Docstoc

33

Document Sample
33 Powered By Docstoc
					Provider-induced Asymmetric Information in the
Insurance Market




Larry Y. Tzeng*
Jennifer L. Wang**
Kili C. Wang***
Jen-Hung Wang****




* Professor, Finance Depart ment, Nat ional Taiwan University.
** Associate Professor, Risk Management and Insurance Department, Nat ional Chengchi Un iversity.
*** Associate Professor, Risk Management and Insurance Depart ment, Shih Ch ien Un iversity.
**** Assistant Professor, Finance Depart ment, Shih Hsin Un iversity.

                                                      1
Provider-induced Asymmetric Information in the
Insurance Market

                                            Abstract
This paper examines the existence of provider-induced asymmetric information in the
insurance market. The empirical data on comprehensive automobile insurance in Taiwan
provide a unique opportunity to test our hypothesis. Consistent with this hypothesis, we find
evidence that providers do induce asymmetric information problems. Our empirical results
show that the conditional correlation between the coverage level and the occurrence of a claim
is higher for insurance policies sold through dealer-owned agents than for those sold through
other marketing channels.

Key words: asymmetric information, automobile insurance, dealer-owned agents, marketing channel.


1. Introduction


     Rothschild and Stiglitz (1976) pioneered the study of asymmetric information problems in


the insurance market. In the three decades since, their work has inspired many researchers who


continue to provide ingenious theoretical findings. The theoretical papers on asymmetric


information that followed Rothschild and Stiglitz (1976) include Wilson (1977), Miyazaki


(1977), Grossman (1979), Shavell (1979), Riley (1979), Radner (1981), Holmstrom (1982),


Dionne (1983), Rubinstein and Yarri (1983), Crocker and Snow (1986), Cho and Kreps (1987),


Cooper and Hayes (1987), Hellwig (1987, 1988), Arnott and Stiglitz (1988), Hosios and Peters


(1989), Hoy (1989), Mookerjee and Png (1989), and Abreu, Pearce, and Stacchetti (1990).



                                                2
However, until recently, relatively few empirical studies devoted to this issue.


     As discussed by Chiappori and Salanie (1997), data from insurance companies are


well-suited for studies of asymmetric information, because they not only record both the


coverage and the claim amounts but also provide information on many characteristics of


individuals. Some recent papers used empirical data in alternative insurance markets to


investigate asymmetric information problems. In the life/health insurance market, Cawley and


Philipson (1999), Cardon and Hendel (2001), and Finkelstein and Poterba (2000) examined


respectively the US life insurance and health insurance markets and the UK annuity market. In


the property/liability insurance market, Puelz and Snow (1994), Chiappori and Salanie (2000),


and Dionne, Gourieroux, and Vanasse (2001) studied the automobile insurance market by


using respectively data from the US, France, and Canada. These studies successfully


constructed a bridge between the theoretical world and real practices, helped people further


understand asymmetric information problems.


     Most of these empirical studies focused on asymmetric information between the insurer


and the insured. Only a few investigated asymmetric information caused by providers. Polsky


and Nicholson (2004) investigated the risk differences of enrollees and medical expenditures


between HMOs and non-HMOs; Newhouse (1996) also investigated asymmetric information


problems in these different organizations. Without a doubt, the provider ’s asymmetric

                                                3
information problems—e.g., the existence of moral hazard in health insurance—raised major


concerns in real insurance practices. However, an empirical testing of asymmetric information


caused by providers might be difficult because it requires data from both the insurance


companies and the providers.


      We first examine whether a positive relationship exists between coverage and the


occurrence of a claim. If there are asymmetric information problems, we should observe a


positive correlation between them. 1 One important conclusion of Rothschild and Stiglitz (1976)


is that a separating equilibrium could exist in the insurance market. In this case, insurance


companies offer a variety of products to attract different types of insured, since the companies


may not have enough information to identify the insured’s risk types. Thus, in this equilibrium,


high-risk individuals choose higher-coverage insurance and low-risk individuals choose


lower-coverage insurance. On the other hand, it is also well known that high insurance


coverage could induce the insured’s moral hazard problem in the insurance market. An


individual with high insurance coverage might drive less carefully, since most of the loss


would be compensated.


      In Taiwan, it is widely believed that comprehensive automobile insurance coverage has


long suffered from asymmetric information problems. According to Wang (2004), alternative

1
  We can observe this directly fro m the unconditional correlat ion. To control the heterogeneity of the sample, in
our emp irical results, we will test the conditional correlat ion after controlling for the related variables.

                                                          4
products could be designed to cope with these problems in this market. In this paper, we intend


to use comprehensive automobile insurance data from a large insurance company in Taiwan to


examine whether this problem might be induced by the providers. The data from the


automobile insurance market in Taiwan provide a unique opportunity to investigate


provider- induced asymmetric information problems, since more than 40 percent of automobile


insurance policies are sold through dealer-owned agents.


      However, why would asymmetric information be induced by dealer-owned agents?


Generally speaking, the policies sold through dealer-owned agents might include a larger


percentage of high-coverage policies; and those who purchase insurance through dealer-owned


agents might include a greater number of high-risk drivers. On the one hand, dealer-owned


agents may have an incentive to promote higher coverage to high-risk customers, since


contracts with higher coverage are more expensive and the dealerships are rewarded with a


commission that is a fixed percentage of the insurance premium. Meanwhile, high- risk


customers 2 may bring them more revenues from repairing cars because of accidents in the


future. On the other hand, the high-risk insured also have an incentive to purchase insurance


through the dealer-owned agents. One reason for this is because dealer-owned agents have

2
  Dealer-o wned agents may understand more about the risk type of their customers t han other agents or sellers of
insurance. They may have longstanding relationships with their customers fro m selling cars. Otherwise, they
could also predict the risk type of their customers not only from the indiv idual characteristics of the insurance
contracts, but also by observing the customers ’ preferences and needs regarding a vehicle when they choose the
vehicle.

