The asymmetric information problem within the health insurance market : an empirical study on the Dutch healthcare system

(1)

The asymmetric information problem within the

health insurance market

An empirical study on the Dutch healthcare system

Beyza Özkul 10573089

Supervised by dhr. dr. A. P. Kiss

(2)

1

Statement of Originality

This document is written by Beyza Özkul who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

2

2.4 Hypothesis 8 3. Methodology 10 3.1 Research setting 10 3.2 Data collection 11 3.3 Data variables 11 3.3.1 Dependent variables 11 3.3.2 Independent variables 15 3.3.3 Control variables 15 3.3.4 Interaction terms 17 3.4 Empirical design 17 3.4.1 Empirical model 17 3.4.2 Difference measurement 18 4. Results 19

4.1 Testing the hypothesis 20

4.2 Additional analysis 23

5. Discussion 26

5.1 Findings 26

5.2 Contributions and implications for the literature 27

5.3 Practical implications 27

5.4 Limitations 27

5.5 Directions for future research 28

6. Conclusion 29

References 30

Appendix A – Summary statistics: Interaction terms 32

Appendix B – Regression analysis on different cost categories 33 Appendix C – Summary statistics of different variables on the logarithm 36

of the difference between the optimal contract and the chosen contract

Appendix D – Regression analysis on the logarithm of the difference 37 Between the optimal contract and the chosen contract

(4)

3

1. Introduction

Asymmetric information problems are highly discussed among many researchers. These problems, pioneered by Arrow (1963) and Pauly (1968), are central to modern economic models of insurance. The Dutch health system provides several methods to minimize these asymmetric information problems. Two types of contracts are known to prevent adverse selection, i.e. the standard health insurance and the additional health insurance. Within the standard health insurance, deductibles are present to inhibit the moral hazard effect. Previous empirical studies have investigated different insurance markets and generated mixed results about the presence of information asymmetry. For instance, Cutler and Reber (1998) and Fang, Keane, and Silverman (2008) found asymmetry in the health insurance market, while Cardon and Hendel (2001) did not.

The aim of this thesis is to investigate asymmetric information within the Dutch health insurance market by providing an empirical study. The main emphasis lies on the voluntary deductibles and in which way this part of standard health insurance affects the decision-making and risk-taking of policyholders. In order to find a positive coverage-risk correlation, the total declared costs at yearend are regressed on the choice for voluntary deductibles. In this case, the choice of a voluntary deductible is used as a risk measurement. Two types of additional regressions will be performed to disaggregate moral hazard and adverse selection. Firstly, a proxy for the healthiness of insureds will be added to capture the adverse selection effect. After that, the same regression will be performed, except for the fact that the variables previously used are regressed on the specific cost categories. The results will be used to conclude which asymmetric information effect influences the ex post total declared costs.

All performed regressions show a negative relation between the amount chosen for deductibles and the total declared costs. This corresponds to a positive coverage-risk correlation and shows the presence of information asymmetry within the Dutch standard health insurance market. After adding the residual of the general practitioner costs as an independent variable, the remaining effect can be explained as ex ante moral hazard. The strongest moral hazard effect is found in the costs made for specialist medical care and pharmaceutical care, as policyholders with voluntary deductibles spend a minimum of 50 percent less on these categories. The moral hazard effect is significantly not different from zero and therefore negligible for costs made abroad and medical transportations.

An insurance contract with deductibles is seen as the optimal contract if the probability of accidents can be influenced by the individual (Hölmstrom, 1979). The main study showed

(5)

4 that even with deductibles, information asymmetry is still present. An additional analysis was performed to learn about the optimal contract within this setting. This analysis showed that individuals generally do not choose for the optimal contract. The most common mistake is that individuals choose to insure themselves more, by choosing the zero voluntary deductibles contract, even though the 500 contract is more beneficial. Choosing the optimal contract, if it existed, would lead to the good risk individuals being driven out of the market, which is in line with Akerlof’s adverse selection theory (1970).

Below, in section 2, related literature that is relevant to the empirical analysis will be reviewed. In section 3, the methodology is explained. Furthermore, in section 4 the results are shown and discussed, after which the findings will be discussed in section 5. Finally, the conclusion is given in section 6.

(6)

5

2. Literature review

In this part, previous literature concerning information asymmetry within different insurance markets will be reviewed. The first part includes theoretical literature, where types of adverse selection and moral hazard are discussed. Following that, empirical research is included to verify the theoretical literature.

2.1 Theoretical research

2.1.1 Adverse selection

Akerlof (1970) discussed adverse selection within the second-hand automobile market. He does so using his well-known “lemons” example, where the lemons are the bad quality cars in the market. Buyers cannot differentiate good quality from bad quality beforehand, so both are sold at the same price. As a result, sellers of good quality cars are unwilling to sell and are eventually driven out of the market. If we compare this to the (health) insurance setting, the bad risk individuals drive out the good risk individuals as the price for insurance rises. This is an example of market inefficiency. Eventually, this inefficiency will lead to the elimination of the insurance market.

The main problem is that the insurer does not have complete information about the individuals. Spence (1978) argues that through a menu of insurance contracts the risk type of the individuals will be revealed, and this could be a solution to the asymmetry problem. Crocker and Snow (1986) discuss risk classification as another solution. Risk classification is the system in which insurers use previously observed characteristics to categorize individuals in risk groups and charge different premiums. Furthermore, Dionne and Doherty (1994) show that experience rating can lead to a separating equilibrium in multi-period contracts. High-risk policyholders will show their true risk type in period 1, and the insurers will use this information to change the level of the premium. But even if the insurer and the policyholder have the same information and can gain new, identical information over time, an asymmetric information problem may still exist. This could be due to the fact that the individual learns faster about his risk type. This phenomenon is called asymmetric learning (Cohen and Siegelman, 2010).

