1. Introduction Rational choice theory seeks to explain differences in financial behavior by looking at deviations between individual investors. Previous research finds that resources of individuals, such as income and wealth, or demographic variables like gender, age and race are deviations between individuals that explain differences in portfolio behavior (see e.g. Bertaut and Starr-McCluer, 2002; Carroll, 2002; Guiso et al., 2002; Guiso and Paiella, 2008; Christelis et al., 2010; Love and Smith, 2010; Dohmen et al., 2011; Christiansen et al., 2015). The explicit role of health in explaining individual portfolio choice has only recently come into consideration for the first time (Rosen and Wu, 2004). Health might have an explanatory role in portfolio choice, since individuals may adjust their exposure to financial risk in response to the exposure to health risk as a background risk (Atella et al., 2012). Health status is an interesting and relevant factor to look into. Rosen and Wu (2004) stated this nicely: “Because health tends to deteriorate with age and older people control a disproportionate amount of total wealth, it seems particularly pressing to understand how poor health affects portfolio allocation decisions”. This is especially true in the ageing societies of the west.
Rosen and Wu (2004) used data from the Health and Retirement Study (HRS) and found that households in poor health were less likely to hold risky assets, ceteris paribus. Subsequent research confirms that health status has a role in explaining stock market participation, but there is disagreement on whether this relationship is direct or indirect (see e.g. Berkowitz and Qiu, 2006; Fan and Zhao, 2009). Overall, the findings on the relationship between health and stock market participation are very heterogeneous.3
– to assess the relationship between health status and stock market participation, and to assess the role of behavioral health risk in this relationship. –
The survey of Health, Ageing and Retirement in Europe (SHARE) will be used as data source to research this relationship. SHARE provides unique data of tens of thousands of individuals above the age of 50 throughout Europe. Data includes health factors, socioeconomic status, demographic variables, and detailed information on stock market participation.
In chapter 2, the existing literature on the relationships between health variables and stock market participation is reviewed. Chapter 3 discusses the data source, data transformation and construction of variables in this research. Chapter 4 explains the empirical framework. The results of the analyses are presented in chapter 5, and chapter 6 provides the discussion and limitations. The paper is concluded in section 7.
2 The Literature
The first section of the literature review discusses how the health status of an individual affects his or her wealth and the basic financial choices that an individual makes. The second part reflects on how health relates to portfolio choice. Section 3 discusses different measures of health that could be adopted, and how these different measures relate to portfolio choice. Section 4 concludes on the literature review. 2.1 Health, Wealth, and Basic Financial Choices Stock market participation is a choice that becomes relevant when you actually own wealth. If you earn just enough to come about every month, all of your income will be used for the consumption of life’s basic necessities. In such a situation, stock market participation would not be a choice, it would be nearly impossible. Therefore, to understand the relationship between health and portfolio choice, the relationship between health, wealth and basic financial choices will first be discussed.
earn more, compared to less healthy people. The second factor that leads to wealth decline are medical expenses that arise as a result of poor health.
concept in the context of health and stock market participation, and no results of the effect of behavioral health risk on portfolio choice are available. Late-in-life risky health behavior such as smoking or excessive drinking is obviously health deteriorating (see e.g. McGinnis and Foege, 1993). Hence, behavioral health risk could have a negative indirect effect on stock market participation through health status. A second way in which risky health behavior could affect portfolio choice is through risk aversion or as proxy of risk aversion. Behavioral health risk could be seen as an individual taking risk. Modern portfolio theory states that the more risk averse an investor is, the safer a portfolio he or she will choose (Markowitz, 1952). This approach would lead to assume that individuals that engage in risky health behavior would choose more risky portfolios. Hence, there could be a composed effect of risky health behavior on stock market participation through multiple factors.
risky health behavior in explaining stock market participation. Risky health behavior might have a direct, indirect or composed effect on portfolio choice. 3. Data 3.1 Data Source This paper uses the Survey of Health, Ageing and Retirement in Europe (SHARE4) as data source. SHARE consists of multiple data-waves over time. For this research, wave 1 (2004) and wave 3 (2009) are used. SHARE offers data on individuals and households with at least one member of the household being 50 or older. Participants in SHARE wave 1 and wave 3 come from 14 different countries across Europe; Austria, Germany, Sweden, Netherlands, Spain, Italy, France, Denmark, Greece, Israel, Switzerland, Czech Republic, Poland and Belgium. Partners of the head of the household also participated in the survey, providing me with the opportunity to use an extensive dataset at individual respondent level. The SHARE questionnaires cover huge amounts of information. In the context of this research, information about household resources, asset allocation, health and socio-economic characteristics are specifically interesting. Some papers that also used SHARE and are linked to my research topic include; Atella et al. (2012), Bressan et al. (2014) and Angelini et al. (2016). 3.2 Construction of Sample The final sample contains data from two different waves. A wave consists of different data subsets which are separately available at the SHARE platform. First, these subsets have to merged based on unique ID codes for individual respondents. Secondly, the two waves have to merged based on the unique identification numbers to arrive at the total sample. Wave 1 provides information on health characteristics, financial asset allocation, and socio-demographic characteristics. This wave has 22,157 unique respondents for the variables used in this research. Wave 3 (SHARELIFE) contains information about early life health characteristics and parental influences during childhood. This wave consists of 27,614 respondents. After merging the waves, a total sample remains of 16,657 unique observations.
