Supplementary health insurance and stock market participation
University of Groningen
Faculty of Economics and Business
MSc Finance
Master´s Thesis
January 2019
Name: Matúš Bilka
Studentnr: s3571394
Supervisor: dr. S. S. H. Eriksen
Abstract
Due to the continuously rising costs of health care, medical expenses start to represent an important form of financial risk. Individuals are taking out supplementary health insurance to decrease the risk of unexpected medical expenditures (the background risk). According to the economic theory individuals exposed to background risk are less willing to take other forms of risk. This thesis studies the relationship between the supplementary health insurance and the willingness of individuals to hold risky assets. Research is focused on the residents aged 50+ in seventeen European countries. All results are based on the wave 6 of Survey of Health and Ageing and Retirement in Europe (SHARE) database. The positive correlation between supplementary health insurance and stock/bond holding was found. Individuals with supplementary insurance are more likely to hold risky assets than those without it. Results are robust and indicate a significant link between supplementary health insurance and the investment behavior of the older Europeans.
Keywords: stock market participation, bond market participation, supplementary health insurance
2 1. Introduction
The increased longevity of European population makes a decision about the use of financial resources focused on smoothing of consumption during retirement very important. At the same time, the costs for health care are continuously rising and medical expenses start to represent an important form of financial risk. To the most endangered categories belong older people who are more likely to suffer from various diseases, often chronic. Because of the potentially high medical costs, many individuals take out supplementary health insurance.
To avoid welfare loss Cocco, Gomes, and Maenhout (2005) recommend participation in risky investments. However, stock market participation of European population of 50+, with exception of few countries, is considerably low. One of the explanations proposed was that transaction and information costs together with entrance barriers (fees, limits etc.) demotivated potential investors. In the last years, a boom of the internet reduced abovementioned costs significantly and made stock market participation easier. However, according to wave 6 of the Survey of Health, Aging and Retirement in Europe (SHARE) database, the participation in certain European countries is still less than 5%.
In the last decades, due to the increased popularity of behavioural finance, many authors tried to explain why some individuals are more likely to invest than the others and how their portfolio compositions change based on different factors. Many studies are focused on the examination of what are the determinants and how they affect the portfolio composition. Some of the most common determinants connected with investment choices are job occupation, financial literacy, education, wealth and risk attitudes. However, some seemingly less financial and more demographic and social aspects are also considered to determine investment strategies. Age, gender, sociability represented by community effects, word-of-mouth communication or even driving behaviour can be mentioned as an example.
There is a growing body of literature that recognises the importance of health risk as a determinant of portfolio composition choice. However, papers are not unified as to what should be tested. Some authors focused on objective health status and health shocks while others are more interested in subjectively assessed perceived health (Edwards, 2008; Atella, Brunetti, and Maestas, 2012).
3
of the United States. They estimated that supplementary insured are 7.1 percentage points more likely to hold risky assets if they are enrolled in less and 13.0 percentage points in more extensive supplementary health insurance.
Atella, Brunetti, and Maestas (2012) define medical expenditure risk as a function of the health risk and health insurance coverage. They suggest that whenever health risk leads to higher out-of-pocket medical expenditure risk, the financial risk of the household is increased, and two possible scenarios might occur in the sense of financial decisions. Firstly, there might be an increase in savings. Secondly, a financial portfolio will be changed, with the aim to decrease overall exposure to financial risks. In their work, they also showed that countries with more extensive National Healthcare Systems have higher stock market participation.
We are aware of the differences in National Healthcare Systems of European countries. These differences imply that supplementary health insurance is not covering exactly the same treatments everywhere. However, people with supplementary health insurance in any country are, with respect to out-of-pocket medical expenditure risk, relatively better off in comparison to the population without it. Based on the country, supplementary health insurance is usually decreasing the costs paid by the patient for e.g. prescribed drugs, dental care, optics, physiotherapist, additional check-ups, psychological help, private hospital treatment, and alternative medicine. The risk of unexpected medical expenditures is decreased by any supplementary health insurance owned. This was shown in the paper of Goldman and Maestas (2013), where they concluded that for an American sample both supplementary health insurances tested (less or more extensive), led to the higher probability that one will hold risky assets.
In this thesis we will estimate the relationship between supplementary health insurance and risky assets holdings, using wave 6 of the SHARE database containing data of the European population aged 50+. Much research has been conducted in the US on this topic, e.g. Goldman and Maestas (2013), Ayyagari and He (2017) or Lee (2018). Based on our knowledge there was no similar study done for Europe. We believe that conducting research for European countries would bring some more light into the topic. This thesis differs from the study of Goldman and Maestas (2013) in two aspects. Firstly, it is focused on the European population, living in different historical, economic, social and political climate in comparison with Americans and therefore the obtained results can be different. Secondly, we are testing the relation between supplementary health insurance and stock and bond market participation separately.
4
opposite, this would imply that people without supplemental health insurance, facing relatively higher medical expenditure risk, should be less likely to invest in stocks in comparison to respondents with better covered medical expenditure risk.
Research Question: Does the supplementary health insurance ownership positively affect the decision about stock market participation in the European countries?
This leads to the following hypotheses:
H1.0: Individuals with supplementary health insurance are not more likely to hold financial assets than individuals without it.
H1.1: Individuals with supplementary health insurance are more likely to hold financial assets than individuals without it.
If hypothesis 1.0 is rejected, then supplementary health insurance is an important determinant of the stock and bond market participation. Because we are using cross-sectional data, we are not able to test the causality. The SHARE data are suitable for testing if supplementary health insurance and stock market participation are related. If the positive relationship is found, there would be the possible implication for risk diversification. Furthermore, it would contribute to the explanation of the market participation puzzle. Obtained results may expand the knowledge of why some people operate in equity markets and others do not.
To evaluate the relation of the supplementary health insurance to the probability of risky assets holding, we will estimate the association with stocks and bonds separately. We expect that the supplementary insurance will be positively related to both of them. However, the correlation should be weaker for bonds, because they are believed to be less risky than stocks.
H2.0: The e
ffec
t on assets held, is significant and similar in magnitude for stocks and bonds.H2.1: The effect on the assets held, will be stronger for stocks than for bonds. Our results suggest that supplementary health insurance has a positive influence on risky investments. People with supplementary health insurance have a significantly higher probability of risky assets holding. However, the results do not provide enough evidence that bond investments are less correlated to the supplementary health investment than stocks.