                                                        5
stronger bargaining power enabling the high-risk insured to obtain a “better” deal on their


contracts. This is especially the case when they consider the ongoing purchase of contracts in


subsequent years. In such cases, high-risk customers might be more likely to be involved in an


accident, and thus they should be charged a higher premium as a penalty if an accident really


does occur and a claim is made in the previous year. In practice, the subsequent contracts sold


by dealer-owned agents seldom reflect the punishments recorded in accident records 3 . The


other reason is that dealer-owned agents may promise to provide “better” service for the


insured when they have their cars repaired. Thus, dealer-owned agents may attract more


high-risk insured to purchase high-coverage policies.


       Moreover, high-coverage policies sold through the dealer-owned agents may result in


more claims for insurance companies. Repair shops owned by dealer-owned agents may have


an incentive to augment the work to increase their revenues, especially for those car owners


who not only repair the cars at their repair shops, but who also purchase insurance from them.


On the one hand, they are very clear about who has high coverage contracts and how those


high coverage contracts can cover the loss from an accident, as compared with the case of an


insured who is without any dealer-owned agent standing by him. On the other hand, only


repair shops can really comprehend how damaged the car is from an accident, and how much


3
    This description comes fro m an interview with a manager of an insurance company.

                                                        6
work is needed to restore the car. Because repairing a car is such a professional task, if the


insurance companies want to audit the claim, they should devote more efforts and funds to this


channel than to other channels. When the cost of the audit is too high to cover the benefit


derived from auditing the claim, the insurance companies will not bother to audit the claim 4 .


This is one of the reasons why the insurance companies will devote less effort to auditing the


claims resulting from the contracts sold by the dealer-owned agents. The other ironic reason is


that insurance companies usually have to tolerate this type of corruption between the insured


and the supplier simply to avoid losing business, since repair shops owned by dealer-owned


agents are the major distribution channels for automobile insurance in Taiwan. 5 In some cases,


insurance companies may even pay claims under certain amounts without performing an


inspection. Thus, dealer-owned agents could have both the motive and the ability to lie and


induce the over- use of car-repair expenditures from insurance claims. According to Alger and


Ma (2003), dishonest dealer-owned agents and repair shops owned by dealer-owned agents


(i.e., the providers) always lie when the insurance contracts are not collusion-proof. Since


insurance companies may audit them less, they should be more likely to induce more


augmented claims involving higher coverage than in the case of insurance sold through other
4
  This is also true when the insurance companies underwr ite. If dealer-owned agents comprehend the type of their
customer, and if they hide some of the underwrit ing information, the insurance companies will not necessarily
underwrite clearly and costs may exceed benefits.
5
  The main distribution intermed iaries of automobile insurance in Taiwan are the direct writers and dealer-owned
agents. Dealer-o wned agents write more than 40 percent of the automobile insurance policies in Taiwan.
Therefore, repair shops owned by dealer-owned agents have very strong bargaining power in claim settlements.

                                                       7
channels.


     Therefore, we hypothesize that automobile insurance policies sold through dealer-owned


agents might suffer from more severe problems of asymmetric information. We expect the


conditional dependence between coverage choice and claim occurrence of the insured to be


greater in the group from the channel of dealer-owned agents than when other channels are


involved.


     The remainder of this paper is organized as follows. Section 2 gives an outline of Taiwan


automobile insurance market. Section 3 describes the data and the methodology used. In


Section 4, the main empirical results are presented and discussed. Section 5 concludes the


paper and provides recommendations for further research.




2. Automobile Insurance in Taiwan

     In Taiwan, automobile insurance accounts for about 50% of the insurance premium

volume in most property- liability insurance companies; and has occupied the largest market

share of property- liability insurance market. To fix the idea, we described in the following the

market based on data in 2005. There are approximately 6.7 million car owners purchasing

automobile insurance from 25 insurance companies. The annual premium for automobile

insurance was about $1.84 billion and the incurred loss was $1.14 billion. Three types of

automobile insurance have been observed in the market: compulsory liability, supplementary

liability, and comprehensive coverage for damage.

                                                8
     Compulsory automobile liability insurance only covers liability for bodily injures

including death and medical expenditure. The total premium for compulsory automobile

liability insurance is about 0.6 billion and the total incurred loss is about 0.4 billion. The

coverage and rating are fully regulated, and moreover the liability for bodily injures in

automobile accident is mandated as strict liability in the limit of compulsory automobile

liability insurance. The coverage includes 42,000 death benefit, and up to 6,000 medical

expense compensation.

     Supplementary liability insurance contains coverage for both liability of property damage

and bodily injury and is purchased voluntarily. The liability for property damage and that for

bodily injures above the limit of compulsory automobile liability insurance are on at-fault base.

The total premium for supplementary liability insurance is about 0.51 billion and the total

incurred loss is about 0.32 billion.

     The third type of automobile insurance is comprehensive coverage of automobile

insurance that provides the coverage for property damage on automobile. The co mprehensive

coverage of automobile insurance is also purchased voluntarily, but accounts for about 42% of

total automobile insurance premium volume and has occupied the largest market share of

automobile insurance market in 2005. The total premium for comprehensive coverage of

automobile insurance is about 0.73 billion and the total incurred loss is about 0.42 billion.