The SHARE surveys are filled in by individual respondents, but the individuals are also part of a household. Some of the variables in my sample are answered individually, and some are answered on household level. This paper focusses on the relationship between health characteristics and stock market participation on an individual level, and the variables are amended in such a way. Detailed information is found in subsequent chapters. In line with Atella et al. (2012), respondents below the age of 50 (558 observations) and above the age of 90 (50 observations) are excluded from the sample for homogeneity purpose. The lower bound is set because the SHARE survey is designed for a population study of people older than 50, and the respondents below that age are just partners of the participants (Angelini et al., 2016). Observations linked to negative amounts of household resources are excluded from the sample as well (51 in total). Questions that were answered with “Refusal” or “Don’t Know” are reclassified as missing values. The last step before arriving at the final data set is the list wise deletion of all missing values for one of the variables in the dataset. The final sample is composed of 10,728 respondents5. In cleaning the dataset, respondents from the countries Israel, Czech Republic and Poland are eliminated. The descriptive statistics of the sample are discussed in Chapter 5 Data Analysis. An overview of the variables in the sample, with SHARE coding and explanation, can be found in appendix A. 3.3 Construction of Variables
private pension precaution. Previous literature on portfolio allocation (see e.g. Guiso et al., 2002; Rosen and Wu, 2004; Atella et al., 2012) use this wealth allocation breakdown to organize the asset classes in three groups; “safe”, “fairly safe”, and “risky”. An individual has a “safe” portfolio when all his or her wealth sits in bank or savings accounts. “Fairly safe” assets are bonds, mutual funds, individual retirement accounts, contractual savings for housing and life insurances. Lastly, stocks are considered to be “risky” assets. This approach is used to generate two variables; (1) Direct stock market participation and (2) Indirect stock market participation. Table 1 shows which asset classes determine the dependent variable. For example, if the individual holds mutual funds and/or an individual retirement account, the dependent variable “Indirect stock market participation” will take value 1. If the individual does not hold stocks, the dependent variable “Direct stock market participation” takes value 0. Both dependent variables are binary. Table 1: Dependent variables and asset classes Dependent variable Asset classes Direct stock market participation Stocks Indirect stock market participation Mutual funds, Individual retirement accounts 3.3.2 Construction of Independent Variables As mentioned in the literature review, the base independent variable in my study is perceived health status. SHARE offers the variable perceived health by asking the question: “Would you say your health is (1) very good, (2) good, (3) fair, (4) bad or (5) very bad”.6 A dummy variable “Poor perceived health” is generated that takes value 1 for when the question is answered with fair, bad or very bad, and takes value 0 for the answers very good and good. Mental health status is used as replacement of perceived health as sensitivity check in the “additional analyses” section of this paper. Mental health status is based on the EURO-D scale, that consists of twelve items; depression, pessimism, suicidality, guilt, sleep, interest, irritability, appetite, fatigue, concentration, enjoyment and tearfulness. Each ‘negative’ item, such as guilt and irritability, is marked with value 1 if present and 0 otherwise. For the
‘positive’ items, such as enjoyment, this is the other way around. Hence, the EURO-D scale ranges from 0 to 12. The variable poor mental health takes value 1 if the EURO-D scale is greater than 3, and 0 otherwise (Prince et al., 1999).
