Docstoc

Trends in Predictive Modeling Topics

Document Sample
Trends in Predictive Modeling Topics Powered By Docstoc
					                                                                              7/5/2011




         Trends in Predictive Modeling


                       June 30th, 2011
                    Gary Wang, FCAS, MAAA
                    G    W




                                        Experience the Pinnacle Difference!




Topics

 Data
 Applications of predictive analytics
 Predictive modeling techniques




                                                                      2




                                                                                    1
                                                            7/5/2011




                          Data




                                                        3




Auto Data – Traditional Factors


  Driver                         Policy
    Class d
    Cl code                        Limits/deductible
                                   Li it /d d tibl
    Vehicle use                    Multi-car discount
    Mileage to and from            Territory
    work                         SDIP
    Annual mileage                 Prior chargeable
  Vehicle                          accidents
    Symbol                         Prior violations
    Model year
    Passive restraint

                                                        4




                                                                  2
                                                         7/5/2011




Auto – Expanded Use of New Risk
Characteristics

  Driver                   Policy
    Age/gender/marital
    A / d /       it l       Prior limits
                             P i li it
    status                   Full coverage
    Education                Driver/vehicle matrix
    Occupation               Payment plans
  SDIP                       Household composition
          non chargeable
    Prior non-chargeable     Prior lapses
    accidents                Prior endorsements
    Time since               Insurance score
    accident/violation


                                                     5




Homeowners Data – Traditional Factors


  Non-coverage based       Coverage based rating
  rating factors           factors
    Amount of insurance      Deductible
    Protection class         Loss settlement
    Construction type        Additional limits
    Protective device
    discount
    Territory
    Number of families



                                                     6




                                                               3
                                                      7/5/2011




Homeowners - Expanded Use of New Risk
Characteristics

  Policy                      Homeowner
     Prior l i history
     P i claim hi t             A
                                Age
     Payment plans              Education
     Payment history            Occupation
     Lapses/cancellations       Insurance score
     Endorsement history
     Presence of children
     Number of mortgages
     Number of adults
     Pools
     Trampolines
                                                  7




New Data

  Telematics
  Vehicle characteristics
  Vehicle ownership history
  Property characteristics
  Property Hazard Elements




                                                  8




                                                            4
                                              7/5/2011




Telematics

  Data
     Mileage driven
     Time of day
     Speed
     Braking
     Sharpness of turns
     Where you drive
   pp
  Application
     Auto insurance rating
     Encourage safety



                                         9




The Current Usage-Based Insurance
Environment in the U.S.

  Progressive Insurance – SnapshotSM
     Now in over 30 states
  Allstate Insurance – Drive WiseSM
     Illinois – Launched December 2010

  Mileage Based Discounts:
     GMAC – 35 states
     State Farm – 5 states
     Auto Club – California




                                         10




                                                    5
                                                                                                                                                                                                                                              7/5/2011




                              Vehicle Characteristics
                               Data
                                          Daytime running lights                                                                 Cylinders
                                          Electronic stability control                                                           Driving wheels
                                          Weight                                                                                 High performance code
                                          Engine size                                                                            Transmission
                                          Segmentation                                                                           Wheel base
                                          Cubic inch displacement                                                                Height
                                          Body type                                                                              Width

                                pp
                               Applications
                                               Liability symbols
                                               Enhanced physical damage classification
                                               New model classification



                                                                                                                                                                                                       11




                                                           Example Company Vehicle Classification Job
                                                            Run 2 Model 1 - Collision Pure Premium - Smoothed standard risk premium model (single claim type)


                            1.2                                                                                                                                                                        2000000
                                                                                  149%

                            0.8                                                                                                                                                      82%
                                                                                                 62%                                          60% 65%
                                                                                                         40% 40%           43%                                                 48%
                                                                            36%           31%                                                               32%                                        1500000
                            0.4                                                                                                                                          18%
                                                                                                                                        11%                                                       9%
                                                                                                                                                                                                                 E x p o s u re (y e a rs )
L o g o f m u lt ip lie r