     Car owners can choose from three types of comprehensive coverage: type A covers all

risks, type B covers selected risks, and type C only covers damage in a collision involving two

or more vehicles. Only type A coverage was offered before 1995. To deal with the asymmetric

information problem, insurance companies started to offer type B coverage in 1995 which

excludes losses whose causes are hard to determine. In response to the increasing loss ratios,

type C coverage was introduced in 1999. These three types of comprehensive coverage in

                                                  9
Taiwan provide us with a unique opportunity to study whether asymmetric information exists

in the island’s insurance market.

     It should be pointed out that consumers can choose different deductible levels for their

comprehensive coverage. The most popular type of deductibles for types A and B coverage is

an increasing deductible system in which the deductible is respectively $85, $140, and $200

for the first, the second, and the third and above claims. The premium of this type of an

increasing deductible system accounts for almost 95% of the total written premium. Type C

coverage does not have any deductible provision. It is also worth noting that the changes in the

amount of deductible and the adjustment rate of experience rating played important roles to

reduce the loss frequency and loss ratio. The adjustment rate for experience rating for both

type A and type B was changed periodically according to the loss ratio. We also find more

types of deductible were offered by insurers in the market after 1996. Especially, the increasing

deductible system may help to control moral hazard, but is much less likely to influence the

adverse selection problem.




3. Data and Methodology


     Our data set is from a large Taiwan automobile insurance company, which controls over


30 percent of the market share of Taiwan automobile insurance. The data offer individual- level


information of insurers/cars and the distribution channel information of policies for 1999-2000.


We cannot collect national data with such full information. Though, we believe that the data


from that company should be representative enough. Specifically, the research data included



                                               10
respectively 61,642 and 64,234 observations in 1999 and 2000. As the type C comprehensive


coverage was first introduced to Taiwan in 1999, the data we explored is in the early period of


the product introduction, wherein the market’s learning effect is likely small.


     To conduct the empirical testing of whether the asymmetric information problems existed


in the market, we follow the method developed by Dionne, Gourieroux and Vanasse (2001).


The asymmetric information problems are tested by a two stage regression to examine the


conditional independence, in each risk class of the insureds, between a measure of individual


risk and the accident probability. Following the literature and adjusting the Taiwan situation,


the choice of coverage is used as our measure of individual risk, and the claim probability as


proxy of accident probability (more details later).


     Specifically, in the first stage we estimate the occurrence of a claim using a probit


regression:

     Prob(accidenti  1 X 1i )  ( X 1i  )                                                   (1)


where X 1i is the vector of variables for information about the characters of the insured and


the characters of the car of the insured,  is the regressor coefficient vector, and  is the


standard normal cumulative distribution function.


     In the second stage, the choice of coverage is regressed, again by a probit regression,




                                                11
upon the information about the insured that are observable to the insurers ( X 2i ) 6 , the claim


dummy as proxy of accident event ( accident ), and the ex ante expected accident dummy

     ˆ
( acci dent ):

                             ˆ
      Prob(cov eragei  1 acci denti , accidenti , X 2i )
                                                                                                               (2)
                ˆ
       ( 1acci denti   2 accidenti  X 2i  3 )

The expected accident dummy is derived from Equation (1). It is incorporated into the second


stage regression to take care of potential omitted-non linear effect of the initial exogenous

variables (here X 2i ). The null hypothesis that  2  0 affirms that there is conditional


independency between the occurrence of a claim and the choice of coverage; by contrast a

significantly positive  2 suggests that there is still residual information asymmetry problem

after taking into account the control variables X 2i .


      In a further exploration, we modify the basic model to test whether the insurance policies


purchased through the dealer-owned agents suffer more severe asymmetric information. A

dummy of channel ( Di ) is put into the second stage. Di  1 when the contract is sold by the


dealer-owned agent, otherwise Di  0 . Thus,




6
   X 2i include all variab les in X 1i except fo r the area du mmy variables ( arean , areas , areaeast ) and
city variable ( city ). The reason why these latter variables are included in the regression which estimates the
occurrence of a claim, but are not included in the regression which estimates the choice of coverage, is that the
location factors could really affect the occurrence of an accident. However, the insurance companies in Taiwan
still do not take them into consideration when they calculate the premiu m for the contracts until then. So, the
choice of coverage will not be affected by them.

                                                         12
                                   ˆ
            Prob(cov eragei  1 acci dent i , accidenti , accidenti  Di , X 2i )
                                                                                                             (2’)
                        ˆ
              ( 1 acci dent i   2 accidenti   3 accidenti  Di  X 2i  4 )

A significantly positive  3 suggests that there is additional (residual) information asymmetry


problem in the channel of dealer-owned agents (taking into account the control variables

X 2i ).


      Now we talk about the variables in details. First, we dichotomize the coverage of the

insurance contracts ( cov erage ) by the scope of risk a contract covers (thus not exactly the


amount of indemnity, as in most theoretical models of asymmetric information). Among the


three kinds of coverage in Taiwan’s comprehensive automobile insurance, the covering scopes

are ordered by. from high to low, type A, B and C contracts. Dichotomizing cov erage to


accommodate the probit model, we combine type A and B contracts as the high coverage and


type C contract as the low coverage. There is a concern that type C contract is without


deductible but type A, and most of type B 7 , contracts are with deductible. Thus if one focuses


only on the car collision coverage and ignores for the moment that type A (and B) covers


additional risk types, one may argue that in view of deductibles the coverage of type C contract


is higher than that of type A (and B) contracts! While the effectiveness of this argument is quite


disputable (because the sole focus on car collision coverage would be valid only under very


7
   Fro m an interview with the insurance company that is our source of data, most of the type B contract policies
are with increasing deductible and the percentage of the policies in type B contract which are without or with
fixed deductible is below 1%.

                                                        13
restrictive assumptions), to circumvent the concern we simply conduct a deductible adjustment


before our empirical test, as follows.


