibility,we would utilize a larger scale parameter which will result in a final pattern that is close to the“slow”benchmark.33E

53、xperience Rating(400K xs 100K)Utilizes the credibility weighted LDFs.Also makes use of any limits drift adjustment.34Experience Rating(500K xs 500K)For the higher 500K xs 500K layer,the experience is volatile and not fully credible.In this case,the experience indication is credibility weighted with

54、an exposure rated relativity selection.35Measuring Benchmark Distortions:Three Year Pandemic and Heightened Inflation View36A.Benchmarking analysis framework Benchmarking components External forces disruptions pandemic 2020,inflation impacts 2021-22 Assessing confidence and avoiding overconfidenceB.

55、Tools to assess the disruption Frequencies with on-level premium Average reportings and settlements Loss development factors pre/during pandemic Closure ratios Adverse development-exanteC.Impact Analysis Experience through 12/31/2022 Holistic view:frequencies,severities,loss ratios focus on GL Impac

56、t on Commercial Lines Where to now?Measuring Benchmark Distortions:Three Year Pandemic and Heightened Inflation ViewAgenda37A.Benchmarking Framework Trends Severities,frequencies,exposures Ground-up and Excess Loss Development Factors Reporting and payment patterns Closure ratios Rate Changes Loss C

57、osts Ground-up and ILFs State/Hazard/Class Differentiations External forces disruptions Resulting expected loss ratiosGoal:Confident entry/exit decisions,anticipating competitive market cycle changes 38Benchmark Assessment MatrixEstimating Confidences Pre-Pandemic-IllustrativeSource:Adapted from IT2

58、 Intermediate/Advanced-CARe May 2014(JBuchanan)12345678TrendsState/Ground UpExcessLoss Devt FactorsHazard/SeverityFrequencyExposureSeverityFrequencyGround UpExcessSublineCasualtyllllPropertylllSpecialtyl9516 WhereRate ChangesLoss CostsExternal Loss RatiosIn thePrimaryReinsuranceGround-upI

59、LFsForcesPrimaryReinsuranceCycle?CasualtyllllllPropertylllllSpecialtyConfidence:GoodlMediumSomeMinimal39As part of an annual or quarterly Best Practices framework,after gathering all relevant internal and external information,it is useful to assess all actuarial benchmarking components.And how confi

60、dent you are in each.Some for example like LDFs and rate changes may feel quite confident,if no major disruptions.While others like ILFs may feel less confident in times of high and unknown social inflation and litigation financing impacts.Pandemic and Inflation Impact:Questions What are the base-li

61、ne expectations?How much have they been distorted?What does the recovery shape look like?What are the expectations for 2023/24?How confident are we in this assessment?40B.Tools to assess the disruption Review loss and premium triangles Calendar/accident quarter Loss development factor distortions Di

62、storted diagonals Frequency ratios Average severities Closure ratios Cumulative,available to be closed,incremental Adverse development-exante41Total General Liability Raw Data TrianglesReviewing overall GL triangles,focusing on pre and post Covid onset,can see even with relatively stable and increas

63、ing premium base,that claim counts are way down,but severities at a heightened level.Both significantly higher than longer term trends.Source:GL SOLM-Qtr at 12/31/202242Total General Liability Overall LDFsCan see lengthening impact on total LDFs,including affecting most recent evaluation of all acci

64、dent years.And affecting both total reporting and payment patterns.But the story goes much deeper than impact on LDFs.Source:SOLM-Qtr at 12/31/2022(SOLM-annual for 2016&prior)43Total General Liability Frequency Ratios-OLEPEven after on-leveling the premium used as an exposure base,the total ground-u

65、p frequencies remain significantly down,with no indicated reversal yet or reversion to normalcy through 12/31/2022.Would want to compare against any overall downward frequency trend including impacts of increasing deductibles and size of claim,before making any full assessments.Source:GL SOLM-Qtr at

66、 12/31/202244Average Qtr Severity Trends YtY through 12/31/2022 GL GU,XS 25k,CAuAverage severity trends are up significantly beyond normal long-term averages.For example,for Total GL,long term pre-pandemic severity trends were about 5.7%and about 7.7%since the start of the pandemic.Total CAu severit

67、y trends also increased by about 2%from before and after the start of the pandemic.45Source:GL and CAu SOLM-Qtr at 12/31/2022Total General Liability Closure Ratios#1Reviewing a standard closure analysis of cumulative closed to incurred claims,indicates that there still remains slower than average se

68、ttlements.Catchup to more normal levels has not yet occurred.Source:GL SOLM-Qtr at 12/31/202246Total General Liability Closure Ratios#2An alternative closure analysis of reviewing closed claims divided by available to be closed from prior quarter shows a similar pattern,but with a bit more catching