                                                      0%
                                   -12%                           -8% -5%
                              0           -18% -17%        -14%                                                     -15%                                                                   -15%
                                                                                                                                 -21%
                                                                                                                                                                                                       1000000
                            -0.4


                            -0.8
                                                                                                                                                                                                       500000

                            -1.2                                                                                                                                  -75%


                            -1.6                                                                                                                                                                       0
                                    1      2    3     4     5     6    7     8       9     A       B     C     D     E     F       G     H     J     K       L     M     N     P     R      U     X

                                                                                                    MSEGMENTATION_CODE



                                                                                  Onew ay relativities       Unsmoothed estimate        Smoothed estimate


                                                                                                                                                                                                       12




                                                                                                                                                                                                                                                    6
                                                             7/5/2011




Vehicle History - CARFAX

  Vehicle damage history
  Emissions test results
  Odometer readings
  Lease history
  Use history (rental, commercial)
  Number of prior owners
  Ownership length




                                                        13




Number of Owners


  2.00
  1.80
  1 80
  1.60
  1.40
  1.20
  1.00
  0.80
  0.60
  0.40
  0.20
  0.00
           1          2           3          4     5+

                 BI   PD   COLL       COMP   MED


                                                        14




                                                                   7
                                                           7/5/2011




Potential Damage

                                        1.10
 1.10
 1.08
 1.06
 1.04
 1.02                1.00
 1.00
 0.98
 0 98
 0.96
 0.94
                    N                   Y


                                                      15




Property Characteristics
  Property                     Real Estate elements
     Total Living Area            Mortgage value
     Year Built                   Market value
     Number of Stories/Style      Interest Rates
     Number of Families           Loan terms
     Foundations
     Finished Basements
     Exterior Wall
     Roofing
     Number of Baths
     Fireplaces
     Swimming pools
     Trampolines


                                                      16




                                                                 8
                                           7/5/2011




Total Living Area - Severity




                                      17




Number of Mortgages

                               1.32
  1.40

  1.20
             1.00
  1.00

  0.80

  0.60

  0.40

  0.20

  0.00
             0                 1


                                      18




                                                 9
                                                 7/5/2011




Hazard Elements Available

  Distance to Fire Station
  Distance to Coast
  Flood zone
  Brush Fire
  Earthquake
  Elevation
  Sink hole




                                            19




Data Elements Being Considered & Rejected

  Auto
     Passive restraint discount
     Anti-theft discount
     ABS discount
     Vehicle use
     Annual mileage
  Home
            p
     Public protection class
     Protective device discount
     Multi-line discount



                                            20




                                                      10
                                            7/5/2011




                       Applications




                                       21




Applications of Predictive Modeling

  Target market analysis
  Underwriting
     Straight through processing
     Target report ordering
  Pricing
     Classification plan development
     Territory definitions
     Vehicle classification




                                       22




                                                 11
                                                                         7/5/2011




New & Expanded Applications

 Automated underwriting/re-underwriting
 Target marketing
 Customer Response Analyses
    New business conversion
    Retention
 Claims
 Price optimization




                                                                   23




Underwriting
                                          New Business
                                           Underwriting
                                                           Renewal
 Analyses                      Data                       Underwriting


    Selection/rejection
    S l ti / j ti                 Hi t i l underwriting
                                  Historical d      iti
    Action indicators             actions
    Vehicle inspection/re-        Underwriting criteria
    inspection                    MVR report data
    Home inspection/re-           CLUE report data
    inspection                              p         p
                                  Home inspection reports
                                  External data feeds
                                      MSB
                                      Carfax
                                      Demographic

                                                                   24




                                                                              12
                                                                                                      7/5/2011




Selection/Rejection
 Higher Cost                                 Increased Risk
 Risks, Lower                                   Appetite
  Credibility                                                              Traditional Risk
                                                                               Targets
                                                                                   g
                     Auto Underwriting Score




                  Present Factor   Indicated Pure Premium Relativity   Proposed Factor           25