      Recall that all the type C contract policies are without deductible, all the type A contract


policies are with increasing deductible design8 , while most of type B contract policies are with


increasing deductible design. We, ignoring the few type B contract policies that are without


deductible or with fixed deductible, artificially raise the claim amount of type C contract by


NT$ 30009 to define the claim for a policy. 10         11




      Next we discuss the variable accident . First of all, it should be noted that not all car


accidents are necessarily claimed while we can observe only the claims. However, this should


cause no problem as what the insurers really care is the probability of claimed accidents.


Second, we do not count all claims into the variable accident but only claims that involve a


collision with two or more cars, taking care of the potential bias from coverage disparity: Since


types A and B contracts have broader coverage than type C contracts, the insured with types A


or B contracts would naturally file claims that are not covered in type C contracts. While such
8
   The deductible amount is NT$ 3000 fo r the fisrt time claim, NT$ 5000 for the second time claim, and NT$
7000 fo r the third time claim and fro m that time on.
9
   Because we are lack of the exact deductible data in our sample, the only way we can do is using the average
deductible amount as a proxy for the real deductible amount. In increasing deductible contract, the deductible
amount is NT$ 3000 for the fisrt time claim, NT$ 5000 for the second time claim, and NT$ 7000 for the third
time claim and fro m that time on. And the average claim t imes for ever-claimed policies in our samp le are about
1.3. So, we treat NT$ 3000 as the average deductible amount for type A and B contracts.
10
    We also conduct all the emp irical tests using cov erage without deductible adjustment. The results are
qualitatively the same as those reported in this paper.
11
    We also test other ways to dichotomize high coverage versus low coverage. Thus, we pair-wisely classify type
A versus B, A versus C, and A versus C contracts. Not reported here, all the results are qualitatively the same as
the results reported in the paper.

                                                        14
asymmetry in claim types entails interesting and important issues to be addressed, it is


arguably an information asymmetry problem. Thus, we will focus on car collision accidents.


Last, there is also a potential bias from the differing deductible design in the various policy


types. Suppose for argument that there were the same number of car collision accidents in both


type A (or B) and C contracts. However, some accidents of the type A (or B) insured that are of


small loss amount would not be claimed due to the deductible, and as a consequence, one


might observe superficially more claims in the group of insureds purchasing type C contracts.


We try a deductible adjustment as follows to take care of this concern. Under the version of


deductible adjustment, accident  1 if a claim caused by a collision is filed under a type A or


type B contract policy; but if such a claim is under a type C contract policy, the claim amount


must be above NT$3000 to have accident  1. In all other cases we set accident  0 . This


deductible adjustment is not necessarily adequate as the choice of deductible is itself


endogenously self- selected by the insureds. Anyway, notice that without adjustment, the


existence of deductible disparity might bias toward a negative relation between the choice of


coverage and the probability of claim as we define it (that is, car collision claims would tend to


be more likely found in the type c contracts). Thus a result of significantly positive correlation


should be immune even no deductible adjustment is made.


     Further, it is curious that whether the existence or degree of asymmetric information is

                                                15
different in various monetary thresholds of claimed amount as arguably the insurance


companies should audit more stringently the claims with higher claim amount. To explore this


question, we further construct various accident variables by three monetary thresholds: a


base where any claim is counted, a moderate threshold where only claims with claim amount


higher than NT$10,000 are counted, and a high threshold where only claims with claim


amount higher than NT$20,000 12 are counted.


       Table 1 gives the list of the (controlled) independent variables that are observable to the


insurance companies, except for the area dummy variables and city variables.




                                            (Table 1 about here)




4. Empirical Results and Discussion


       The summary statistics for all the variables are displayed in Tables 2 to 5. The means of


the claim amounts and the coverage levels from the dealer-owned agents are much greater than


those from other marketing channels in both years.




                                          (Tables 2-5 about here)


       From the tables, there is much higher percentage of new cars (with age within one year)

12
     Accordingly, the three monetary thresholds for type C contract are NT$ 3000, NT$ 13000, and NT$ 23000.

                                                       16
in the channel of dealer-owned agents than others. New cars are over 80% in the channel of


dealer-owned agents in each sample year, with the brand-new cars being over 70%. Meanwhile,


new cars are less than 50% in the other channels in each sample year. There are some concerns


about this phenomenon. The new-car owners may be more willing to purchase high coverage


for their new cars (empirically, it is the case). Besides, these owners may be more likely new


drivers, who may be unfamiliar about driving their new cars yet, and consequently the


probability of incurring collision accident may be higher. In this case, there is likely a higher


(unconditional) correlation between high coverage and accident in the new car group. Since


this group is more heavily concentrated in policies sold through the dealer-owned agents, so


we may find the same pattern of (unconditional) correlation in the group of policies sold


through the dealer-owned agent channel. Though the above argument does not necessarily


apply for conditional correlation (the focus of our tests), such concerns perhaps still needs to


be well taken (more later).


     The empirical results of our base regression are given in Table 6. All the coefficients of

 2 are significantly positive, no matter under what threshold or in which year. The evidence

supports that there are asymmetric information problems in Taiwan automobile collision


insurance market.


                                      (Table 6 about here)

                                                17
     Furthermore, we find that the coefficient  2 is least when the monetary threshold is


highest (above NT$ 20000). This result may reflect that the insurer pays more attention in


auditing a claim with large monetary amount.


     Dionne, Gourieroux, and Vanasse (2001) and Chiappori and Salanie (2000) found no


evidence to support the existence of asymmetric information using data from respectively


Canadian and French automobile insurance markets. Our results by contrast support the


existence of such asymmetric information in the Taiwanese car collision insurance market.


This interesting contrast may be more than a pure difference of markets, but reflect the


differing maturity of markets.