5 2. Review of the literature
More than 30 years ago Mehra and Prescott (1985) proposed that returns on equity are too high with respect to the short-term treasury bills and this disproportion cannot be explained by standard equilibrium asset pricing model. They have found that the model would predict the risk premium of 0.35%, however, the market showed 6% on average during the period of 1889 - 1978. This phenomenon was called the equity premium puzzle. In the following years the surprisingly low stock market participation of the households was documented (Mankiw and Zeldes, 1991; Haliassos and Bertaut, 1995). Vissing-Jorgensen (1998) reported that the limited stock market participation should be considered an important aspect of the equity premium puzzle.
In the academic literature, the limited stock market participation is often referred to as the stock market participation puzzle. Part of the literature emphasised the importance of market imperfections as a possible cause of limited stock market participation. Bogan (2008) tested if low stock market participation can be a result of market frictions. With a boom of online trading transaction costs, information costs and limited access barriers were mitigated, leading to the possibility to test the influence of transaction costs on stock market decisions of individuals. His results showed that the participation rates of internet owning households raised substantially in comparison to those without internet, supporting the hypothesis that transaction and information costs are important determinants of stock market participation puzzle.
Other researchers tried to explain why some people participate in the stock market and others do not, with the use of individual traits rather than market frictions. Shefrin and Statman (2000) proposed behavioural portfolio theory and following studies showed how the rational investor changes his portfolio from bond to stock and then back depending on age (Binswanger, 2011). The pattern of decreasing exposure to stocks after retirement was also shown in the recent paper of Fagereng, Gottlieb, and Guiso (2017). Other examples of determinants of financial market participation behaviour tested, beside the age, included for instance sensation seeking derived from number of speeding tickets or overconfidence (Grinblatt and Keloharju, 2009), neighbourhood and social ties (Brown, Ivković, Smith, and Weisbenner, 2008; Hong, Kubik, and Stein, 2004; Hong, Kubik, and Stein, 2005), financial literacy (Gaudecker 2015; Balloch, Nicolae, and Phillip, 2014) and cognitive abilities (Christelis, Jappelli, and Padula, 2010).
6
many individuals. They also concluded that purchasers of life insurance are more likely to own stocks or invest to mutual funds.
One stream of literature is trying to connect objective health status, with financial risk-taking. Rosen and Wu (2004) tested the relationship between health and stock ownership and their conclusion was that households with ill members tend to invest less relatively to healthy ones. Berkowitz and Qiu (2006) decided to delve deeper into the topic and suggested that the health status and portfolio composition is linked not directly, but there is a link through financial wealth as intermediator. They have tested an impact of a sudden change in health status on the financial and non-financial wealth. They concluded that health is an important factor determining the wealth status and therefore indirectly also the financial portfolio composition. On the other hand, the study reported by Fan and Zhao (2009) suggested that the relationship between health and wealth does not have to be causal and can be driven by heterogeneity (an inter-individual variability in the population), arising from unobserved third factors. In their work, authors have used the longitudinal dataset in combination with the fixed-effects model, which allows to take into account and difference out unobserved heterogeneity. Estimates of Fan and Zhao (2009) also indicate that health shocks lead to the shift from risky to safer portfolio choice. However, the significant effect was observed only for two health indices (heart attack or stroke history and physical function) while for the other two (chronic conditions and work-related limitations) the effect was insignificant.
Also, Love and Smith (2010) tested if health is causally affecting portfolio choice, or if this effect is the result of heterogeneity, caused by unobserved characteristics such as risk attitudes, impatience, information, and motivation. Their conclusion is in favour of heterogeneity, signalling that health is not important as a determinant of portfolio choice.
Other authors tried to use perceived health or health insurance coverage, instead of health shocks, as a determinant of portfolio choice. This conception is based on the relation of background risk to financial risk. As defined in Guiso and Paiella (2008) background risk is a risk, which cannot be fully insured or avoided. Gollier and Pratt (1996) suggested that individuals facing background risk are less likely to have demand for risky assets. Their work was based on the idea that even if various forms of risks are independent on each other when deciding, we have to take into account all of them. Based on Goldman and Maestas (2013) it is not possible to be fully insured against medical expenditure risk and people are not in control over it and therefore, it can be perceived as a background risk. In further text, medical expenditure risk will be addressed as background risk.
7
retirement. In connection with health insurance Edwards (2008) mentions that association between insurance coverage and portfolio choice depends on the utility brought by coverage and states that additional health insurance does not decrease significantly the health risk and additional insurance has no effect. However, according to Goldman and Maestas (2013), if health risk leads to higher medical costs, not covered by insurance, portfolio allocation changes with the aim to decrease the whole financial exposure. To put it simply, if one financially related risk goes up, agents are trying to offset this new risk by decreasing other types of financial risks. This concept was analysed for example in the study of Gollier and Pratt (1996) focused on risk vulnerability and background risk. Evidence from the United States showed that higher health care costs as a contributor to the financial risk can lead even to the bankruptcy (Cunningham, 2008). Atella, Brunetti, and Maestas (2012) concluded, that insurance coverage is linked to the background risk arising from the health risk. The health risk is imposing direct and indirect costs, which can be, to some extent, decreased by insurance. In their study some other key facts are stated:
1. Risky assets holdings depend more on perceived, not objective health.
2. Households are forward-looking – opinions about future health state and risks are important.
3. Countries with weaker insurance coverage have higher background risk and lower investments to the risky assets.
Atella, Brunetti, and Maestas. (2012) used an index composed of behavioural, asymptomatic and strength variables as a measure of health risk. That was criticised by Courbage, Montoliu-Montes, and Rey (2017) in the paper about decreasing absolute risk aversion (firstly introduced by Arrow, 1970), in which authors tried to connect financial and health risk as well. The main concern is focused on biased index and endogeneity (unobserved determinants of risky health behaviour, might be correlated with financial risk aversion) and therefore they use a subjective answer to the questionnaire as a measure of perceived health.
8
suggested that less risk-averse people are affected by optimism when they make stock investment decisions, but for risk averters optimism or pessimism does not matter.
Goldman and Maestas (2013) investigated the relationship between supplemental insurance policies and stock holding in the United States and concluded that lower medical expenditure risk leads to higher odds of stock holding. They have included two different levels of supplementary health insurance (Medicare HMO and Medigap/employer) and showed that the probability to hold risky assets was increasing with the level of insurance coverage. The study resulted in findings supporting the relation between supplementary health insurance, offsetting the background risk, and riskiness of the portfolio. That is in line with Pang and Warshawsky (2010) who found that the higher uninsured health expenditure risk, the safer the portfolio composition.