(7)

6 2.1.2 Moral hazard

Winter (1992) states that moral hazard is the unfavourable effect on individuals’ incentives to avoid losses. For example, policyholders of gadget insurances tend to be less careful with their mobile phone when they know that damages due to accidents that could possibly occur will be reimbursed by their insurer. Researchers often differentiate between ex ante and ex post moral hazard. Ex ante moral hazard is the incentive that is not taken to avoid expected costs when the insurance coverage is higher. Ex post moral hazard, on the other hand, refers to the situation in which the costs have already been made, but the policyholder has the incentive to declare or spend more than necessary (Picard, 1996).

Additionally, deductibles are frequently used in accident insurance markets. Hölmstrom (1979) argues that a deductible is the optimal policy if the insured’s action affects the probability, but not the size of an accident. If the size can be affected, but the probability of the loss cannot, the ideal contract holds full coverage for small losses, and less coverage for larger losses (Winter, 1992).

2.2 Empirical research

As mentioned before, Dionne (2010) points out that it is highly recommended to empirically verify the asymmetric information theory provided by researchers like Arrow (1963) and Pauly (1986). The main motivation was to distinguish theoretical facts from quantitative facts. As a result, the literature is applied in various insurance markets, e.g. the automobile insurance market, life insurance market, and health insurance market. Many scientists use the correlation between the insurance coverage and the individual risk to determine whether information asymmetry is present. A positive coverage-risk correlation implies an information problem (Cohen and Siegelman, 2010).

Moreover, Cohen and Siegelman (2010) discuss a few papers and show that within the same market different conclusions can be reached. For instance, Cutler and Zeckhauser (1998) do find adverse selection within the health insurance market, whereas Fang, Keane, and Silverman (2008) do not find this information asymmetry. Instead, they found that individuals with a better health status tend to purchase the additional Medigap insurance, and therefore have less personal risk compared to non-purchasers (Fang, Keane, and Silverman, 2008). Going back to Cutler and Zeckhauser (1998), who did find an information inefficiency (i.e., adverse selection), several insurance policies offered by Harvard University and the Group

(8)

7 Insurance Commission (GIC) of Massachusetts were reviewed. Both authorities provide multiple insurance contracts individuals can choose from. In the article, they assume that people choose a contract based on a cost-benefit calculation, in which frequent health users demand a higher coverage. Nevertheless, general preferences will also influence the decision making, which may be negatively related. When these preferences overshadow the cost-benefit calculation, there will be no adverse selection. To exclude this statement, Cutler and Zeckhauser examine the causes of the dis-enrolment from the extensive package. They concluded that these employees were younger and healthier than the employees who stayed in the extensive package. In 1997, the extensive contract became untenable and was cancelled due to the adverse selection problem. For GIC the same problem occurred. They found that in the standard healthcare contract, 25 percent of the population is over 45 years old. In the extensive healthcare contract, however, 50 percent is 45 years or older. They also found that in the extensive contract more heart attacks and pregnancies took place. Cutler and Zeckhauser conclude that adverse selection is present and is the reason as to why relatively healthy people opt for the standard healthcare contract. From this article we can conclude that adverse selection is not only verified within the theory, but also in empirical settings.

2.3 Disentangling adverse selection and moral hazard

The main focus of Cutler and Zeckhauser’s paper was adverse selection. However, adverse selection and moral hazard both result in the same effect, and are therefore hard to distinguish from one another in empirical research. Cohen and Siegelman (2010) deduct three ways in the significant literature testing of moral hazard.

Firstly, a randomized or a natural experiment, which is an exogenous reason that leads to different insurance coverage or changes in policyholders’ setting. The assumption with this kind of experiment is that the exogenous shock only changes the policyholders’ behaviour and not their underlying risk. Consequently, these changes in behaviour can only be explained by “hidden actions” and are a result of moral hazard. Manning, Newhouse, Duan, Keeler, and Leibowitz (1987) find results that are consistent with this (ex post) moral hazard theory. In their experiment, different levels of coverage were randomly assigned to individuals. Individuals with a higher coverage are expected to have higher spending, and therefore are costlier to the insurer. This effect is what Manning et al. find. These kinds of randomized experiments are often impossible or too expensive to conduct. Researchers therefore use natural experiments

(9)

8 such as exogenous reasons caused by new policies or regulations. An example of such a natural experiment would be the study performed by Chiappori, Durand, and Geoffard (1998). The French health insurance replaced the full coverage with a 10 percent co-payment. They found moral hazard in some dimensions, e.g. the introduction of the co-payment did not affect the on doctor office visits, but affected the doctor home visits.

Secondly, dynamic properties of moral hazard and adverse selection are used to distinguish the effect. This is possible due to the different relationships the two conditions have between past and future risk. In the automobile insurance Abbring, Chiappori, Heckman, and Pinquet (2003) provide a model that uses dynamic data. They use a “bonus-malus” system in which the height of the future premium depends on the prior claim history. Moral hazard should therefore result in a negative correlation between the number of previous claims and accidents in a subsequent year. Adverse selection is a reflection of the policyholder’s risk type and consequently claims in the history and claims in the future should be positively correlated. Abbring et al. do not find moral hazard in the automobile insurance market in France, where this “bonus-malus” system is obligatory.

Finally, to untangle moral hazard and adverse selection, the focus is on disaggregating the coverage-risk correlation. This method is used for a static, single period analysis. Cohen (2005), for example, pursued this approach. After finding the positive coverage-risk correlation under all policies, she shows that this correlation interacts with characteristics which are easier to explain under adverse selection. In her paper she concludes that the correlation only exists for policyholders who have three or more years of driving experience. This is consistent with adverse selection, where their risk type is discovered after three years of experience. Moral hazard cannot conclude this, because expected losses causes experienced drivers to take precautions, but not the new drivers.

2.4 Hypothesis

In this thesis, I will research the effect of voluntary deductibles within the Dutch healthcare system on individual risk, measured in total declared costs. To find the presence of information asymmetry, I will look for a positive coverage-risk correlation. In my case, this will be a negative correlation between the height of the deductible and the total declared costs at yearend. Adding a higher voluntary excess to the standard deductible will reveal the risk type of an

(10)

9 individual, which will be relatively lower. This will result, according to the theory, to lower costs.

