Territory Analysis U d t Updates to the Traditional Methods t th - PowerPoint PPT Presentation

Territory Analysis U d t Updates to the Traditional Methods t th T diti l M th d CAS RPM March 22, 2011 Sandra Ross, FCAS, MAAA, CIC Experience the Pinnacle Difference!

Agenda � State of territory definitions today � Reasons for modifying territories R f dif i i i � Considerations � Processes � Processes � Data � Availability and collection � Capping � Smoothing � Combining C bi i � Clustering � Selecting g 2

Current Definitions � Current sets � Often outdated � Often outdated � Uniform across product/policy � Less than optimal match of exposure � Developed in less than optimal ways � Technique � Basis for definitions � Tweaked over time � Possibly leading to: � Misclassification Misclassification � Misinterpretation of other factors � Adverse selection 3

Changing Landscapes � Anyone else notice where there used to be a crop planted there is now a subdivision or a strip mall? there is now a subdivision or a strip-mall? � Over a 20-year period (1970-1990), the 100 largest urbanized areas in the United States sprawled out over an additional 14,545 square miles. That is more than 9 million acres of natural habitats, farmland and rural areas that have been converted to subdivisions shopping centers etc converted to subdivisions, shopping centers, etc. � What has happened since 1990? � Increased population density � Increased vehicle density � More new homes � Less populations in cities, more abandoned homes L l ti i iti b d d h 4

Indianapolis Pop Chg 4/1/00 - � 14 largest city in the 14 largest city in the County County City City Population Population 7/1/09 7/1/09 Marion Indianapolis 785,597 0.5% U.S. according to 2010 Remainder 105,282 33.2% Census Total 890,879 3.5% B Boone 56 287 56,287 22 1% 22.1% � 3 rd largest in the Hamilton 279,287 52.8% Hancock 68,334 23.4% Midwest Hendricks 140,606 35.1% � One of the fastest Morgan 70,876 6.3% Johnson 141,501 22.8% growing regions in the Shelby 44,503 2.4% All Other All Other 4 730 840 4,730,840 26.2% 26 2% Midwest. d Indiana 6,423,113 5.6% http:\quickfacts.census.gov as of 3/3/11 5

Geographic Rating � Goal is to isolate variables to explain risk � Use variables to segment property insured, coverage U i bl i d selections and insured characteristics � Territory is used to explain differentiation in risk not picked up Territory is used to explain differentiation in risk not picked up by other rating variables and to explain geographic differences � Geographic difference can be due to � Population and vehicle density � Theft/crime rates � Hazards � Hazards � Differences in mix of business � Properties insured � Vehicles driven 6

Deriving Territory Definitions � Territory definition Territory definition analysis is driven by a lot of numbers, analysis, statistical techniques, etc. � However, there are still many areas where actuarial judgment actuarial judgment plays an important roles 7

Upfront Considerations � State regulations � Available data � Ex. OH must rate by city y y � Internal � Types of analysis � External � Historical events � Total state/line � By coverage/peril B / il � Desire to remove or adjust D i t dj t for them � Contiguous or not � Specific concerns � Basis for analysis y � Management � Zip Code � Sales � Census Tract � Competitive pressures � Competitive pressures � Other Other and competitor � System capabilities boundaries 8

Increased Segmentation in Definitions � Auto � Territories by coverage � Territories by coverage group � Territories by peril for Comprehensive y p p � Home � Territories by peril � Territories by peril group � Territories by peril group � Territories by coverage � Loss Components � Pure Premium i � Frequency � Severity 9

External Data � Historical Insurance Industry data � ISO � ISO � HLDI � Hazard data providers � Census and other governmental d h l data � Housing density � Traffic density � Crime statistics � Accident statistics � Accident statistics � Home values � Catastrophe Model Output 10

Basis for Data � Statistics by S a s cs by � County � Zip Code � Census Block � Census Tract � Address � Address � Location � Longitude g � Latitude � Adjacency 11