Ayyagari and He (2017) investigated the effect of the reduction in prescription drug spending risk in the United States after Medicare Part D was introduced. Their results are consistent with Goldman and Maestas (2013). The decrease in health expenditure risk represented by drug prescription spending led to the increase of investments into risky assets. In the same year, Lee (2018) examined the effect of health insurance on portfolio choice in the United States after the implementation of the Affordable Care Act (ACA). ACA was aimed at private insurers and forced them to accept older dependants to stay on their parent’s insurance plans till 26 years. Results are the same as in the previous studies. The lower the risk of out-of-pocket expenditures, the higher is the likelihood of investing in risky assets.
Another study from the United States, published by Angrisani, Atella, and Brunetti (2018), suggests not only that the reduction in background risk through insurance increases the investment activity, but also that the effect of poor health on investment decision is eliminated by implementing Medicare. Before the implementation of Medicare, households with poor health were less likely to hold stocks. After implementation, this difference between individuals with poor and good health mostly disappeared.
9 3. Methodology
3.1. Database
Our research will be based on the data obtained from the Survey of Health, Aging and Retirement in Europe (SHARE)1. The target of the survey are the Europeans older than 50 years. The number of countries included in the SHARE survey is increasing with each wave. The most current wave at the time of writing this thesis is wave number 6 conducted in 2015. Eighteen countries are included – 17 European (Austria, Belgium, Croatia, Czech Republic, Denmark, Estonia, France, Germany, Greece, Italy, Luxembourg, Poland, Portugal, Spain, Sweden, Switzerland, and Slovenia) and Israel. For our study, we are interested in population within European countries only. However, the results for all countries participating in wave 6, including Israel, are reported in Appendix N. All the data are obtained with the use of a questionnaire and the result is an extensive multidisciplinary database spreading through many different modules. That makes it possible to test varying relations between health care and finance.
Data are obtained from 6 modules of SHARE – Assets, Cognitive Function, Demographics, Health Care, IT module and Activities. All the data are used on the individual level.
SHARE waves 5 and 4 served to fill in the data missing in wave 6. The reason is that some modules, such as Cognitive Functions or part of the Demographics, are not reported for each wave. For example, the respondent´s answer is recorded only in the time of his first contact with the SHARE survey or when an answer is a subject of change.
3. 2. Variables
3. 2. 1. Stock market participation
To test for a relationship between stock market participation (SMP) and supplementary health insurance, the dichotomous variable stock market participation will be used as a dependent variable. This variable will include direct and indirect stock market participation together.
We will use data from the Assets module of SHARE wave 6. Direct participation is reflected through the question:
“Do you currently have any money in stocks or shares that are listed or unlisted on the stock market?”
Possible answers are “yes” or “no”, what makes it suitable for creating a binary variable. Indirect market participation is determined through the set of two questions: a) “Do you currently have any money in mutual funds or managed investment accounts?”
10
b) “Are these mutual funds and managed investment accounts mostly stocks or mostly
bonds?”
Possible answers for question a) are “yes” or “no”, and for b) outcomes are “mostly stocks”, “half stocks and half bonds” or “mostly bonds”. To follow our research question, the knowledge about exact amount of money invested in stocks is not necessary and therefore both combinations – “yes”, “mostly stocks” and “yes”, “half stocks and half bonds” – can be taken as equal.
Final binary variable stock market participation is constructed as follows: direct, indirect or both takes values of 1, no participation is represented by 0. A similar approach was used for example by Angelini and Cavapozzi (2017) or Arts (2018). Former authors included another form of indirect market participation through retirement funds. However, as Arts (2018) points out, the decision about retirement fund allocation is not made by individual investors and therefore is also not included in our stock market participation variable. We do not include intensive identification of market participation through the actual amount invested. Including intensive participation would lead to smaller sample size due to unreported values. For testing of the relationship between insurance and investments, intensive participation is not necessary.
3. 2. 2. Bond market participation
For estimating the relation between supplementary health insurance and bond market participation (BMP), we will use investment into the bonds as a dependent variable. Construction of the binary variable which takes value 1 if the respondent owns bonds and 0 if he does not, is similar to the construction of stock market participation variable. Response to two questions from the SHARE questionnaire is used – if respondent owns the bonds directly or if he invests in bonds indirectly through mutual fund portfolio. We do not make a distinction between direct and indirect bond investment. The binary variable takes value 1 for both cases. For the same reason as for stocks, we do not include intensive identification of the bond investments. We would like to point out, that those invested in mutual funds with an equal proportion of stocks and bonds are included in both variables. The reason is, that the mean for bond market participation would fall below 0.1 if equally allocated mutual funds are excluded, making the marginal effects from the Two Stage Least Square (TSLS) questionable.
3. 2. 3. Supplementary health insurance
11
With the use of SHARE wave 6 Health care module, which contains a “yes” or “no” question about supplementary health ownership we are able to obtain dichotomous variable with 1 signaling that a respondent has supplementary insurance and 0 referring to those without it. In the SHARE questionnaire, respondents are asked:
“Do you have any supplementary health insurance that pays for services not covered by your basic health insurance/national health system/ third-party payer?”
This allows us to include extensive supplementary health insurance ownership. For intensive identification, we would need to have data about the actual amount paid as a premium and the amount saved due to reimbursements, currently not available in the SHARE database.
Although there are differences in health care systems across European countries, in general, those with supplementary health insurance should be always relatively better secured for the medical expenditure risk than the ones who do not have it. As depicted in Fig. 1, the share of the population with supplementary health insurance varies a lot amongst the countries. Inhabitants of all included countries have an opportunity to take supplementary health insurance, which improves the coverage of medical expenditures. Therefore, the main objective of the thesis can be pursued even with respect to the differences in the healthcare systems.
Fig. 1: Supplementary health insurance in European countries
12
We need to be aware of possible endogeneity between supplementary health insurance and stock market participation. Supplementary health insurance as an endogenous variable contains the unobservable error term attached to it. This error term will be correlated with the error term of the regression in which supplementary insurance acts as an explanatory variable. As a result, the estimated coefficient would be inconsistent. Endogeneity in the model can possibly arise in our model from an omitted variable, data measurement error or self-selection. To overcome possible omitted variables, we have included many relevant determinants which have been found to affect investment decisions in the previous research papers.