Both adverse selection and moral hazard will cause this positive relationship. To explain the correlation that is found, another correlation must be measured to identify the effect. This will be done with the approach from Cohen (2005), because the data obtained is for a single period, i.e. only 2014. She uses disaggregating the coverage-risk correlation to explain which effect is the cause of the positive relation.

(11)

10

3. Methodology

The data, the variables used, and the model will be explained in this section.

3.1 Research setting

In 2008 the Dutch healthcare system was reorganized and two kinds of insurance are known: the standard health insurance and the additional health insurance. The government decides on the coverage provided by the standard contract. All insurance companies are non-governmental organizations and offer the same standard package. Beside this contract, which is compulsory for every Dutch citizen, the insurance company may offer some other additional packages. For each of the insurance contracts, the insured pays a monthly premium. For only the standard insurance, a deductible has to be paid up to a certain amount if there are actual costs made. This required amount started at 150 euros in 2008. In 2013 the compulsory deductibles increased from 220 to 350 euros and now in 2016 it equals 385 euros. The voluntary deductibles exist since the reorganization of the healthcare system and have never been changed over time.

In the year 2014, which will be investigated in this study, the compulsory deductibles equalled 360 euros. The premium for 2014 was 99.50 euros per month. Along with the required deductibles, the insureds can choose a voluntary deductible each year. On top of the 360 euros, they can add an amount of 100, 200, 300, 400, or 500 euros. After choosing one of the deductible add ups, the policyholder receives a discount on the monthly premium. Together with choosing no voluntary deductibles, this splits the policyholders in six different categories. Within the same year, the insured cannot change his or her choice for the voluntary deductibles. After choosing one of these options, the monthly premium can be calculated as shown in table 1.

Table 1 - Premiums and deductibles for each voluntary deductible of choice

Monthly Discount Monthly Premium Yearly Premium Total Deductibles Max. Amount Difference with 0 0 € - € 99.50 € 1,194.00 € 360.00 € 1,554.00 € - 100 € 3.40 € 96.10 € 1,153.20 € 460.00 € 1,613.20 € 59.20 200 € 6.50 € 93.00 € 1,116.00 € 560.00 € 1,676.00 € 122.00 300 € 9.50 € 90.00 € 1,080.00 € 660.00 € 1,740.00 € 186.00 400 € 12.30 € 87.20 € 1,046.40 € 760.00 € 1,806.40 € 252.40 500 €25.00 € 74.50 € 894.00 € 860.00 € 1,754.00 € 200.00

(12)

11

3.2 Data collection

The data used for this empirical study are collected by ONVZ Zorgverzekeraar, located in Houten, The Netherlands. This health insurance company is medium-sized and has around 450,000 insureds, which was the original number of observations. The voluntary deductibles that are discussed in this thesis will only affect the people with a deductible, therefore people under the age of 18 will be left out. In order to make sure the outliers will not affect the results, 1 percent of the biggest observations for declared costs will also be left out. This results in approximately 330,000 observations, which were all fully anonymized before the company provided me with the data.

First of all, the data covers the choice of voluntary deductibles made in a single period, the year 2014. Furthermore, the total amount declared per cost category and the amount paid by the policyholder for the deductibles are known. The data also provides the information if the policyholder has chosen an additional health insurance contract, or dental health insurance contract. For each participant, age, gender and year of entrance is known as well.

3.3 Data variables

3.3.1 Dependent variables

All dependent variables were originally expressed in euros. These amounts could increase considerably, and therefore the logarithms of these amounts are taken. Values of zero within all the dependent variables are replaced with 1 so that the logarithms are equal to zero.

Total Declared Costs. The total declared costs that influence the deductibles are used as a risk measure. Not included are the costs made for the general practitioner, antenatal care, postnatal care, and most costs under the category “others”, e.g. preventive care like quitting smoking programs.

(13)

12 To explain the correlation found, the policyholder’s characteristics are also regressed on the individual cost categories. The costs made in the following categories are also used as dependent variables. A few examples or descriptions will be given to explain these different cost categories.

Pharmaceutical Care Costs. Within this risk measurement, we find registered medicines and dietary preparations.

Figure 2: Average pharmaceutical care costs per age category

Dental Health Care Cost. This cost category differs from the dental health insurance, because not all expenses made within this category qualify for reimbursements. The insured’s expenses are considered to be eligible for reimbursements if the treatment is the cause of a growth disorder, the consequences are the result of a mental or physical condition unrelated to dental health, or due to another medical treatment.

(14)

13 Specialist Medical Care Costs. Costs that fall within this cost category are mostly made by specialists, e.g. oncologist, cardiologist, and orthopaedist.

Figure 4: Average specialist medical care costs per age category

Allied Health Care Costs. The head of expenditure within this cost category includes visits to the physiotherapist, but also the speech therapist.

Figure 5: Average allied health care costs per age category

Medical Appliances Costs. Test materials for diabetes patients, professional wigs, and artificial parts of the body, e.g. eyes, legs, and arms.

(15)

14 Medical Transportation Costs. Medical transportation by ambulance, taxi, and private car are a few of the kinds of transportation costs that are involved within the deductibles. For example, these costs occur when travelling to hospitals for a kidney dialysis or in case of emergency.

Figure 7: Average medical transportations costs per age category

Mental Health Care Costs. Visits to the psychologist and psychiatrist are examples that influence the deductibles to be paid.

Figure 8: Average mental health care costs per age category

Abroad Costs. The costs made in case of illness or an accident during time spent outside the Netherlands on vacation or for work occasions. Another case is, if the insured wants to undergo a treatment outside the Netherlands which is covered by the standard health insurance.

(16)

15 3.3.2 Independent variables

The choice for one of the voluntary deductibles is given by dummy variables. If the values of all these dummy variables are equal to zero, the selected individual has chosen no voluntary add up, and the deductibles are equal to 360 euros. The dummy variables for the deductibles are a coverage measurement, where a higher voluntary deductible stands for a lower coverage.

