Discussions on

1
Discussions on Decomposing Automobile Insurance
Policy Buying Behavior Evidence of Adverse
Selection byChu-Shiu Li, Chwen-Chi Liu and
Jia-Hsing YehTong YuUniversity of Rhode
Islandtongyu_at_uri.eduARIA, August 6, 2007
2
Summary

• Issue
• To present evidence on the presence of adverse
  selection
• More specifically, to see if there is an positive
  relation between risk and insurance purchase
• Data
• Coverage and claim information of Taiwan auto
  insurance in years 2002 and 2003, facilitating
  two sets of analyses
• High-coverage policy (comprehensive policy)
  versus low-coverage policy (collision only)
  both without deductible
• Policy without deductible versus policy having
  deductible

3
Summary

• Testable Conditions
• A positive link between insurance claims and
  subsequent coverage
• A negative link between insurance claims and
  subsequent deductible choice
• Specific Cases favoring Adverse Selection
• 1. L in year t, no loss, L in year t1
• 2. H in year t, no loss, L in year t1
• 3. L in year t, loss, H in year t1
• 4. H in year t, loss, H in year t1

4
Summary

• Results
• T 6 Prob(LC in 03LC in 02) is positively
  related to the No_Claim dummy of 2002
  (NoClaim_02)
• T 7 Prob(LC in 03HC in 02) is negatively
  related to NoClaim_02
• T 8 Prob(HD in 03HD in 02) is positively
  related to NoClaim_02
• T 9 Prob(HD in 03LD in 02) is negatively
  related to NoClaim_02
• Results are obtained after controlling for some
  characteristics of insured and auto, e.g., age,
  gender, car age, expected losses of a
  policyholder, etc
• Carefully describe the procedure to compute
  expected loss, e.g., ENoClaim_02

5
Minor Suggestions

• Also look at the group having high coverage in
  2003
• Perform an unconditional test examining coverage
  choice and prior-year claim experience
• Need discuss the benefit of decomposing year t
  insured type
• Compare the results across various groups

6
Major Issue

• Risk ? Loss Experience

7
Major Issue

• Risk ? Loss Experience
• Loss experience is not private information to
  policyholder. It is available to insurers as well
• Hard to conclude the finding is supportive to
  adverse selection
• Test against alternative hypotheses learning and
  habit persistence

8
Direct Test on Adverse Selection

• Develop a model to compute the price of each
  insurance contract in year t1
• Look at insurance purchase in the over- and
  under-price groups respectively
• Underlying assumption Risk is quantifiable
• Feasible??

9
Solution 1 Estimate Risk

• Get claim information for more years. Say 5
  years, L1, L2 , L3 , L4 , and L5.
• Test Prob(C2C1) as a function of insureds
  subsequent loss experience Li
• Underlying assumption Insurers have better
  information on their own future losses than
  insurers

10
Solution II Get around Risk

• Identify insured factors potentially correlated
  with insureds AS incentive but uncorrelated with
  insurance price, e.g., income, education
• Test if the loss and coverage relationship
  differs across insured groups with different
  values of insured characteristics
• Specifically, interact loss experience with some
  of the control variables used in the regressions

11
Conclusions

• Smart idea, neat data, good potential
• The authors need to differentiate adverse
  selection from competing hypotheses
• Risk ? Loss Experience