69、up done in the earliest accident quarters to longer term averages.Source:GL SOLM-Qtr at 12/31/202247Total General Liability Closure Ratios#3This closure ratio,which requires triangulation estimates to ultimate and using that as a base,can see a bit more clearly the impact of the onset of Covid in 20

70、20Q1 affecting most of the calendar quarters due to shutdown of claims activities and courts.Inventories are again starting to be cleared up.Source:GL SOLM-Qtr at 12/31/202248Total Commercial Auto and Personal Auto-Closure Ratios#1 and#349Commercial Auto-Paid IndemnityPersonal Auto-Paid IndemnityFor

71、 CAu and PAu,can see rather clearly the cumulative slowdown over 3 to 6 quarters to longer-term averages,and the attempts being made to catch up.There does appear to be some residual slowdown occurring in the first quarter evaluation even in the most recent 6 or so quarters.Claims departments“Robbin

72、g Peter to close Paul”?Source:CAu SOLM-Qtr at 12/31/2022&PAu SOLM-Qtr at 9/30/2022Calculating Ex-Ante latest 7 qtr VWAThis exhibit shows how“ex-ante”or reserve runoff calculations are produced.This calculation,which rolls back each of the LDF sets to estimate what would have known at the time,to giv

73、e one of the best actual vs expected early warnings of lengthening LDFs.In the highlighted cell,the 2.108 LDF experienced for 2022Q3,is higher then the prior 7 qtr LDFs average of 2.001,producing adverse development of 22.5M for that cell.Source:GL SOLM-Qtr at 12/31/2022All GL Reserve Run-off Test 1

74、2/31/2022-Ground-upGeneral Liability ExAnte Reserve RunoffComparing to initial selected excess loss ultimates at 3 months using a mechanical 7-year average,produces adverse development across all quarters since 2020Q2.51Source:GL SOLM-Qtr at 12/31/2022All GL Reserve Run-off Test 12/31/2022 BI xs 25k

75、General Liability ExAnte Reserve RunoffSimilar to total GL GU,BI claims excess of 25k have for developed adversely for almost all quarters since 2020Q2.52Source:GL SOLM-Qtr at 12/31/2022C.Main Impacts Severities up beyond normal increases Frequencies down significantly below pre-pandemic,also below

76、normal base-line decreases Adverse development Delayed closures and catch-up settlements 1st evaluation claims:maybe“Robbing Peter to Close Paul”Increasing loss ratios Concern for future:if average severities remain high,frequencies revert closer to pre-pandemic,closure catch-up continues to occur,a

77、nd adverse development continues loss ratios could significantly increase soon as the pandemic abates53Recent Trends Impacted by Covid/Inflation Total GL2017 through 2022 Year-End-NominalGL showed a 29%frequency reduction in 2020 due to Covid,with similar depressed level in 2021 and further reductio

78、n in 2022.Average severities increased in 2020,2021,and 2022 by about 11%each year,compared to the 6-7%trends that we had been seeing in the past.Questions:how long will it take for the frequencies to return to normal or new normal levels?how much of this heightened inflation is expected to continue

79、 into 2023 and beyond?54NB:mechanical selection for LDFs of last 7 qtr VWA used in projections from GL SOLM-Qtr at 12/31/2022.No tail beyond 2017 supplied.Indemnity Only uses ISO MarketWatch 6/30/2022 rate changes Recent Trends Impacted by Covid/Inflation Total GL2017 through 2022 Year-End On-levelO

80、n an On-Level basis,GL showed a 28%frequency reduction in 2020 due to Covid,with a slight increase in frequency in 2021 and similar level in 2022.This slight frequency increase coupled with the 11%severity increases in recent years has led to increasing on-level loss ratios to about pre-pandemic lev

81、els in 2022.If severities continue to stay high and frequencies return closer to pre-pandemic levels,loss ratios may continue to rise.55NB:mechanical selection for LDFs of last 7 qtr VWA used in projections from GL SOLM-Qtr at 12/31/2022.No tail beyond 2017 supplied.Indemnity Only uses ISO MarketWat

82、ch 6/30/2022 rate changes As observed previously,in 2020 there was a significant frequency reduction driving a significant loss ratio reduction.For severity,we see YTY changes significantly higher than in the past with increases above 10%in 2020-2022.This large increase in severity,paired with a par

83、tial rebound in frequency led to an increase in loss ratio in 2021 and 2022 to higher than pre-pandemic levels.NB:mechanical selection for LDFs of last 7 qtr VWA used in projections from CAu SOLM-Qtr at 12/31/2022.No tail beyond 2017 supplied.Indemnity Only uses ISO MarketWatch 6/30/2022 rate change

84、s Recent Trends Impacted by Covid/Inflation Total CAu2017 through 2022 Year-End On-LevelRecent Trends Impacted by Covid/Inflation Total CP2017 through 2022 Year-End On-LevelCP showed a 14.5%on-level frequency reduction in 2020 due to Covid,with similar depressed level in 2021 and further reduction i