100. = 1 if the value of the voluntary deductibles is equal to 100 0 otherwise

GP. The following regression has been done to calculate the residual for the costs made for the general practitioner:

𝐺𝑃# = 𝛼&+ 𝛽)𝐹𝑒𝑚𝑎𝑙𝑒# + 𝛽/𝐴𝑔𝑒 𝐶𝑎𝑡𝑒𝑔𝑜𝑟𝑦 𝐷𝑢𝑚𝑚𝑖𝑒𝑠# + 𝜀#

The residual is used as a proxy for the healthiness of an individual. The costs that are made for this category are not involved in the measurement of deductibles that are to be paid by the insured. If the individual feels sick, there is no financial argument that will hinder to not going to the GP. This variable should capture the adverse selection effect within the choice for the deductibles. Individuals with a higher risk will visit the GP more often and therefore make higher costs. The difference in total declared costs at yearend will be explained by the healthiness of the insured, and not by the choice for one of the deductibles.

3.3.3 Control variables

The age of the policyholders is included in order to capture the costs made due to change in age. For this a system of dummy variables is created, where the base category is an age between

(17)

16 18 and 29 years old. Meerding et al. (1998) discussed that costs for healthcare slowly rose until the age of 50, and after that increased exponentially until the oldest age category of over 90 years.

30-44. = 1 if the age is between 18 and 29 0 otherwise

Additional Insurance. A dummy variable to control for the healthiness of

policyholders. High risk individuals will choose an additional insurance alongside their compulsory standard health insurance. This fits within the adverse selection theory of Akerlof (1970).

AI. = 1 if the individual has an additional insurance 0 otherwise

Dental Health Insurance. This control variable is added to investigate the relation between the regular health and the dental health. Dental health insurance is a specific form of the additional insurance The same relation is expected as in the additional insurance, but with a lower intensity. This variable is expressed as a dummy variable.

DHI. = 1 if the individual has a dental health insurance 0 otherwise

Year of Entry. This data was obtained from a specific system of the company, and covers only entrance in the period this system was used. Even if entrance was earlier, 2006 is mentioned as year of entry. Therefore, the period of entry covered is from 2006 until 2014. This variable may be a measurement of insured satisfaction.

(18)

17 3.3.4 Interaction terms

A few interaction terms are included to see if females from different age categories and with different choices of deductibles, make significantly more or less costs compared to males. The following interaction terms are used to measure the gender effects:

Female*Age Category (i.e. 30-44, 45-59, 60-74, 75-89, and 90-105); Female*Deductibles (i.e. 100, 200, 300, 400, and 500).

3.4 Empirical design

3.4.1 Empirical model

Cohen and Siegelman (2010) review the following model as commonly used by researchers to investigate whether or not a coverage-risk correlation is present:

𝑅𝑖𝑠𝑘_# = 𝛼_&+ 𝛽₎𝐶𝑜𝑣𝑒𝑟𝑎𝑔𝑒_#+ 𝛾_/𝑋_# + 𝜀_#

Where 𝑅𝑖𝑠𝑘# is the dependent variable that represents the ex post realization of the insured’s

risk, 𝐶𝑜𝑣𝑒𝑟𝑎𝑔𝑒_# is an independent variable representing the insured’s choice of coverage, and 𝑋_# are other characteristics of the insured that may be relevant for classifying the ex post risk. Based on the literature by Cohen and Siegelman (2010) the model that will be analysed to find this coverage-risk correlation, is defined by:

ln (𝑇𝑜𝑡𝑎𝑙 𝐷𝑒𝑐𝑙𝑎𝑟𝑒𝑑 𝐶𝑜𝑠𝑡𝑠)_#

= 𝛼_& + 𝛽₎100_# + 𝛽_K200_# + ⋯ + 𝛽_N500_# + 𝛾_/𝐶𝑜𝑛𝑡𝑟𝑜𝑙 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 + 𝛿_/𝐼𝑛𝑡𝑒𝑟𝑎𝑐𝑡𝑖𝑜𝑛 𝑇𝑒𝑟𝑚𝑠 + 𝜀_#

For the specific cost categories, the formula is as specified below:

ln (𝐶𝑜𝑠𝑡 𝐶𝑎𝑡𝑒𝑔𝑜𝑟𝑦)_# =

= 𝛼_&+ 𝛽₎100_# + 𝛽_K200_# + ⋯ + 𝛽_N500_#+ 𝛾_/𝐶𝑜𝑛𝑡𝑟𝑜𝑙 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 + 𝛿/𝐼𝑛𝑡𝑒𝑟𝑎𝑐𝑡𝑖𝑜𝑛 𝑇𝑒𝑟𝑚𝑠 + 𝜀#

(19)

18 The independent variable 𝐺𝑃_# is also added to some models, to investigate whether the effect on the ex post health expenditures are influenced by this variable. This variable is also added to see if the choice for one of the deductibles is affected by this illness measurement.

3.4.2 Difference measurement

To determine if adding the illness measurement leads to significant differences within the choice for voluntary deductibles, a Z-test is done. The following line presents the calculation of the Z-value:

𝑍 = 𝛽V#WX YZ − 𝛽V#WX\]W YZ 𝜎V#WX\]W YZ_{+ 𝜎}V#WX YZ

The Z-values of 1.645, 2.055, and 2.33 correspond to significance levels of respectively 5, 2, and 1 percent.

(20)

19

4. Results

In this part, the hypothesis made in section 2 will be tested. After that, results of an additional analysis will be given.