     Underwriting systems can serve as an essential tool for insurance companies to overcome


asymmetric information problems. But a well established underwriting system takes much


time and experience. First of all, it must be based on a reliable data set. For newly-written


business, insurance companies do not have enough data to explore information related to the


insured’s risk. Insurance companies would need to collect more useful information year by


year and in the meanwhile learn to use the data to classify the insured. Eventually, they may


adopt various tools to control the asymmetric information problems well enough. The market


would then entail the statistical results rejecting the existence of asymmetric information, as



                                                18
found by Dionne, Gourieroux, and Vanasse (2001) and Chiappori and Salanie (2000).


Compared to the insurance markets in France and Canada, the insurance market in Taiwan is


still in an emerging- market stage. Specifically, some critical underwriting factors used in


well-established insurance markets have not been employed in the Taiwan market. For


example, driving records are not yet used for underwriting and pricing because insurance


companies do not have access to the database of driving records. Experience rating is


introduced only in recent year and the system is primitive and crude. To make things worse, it


is difficult to date for Taiwan insurance companies to rigorously implement even such a crude


experience rating system. High- level executives reveal that insurance companies in Taiwan


frequently give the insured credit for their experience rating but waive the penalties in relation


to experience rating due to both marketing competition and pressure from the dealer-owned


agents. In a word, in an emerging market such as the Taiwan comprehensive automobile


insurance market, insurance companies may have no the expertise and sophistication to well


control the asymmetric information problem. And it is more likely to find evidence supporting


the theoretical predictions of asymmetric information models.


     Table 7 gives the result of our regression that examines the additional asymmetric

information problem in the car-dealer channel. All the coefficients of  3 are significantly


positive, suggesting that the policies written through dealer-owned agents exhibit additional

                                                19
asymmetric information problems. It is interesting to note that all the coefficients of  2 are


still significantly positive, suggesting that polices sold through channel other than


dealer-owned agents still exhibit asymmetric information problems. Thus the asymmetric


information problems identified in our base regression come not only from the dealer-owned


agents.




                                         (Table 7 about here)




     Above we mentioned about a concern about potential new-car bias in the group of


policies written through dealer-owned agents. To address this concern, we separate the full


sample into new-car (one year and within one year old cars) and old-car (over one year old


cars) sub-samples. Then, we test whether there is additional asymmetric information


phenomenon in dealer-owned agent channel in each sub-sample. The evidence shows positive


correlation in dealer-owned agent channel in each sub-sample.


     We also control for the threshold of the amounts claimed in the empirical tests. The

results are given in Table 7. Again, the coefficients  2 and  3 are least when the monetary


threshold is highest.


     The evidence coincides with our inferences regarding the auditing and underwr iting

                                                20
tendencies of insurance companies. The insurance companies might be less willing to audit or


underwrite when the costs exceed the benefits of doing so. Thus they will apply less stringent


auditing and underwriting to insurance policies sold through dealer-owned agents. However,


for claims involving larger monetary amounts, it is reasonable that insurance companies will


pay more attention to examine the claims, and the effort is likely to reduce the


provider- induced moral hazard.




5. Conclusion


     In this paper we examine the existence of asymmetric information in the Taiwan


automobile insurance market. We think that Taiwan data is interesting in that the Taiwan


automobile insurance market provides a unique opportunity to test our hypothesis. Because


dealer-owned agents could induce both adverse selection and moral hazard problems in the


automobile insurance market, the asymmetric information problems might be more severe in


policies sold through the dealer-owned agents. Our empirical results show that the conditional


dependency between coverage and claims is indeed significantly more higher when the


policies are written through the channels of the dealer-owned agents, supporting the hypothesis


that providers (additionally) induce asymmetric information problems.


     While Chiappori and Salanie (2000) and Dionne, Gourieroux, and Vanasse (2001) found

                                               21
no evidence in data from French and Canadian insurance markets, our results support the


existence of asymmetric information in the Taiwanese automobile insurance market. In an


emerging insurance market, such as the comprehensive automobile insurance market in Taiwan,


the existence of asymmetric information problems may result from either imperfections in the


underwriting or pricing systems or an inability to implement effective underwriting systems. In


addition, provider- induced asymmetric information problems may make it more difficult for


insurance companies to implement effective underwriting systems, since dealer-owned agents


control the major market share for the comprehensive automobile insurance market in Taiwan.


Thus, the imperfections in the underwriting systems of insurance companies might also be the


critical reason why we observe the existence of provider-induced asymmetric information. Our


results definitely demonstrate the need for further studies on asymmetric information problems


in different markets as well as in different countries.




                                                 22
References

Abreu, D., D. Pearce, and E. Stacchetti, “Toward a Theory of Discounted Repeated Games
   with Imperfect Monitoring.” Econometrica, Vol. 58 (1990), pp. 1041-1064.


Alger, I., and Ching-to A. Ma, “Moral Hazard, Insurance, and Some Collusion.” Journal of
   Economic Behavior & Organization, Vol. 50 (2003), pp. 225-247.


Arnott, R., and J.E. Stiglitz, “Randomization with Asymmetric Information.” Rand Journal of
   Economics, Vol. 19 (1988), pp. 344-362.


Cardon, J.H., and I. Hendel, “Asymmetric Information in Health Insurance: Evidence from the
   National Medical Expenditure Survey.” Rand Journal of Economics, Vol. 32 (2001), pp.
   408-427.


Cawley, J., and T.J. Philipson, “An Empirical Examination of Information Barriers to Trade in
   Insurance, American Economic Review, Vol. 89 (1999), pp. 827-846.


Chiappori, P., and B. Salanie, “Empirical Contract Theory: The Case of Insurance Data.”
   European Economic Review, Vol. 41 (1988), pp. 943-950.