Beside of the omitted variables, endogeneity in our model will likely arise from the simultaneous decision making about stock market participation and supplementary health insurance coverage. As Goldman and Maestas (2013) pointed out, decisions about stock market participation and supplementary health insurance are likely to be made simultaneously. To overcome this problem, the usage of instrumental variable approach needs to be applied. Identifying a suitable instrument is not easy. Surprisingly, satisfaction with the national health care coverage, as the instrument in which we put the most hope, was found out to be weak (Weak instrument test = 0.199) (Appendix M). After further testing we have managed to find an instrument which is strong enough – dentist (Weak instrument test = 39.89) (Appendix G). This variable is binary, taking values 1 if individual visited a dentist in the last year and 0 if he did not. In many European countries, such as Denmark, Belgium or Germany, dental care is not fully included in national health care systems in such an extent as other doctor visits. Instead, they are part of a supplementary health insurance packages. Therefore, we assume that there will be a correlation between the supplementary health insurance dummy and dentist visit. This assumption will be later tested by Weak instrument test. The second assumption about no correlation of the instrument to the outcome (SMP or BMP) cannot be tested when only one instrument is used, however, based on the theory we do not see any reason why a visit to the dentist should be influencing investment decisions beside of the indirect link through supplementary health insurance.
Even after the use of an instrumental variable approach, the results should be interpreted with caution. The data from the SHARE database are likely to contain the measurement errors, causing the additional source of endogeneity.
3. 2. 4. Other explanatory variables Gender
13
included as a control variable also in many other studies interested in stock market participation, such as Goldman and Maestas (2013), Angelini and Cavapozzi (2017), Atella, Brunetti, and Maestas (2012). We will take into account gender differences through the explanatory variable gender which takes values 1 for male and 0 for female respondents.
Age
Another explanatory variable, which is commonly used in literature, is age. The relation between stock market participation and age was stressed for example by Binswanger (2011). He suggests that optimal investor should invest more in stocks with growing age till he/she reaches middle age and then continuously decreases the exposure to the stocks. If this is a case, we would expect a negative coefficient for the age variable, because our sample is focused on the population above 50. Age will be incorporated in as a logarithm of age in years and, the same as gender will be taken from the module Demographics. Wave 6 data were collected in 2015 and therefore we will count age as the difference between 2015 and the year of the respondent’s birth. Transforming age into logarithm makes this variable continuous and compresses the span of the values.
Although the SHARE database is focused on respondents above 50, in our final sample we have some observations with a reported age below 50. Not to lose observations, we enclose these respondents to the group of 50 years old as the youngest possible.
To take into account the possibility that the relationship between stock market participation and age is not linear, we will include also the square of age. By this approach, we will be able to capture the chance that the relation is turning or slowing at some age and therefore, is not linear.
Health
14 Internet usage
The internet is decreasing transaction costs and costs of information and therefore can play a substantial role in investment decisions. Bogan (2008) concluded that households with internet are more likely to participate in the stock market. IT module of the SHARE database includes the question of whether the person using the internet in last 7 days:
“During the past 7 days, have you used the Internet, for e-mailing, searching for information, making purchases, or for any other purpose at least once?”
That allows us to incorporate binary variable for internet usage, taking values 1 for internet users and 0 for non-users. We expect a positive coefficient for this variable.
Financial literacy
Financial literacy is considered another important determinant of stock market participation. The relationship was shown for example in the papers of Gaudekker (2015), Balloch, Nicolae, and Philip (2014) or recently by Arts (2018). Arts in his paper used the same database as is used for this thesis and therefore we will apply the same method to construct an ordinal variable measuring financial literacy level of respondents.
Cognitive skills module of the SHARE database includes 4 numerical questions with a different level of difficulty:
a) “If the chance of getting a disease is 10 percent, how many people out of 1,000 would
be expected to get the disease?”
b) “In a sale, a shop is selling all items at half price. Before the sale, a sofa costs 300.
How much will it cost in the sale?”
c) “A second-hand car dealer is selling a car for 6,000. This is two-thirds of what it costs
new. How much did the car cost new?”
d) “Let's say you have 2,000 in a savings account. The account earns ten percent interest each year. How much would you have in the account at the end of two years?”
At first, the respondent is asked to answer the question a) which is considered to be intermediate. If the answer is incorrect he will proceed to the question b), considered to be easy, and if he replies correctly his financial literacy is 2. If the answer is incorrect for question b) he scores financial literacy of 1. If the respondent gave the correct answer to a) he will be questioned by c) and if he replies incorrectly he has financial literacy of 3. If both a) and c) are answered correctly, final question d) is asked. Fail to solve d) leads to the literacy of 4. Maximum possible score is 5, which is received for answering correctly a), c) and d).
15
and 4 (Börsch-Supan 2018a, 2018b, 2018c). For a small number of respondents, answers are recorded more than once. Where this is a case, the most recent answers are used.
Risk aversion
Effect of risk aversion on the financial decision making is well documented in the literature. The direct effect can be drawn based on the standard portfolio theory – more risk averse the individual is, the lower portion of his wealth is willing to invest into risky assets (Laakso, 2010). Besides affecting the share of wealth invested in risky assets, some individuals are deciding not to participate in the stock market due to the risk aversion (Dimmock and Kouwenberg, 2010).
Risk aversion is already taken into account in the SHARE survey. We will use the value reported by respondents. There are four possible answers coded 1 to 4, with 1 indicating the lowest level of risk aversion and 4 indicating the most risk-averse people. Almost three-quarters of the respondents indicated the highest level of risk aversion. We will transform this categorical variable to the binary. Constructed binary variable will take values of 1 for those who have high risk-aversion and 0 for those willing to take the average or higher financial risk if rewarded by average or higher compensation. Risk aversion was used in the same way also by Angelini and Cavapozzi (2017).
Marital status
In the study of Goldman and Maestas (2013) marital status is used to explain the difference in stock participation between married and unmarried individuals. Their results suggest that married people tend to own more assets than divorced or widowed respondents. As the study of Goldman and Maestas is close to our one in subject, we will include an explanatory variable for marital status as well.
A binary variable is constructed from an originally categorical variable as follows: value 1 is assigned to married people living together or separately and to those in a registered partnership. 0 stands for never married, divorced or widowed. Similar to financial literacy, values are reported once and only further change is reported in more recent waves of the SHARE database. Waves 5 and 4 were used to fill in the data missing in wave 6. In cases where respondents reported the change in their marital status, we used the most actual information.