Table 2 - Summary statistics

Variable Number of Obs. Mean Standard Deviation Minimum Maximum Control Variables Age 330,666 49.3671 17.0314 18 105 30-44 330,666 0.2288 45-59 330,666 0.3280 60-74 330,666 0.2173 75-89 330,666 0.0655 90-105 330,666 0.0058 Female 330,666 0.4879 Additional Insurance 330,666 0.7905 Dental Health Insurance 330,666 0.5240

Year of Entry 330,666 2007.982 2.9869 2006 2014 Independent Variables 100 330,666 0.0097 200 330,666 0.0236 300 330,666 0.0101 400 330,666 0.0041 500 330,666 0.1193 General Practitioner 330,666 0.0000 101.1493 -346.36 5207.87 Dependent Variables ln (Total Declared Costs) 330,666 4.6861 2.9493 0 9.9848 ln (Pharmaceutical Care) 330,666 3.2726 2.4482 0 9.9707 ln (Dental Health Care) 330,666 0.1372 0.8866 0 9.5379 ln (Specialist Medical Care) 330,666 3.6360 3.2581 0 12.4429 ln (Allied Health Care) 330,666 0.2386 1.1573 0 9.4150 ln (Medical Appliances) 330,666 0.6072 1.7843 0 9.4150 ln (Medical Transportation) 330,666 0.1630 1.0238 0 9.3424 ln (Mental Health Care) 330,666 0.3227 1.5032 0 9.9659 ln (Abroad) 330,666 0.1637 0.9531 0 9.9659

(21)

20

4.1 Testing the hypothesis

The hypothesis predicts that higher deductibles lead to less ex post declared costs. A log-linear regression is used to interpret the resulting coefficients more easily. The dependent variable is transformed into a logarithm. Therefore, all policyholders that did make zero costs within the year 2014 are replaced with costs of 1 euro, and consequently, the outcome of the logarithm will be zero. Table 2 gives an overview of all variables.

The summary statistics make clear that the base category for age (18-29) comprises around 15 percent of the total policyholders. Most of the insureds are found in the category 45-59 years old, approximately 110,000 people. Moreover, the additional insurance and the dental health insurance are both chosen more than 50 percent of the times, which may reveal hidden characteristics of the insured. Furthermore, table 2 shows that when people pick a voluntary deductible, most of them tend to select the highest category of 500, namely around 12 percent. As shown in table 1, choosing the highest voluntary deductible is relatively the most beneficial choice.

In table 3 we find the results of the regression on the total declared costs, and additionally the model with the residual of the GP regression is included as an explanatory variable. In both models the interaction terms are involved (see appendix A). Females do not significantly make less costs compared to men within the same voluntary deductibles category. Furthermore, we see the coefficients for every age category. As people get older, they make more costs. In the table is shown that the costs increase the most as insureds get in the age category of 90-105 years old. The outcomes are similar to the findings of Meerding et al. (1998). People who have an additional insurance package make more costs. Furthermore, every year extra an individual is insured, leads to significantly less costs.

We see that the choice of voluntary deductibles has a significantly negative effect on the total declared costs at yearend. The higher the deductibles get, the less costs are made by the insureds. The highest voluntary deductibles make around 100 percent less costs compared to people without voluntary deductibles. However, one contradiction is that people with an amount of 300 on voluntary deductibles make less costs compared to people who have chosen 400. This outcome is also significant.

(22)

21

Table 3 - Regression analysis with various variables on the logarithm of the total declared costs

The unstandardized coefficients are given, between parentheses are the robust standard errors. ***P < 0.01 **P < 0.02 *P < 0.05

Model ln(Total Costs) ln(Total Costs) Difference 30-44 0.1734*** (0.0239) 0.1786*** (0.0225) 44-59 0.9523*** (0.0223) 0.9431*** (0.0211) 60-74 2.5174*** (0.0228) 2.5037*** (0.0215) 75-89 3.6026*** (0.0273) 3.6424*** (0.0283) 90-105 4.0159*** (0.0789) 3.6961*** (0.1280) Female 0.8462*** (0.0313) 0.8413*** (0.0293) 100 -0.6990*** (0.0702) -0.5977*** (0.0663) 0.1013 (0.1365) 200 -0.7856*** (0.0434) -0.7061*** (0.0410) 0.0795 (0.0844) 300 -1.0445*** (0.0652) -0.9231*** (0.0612) 0.1214 (0.1264) 400 -0.9960*** (0.0982) -0.8590*** (0.0930) 0.1370 (0.1912) 500 -1.0626*** (0.0211) -0.9787*** (0.0199) 0.0839* (0.0410) General Practitioner 0.0086*** (0.0002) Additional Insurance 0.5580*** (0.0179) 0.4512*** (0.0172) Dental Health Insurance 0.0551*** (0.0142) 0.0297* (0.0135) Year of Entry -0.0437*** (0.0017) -0.0430*** (0.0016) Constant 90.8305*** (3.5028) 89.4636*** (3.2791)

Interaction Terms Yes Yes

R² 0.1662 0.2532

Number of Observations

(23)

22 In order to test if illness influences the total declared costs, the variable GP is added to the regression. One of the reasons this variable is added is to see if it captures the effect of adverse selection within the choice for the deductibles. A euro increase in costs for the general practitioner leads to almost a 1 percent rise of the total declared costs. The coefficients of the deductibles are lower and significant. Furthermore, the table shows that the differences by adding the GP variable are only significant for the 500 contract. However, adding GP to the model does not capture the total coverage-risk correlation. The remaining coefficient must be explained otherwise.

The choice of deductible is regressed to the different cost categories (see appendix B). The GP variable is also included in this model. We see three effects for the deductible coefficients per cost category. Firstly, the percentages which lower the total declared costs stay more or less the same compared to the total costs. This is the case for pharmaceutical care and specialist medical care. Secondly, percentages which are lower compared to the total costs, e.g. dental health care, allied health care, medical appliances, and mental health care. Thirdly, a decrease after which the percentages are not significantly different from zero, and therefore negligible compared to the percentages that affect the total costs. This effect is seen by medical transportation and costs made abroad.

If the coefficients get closer to zero, it means the asymmetric information problem also decreases. If we assume that the GP variable is a good proxy for determining the illness, the adverse selection problem is explained by this variable. The remaining part is due to moral hazard. Moral hazard can be divided in two dimensions, ex ante and ex post. Ex ante moral hazard is related to the cautiousness of the insured before the costs have occurred (Winter, 1992). Ex post moral hazard refers to the actions made after the costs have occurred, e.g. fraud (Picard, 1996). The individual is not involved with the reimbursement of costs, because this is done directly from the care provider to the insurance company. This means the remaining part is due to ex ante moral hazard.