Industry Data � ISO � HLDI ISO HLDI � Auto � Auto � By coverage � Free to members � Cat indicators � More than 25 years � Home � By coverage � By cause of loss � By cause of loss � Comprehensive broken � By coverage into fire, theft, glass and � Cat indicators other other � Data by zip � Data by zip 12

How much data is necessary? � Non-catastrophe � Generally 5-10 years depending on credibility of data G ll 5 10 d di dibilit f d t � Catastrophe � Much longer periods if available � Much longer periods if available � HLDI provides over 25 years � Cat Modelers � Represents hundred’s of years of experience and forecast of future events 13

Accounting for Catastrophes � Company data � Cat model data p y � Usually cat and x-cat � AIR and RMS models available � Wind/Hail models � May not coincide with M t i id ith � Winter storm models industry coding � Hazard data � ISO � Sinkholes Sinkholes � Cat and x-cat data � Distance to coast � HLDI � Comprehensive other than Fire, Theft and Glass 14

Data Adjustments to Consider � Eliminate effect from all other rating variables � Capping C i � Smoothing � Possible clustering of partial components to add further � Possible clustering of partial components to add further smoothing (i.e. cluster cat component before combining with non-cat) � Normalizing � Inflationary adjustments � Weighting together of various data sources W i hti t th f i d t 15

Modeling Output � Cat modeler output Cat modeler output can provide very different results 16

Historical Experience � Model results can also be quite different from historical experience 17

Capping � Used at various places in process � Average rating factors A i f � Large individual losses � Large events or catastrophes � Large events or catastrophes 18

Territories by Coverage and Peril � Separate definition sets by coverage or peril provide more optimal rate classification and factors p � Geographic location may not uniformly impact coverage or peril � Similar process for frequency/severity separate analysis � There are ways to develop territory sets by coverage or peril and combine the sets into one consolidated set � May ease systems implementation � May ease systems implementation 19

Auto Components Liability & Liability & Comprehensive h i Collision Company Company Industry Industry Non Cat Non-Cat Cat Cat Company Industry Company Industry 20

Home Components Non-Weather Weather Liability Fire, water, theft Wind, hail, lightning and water other property other property Company Industry Non-Cat Cat Company Industry Company Industry Cat Modelers Winter Wind/Hail Storm Storm 21

Average Rating Plan Factors � Adjustment of historical � Rating variables such as: djus e o s o ca a g a ab es suc as experience to a common � Age of driver level � Insured Value of Homes � Removes distributional � Protection Class biases from the � Deductible underlying data underlying data � Discounts � Discounts � Claims surcharge � Assisted by generalized linear models 22

Smoothing � Data at the basic element level lacks “credibility” a a a e bas c e e e e e ac s c ed b y � Smoothing process allows inclusion of more localized data rather than statewide information � Results in a rate or rate relativity for each individual zip code based upon the data within that zip code modified as necessary to include a significant number of t i l d i ifi t b f observations 23

Smoothing � Key smoothing variables � Predictive value of local data P di ti l f l l d t � Identification of complement data � How many observations are required to smooth y q � How far to allow smoothing search to continue � Many equations are available to combine local data with surrounding information d f � Exposure Weighted Average � Straight Line Declining Distance formula � Straight Line Declining Distance formula � Squared Declining Distance formula � Werland-Christopherson Method 24

Smoothing Considerations � State Borders and Corners � Use of smoothing across state boundaries � Potential separate smoothing of urban and rural areas � Distance based smoothing process or contiguous based b d h b d smoothing process 25

“Neighboring” 26

Smoothing Impact Unsmoothed data Smoothed data 27

Smoothing Impact 10 10 20 30 30 40 50 50 60 70 70 80 0 1 35 69 103 137 171 205 240 Input Data 274 308 342 376 410 444 478 512 546 580 Smoothed Data 614 648 682 716 750 784 818 852 886 920 954 988 28

Clustering Process � Grouping of areas based on similarity of oup g o a eas based o s a y o statistics � Begin with most detailed data and combine – bottom up approach � Comparison can be based on percentage or value differences l diff � Contiguity can be a constraint � Summitt TM � Summitt TM 29