Chiappori, P., and B. Salanie, “Testing for Asymmetric Information in Insurance Markets.”
   Journal of Political Economy, Vol. 108 (2000), pp 56-78.


Cho, In-Koo, and D.M. Kreps, “Signaling Games and Stable Equilibria.” Quarterly Journal of
   Economics, Vol. 102 (1987), pp. 179-221.


Cooper, R., and B. Hayes, “Multi-period Insurance Contracts.” International Journal of
   Industrial Organization, Vol. 5 (1987), pp. 211-231.


Crocker, K.J., and A. Snow, “The Efficiency Effects of Categorical Discrimination in the
   Insurance Industry.” Journal of Political Economy, Vol. 94 (1986), pp. 321-344.


Dionne, G., “Adverse Selection and Repeated Insurance Contracts.” Geneva Papers on Risk
   and Insurance, Vol. 8 (1983), pp. 316-333.


                                                23
Dionne, G., C. Gourieroux, and C. Vanasse, “Testing for Evidence of Adverse Selection in the
   Automobile Insurance Market: A Comment.” Journal of Political Economy, Vol. 109
   (2001), pp.   444-455.


Finkelstein, A., and J.M. Poterba, “Adverse Selection in Insurance Markets: Policyholder
   Evidence from the U.K. Annuity Market.” Working Paper no. 8045, NBER, 2000.


Grossman, H.I., “Adverse Selection, Dissembling and Competitive Equilibrium.” Bell Journal
   of Economics, Vol. 10 (1979), pp. 336-343.


Hellwig, M.F., “Some Recent Developments in the Theory of Competition in Markets with
   Adverse Selection.” European Economic Review, Vol. 31 (1987), pp. 319-325.


Hellwig, M.F., “A Note on the Specification of Interfirm Communication in Insurance Markets
   with Adverse Selection.” Journal of Economic Review, Vol. 31 (1988), pp. 319-325.


Holmstrom, B., “Moral Hazard and Term.” Rand Journal of Economics, Vol. 13 (1982), pp.
   324-340.


Hosios, A., and M. Peters, “Repeated Insurance Contracts with Adverse Selection and Limited
   Commitment.” Quarterly Journal of Economics, Vol. 2 (1989), pp. 229-254.


Hoy, M., “The Value of Screening Mechanisms Under Alternative Insurance Possibilities.”
   Journal of Public Economics, Vol. 39 (1989), pp. 177-206.


Miyazaki, H., “The Rate Race and Internal Labour Market.” Bell Journal of Economics, Vol. 8
   (1977), pp. 394-418.


Mookerjee, D., and I. Png, “Optimal Auditing, Insurance and Redistribution.” Quarterly
   Journal of Economics, Vol. 3 (1989), pp. 415-440.


Newhouse, J.P., “Reimbursing Health Plans and Health Providers: Efficiency in Production
   Versus Selection.” Journal of Economic Literature, Vol. 34 (1996), pp. 1236-1263.


Polsky, D., and S. Nicholson, “Why Are Managed Care Plans Less Expensive? Risk Selection
   Utilization, or Reimbursement.” The Journal of Risk and Insurance, Vol. 71 (2004), pp.

                                             24
   21-40.


Puelz, R., and A. Snow, “Evidence on Adverse Selection: Equilibrium Signaling and
   Cross-subsidization in the Insurance Market.” Journal of Political Economy, Vol. 102
   (1994), pp. 236-257.


Radner, R., “Monitoring Cooperative Agreements in a Repeated Principal- Agent
   Relationship.” Econometrica, Vol. 49 (1981), pp. 1127-1148.


Riley, J.G., “Informational Equilibrium.” Econometrica, Vol. 47 (1979), pp. 331-359.


Rothschild, M., and J.E. Stiglitz, “Equilibrium in Competitive Insurance Markets: An Essay on
   the Economics of Imperfect Information.” Quarterly Journal of Economics, Vol. 90 (1976),
   pp. 629-649.


Rubinstein, A., and M.E. Yarri, “Repeated Insurance Contracts and Moral Hazard,” Journal of
   Economic Theory, Vol. 30 (1983), pp. 74-97.


Shavell, S., “On Moral Hazard and Insurance.” Quarterly Journal of Economics, Vol. 93
   (1979), pp. 541-562.


Wang, Jennifer. L., 2004, “Asymmetric Information Problems in Taiwan’s Automobile
   Insurance Market: The Effect of Policy Design on Loss Characteristics.” Risk Management
   and Insurance Review, Vol. 7 (2004), pp. 53-71.


Wilson, C.A., “A Model of Insurance Markets with Incomplete Information,” Journal of
   Economic Theory, Vol. 16 (1977), pp. 167-207.




                                              25
                                                          Table 1
                                           Definitions of the Variables
Variab le          Definition

Dependent variables:
coverage(   y)     a dummy variable that equals 1 when an individual chooses a type A or B policy, otherwise it equals 0

accident(   z)     a dummy variable that equals 1 when an individual’s claim is caused by a collision and the claim amount

                   is above the threshold amount, otherwise it equals 0

Independent variables:

Premium            amount of the premium for each policy.