To explain this, a few examples are given. If the coefficient gets closer to zero, people do not want to or cannot make an effort to avoid these costs. Under medical transportation one can find transportation by ambulance, which is unavoidable in case of emergency and in most cases fills the complete deductible. Costs made abroad is the other category where the asymmetric information problem becomes negligible, which mainly includes accidents that cannot be avoided and must be immediately attended to. The categories where the coefficients stay around the same number are pharmaceutical care and specialist medical care. If the insured has to pay for the healthcare himself, he may choose not to buy the drugs and live with the pain

(24)

23 in case of pharmaceutical care. In the case of a treatment by a specialist, for example removing one’s wisdom teeth, one can decide to delay this treatment or to not get it at all.

4.2 Additional Analysis

Since in most cost categories choosing the highest voluntary deductible leads to making less costs, it seems useful to consider if choosing an even higher voluntary deductible will lead to making even less costs. To say something about the optimal contract, an additional analysis will be done. Table 1 shows the minimum and maximum amounts to be paid by the insured per chosen voluntary deductible. From this we can calculate that the optimal contract can be obtained when choosing a voluntary deductible of 500, if the total declared costs do not exceed 660 euros. If the total declared costs are equal to 660, the insured is indifferent between the 0 and 500 contract. If the costs are higher than 660, choosing no voluntary deductible leads to the optimal outcome.

The tables 4 and 5 show the percentages of insureds for the different optimal contract and contract of choice. This is split into two age categories: younger than 45 years old and 60 years or older.

As stated in the tables 4 and 5, more people choose the optimal contract in the older category compared to the younger category of under 45 years old. This could have two possible reasons. In the first place, this could mean that people learn over time from the choices they have made in the past. As these individuals get older, more experience and knowledge lead to a contract that matches their risk type better. Another reason could be that the costs they make, matches their contract of choice as they get older. The best contract if your costs are high is the contract with no voluntary deductibles. Table 5 shows that the contract with zero voluntary deductibles is the optimal one for around half of the people of 60 years or older. This is only 26 percent for the people in the category “< 45 years”. Assuming that people do not think carefully before making a choice and just choose this contract throughout their lifetime, the older they get the more this contract suits them. This is the case if individuals make more costs when they get older, which is indeed proven by the previous regression shown in table 3.

(25)

24

Table 4: Percentages for the optimal and chosen contract for people under the 45 years old

Optimal Contract < 45 years

Chosen Contract 0 100 200 300 400 500 Total 0 23.42% 0% 0% 0% 0% 57.07% 80.50% 100 0.20% 0% 0% 0% 0% 1.03% 1.24% 200 0.30% 0% 0% 0% 0% 1.69% 1.99% 300 0.14% 0% 0% 0% 0% 0.90% 1.04% 400 0.04% 0% 0% 0% 0% 0.24% 0.28% 500 1.49% 0% 0% 0% 0% 13.48% 14.97% Total 25.60% 0% 0% 0% 0% 74.40% 36.60%

Table 5: Percentages for the optimal and chosen contract for people above the 60 years old

Optimal Contract > 60 years

Table 6: Percentages for the optimal and chosen contract for females under the 45 years old

Optimal Contract Female

(26)

25

Table 7: Percentages for the optimal and chosen contract for males under the 45 years old

Optimal Contract Male

Furthermore, tables 6 and 7 show the differences between gender. The insureds under 45 years old will be used to compare these two categories. The tables show that males opt for the optimal contract more often than females. This is the case in the 0 and 500 contracts, which could mean that males are more aware of their risk type compared to females.

In all of the categories, the conspicuous mistake is that people choose the 0 contract, even though the 500 contract is the optimal one. Choosing the 500 contract will lead to a discount of 300 euros up to €660 on total declared costs. When exceeding 660 euros, the maximum loss will be 200 euros.

In appendix D the results of the following regression are shown:

ln (𝐷𝑖𝑓𝑓𝑒𝑟𝑒𝑛𝑐𝑒)# =

= 𝛼_&+ 𝛽₎100_# + 𝛽_K200_# + ⋯ + 𝛽_N500_#+ 𝛾_/𝐶𝑜𝑛𝑡𝑟𝑜𝑙 𝑉𝑎𝑟𝑖𝑎𝑏𝑙𝑒𝑠 + 𝛿_/𝐼𝑛𝑡𝑒𝑟𝑎𝑐𝑡𝑖𝑜𝑛 𝑇𝑒𝑟𝑚𝑠 + 𝜀_#

The coefficients verify the conclusions found in tables 4 to 7. The higher the amount for deductibles chosen, the smaller the difference between the optimal and chosen contract becomes. This would mean that an even higher voluntary deductible will lead to a more optimal contract for some individuals. The findings are in favour of the adverse selection theory by Akerlof (1970). He states that the adverse selection problem will lead to driving out all the good risk individuals of the market. For people with no ex post declared costs the optimal contract will be a contract with no monthly premiums and just deductibles. In other words, they will not take an insurance. This is indeed not possible due to the compulsory nature of the Dutch health insurance system. To conclude, for high risk individuals a contract with no voluntary deductibles is the optimal one. For low risk individuals, a contract with as high as possible deductibles is optimal, or in this case the excess of 500.

(27)

26

5. Discussion

5.1 Findings

The aim of this paper is to answer the question if voluntary deductibles within the Dutch healthcare system influence the asymmetric information problem. The results support that information asymmetry is present in the market for standard health insurance. All the regressions, including the regression for the total costs and the regressions for the specific cost categories, show that deductibles have a significant negative relation with respect to declared costs at yearend. This corresponds to the theory of Cohen and Siegelman (2010), where a positive coverage-risk correlation is present in case of information asymmetry.