Social connectedness
16
between strong and weak social ties and their influence on investment decisions. Strong ties are represented by interactions with family and friends, while weak by involvement in various social activities or clubs. Their results suggest, that strong ties are not significantly increasing the propensity to hold stocks. On the other hand, weak ties are providing new information and therefore are an important stimulus for investment decision making.
To include weak social ties into our model, a new variable is constructed. This measure is based on the number of activities done in the last year. For each out of five activities answer is either 1, signaling involvement, or 0 if the activity was not done in the last year. Included activities are 1) voluntary or charity work; 2) educational or training course; 3) sport, social or another kind of club; 4) taking part in the political or community-related organization and 5) playing cards or games such as chess. A binary variable is created to distinguish individuals participating at least in one out of the five possible activities, from those not participating in any of them. Atella, Brunetti, and Maestas (2012) used, besides the control for the participation on the social activities, also a binary variable indicating participation on the religious events. However, they did not find it significant in any of the models so we have decided not to use this control.
Wealth
Many studies provide evidence about the effect of wealth on investment activity. To incorporate this effect, we will include categorical variable based on the value of gross financial wealth.
Gross financial wealth = bank accounts + stocks + bonds + mutual funds + individual retirement accounts + contractual savings for housing + life insurance policies (1)
The abovementioned measure is based on the paper of Angelini and Cavapozzi (2017). To get rid of outliers and to take into account differences in salaries and living costs amongst European countries, gross financial wealth is transport into a categorical variable. For each country, respondents are distributed into 5 bands. As borders, 20th, 40th,
60th and 80th percentiles are used. Dummy variables for quintile 2, 3, 4 and 5 are used. The
first quintile goes to the constant. This procedure is adapted from Goldman and Maestas (2013).
3. 2. 4. Country dummies
17 Fig. 2: Rates of stock market participation
The rates of stock market participating individuals in 17 European countries. Column “Europe” represents the weighted average of these countries. Numbers are based on data from SHARE wave 6. Stock market participation includes both – direct and indirect stock investing.
Fig. 3: Rates of bond market participation
The rates of individuals investing in bonds in European countries. Column “Europe” represents the weighted average of these countries. Numbers are based on data from SHARE wave 6. Bond investments include both – direct and indirect bond holdings.
3.3 Regression model
18
levels. To comply with existing literature, we expect a positive correlation between stock market participation and supplementary health insurance.
Specification of the benchmark OLS model:
SMPi = α1 + β1INSi + β2X´i + 𝜔COUNTRY´i + 𝜀𝑖 (2)
SMP stands for stock market participation, INS for dummy signaling supplementary health insurance and α1 is constant. X represents the vector of exogenous
explanatory variables used in the model – logarithm of age, age squared, gender, internet usage in last week, wealth represented by dummies representing wealth quintiles, participation in social activities, financial literacy, financial risk aversion, self-reported health and marital status signaling if person is currently married (registered partnership is taken as equivalent) or not. COUNTRY stands for the vector of country dummies incorporated to control for country fixed effects. The error term is represented by 𝜀
.
Subscript (i) marks individual respondents. OLS will be run twice and for the second specification, SMP will be replaced by BMP representing dummy for bond market participation.
As mentioned above, SMP (as well as BMP) and supplementary health insurance are likely to be endogenous due to the measurement error and self-selection. To test if our expectations about self-selection endogeneity are correct, Seemingly Unrelated Regressions (SUR) were regressed. SUR estimates two probits simultaneously, the first with stock market participation and the second with supplementary health insurance as a dependent variable. As a result, it provides us with the coefficient Rho which measures the correlation between residuals of the two probits. If Rho is significant, self-selection endogeneity is present and instrumental variable (IV) approach is needed. This is caused by common omitted variables, influencing both decisions – about the stock market participation and the supplementary health insurance take out. Results for SUR with the use of stock market participation and supplementary health insurance dummies as dependent variables are reported in Appendix D. Rho is reported to be statistically significantly different from 0 with the value of 0.077, showing that residuals of these two probits are correlated. This provides evidence of the endogeneity arising from self-selection and justifies the use of the IV procedure. In Appendix E results of SUR with the bond market participation and supplementary health insurance dummies are reported. Rho takes a value of 0.097 and it is significant at the 1% level.
19
Hausman test will show if the use of IV instead of OLS is justified and Weak instrument test will provide evidence if chosen instrument for IV regression is strong enough.
The IV regression to run will be Two-Stage Least Squares (TSLS). The choice of this method is based on the nature of our variables and endogeneity issues. We are aware that binary logistic model is more reliable than Ordinary Least Squares, for regressing binary dependent variables, due to the normality of distribution and forecasting power. OLS does not put any restrictions on the fitted values of instrumented supplementary health insurance, which can be out of the 0 to 1 interval. However, based on the econometric literature, it is not allowed to mimic TSLS and use probit in the first stage. Mimicking would lead to the so-called forbidden regression. Forbidden regression arises because only OLS regression in the first stage is producing fitted values which are not correlated with the residuals (Greene, 2008).
Econometric literature argues, that the use of TSLS is appropriate and still producing robust consistent results even when used with a binary dependent variable and binary endogenous explanatory variable (Angrist and Pischke, 2008). Also, Grondijs (2015) in his work compared different regression methods for estimating the model with endogenous variable and in discussion recommends using TSLS even when outcome and treatment are both binary variables. He argues that even though the model might be misspecified, it still allows to adequately estimate the difference between the two groups if the instrument is valid. When interpreting the results, caution is needed. TSLS estimates are showing a local average treatment effect. This means the effect for population induced by instrument to take up the treatment. Therefore, the results from instrumental variable regressions are instrument sensitive and might produce different results under different instruments used.
Regression for TSLS is specified as follows:
INSi = α2+ β1Di + θX´i + μCOUNTRY´i+ 𝜕i
SMPi = α3 + β2INShati + φX´i + υCOUNTRY´i +κi (3) Binary variable SMP stands for stock market participation, INS is a dichotomous
variable for supplementary health insurance and variable INShat represents fitted values of supplementary health insurance obtained from the first stage of TSLS. D represents the instrument dentist. Constants are included in our model and denoted as α1 and α2. X
represents the vector of exogenous explanatory variables used in the model, which are the same as for OLS estimation. COUNTRY stands for the vector of country dummies. Error terms are represented by 𝜕 and κ
.
Subscript (i) marks individual respondents.20
bonds) will be used instead of SMP. This allows us to compare the effect on different financial assets.
