Mining Insurance Data to Promote Traffic Safety and Better Match Rates to Risk

1
Mining Insurance Data to Promote Traffic Safety
and Better Match Rates to Risk

• 2002 CAS Seminar on Ratemaking
• Greg Hayward

2
A Photo of the Data Mart
3
The Data Mart Opens New Opportunities !!
37 Terabytes of Data Storage Capacity 3.7 x
1013 bytes
4
Data Mart Issues

• Data What, When, How Often, How Long
• Data Dictionary Source, Codes, Edits, Validation
• Optimize Access Spends
• Quality Control

5
Main Data Mart Menu
6
A Simple Data Mining Illustration

• Select some variables to explore
• Select some target variables
• Query the Data Mart
• Download the data into an Excel Pivot Table
• Lets look at some real data in Excel

7
Step-Wise Regression

• Identify
• Transform
• Search
• Explore
• Quality Control
• Develop
• Derive
• Run Regression

8
Data Mart Provides for More Sophisticated Analysis

• Interactive Multi-Variable Rate Factor Analysis

9
Concept

• Multi-variable approach produces indicated
  factors simultaneously rather than producing each
  rating variable set of factors successively one
  after another
• Multi-variable approach takes into account the
  interaction between the different rating
  variables

10
Method

• Collect detailed data from the Data Mart
• Calculate single way indicated factors for each
  rating variable
• Compare the single way indicated factor to the
  current factor and use this to adjust the premium
  for each combination of rating variables

11
Method (Contd)

• Once we have the adjusted premium for each
  combination of rating variables, sum up the
  premium for each specific rating variable and
  recalculate the indicated factor for that
  variable
• Keep repeating the steps above until you have
  equal loss ratios for each rating variable

12
Putting the Technology to Use

• Vehicle Safety
• Dangerous Intersections
• Child Passenger Safety
• Teenage Driver Safety

13
Vehicle Safety Discount

• Collect data for Personal Injury Protection and
  Medical Payments Coverages for each make , model
  and body style of vehicle using Vehicle
  Identification Numbers (VIN)
• Loss ratios at uniform rate levels are calculated
  for each vehicle type by model year and compared
  to the average loss ratio for all vehicles of
  that particular model year to calculate an index

14
Vehicle Safety Discount (Contd)

• Resulting Vehicle Safety discounts assigned to
  each make, model and body style of vehicle were
  developed from real world data as to how well all
  the safety features of the vehicle, in
  combination, protected the occupants
• Collecting data in this detail allows us to more
  equitably distribute premiums to our
  policyholders and allows us to share this
  information with the auto manufacturers to
  promote safer vehicles

15
Dangerous Intersections

• Develop a query to sort the millions of claims
  into accident location
• Focus on the intersections with the most claims
  to search for additions and deletions
• Score each claim based on the property damage
  severity and whether the crash involved injury
• Adjust the score to a common baseline using the
  companys market share of vehicles injured (to
  ZIP Code level)
• Rank order the intersections by score
• Share results with public, community leaders, and
  traffic safety engineers.

16
Child Passenger Safety

• Largest single research project devoted
  exclusively to pediatric motor vehicle injury
• Partnered with the Childrens Hospital of
  Philadelphia
• Multi-disciplinary research insurance,
  medicine, biomechanics, engineering, health
  education, behavioral science
• Combines our insurance data with in-depth
  interviews, on-site crash investigations, and
  computer crash simulations
• Share results with parents, medical providers,
  auto manufacturers, and NHTSA

17
Teenage Drivers

• Attempt to study experience and attitude
• Experience time, practice, and exposure to
  driving situations
• Attitude patience, not being aggressive, not
  taking unsafe risks, not becoming distracted
• Worked with American Driver and Traffic Education
  Association
• Put Teenage Drivers through agent/parent/driver
  team program
• Mine the data from the experiment in great detail

18
(No Transcript)
19
(No Transcript)