Both adverse selection and moral hazard cause the positive coverage-risk correlation. To disaggregate the two effects, a few additional regression analyses are implemented which is comparable to the approach of Cohen (2005). The first method involved adding an explanatory variable: the residual of the costs made for the general practitioner. Assuming that this is a good proxy for the illness of individuals, it must attract the hidden characteristics that influence the ex post total declared costs. A significant positive correlation between the declared costs and the independent variable GP is found. Assuming that this variable captures the adverse selection effect, the remaining part of the deductibles coefficients that influences the costs made, is due to moral hazard. Previously, the reason why ex post moral hazard is not possible for the insured was explained, which is because the insured is not involved in the process of reimbursement. To explain ex ante moral hazard, the same model is regressed on the specific cost categories. These resulted in three different sorts of coefficients, which remain the same, decrease, or are not significantly different from zero compared to the former regression. In section 4.1 an explanation as to why this could be the case is given.

To find characteristics that influence the optimal decision making, an additional analysis is done. This results in the following findings. As age increases, better choices are made. This could be due to learning over time, or that the chosen contract fits the individual more as they get older. Furthermore, results show that males make better decisions than females. Moreover, the deductibles show that the adverse selection problem found by Akerlof (1970) is present. It is beneficial for good risk individuals not to buy insurance. In fact, these individuals are driven out of the market by the bad risk individuals.

(28)

27

5.2 Contributions and implications for the literature

Up until now, a lot of searching for the information asymmetry problem within different insurance markets is done (Cohen & Siegelman, 2010). In line with Cutler and Zeckhauser (1998), this study finds a positive coverage-risk correlation and therefore asymmetry within the health insurance market. However, there are some differences between those studies and this study. While Cutler and Zeckhauser (1998) find adverse selection, but do not include the fact that moral hazard has the same effect, this study tried to disaggregate the two effects by adding various changes to the original model. The results of this study contribute to the assumption that deductibles do not completely counteract the moral hazard problem, which is seen as the optimal contract if people can influence the probability of accidents to happen (Hölmstrom, 1979). However, further research is necessary to investigate the reason(s) underlying the positive coverage-risk correlation.

5.3 Practical implications

The findings of this study are relevant for parties involved within closing insurance contracts. Individuals, especially, can learn from the findings in this study. The results show that low risk individuals can benefit from choosing a contract with high deductibles. By way of contrast, only 15 percent of the policyholders younger than 45 choose for this optimal contract, whereas around 75 percent could benefit from choosing the contract with the highest voluntary deductibles.

5.4 Limitations

The results of this study may be valuable, though it has several limitations. The first limitation is internal validity. We assume that the costs made for visits to the general practitioner is a good explanatory variable to capture the adverse selection effect. Table 3 shows that adding GP as an independent variable leads to insignificant differences in most cases. An empirical explanation as to why the remaining results are due to the moral hazard problem is already given, when in fact another explanation could also be given. If this is not a good proxy, the coefficients still represent an adverse selection problem.

(29)

28 The second limitation is the assumption of individuals being equally risk averse, which is also assumed in the basic adverse selection model by Akerlof (1970). Personal characteristics could influence the choice for deductibles. For example, the magnitude in minimum and maximum costs at yearend is the highest for the voluntary deductible of 500. Even though an individual is of good risk, his risk aversion will force him to choose lower deductibles.

The last limitation can be found in another unobservable difference among policyholders. This is how many value individuals attach on the time value of money. More present orientated individuals could attach more value for discount on premium now than, potentially, higher deductibles in the future (Cohen & Siegelman, 2010). This could be the case for students, who do not have that much to spend and want to save on their monthly spending.

5.5 Directions for future research

The limitations of this study lead to several opportunities for future research. First of all, future research could look for a better explanatory variable to capture the adverse selection effect. Additionally, the study could also involve the additional insurance contract to see if there is a matter over overuse within the insurance market. Within the additional insurance we see more luxury treatments, e.g. teeth whitening and other aesthetic surgical interventions.

(30)

29

6. Conclusion

The aim of this study was to investigate the effect of voluntary deductibles within the Dutch healthcare system on information asymmetry. All performed regressions show the presence of asymmetric information within the Dutch standard health insurance market. After adding the residual of the costs made for the general practitioner as independent variable, the remaining effect is explained as ex ante moral hazard. However, the results of this paper show that a contract with deductibles, which is seen as the optimal contract if the probability of accidents can be influenced by the individual, is not reducing the complete asymmetric information problem. The paper also shows that individuals generally do not choose the optimal contract. Choosing the optimal contract, if it existed, would lead to the good risk individuals being driven out of the market. Two important limitations of this paper are internal validity, caused by using the residual of the costs for GP as an explanatory variable, and the market that is tested where some theoretical assumptions are made that may not be accurate in a real life setting. Furthermore, this study provides several directions for future research. An example of opportunities for future research includes investigating the same model, but with a database that is more detailed. Overuse of healthcare and another proxy for illness of individuals can be found in this way.

(31)

30

References

Abbring, J., Chiappori, P.-A., Heckman, J., & Pinquet, J. (2003). Adverse selection and moral hazard in insurance: Can dynamic data help to distinguish? Journal of the European Economic Association, 1(2-3), 512-521.

Akerlof, G. A. (1970). The market for “lemons”: Quality uncertainty and the market mechanism. The Quarterly Journal of Economics, 84(3), 488-500.

Arrow, K. J. (1963). Uncertainty and the welfare economics of medical care. The American Economic Review, 53(5), 941-973.

Cardon, J. H., & Hendel, I. (2001). Asymmetric information in health insurance: Evidence from the national medical expenditure survey. RAND Journal of Economics, 32(3), 408-427.

Chiappori, P.-A., Durand, F., & Geoffard, P. Y. (1998). Moral hazard and the demand for physician services: First lessons from a French natural experiment. European Economic Review, 42(3), 499-511.

Chiappori, P.-A., & Salanié, B. (2000). Testing for asymmetric information in insurance markets. Journal of Political Economy, 108(1), 56-78.