Marketing                                                  Marketing           Quote


                                                                       Sale              Cross
                                                                                          Sell
Analyses                                      Data
  Model the likelihood of a                        Internal company information
  potential risk contacting company                External demographic
  for a quote                                      information
  Measure characteristics of                           ZIP code level
  shoppers/quoters                                     Individual/household level
  Measure likelihood of insureds                       demographics
  responding to marketing                          Credit profiles
  initiatives
                                                   Marketing efforts
  Measure the likelihood of a risk
  responding to a cross-sell contact               Focus groups
  Goal: Identify insureds to target                Shopping incentives



                                                                                                 26




                                                                                                           13
                                                                                                7/5/2011




Cross Sell Opportunities
                         Homeowner Experience by Prior Auto Losses

                       1.40                                               1.257
                                                        1 119
                                                        1.119
                       1.20        1.000
   eowner Relativity




                       1.00

                       0.80

                       0.60
                                                   Tradition: Cross-sellingg
Home




                       0.40                           is always good!
                       0.20                         Reality: is this always
                                                            good?
                       0.00
                                   0                    1                  2
                                            Prior Auto Losses
                                                                                           27




Customer Response Analysis

                       Quoting analysis: analysis of the likelihood of a prospective
                       insured obtaining an insurance quote from you
                       Conversion analysis: analysis of the likelihood of a insured
                       that has received a quote purchasing insurance from you
                       Retention analysis: analysis of the likelihood of a current
                       insured renewing with you
                       Cross sell analysis: analysis of the likelihood of a current
                       insured purchasing additional products with you

                                   Sale                                            Cross
                                                            Retention               Sell
                                            Ongoing
  Quote                                     Servicing                   Customer
                                                                         Service


                                                                                           28




                                                                                                     14
                                                                                                                           7/5/2011




                              Customer Response Model Characteristics
                              Traditional Rating Factors                       Agent/Distribution Channel
                                     Class                                     Issues
                                     Territory                                     Satisfaction with Agent/Service
                                     Limit
                                     Li it                                         Distance t A t
                                                                                   Di t      to Agent
                                     Insurance Score                               Urban/Rural
                                     Claims history                                Independent vs. Captive vs.
                                     Violation history                             Direct

                              Account Characteristics                          Market Conditions
                                                                                   Competitive Position
                                     Number of Years Insured
                                                                                   Inflation
                                     Number of Policies
                                                                                   U/W Cycle
                                     Account Size
                                                                                   Reinsurance Pricing
                                     Renewal Pricing Change /
                                     New Business Price                            Market Capitalization
                                     Difference                                    Brand Value (company &
                                                                                   competitors)
                                                                                                                      29




Demand Curve

                                                                 Rate Change


                              1.80

                              1.60

                              1.40
   ative Renewal Likelihood




                              1.20

                              1.00

                              0.80

                              0.60
Rela




                              0.40

                              0.20

                              0.00
                                       1   2    3   4    5   6   7     8       9   10   11   12   13   14   15   16




                                                                                                                      30




                                                                                                                                15
                                                                                                                             7/5/2011




     Claims

                                                                       Adjustment/
   Occurrence                           Report                                                        Settlement
                                                                       Development

•Occurrence Characteristics     •Est. claim settlement value     •Claim development                 •Likelihood of reopen
•Claim fraud                    •Claim assignment                •Claim service providers           •Salvage/subrogation
                                •Early warning indicator         •Claim adjustment procedures       •Customer satisfaction
                                •Estimated cycle time            •Fraud
                                                                 •Claim procedures
                                                                 •Lawsuits/Attorney Involvement


                      •Reporting Lag                    •Contact Lag                     •Settlement Lag
        Data
           Geography (State or Regional Courts)
                 (Inflation,
           Time (Inflation Settlement Lags)
           Claimant Characteristics (Age, Class)
           Insured Characteristics (Vehicle Weight)
           Attorney Involvement
           Preferred Claim Network (Medical, Glass, Auto
           Repair, Attorney)
           Other Claims Features (Arbitration/ADR,
           Settlement Lag)
                                                                                                                   31