21 Table 1: Correlation matrix for the stock market participation sample
The table provides correlation coefficients between variables used in the model for stock market participation. Numbers are based on waves 6, 5 and 4 of the SHARE database. 23,627 observations are used. Correlation coefficient lower than 0.3 represents weak; values from 0.3 to 0.49 moderate and from 0.5 to 0.69 high correlation. Values ≥ 0.7 are considered to be critical.
Supple-mentary health insurance Log of age Age squared Male or female Internet usage in the past 7 days Financial risk aversion Marital status Social activity Wealth quintile Financial literacy Health Health Dentist Dentist Supplementary health insurance 1 Log of age -0.085 1 Age squared -0.084 0.986 1 Male 0.046 0.003 -0.006 1 Use of internet in past 7 days 0.132 -0.424 -0.430 0.103 1 Financial risk aversion -0.093 0.158 0.159 -0.167 -0.273 1 Marital status 0.065 -0.173 -0.184 0.189 0.139 -0.083 1 Social activity 0.123 -0.160 -0.170 0.068 0.335 -0.189 0.070 1 Wealth quintile 0.091 -0.019 -0.025 0.116 0.205 -0.219 0.168 0.159 1 Financial literacy 0.079 -0.165 -0.178 0.172 0.329 -0.219 0.114 0.235 0.207 1 Health 0.149 -0.232 -0.236 0.084 0.292 -0.189 0.115 0.214 0.165 0.203 1 Dentist 0.093 -0.148 -0.159 0.003 0.289 -0.166 0.096 0.246 0.161 0.204 0.163 1
22 4. Data
4.1. Sample construction and summary statistics
For this research, only respondents from European countries involved in SHARE wave 6 will be included.2 Furthermore, we will filter out those with missing data for our dependent variable, explanatory variables, and instrument. Construction of the sample for testing the relationship between stock market participation and the supplementary health insurance is shown in Table 2.
Table 2: Construction of the sample for stock market participation
Construction of the sample for testing the relation between stock market participation and supplementary health insurance. Stock market participation and supplementary health insurance are based on wave 6 of the SHARE database. To complement the missing data for control variables waves 4 and 5 have been used.
Phase Number of respondents
SHARE database wave 6 - 68,231
Excluded Remaining
Excluding Israel 2,035 66,196
Inconclusive observation* 1 66,195
Missing Remaining
Stock market participation** 21,830 44,365
Supplementary health insurance 67 44,298
Age 0 44,298
Gender 0 44,298
Use of internet in last 7 days 3 44,295
Marital status 10,000 34,295
Activities (Social connectedness) 1,006 33,289
Wealth 7,954 25,335 Financial literacy 1,540 23,795 Risk aversion 166 23,629 Health 1 23,628 Instrument: Dentist 1 23,627 Final Sample - 23,627
* Observation Bn-572539-01 - no mutual fund reported, however signaling equal share of stocks and bonds in mutual fund
** Both, indirect and direct participation included
After removing all the observations with missing values, we end up with a sample of 23, 627 individuals from Europe. Descriptive statistics of our sample are shown in Table 3.
23
Table 3: Summary statistics for the stock market participation sample
Summary statistics of variables included. Size of the sample is 23,627. Numbers are based on wave 6, 5 and 4 of the SHARE database. 95% confidence interval for mean was used.
Variable Mean St. dev Min Max
Stock market participation 0.170 0.376 0 1
Supplementary health insurance 0.408 0.491 0 1
Logarithm of age 1.815 0.065 1.38 2
Age squared 4,469.19 1,369.52 576 10,000
Male 0.423 0.494 0 1
Use of internet 0.566 0.496 0 1
Financial risk aversion 0.731 0.443 0 1
Marital status 0.616 0.486 0 1 Social connectedness 0.576 0.494 0 1 Wealth quintile 3.039 1.438 1 5 Financial literacy 3.478 1.043 1 5 Health 0.618 0.486 0 1 Austria 0.074 0.261 0 1 Belgium 0.091 0.287 0 1 Croatia 0.038 0.190 0 1 Czech Republic 0.076 0.264 0 1 Denmark 0.055 0.227 0 1 Estonia 0.123 0.328 0 1 France 0.061 0.240 0 1 Germany 0.083 0.275 0 1 Greece 0.033 0.178 0 1 Italy 0.055 0.228 0 1 Luxembourg 0.033 0.179 0 1 Poland 0.007 0.082 0 1 Portugal 0.029 0.169 0 1 Slovenia 0.088 0.283 0 1 Spain 0.055 0.228 0 1 Sweden 0.054 0.227 0 1 Switzerland 0.048 0.213 0 1 Instrument: Dentist 0.567 0.496 0 1
24
risk-averse. 6 out of 10 respondents considered their health to be at least good. The most frequent nationality in our sample is Estonian and the least frequent is Polish. Summary statistics for all included variables are reported in Table 3.
Table 4: Summary statistics for the bond market participation
Number of observations 23,640. They are based on wave 6 of the SHARE database. Direct and indirect bond investments are included. 95% confidence interval for mean was used. We do not report the summary statistics for independent variables, because the validity test (Appendix B) showed that there are no significant differences between stock and bond participation samples.
Variable Mean St. dev Min Max
Bond investment 0.110 0.313 0 1
To test the relation between supplementary health insurance ownership and the bond market participation, similar sample as for stocks will be used. The only adjustment made is the exclusion of 21 observations with missing values for bond investment activity and the inclusion of 34 observations which had missing values for stock market participation but contained information about investments in bonds. The final sample for the bond market participation contains 23,640 observations. In Appendix B, the validity test can be found, showing that there are no significant differences in independent variables between the sample for stock market participation and bond market participation. Summary statistics of the variable for bond investment are shown in Table 4.
The balance test was conducted on the SMP sample and BMP variable to find out if those, having supplementary health insurance, differ from respondents without it. Results are presented in Table 5. From the balance test, it can be concluded that those having insurance are, on average, a bit younger, more socially active, healthier, richer and they are using the internet more often. Also, gender representation is more equal amongst insured. Insured are less financially risk-averse on average and also participating more in the stock market. This is in line with what we expected. All characteristics are statistically significantly different at the 1% level. For the balance test, we have also included an instrumental variable dentist, which shows us that insured people were on average more prone to visit the dentist. In Table 6, the balance test of the dummy variable for bond investment is reported, to show the difference between insured and uninsured individuals. It can be seen that insured people are more likely to invest in the bonds.