85、n 2022.Average severities increased in 2021 and 2022 by about 25%and 12%respectively,much higher than in prior years.This led to on-level loss ratios getting to higher than pre-pandemic levels in 2022.57NB:mechanical selection for LDFs of last 7 qtr VWA used in projections from CP SOLM-Qtr at 12/31/

86、2022.No tail beyond 2017 supplied.Indemnity Only uses ISO MarketWatch 6/30/2022 rate changes Recent Trends Impacted by Covid/Inflation GL Restaurants&Bars2017 through 2022 Year-End On-levelGL Restaurants and Bars was one of the most impacted class groups for GL over the past 3 years.On-Level frequen

87、cy fell more than 40%in 2020 due to the pandemic,but then saw a 12%recovery in 2021 with slight increase again in 2022.Severity saw a significant increase in 2021 of 30%with similar level in 2022.These frequency and severity impacts led to a sharp drop in on-level loss ratio in 2020 with increases b

88、ack to pre-pandemic levels in 2021 and 2022.58NB:mechanical selection for LDFs of last 7 qtr VWA used in projections from GL SOLM-Qtr at 12/31/2022.No tail beyond 2017 supplied.Indemnity Only uses ISO MarketWatch 6/30/2022 rate changes Residual Trends(ART)GL Restaurants&Bars(incl Covid Adjustment)No

89、minal ultimate loss ratios were adjusted by various development,trend,and on-leveling adjustments.The goal of this analysis is to end up with a straight line of loss ratios(black line on graph)that only exhibit random variations around a mean(process variance).Any remaining trend would be due to not

90、 including enough adjustments(coverage changes,risk management improvements,one-time plateau events in either direction(Great Recession),etc.These are similar adjustments that are relevant to Rate Change Method 5.This is especially important in 2020/2021 and beyond as if it is believed that the Covi

91、d Pause,with its impact on economic and loss activity,will eventually revert back to normal,then there would need to be an explicit adjustment for both the numerator and denominator.This same analysis can be done on other metrics such as frequencies,excess layers,partial loss trends,etc.In this exam

92、ple,there is a generally negative positive trend in the adjusted loss ratios of about 2.65%,with some moderate downward trend from 2015-2017 and then some moderate upward trend since 2017.Therefore,we can conclude that there must be some loss or premium influences that have not been considered.Sourc

93、e:SOLM-Qtr at 12/31/2022(SOLM-annual for 2016&prior)Benchmark Assessment MatrixEstimating Confidences Post Pandemic-IllustrativeSource:Adapted from IT2 Intermediate/Advanced-CARe May 2014(JBuchanan)12345678TrendsState/Ground UpExcessLoss Devt FactorsHazard/SeverityFrequencyExposureSeverityFrequencyG

94、round UpExcessSublineCasualtylPropertyllSpecialty9516 WhereRate ChangesLoss CostsExternal Loss RatiosIn thePrimaryReinsuranceGround-upILFsForcesPrimaryReinsuranceCycle?CasualtyllPropertyllSpecialtyConfidence:GoodlMediumSomeMinimal60Your post pandemic assessment of parameter confidence sho

95、uld reflect any unknowns that may occur as to frequency drop reversals,closures back to normal,inflations impacts,adverse development,size of claim impacts,etc.The confidence levels of some attributes may still remain high,like well monitored rate changes.But others in particular longer tail line fr

96、equencies,excess severities,ILFs,and LRs may suffer due to the additional unknowns.Some of the benchmarks may in essence become“couchmarks”.Mechanical Indication of Trends 12/2022 Post PandemicMetrics for Pre Covid,First 2 Covid years90%CI(Responses)Actualand Heightened Inflation yearLowerUpper1.Tot

97、al GL Annual Severity Change 2015-20194.2%2.Total GL Annual Severity Change .4%3.Total GL Severity Change .7%4.Total GL Annual Frequency Change 2015-2019-4.1%5.Total GL Annual Frequency Change 2019-2021-12.9%6.Total GL Frequency Change 2021-2022-0.2%7.Total CAu Annual Severity

98、Change .3%8.Total CAu Annual Frequency Change 2019-2022-5.8%9.Total CP Annual Severity Change .9%10.Total CP Annual Frequency Change 2019-2022-6.5%61 Actual annual trend indications using SOLM-Qtr mechanical LDFs last 7 quarters Frequency indications use on-level premium 12/31/

99、2022 as base No part of this presentation may be copied or redistributed without the prior written consent of Insurance Services Office,Inc.This material was used exclusively as an exhibit to an oral presentation.It may not be,nor should it be relied upon as reflecting,a complete record of the discussion.Insurance Services Office,Inc.,2023http:/ and Feedback63