     Likelihood of Large Loss
     Decision Tree – Rules Based Approach




                 WHERE HSPTLTX 1, 2 AND HIGHINJ 25, 7... AND UNABLEDY
                  >= 3.5 AND COVPERS 100000, 30000... AND HIGHINJ 22,
                               15... AND LAWSUIT 2, 3




                                                                                                                   32




                                                                                                                                  16
                                                                                     7/5/2011




 Price Optimization
                                                   Customer
             Pricing                                Service
                                                                        Cross
                                                                         Sell
                           Sale

                                       Ongoing
Marketing     Quote                    Servicing      Retention

                        Underwriting
                                                       Claims


                                                            Re-
                                                         underwriting




                                          ?
                                                                                33




 Benefits of Optimization

    More complete picture of your current and prospective
    customers
    More complete picture of pricing impacts on policy retention,
    conversion and premium
    Better specified pricing and financial models
    Increased focus on program stability and profitable growth –
    longer term view
        g      p     g              y                          y
    Integrates pricing more directly with the entire business cycle




                                                                                34




                                                                                          17
                                                             7/5/2011




                        Techniques




                                                        35




Current Practice

  Generalized Linear Modeling (GLM) is still standard
     Separate Frequency and Severity Models
     Auto – models by coverage
     Homeowners – models by peril
     Loss Ratio Models with Tweedie distribution


  Advantages
     Established and tested methodologies
     Ease of analysis and presentation
     Regulatory acceptance of methodology



                                                        36




                                                                  18
                                                                   7/5/2011




Additional Techniques

 Clustering/Segmentation
 Principal Components
 Decision Trees
 Neural Networks
 Ensemble




                                                              37




Clustering/Segmentation

 Unsupervised classification technique
 Focuses on input variables
 Groups data into set of discrete clusters or contiguous groups
 of cases
 Example: group customers into segments for purposes of
 marketing campaigns
 Can be used as a dimension reduction technique




                                                              38




                                                                        19
                                                                       7/5/2011




Penetration by Cluster Segment




                                                                  39




Principal Components

  Mathematical transformation of input variables
  Calculated from the correlation matrix of the input variables
  Can be used as a dimension reduction technique, creates a
  summarized version of the inputs to use in models




                                                                  40




                                                                            20
                                                                                      7/5/2011




 Decision Tree Process
                                                              Subpopulation 3




                                                        Subpopulation 4
                              Subpopulation 1

        Entire
      Population                                        Subpopulation 5


•Y = dependent                Subpopulation 2
variable
 X independent
•X1 = i d    d t                                               Subpopulation 6
variable 1             • Every potential split is evaluated
•X2 = independent      • First split will be split that
variable 2             optimizes split criterion
•Xn = independent      • Process is repeated until no
variable n             additional splits can be made

                                                                                 41




 Neural Networks

     Target layer regression model on a series of derived input,
     called hidden units
     Hidden units (or layers) are regressions on the original inputs
     Target and hidden layers both have activation functions

                    Input layer      Hidden layer Target layer
                      X1                 H1

                      X2                 H2
                                                         Y
                      X3                 H3

                      X4                 H4


                                                                                 42




                                                                                           21
                                                                  7/5/2011




Ensemble

  Creates new model by combining probabilities from multiple
  models
  Produces more accurate results than individual models to the
  extent they disagree
             Decision
              Tree




                                         Final
           Regression     Ensemble     Prediction




             Neural
             Network




                                                             43




Lift Chart Comparison




                                                             44




                                                                       22
                                                            7/5/2011




                     Visit us at pinnacleactuaries.com




Thank You for Your Attention



                 Gary Wang, FCAS, MAAA
                            (309) 807-2331
               gwang@pinnacleactuaries.com




                      Experience the Pinnacle Difference!




                                                                 23

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:7
posted:2/25/2012
language:
pages:23