carage0            a dummy variable that equals 1 when the car is new, otherwise it equals 0

carage1            a dummy variable that equals 1 when the car is one year old, otherwise it equals 0

carage2            a dummy variable that equals 1 when the car is two years old, otherwise it equals 0

carage3            a dummy variable that equals 1 when the car is three years old, otherwise it equals 0

carage4            a dummy variable that equals 1 when the car is four years old, otherwise it equals 0

carage5            a dummy variable that equals 1 when the car is five years old, otherwise it equals 0

carage6            a dummy variable that equals 1 when the car is six years old, otherwise it equals 0

carage7            a dummy variable that equals 1 when the car is seven years old, otherwise it equals 0

carage8            a dummy variable that equals 1 when the car is eight years old, otherwise it equals 0

carage9            a dummy variable that equals 1 when the car is nine years old, otherwise it equals 0

carage10           a dummy variable that equals 1 when the car is ten years old, otherwise it equals 0

carage11           a dummy variable that equals 1 when the car is eleven years old, otherwise it equals 0

sexf               a dummy variable that equals 1 when the owner of the car is female, otherwise it equals 0

married            a dummy variable that equals 1 when the owner of car is married, otherwise it equals 0

city               a dummy variable that equals 1 when the owner of the car lives in a city, otherwise it equals 0
arean              a dummy variable that equals 1 when the car is registered in the north of Taiwan, otherwise it equals 0

areas              a dummy variable that equals 1 when the car is registered in the south of Taiwan, otherwise it equals 0

areaeast           a dummy variable that equals 1 when the car is registered in the east of Taiwan, otherwise it equals 0

catpcd_1            a dummy variable that equals 1 when the car is a sedan and is for non-commercial or for long-term rental

                    purposes, otherwise it equals 0

catpcd_2            a dummy variable that equals 1 when the car is a small freight-truck and is for non-commercial purposes

                    or for business use, otherwise it equals 0

tramak_i            i=n,f,h,t,c, a dummy variable that equals 1when the trademark of the car is the assigned brand, otherwise

                    it equals 0

age2                a dummy variable that equals 1 when the insured is between the ages of 30 and 25, otherwise it equals 0

age3                a dummy variable that equals 1when the insured is between the ages of 60 and 30, otherwise it equals 0



                                                                 26
age4   a dummy variable that equals 1 when the insured is over the age of   60, otherwise it equals 0




                                               27
                                 Table 2
Summary Statistics for Data on Dealer-owne d Agents in 1999
Variable      N         Mean          Std Dev    Minimum   Maximum

accident     21615    0.329216       0.469939        0     1.000000

coverage     21615    0.684201       0.464844        0     1.000000

prem         21615     31997          20406          454   502752

carage0      21615    0.715244       0.451309        0     1.000000

carage1      21615    0.163081       0.369448        0     1.000000

carage2      21615    0.066343       0.248886        0     1.000000

carage3      21615    0.026186       0.159690        0     1.000000

carage4      21615    0.012399       0.110660        0     1.000000

carage5      21615    0.007865       0.088337        0     1.000000

carage6      21615    0.004673       0.068199        0     1.000000

carage7      21615    0.002082       0.045581        0     1.000000

carage8      21615    0.000879       0.029636        0     1.000000

carage9      21615    0.000139       0.011781        0     1.000000

carage10     21615    0.000278       0.016659        0     1.000000

carage11     21615    0.000139       0.011781        0     1.000000

sexf         21615    0.690955       0.462110        0     1.000000

married      21615    0.452324       0.497733        0     1.000000

city         21615    0.483506       0.499740        0     1.000000

arean        21615    0.418136       0.493264        0     1.000000

areas        21615    0.277076       0.447565        0     1.000000

areaeast     21615    0.028221       0.165608        0     1.000000

catpcd_1     21615    0.976313       0.152076        0     1.000000

catpcd_2     21615    0.023687       0.152076        0     1.000000

tramak_n     21615    0.084201       0.277695        0     1.000000

tramak_f     21615    0.112052       0.315438        0     1.000000

tramak_h     21615     0.07814       0.268398        0     1.000000

tramak_t     21615     0.56049       0.496339        0     1.000000

tramak_c     21615    0.092297       0.289452        0     1.000000

age2         21615    0.109785       0.312629        0     1.000000

age3         21615    0.834744       0.371420        0     1.000000

age4         21615    0.018367       0.134277        0     1.000000




                                                28
                               Table 3
Summary Statistics for Data on Non-dealer-owne d Agents in 1999
Variable      N        Mean         Std Dev         Minimum   Maximum

accident     40027    0.219677      0.414033          0       1.000000

coverage     40027    0.537837      0.498573          0       1.000000

prem         40027     28224         17887            212     259011

carage0      40027    0.266120      0.441934          0       1.000000

carage1      40027    0.293102      0.455191          0       1.000000

carage2      40027    0.187523      0.390336          0       1.000000

carage3      40027    0.101806      0.302397          0       1.000000

carage4      40027    0.069028      0.253506          0       1.000000

carage5      40027    0.042596      0.201948          0       1.000000

carage6      40027    0.021486      0.144998          0       1.000000

carage7      40027    0.010493      0.101897          0       1.000000

carage8      40027    0.003972      0.062902          0       1.000000

carage9      40027    0.001949      0.044101          0       1.000000

carage10     40027    0.000999      0.031597          0       1.000000

carage11     40027    0.000150      0.012243          0       1.000000

sexf         40027    0.595773      0.490748          0       1.000000

married      40027    0.602368      0.489415          0       1.000000

city         40027    0.538661      0.498509          0       1.000000

arean        40027    0.488545      0.499875          0       1.000000

areas        40027    0.270717      0.444336          0       1.000000

areaeast     40027    0.040997      0.198287          0       1.000000

catpcd_1     40027    0.969321      0.172450          0       1.000000

catpcd_2     40027    0.025408      0.157362          0       1.000000

tramak_n     40027    0.183576      0.387143          0       1.000000

tramak_f     40027    0.157119      0.363917          0       1.000000

tramak_h     40027    0.083369      0.276442          0       1.000000

tramak_t     40027    0.210533      0.407692          0       1.000000

tramak_c     40027    0.128988      0.335191          0       1.000000

age2         40027    0.119344      0.324197          0       1.000000

age3         40027    0.818548      0.385398          0       1.000000

age4         40027    0.021461      0.144915          0       1.000000




                                               29
                             Table 4
Summary Statistics for Data on Dealer-owne d Agents in 2000
Variable      N           Mean         Std Dev       Minimum   Maximum