25
variable averages seem to be rather different for total database and sample. T-test showed that our sample has means for all variables, besides of wealth, significantly different from the averages of the total database. Therefore, generalizing the results of this research for the total European sample reported in SHARE has to be done with caution.
Table 5: Balance test for insured and uninsured for stock market participation sample
The table shows the differences in characteristics between insured and uninsured individuals in stock market participation sample. ***, **, * represent significance at the 1%, 5% and 10% level, respectively. Numbers are based on the SHARE database, waves 6, 5 and 4. Number of the observations is 23,627.
Variable
Insured Uninsured t-test
(p-value)
Average St. dev Average St. dev
Stock market participation 0.233 0.423 0.127 0.333 20.739*** (0.000) Logarithm of age 1.809 0.064 1.820 0.065 -13.150*** (0.000) Age squared 4,330.96 1,335.53 4,564.31 1,384.48 -13.00*** (0.000) Male 0.451 0.498 0.405 0.491 7.039*** (0.000) Use of internet 0.645 0.479 0.512 0.500 20.574*** (0.000) Financial risk aversion 0.681 0.466 0.765 0.424 -14.122***
26
Table 6: Balance test for insured and uninsured for bond market participation variable
The table shows differences in bond market participation between insured and uninsured individuals. ***, **, * represent significance at the 1%, 5% and 10% level, respectively. Numbers are based on the SHARE database, waves 6, 5 and 4. Number of the observations is 23,640. We do not report the balance test of independent variables, because the validity test (Appendix B) showed that there are no significant differences between stock and bond participation samples.
Variable
Insured Uninsured t-test
(p-value)
Average St. dev Average St. dev
Bond market participation 0.146 0.353 0.085 0.278 14.366*** (0.000)
Number of observations 9,641 13,999
4.2 Univariate analysis for stock market participation
Shares of the respondents participating in the stock market with respect to our set of independent variables are reported in Table 7. Age is excluded due to the large space requirements, however, the graph showing the stock market participation rate with respect to age in years can be found in Appendix F.
Results (Table 7) are showing that insured people are almost one time more likely to hold stocks than uninsured. Other explanatory variables are in line with what the theory suggests – males are holding more stocks than females, internet users more than non-users, richer more than poorer, more financially literate more than less literate, healthier more than unhealthy. Amongst the socially active respondents, 23 out of 100 are participating in the stock market, while amongst those not participating in reported activities only 8.6% are investing in stocks. Univariate statistics show that married people or those in a registered partnership are approximately 50% more likely to invest in stocks than unmarried. Risk-averse individuals are less likely to hold stocks than risk-tolerant.
For age (Appendix F) the pattern is not easily visible. The share of the population investing in stocks is fluctuating between 15% and 20% till the age of 75. Then there is a decline in investment activity.
27
Table 7: Univariate analysis for stock market participation
This table contains univariate analysis for stock market participation (SMP) for all included variables with exception of the logarithm of age and age2. Sample consists of 23,627 observations. Numbers are based on wave 6, 5 and 4 of the SHARE database.
Variable SMP [%] Variable SMP [%]
Supplementary health insurance
No Yes Gender Female Male Usage of internet No Yes Marital status Married Unmarried Risk aversion 0 (risk-tolerant) 1 (risk-averse) Financial literacy 1 (lowest) 2 3 4 5 (highest) 12.7 23.3 12.5 23.1 6.8 24.8 19.6 12.9 36.8 9.7 3.6 8.0 12.6 19.2 30.8 Social activities 0 (not participating) 1 (participating) Self-reported health 0 (bad or fair) 1 (at least good)
Wealth 1 (poorest) 2 3 4 5 (richest) 8.6 23.2 8.7 22.2 2.5 7.4 12.0 23.6 37.4
4.3 Univariate analysis for bond market participation
Univariate analysis for the bond market participation is reported in Table 8. All independent variables used in our model are included in univariate analysis. Age is not included in the table because of the space requirements. Instead, bond market participation rate with respect to the age in years is graphically presented in Appendix F.
Univariate analysis for bond market participation shows the same patterns like that for stock market participation. Supplementary health insured are more often bond owners than uninsured, males more often than females and married more than unmarried. The probability that respondent holds stocks is increasing with wealth and financial literacy and is higher if he/she is an internet user. Not socially participating individuals are less likely to hold bonds. Perceived health and risk aversion are negatively related to the probability that the respondent is a bondholder.
28
hold bonds than risk-tolerant (7.1% vs. 21.4%). For stocks univariate analysis the difference was almost 4 times higher (9.7% vs. 36.8%). Males, in comparison to females, are 1.85 times more likely to hold stocks, however only 1.65 times more likely to hold bonds. The difference in bond market participation between internet users and not users is also lower in comparison to stocks. Probability to hold stocks and bonds, with respect to the supplementary health insurance, shows similar magnitude for both risky assets. Insured are 1.87 times more likely to hold stocks and 1.74 times bonds than those uninsured.
Table 8: Univariate analysis for bond market participation
This table contains univariate analysis for bond market participation (BMP) for all included variables with exception of the logarithm of age and age2. Sample consists of 23,640 observations. Numbers are based on wave 6, 5 and 4 of the SHARE database.
Variable BMP (%) Variable BMP (%)
Supplementary health insurance
No Yes Gender Female Male Usage of internet No Yes Marital status Married Unmarried Risk aversion 0 (risk-tolerant) 1 (risk-averse) Financial literacy 1 (lowest) 2 3 4 5 (highest) 8.5 14.6 8.6 14.2 5.2 15.4 12.7 8.3 21.4 7.1 2.7 5.2 8.3 13.0 18.2 Social activities 0 (not participating) 1 (participating) Self-reported health 0 (bad or fair) 1 (at least good)
29 5. Results
5.1 Multivariate regression for stock market participation
To estimate the relationship between supplementary health insurance and stock market participation, we perform multivariate regression. As a benchmark model, we run the OLS regression with all the explanatory variables and with the fixed country effects. Also, heteroscedasticity adjusted errors are used. This specification is not taking into account the endogeneity of supplementary health insurance. However, the sign of the coefficients provides us with the first indications of the directions in which variables are related to the stock market participation decision. Results for supplementary health insurance are shown in Table 9. Full results for all independent variables can be found in Appendix G. The most of the coefficients are statistically significant at the 1% level. The exception is a marital status which is insignificant on any level of significance commonly used. Signs of the coefficients are in line with what was expected based on the theory (Section 2 and 3). Males, internet users, wealthier, financially literate and socially more active population participate in the stock market more. Risk-averse people are significantly less likely to own stocks. The coefficient for age shows that older people are more likely to own stocks, however negative sign for age squared signals that relationship between age and participation in the stock market is hump-shaped. People with good self-reported health tend to participate more than those with worse perceived health. That is also aligned with the results of other authors (Courbage, Montoliu-Montes, and Rey, 2017). The coefficient for supplementary health insurance is positive and statistically significant, so far supporting our hypothesis 1.1.