Cohen, A. (2005). Asymmetric information and learning in the automobile insurance market. Review of Economics and Statistics, 87(2), 197-207.

Cohen, A., & Siegelman, P. (2010). Testing for adverse selection in insurance markets. Journal of Risk and Insurance, 77(1), 39-84.

Crocker, K. J., & Snow, A. (1986). The efficiency effects of categorical discrimination in the insurance industry. Journal of Political Economy, 94(2), 321-344.

Cutler, D., & Reber, S. (1998). Paying for health insurance: The trade-off between competition and adverse selection. The Quarterly Journal of Economics, 113(2), 433-466.

Cutler, D. M., & Zeckhauser, R. J. (1998). Adverse selection in health insurance. Forum for Health Economics and Policy, 1: Article 2.

Dionne, G. (2012). The empirical measure of information problems with emphasis on insurance fraud and dynamic data. CIRPEE, 12-33.

Dionne, G., & Doherty, N. A. (1994). Adverse selection, commitment, and renegotiation: Extension to and evidence from insurance markets. Journal of Political Economy, 102(2), 209-235.

(32)

31 Fang, H., Keane, M. P., & Silverman, D. (2008). Sources of advantageous selection: Evidence

from the Medigap insurance market. Journal of Political Economy, 116(2), 303-350. He, D. (2008). The life insurance market: Adverse selection revisited. Working paper,

Department of Economics, Washington University in St. Louis. World Wide Web: http://artsci.wustl.edu/~econgr/gradconference/08/DaihengHe.pdf.

Hölmstrom, B. (1979). Moral hazard and observability. Bell Journal of Economics, 10(1), 74-91.

Manning, W. G., Newhouse, J. P., Duan, N., Keeler, E. B., & Leibowitz, A. (1987). Health insurance and the demand for medical care: Evidence from a randomized experiment. American Economic Review, 77(3), 251-277.

Meerding, W. J., Bonneux, L., Polder, J. J., Koopmanschap, M. A., & Van der Maas, P. J. (1998). Demographic and epidemiological determinants of healthcare costs in Netherlands: Cost of illness study. British Medical Journal, 317(7151), 111-115. Pauly, M. V. (1968). The economics of moral hazard: Comment. The American Economic

Review, 58(1), 531-537.

Pauly, M.V. (1974). Overinsurance and public provision of insurance: The roles of moral hazard and adverse selection. The Quarterly Journal of Economics, 88(1), 44-62. Picard, P. (1996). Auditing claims in the insurance market with fraud: The credibility issue.

Journal of Public Economics, 63(1), 27-56.

Spence, M. (1978). Product differentiation and performance in insurance markets. Journal of Public Economics, 10(3), 427-447.

(33)

32

Appendix A – Summary statistics: Interaction terms

Variable Number of Obs. Mean Female*(30-44) 330,666 0.1170 Female*(45-59) 330,666 0.1534 Female*(60-74) 330,666 0.1025 Female*(75-89) 330,666 0.0353 Female*(90-105) 330,666 0.0041 Female*AI 330,666 0.3969 Female*DHI 330,666 0.2609 Female*100 330,666 0.0046 Female*200 330,666 0.0096 Female*300 330,666 0.0040 Female*400 330,666 0.0015 Female*500 330,666 0.0504

(34)

33

Appendix B – Regression analysis on different cost categories

(35)

(36)

(37)

36

Appendix C – Summary statistics of different variables on the logarithm of

the difference between the optimal contract and the chosen contract

Variable Number of Obs. Mean Standard Deviation Minimum Maximum Control Variables Age 330,666 49.3671 17.0314 18 105 30-44 330,666 0.2288 0.4201 45-59 330,666 0.3280 0.4695 60-74 330,666 0.2173 0.4124 75-89 330,666 0.0655 0.2474 90-105 330,666 0.0058 0.0761 Female 330,666 0.4879 0.4999 Additional Insurance 330,666 0.7905 0.4069 Dental Health Insurance 330,666 0.5240 0.4994

Year of Entry 330,666 2007.982 2.9869 2006 2014 Independent Variables 100 330,666 0.0097 0.0982 200 330,666 0.0236 0.1518 300 330,666 0.0101 0.0998 400 330,666 0.0041 0.0637 500 330,666 0.1193 0.3242 General Practitioner 330,666 0.0000 101.1493 0 5207.87 Dependent Variable ln (Difference) 330,666 3.2991 2.7592 0 6.8161

(38)

37

Appendix D – Regression analysis on the logarithm of the difference between

the optimal contract and the chosen contract

The unstandardized coefficients are given, between parentheses are the robust standard errors. ***P < 0.01 **P < 0.02 *P < 0.05

Model ln(Difference) ln(Difference) 30-44 -0.0854*** (0.0170) -0.0887*** (0.0166) 44-59 -0.4466*** (0.0167) -0.4407*** (0.0163) 60-74 -1.6052*** (0.0194) -1.5964*** (0.0189) 75-89 -3.4247*** (0.0279) -2.8917*** (0.0279) 90-105 -3.4247*** (0.0866) -3.2206*** (0.1067) Female -0.0952*** (0.0255) -0.0921*** (0.0250) 100 1.2421*** (0.0245) 1.1774*** (0.0249) 200 1.5214*** (0.0146) 1.4706*** (0.0153) 300 1.2956*** (0.0199) 1.2181*** (0.0214) 400 1.3315*** (0.0312) 1.2439*** (0.0327) 500 -3.4290*** (0.0156) -3.4826*** (0.0160) General Practitioner -0.0055*** (0.0001) Additional Insurance -0.2088*** (0.0141) -0.1406*** (0.0140) Dental Health Insurance -0.0245* (0.0122) -0.0084 (0.0119) Year of Entry 0.0315*** (0.0015) 0.0310*** (0.0015) Constant -58.6969*** (3.0804) -57.8243*** (3.0078)

Interaction Terms Yes Yes

R² 0.1973 0.2378

Number of Observations

The asymmetric information problem within the health insurance market : an empirical study on the Dutch healthcare system