accident     21533      0.414527       0.492652        0       1.000000

coverage     21533      0.767798       0.422247        0       1.000000

prem         21533       33266          23420          458     505170

carage0      21533      0.700274       0.458149        0       1.000000

carage1      21533      0.146612       0.353727        0       1.000000

carage2      21533      0.074583       0.262724        0       1.000000

carage3      21533      0.037199       0.189253        0       1.000000

carage4      21533      0.020620       0.142110        0       1.000000

carage5      21533      0.009381       0.096402        0       1.000000

carage6      21533      0.005526       0.074136        0       1.000000

carage7      21533      0.003390       0.058128        0       1.000000

carage8      21533      0.001393       0.037301        0       1.000000

carage9      21533      0.000060       0.024564        0       1.000000

carage10     21533      0.000139       0.011803        0       1.000000

carage11     21533      0.000279       0.016691        0       1.000000

sexf         21533      0.722612       0.447720        0       1.000000

married      21533      0.551804       0.497321        0       1.000000

city         21533      0.512237       0.499862        0       1.000000

arean        21533      0.412065       0.492218        0       1.000000

areas        21533      0.289648       0.453610        0       1.000000

areaeast     21533      0.028886       0.167490        0       1.000000

catpcd_1     21533      0.980495       0.138295        0       1.000000

catpcd_2     21533      0.019505       0.138295        0       1.000000

tramak_n     21533      0.031579       0.174882        0       1.000000

tramak_f     21533      0.114290       0.318170        0       1.000000

tramak_h     21533      0.090094       0.286323        0       1.000000

tramak_t     21533      0.643710       0.478914        0       1.000000

tramak_c     21533      0.022291       0.147633        0       1.000000

age2         21533      0.095528       0.293949        0       1.000000

age3         21533      0.876701       0.328788        0       1.000000

age4         21533      0.014350       0.118932        0       1.000000




                                                30
                                 Table 5
Summary Statistics for Data on Non-dealer-owne d Agents in 2000
Variable      N        Mean          Std Dev         Minimum   Maximum

accident     42701    0.230323       0.421044          0       1.000000

coverage     42701    0.522798       0.499486          0       1.000000

prem         42701     25913          17063            217     351837

carage0      42701    0.225405       0.417853          0       1.000000

carage1      42701    0.239222       0.426613          0       1.000000

carage2      42701    0.204866       0.403609          0       1.000000

carage3      42701    0.135454       0.342211          0       1.000000

carage4      42701    0.080794       0.272522          0       1.000000

carage5      42701    0.054636       0.227271          0       1.000000

carage6      42701    0.032482       0.177278          0       1.000000

carage7      42701    0.014777       0.120661          0       1.000000

carage8      42701    0.007494       0.086244          0       1.000000

carage9      42701    0.002787       0.052717          0       1.000000

carage10     42701    0.001124       0.033509          0       1.000000

carage11     42701    0.000679       0.026052          0       1.000000

sexf         42701    0.618393       0.485787          0       1.000000

married      42701    0.703496       0.456721          0       1.000000

city         42701    0.546990       0.497793          0       1.000000

arean        42701    0.505140       0.499979          0       1.000000

areas        42701    0.256411       0.436656          0       1.000000

areaeast     42701    0.042973       0.202799          0       1.000000

catpcd_1     42701    0.971125       0.167458          0       1.000000

catpcd_2     42701    0.024215       0.153718          0       1.000000

tramak_n     42701    0.178520       0.382955          0       1.000000

tramak_f     42701    0.142549       0.349617          0       1.000000

tramak_h     42701    0.089272       0.285139          0       1.000000

tramak_t     42701    0.230018       0.420849          0       1.000000

tramak_c     42701    0.124400       0.330041          0       1.000000

age2         42701    0.104260       0.305601          0       1.000000

age3         42701    0.862673       0.344196          0       1.000000

age4         42701    0.019180       0.137159          0       1.000000




                                                31
Table 6 Coefficients of the Correlation Between Coverage and Claims in Automobile
                Insurance Using the Two-stage Method for Equations (2)
                                  First stage: estimate occurrence of claim

                                  Second stage: regress on choice of coverage

              Coefficients                    1                              2
                                            Year 1999
              claim  0                   1.3665***                      0.6779***

              claim  10000               1.5124***                      0.7706***

              claim  20000               2.1312***                      0.5601***

                                            Year 2000
              claim  0                   1.4582***                      0.8169***

              claim  10000               1.6258***                      0.9771***

              claim  20000               1.5359***                      0.7128***



      Note: The significance level of 99% is denoted by ***
            The significance level of 95% is denoted by **
            The significance level of 90% is denoted by *




                                                   32
                                                        Table 7
  Coefficients of the Correlation Between Coverage and Claims in Automobile Ins urance
                             Using the Two-stage Method for Equations (2’)
                    First stage: estimate occurrence of claim

                    Second stage: regress on choice of coverage

Coefficients                    1                              2        3
                                            Year 1999
claim  0                   1.3400***                      0.5296***   0.3895***

claim  10000               1.4937***                      0.6407***   0.3504***

claim  20000               2.1167***                      0.4787***   0.2269**

                                            Year 2000
claim  0                   1.4010***                      0.6613***   0.4633***

claim  10000               1.5792***                      0.8304***   0.4484***

claim  20000               1.5093***                      0.6606***   0.3439***



Note: The significance level of 99% is denoted by ***
     The significance level o f 95% is denoted by **
     The significance level o f 90% is denoted by *




                                                           33

				
DOCUMENT INFO