Provider-induced Asymmetric
Information in the Insurance
Market
Larry Y. Tzeng
Jennifer L. Wang
Kili C. Wang
Jen-Hung Wang
Asymmetric information—
• Adverse selection
• Moral hazard
There is significantly positive conditional
correlation between risk and coverage.
• Risk—claim
• Coverage—high versus low
--Most literatures focus on the asymmetric
information between insured and insurer.
--Few literatures focus on the asymmetric
information caused by provider in health
insurance
--This paper focus on the asymmetric
information caused by provider in
automobile insurance of Taiwan
“Provider” in this paper:
• Dealer-owned agent (DOA)—who sold
automobile as well as automobile
insurance, meanwhile he also owned
the repair house.
• More than 40% of automobile
insurance policies are sold through
DOA.
In year 2000
DOA
Non-DOA
Type A & B
contract
0.78
0.52
Accident
0.41
0.23
Why provider induce more sever
asymmetric information?
Adverse selection—
• DOA sold larger percentage of high coverage
contract
– The commission incentive for DOA to sell high
coverage
– “obtain better deal” motive for insured to
purchase high coverage
• DOA induced greater number of high risk drivers
– “Higher expected repair revenue” incentive for
DOA
– Immune form premium penalty for high risk
insured
– Get “better car repair service” for high risk
insured
Moral hazard—
High coverage policies through DOA
result in more claim
• DOA clear about who has higher coverage
can cover higher loss
• DOA owned more information about car
damage caused by accident
• Insurer audit DOA less (insurer tolerate
DOA more)
Our hypothesis:
Automobile insurance sold through
DOA suffers more severe problems of
asymmetric information
Empirical methodologies
• Chiappori and Salanie’s (2000) approach
--residual correlation from two probit
regression
• Approach similar to Dionne, Gourieroux
& Vanasse (2001)
--two stage method
Chiappori and Salanie’s (2000)
approach
Pr ob(cov erage  1)  X i  c   i
Pr ob(accident  1)  X i  a   i
( X i c )
( X i c )
ˆi  E ( i | yi ) 
yi  (1  yi )
( X i  c )
(  X i  c )
( X i a )
(X i a )
ˆi  E ( i | zi ) 
z i  (1  z i )
( X i  a )
(  X i  a )
n
2
( ˆiˆi )
W

i 1
n
 ˆ ˆ
i 1
2 2
i i
is the correlation coefficient of
Predict:
 A   NA
i
and
i
A robust test
Wˆ i 
ˆiˆi
ˆi ˆi
2
2
let Di  1 when Wˆi  1 ,and Di  0 when Wˆi  1
Pr ob( Di  1)  1  empenoi  XB
Predict:
1
should be significantly positive
Approach similar to Dionne,
Gourieroux & Vanasse (2001)
(a) estimate the occurrence of claim in first stage
Prob (accident i  1 X 1i )   ( X 1i )
Prob(cov eragei  1 acciˆdent i , accident i , accident i  Di , X 2i )
 ( 1acciˆdent i   2 accident i   3 accident i  Di  X 2i  4 )
(b) estimate the choice of coverage in first stage
Prob (cov eragei  1 X 3i )  ( X 3i )
Prob(accident i  1 cov eˆragei , cov eragei , cov eragei  Di , X 4i )
 (  5 cov eˆragei   6 cov eragei   7 cov eragei  Di  X 4i  8 )
Predict:
 3 is
significantly positive
 7 is significantly positive
Empirical Results
• The asymmetric information do exist in
automobile insurance market of Taiwan
• The asymmetric information problems in
insurance written by DOA are more sever
than those through non-DOA channels
– From Chiappori and Salanie’s (2000) approach
– From robust test
– From Approach similar to Dionne, Gourieroux &
Vanasse (2001)
Thank you !
All your comments are
welcome !