In the next step, we make a switch in the model used – from single OLS model, we will move to the TSLS model. Identification of the proper instrument was not easy, several possible variables were tested. We put the most hope into the variable indicating the satisfaction with the National Health Care coverage, however, that was reported to be a weak instrument and Hausman test suggests that the OLS model is consistent. Finally, we identified a dentist as a suitable instrument.
30
considered to indicate an appropriate instrument. We obtained the value close to 40 and therefore we conclude that a dentist as an instrument is not weak.
Results of TSLS regression suggest that respondents with supplementary health insurance are much more likely to invest in stocks. The coefficient of 0.59 shows that, after using a dentist as an instrument, the probability to hold stocks is 59 percentage points higher for supplementary health insured than for uninsured. The effect is significant even at the level of 1%.
As can be seen in Appendix G, after including supplementary health insurance and controlling for endogeneity, some of the explanatory variables have lost their predictive power. Usage of the internet, financial literacy, marital status, social activity as a proxy for weak social ties and health status are not significant.
Table 9: Regressions estimated for stock market participation
The table provides results from benchmark OLS model and TSLS model. Numbers are based on the SHARE database wave 6, 5 and 4. As a dependent variable, stock market participation is used. For both regressions heteroscedasticity adjusted standard errors were used and country fixed effects included. Beside the supplementary health insurance other explanatory variables used are: logarithm of age, age squared, gender, internet usage, wealth dummies, participation in social activities, financial literacy, risk aversion, perceived health and marital status. Standard errors are in parentheses. ***, **, * represent significance at the 1%, 5% and 10% level, respectively. For TSLS the dentist was used as an instrument. Hausman test of endogeneity and Weak instrument test for TSLS model are reported in the bottom section. The table containing coefficients for all independent variables can be found in Appendix G.
Benchmark Ordinary Least Squares
Two Stage Least Squares
Variable Stock market participation Stock market participation
Constant -1.327***
(0.302)
-1.746*** (0.381) Supplementary health insurance 0.041***
(0.007)
0.590*** (0.165)
Control variables Yes Yes
Country fixed effects Yes Yes
F statistic Adjusted R2 Number of observations 257.86*** 0.286 23,627 200.80*** 0.159 23,627 Hausman test
Weak instrument test
31
Directions of statistically significant coefficients are in line with our benchmark model. The coefficient for the logarithm of age suggests that older people tend to own stocks more, however, negative age squared indicate that relation is not linear but hump-shaped. This is in line with the results of the univariate analysis, showing that after reaching the age of ca 75 years, further increase of age leads to the decreasing probability of stock holding. Men are 3.5 percentage points more likely to invest in stocks in comparison to women. Respondents in higher wealth quantiles are more likely to hold stocks, with low increments when moving between the first three quintiles but much higher for the last two. Risk-aversion, as expected, has a negative relation to the stock ownership.
We report significance levels after Šidák´s correction (1967) in the Appendix G. Šidák´s correction is taking into account the possibility, that when many t-tests are run on the same data, the chance that false discovery will be made is increasing with the number of regressors. False discovery can be established as a type I error – incorrect rejection of the null hypothesis of the t-test. Another alternative used in literature is Bonferroni correction (1936), which is a bit more conservative, making the rejection of the null hypothesis after the correction less likely. Dinno (2015) in his article compared this two corrections and stated that differences in significance levels obtained using Šidák and Bonferroni correction are minimal, therefore we have decided to use less conservative and more recent Šidák correction. The coefficient of supplementary health insurance remained significant at the 1% level even after this correction.
Summarized, both specifications, OLS and TSLS, show a positive relationship between supplementary health insurance and the stock market participation. The effect is statistically significant under both regressions used.
5.2 Multivariate regression for bond market participation
To test if supplementary health insurance is also correlated with bond investment, the same procedure as for SMP is applied. In Table 10 we present coefficient and standard error of the supplementary health insurance for benchmark OLS model. Full results for all independent variables are shown in Appendix H. Heteroscedasticity adjusted errors are used and country fixed effects are incorporated.
32
In the next step TSLS, with country dummies and heteroscedasticity adjusted errors is conducted. Short results are presented in Table 10. Full results for all independent variables are shown in Appendix H. Hausman test signals the presence of the endogeneity, rejecting the consistency of the OLS estimates and justifying the use of TSLS. Value of the Weak instrument test above 40 shows that the dentist is not considered to be a weak instrument. Signs and significance levels are the same as for stock market participation model. Marital status is an exception but it is not significant in any of the models. After using the instrument, people with supplementary health insurance are 52.5 percentage points more likely to own bonds in comparison to uninsured. The coefficient is statistically significant at the 1% level. We can conclude that the relation between supplementary health insurance and the bond investments is positive and economically and statistically significant.
Table 10: Regressions estimated for bond market participation
The table provides results from benchmark OLS model and TSLS model. Numbers are based on the SHARE database wave 6, 5 and 4. As a dependent variable, bond market participation is used. For both regressions heteroscedasticity adjusted standard errors were used and country fixed effects included. Beside the supplementary health insurance other explanatory variables used are: logarithm of age, age squared, gender, internet usage, wealth dummies, participation in social activities, financial literacy, risk aversion, perceived health and marital status. Standard errors are in parentheses. ***, **, * represent significance at the 1%, 5% and 10% level, respectively. For TSLS the dentist was used as an instrument. Hausman test of endogeneity and Weak instrument test for TSLS model are reported in the bottom section. The table containing coefficients for all independent variables can be found in Appendix H.
Benchmark Ordinary Least Squares
Two Stage Least Squares
Variable Bond market participation Bond market participation
Constant -1.552***
(0.266)
-1.927*** (0.340) Supplementary health insurance 0.034***
(0.006)
0.525*** (0.143)
Control variables Yes Yes
Country fixed effects Yes Yes
F statistic Adjusted R2 Number of observations 116.07*** 0.189 23,640 97.01*** 0.083 23,640 Hausman test
Weak instrument test