Three binary independent variables reflect late-in-life risky health behavior of the individual. The first, “Individual smokes” is a binary variable that takes value 1 if the individual answered the question “Do you currently smoke” with “Yes”. The variable takes value 0 for other answers. The same applies to the variable “Individual drinks”, which takes value 1 if the individual drinks more than two glasses per day at least five days per week, and the variable “Individual is physically inactive”, which takes value 1 if the individual says to be physically inactive. Two binary independent variables reflect risky health behavior of the parents to which the individual was exposed early in life. “Parents smoked” and “Parents drank heavily” take value 1 if one or both of the parents smoked or drank more than 2 glasses per day during childhood, and 0 otherwise. 3.2.3 Construction of Control Variables A set of control variables that might cause omitted variable biases will be added to the equation. These controls aim to address the confounding variables in the investigated relationship. An extensive set of demographic variables will be used as controls, based on previous research (Bertaut and Starr-McCluer, 2002; Carroll, 2002; Guiso et al., 2002; Guiso and Paiella, 2008; Christelis et al., 2010; Love and Smith, 2010; Dohmen et al., 2011; Atella et al., 2012; Bressan et al., 2014; Christiansen et al., 2015; Angelini et al., 2016). Age is included in linear and quadratic form to account for life-cycle effects (Atella et al., 2012). Gender and marital status are included to capture differences between males and
account for the bequest motive7 in relation to portfolio choice, the expectation of leaving an inheritance are included (Hurd, 2002; Bressan et al., 2014). Christelis et al. (2010) find that high cognitive skills in early life have positive explanatory power in portfolio choice. To capture this effect, two control variables for cognitive skills during childhood are generated. SHARE offers two variables on relative school performance of the individual in childhood, both of which are included as binary control variables. The binary variable takes value 1 if the individual answered the question “when ten years old, relative to your schoolmates, how good were you in mathematics (language)” with “much better” or “better”, and 0 otherwise. To account for cross-country variance in portfolio choice, country dummies are included as control variables. Finally, household resources are controlled for by including gross household income, gross household financial assets and household real assets. These variables are numeric and should be tested for normality before including them in the analysis. In expectation, the distribution of wealth data has fat right tails, since a small group of people hold a disproportioned amount of wealth. Table 2 shows that the household resource variables are indeed highly skewed. After using the natural logarithm technique to transform the variables, the distribution is sufficiently symmetric to incorporate the log variables in my analysis.
Table 2: Testing numeric control variables for normality
Variable Skewness Kurtosis Observations
Household gross income 21.45 1105.62 10,767 Household gross financial assets 16.80 705.46 10,767 Household real assets 27.46 1253.42 10,767 Log Household gross income -0.32 3.95 10,767 Log Household gross financial assets 0.11 1.89 10,767 Log Household real assets -0.98 2.88 10,767 Note: As a generally accepted rule of thumb, a distribution is highly skewed if Skewness is less than -1 or greater than 1. A distribution is moderately skewed if Skewness falls within the -0.5 to -1 and 0.5 to 1 range. If Skewness is between -0.5 and 0.5, a distribution is approximately symmetric.
4. Empirical Framework
The empirical strategy of this paper aims to explore the relationship between health status and portfolio choice, and how late-in-life risky health behavior and early-in-life exposure to risky health behavior affects this relationship. A logistical regression model8 is used that includes relevant variables step by step, to see how the associations between the key variables are affected. The results of the regression analyses will be interpreted with odds ratios. The logistical regressions will be performed in STATA 15. 4.1 Unconditional Relationship
Based on correlation matrices and t-tests of the independent variables (perceived health, individual smokes, individual drinks, individual is physically inactive, parents smoked during individual’s childhood, and parents drank heavily during individual’s childhood) and the dependent variables (direct stockholdings and indirect stockholdings), the association of the relationships can be estimated. Without any control variables included, this approach will suffer from the omitted variable bias and will not give a fully accurate view as the role of other explanatory variables are hidden. However, it does give an indication. 4.2 Conditional Relationship After examining the descriptive statistics, correlations and t-tests, I will focus on the heart of the game with a framework that examines the relationship of my key variables, conditional on the set of control variables. This paper uses the following cross-sectional direct relationship between the dependent variable !, which is a binary variable for direct stock market participation, and the independent variables on the right-hand side of the equation: 1 !$ = & + ()*+$+ (,-.$+ (/-0$+ (1-2$+ (3*.$+ (4*0$+ 5)61$7+ 5 ,62$7+ 5/0$7+ 9$ where the subscript : represents the individual investor. & is the intercept, perceived health status is measured by *+. Individual smokes, individual drinks and individual is physically inactive are captured by -., -0, and -2, respectively. Parents smoked and parents drank heavily are measured by *. and *0. The standard demographic control variables are
variables if not added to the equation. If not added, the model attributes the effect of risky health behavior to the variable perceived health. My expectation is that individuals who engage in risky health behavior are fully aware that this is health deteriorating and therefore rate their perceived health lower than individuals who do not engage in risky health behavior. The third hypothesis is therefore: “Individual risky health behavior is negatively associated with stock market participation, and adding individual risky health behavior decreases the magnitude of the negative association between perceived health and stock market participation.” Tranche 3 is estimated with equation (1C):
1< !$ = & + ()*+$+ (,-.$+ (/-0$+ (1-2$+ 5)61$7+ 5,627$+ 5/0$7+ 9$
4.2.3 Conditional Relationship – Tranche 4
The last tranche includes parental risky health behavior to which individuals were exposed during childhood. It is assumed that ‘parents smoked’ and ‘parents drank heavily’ would be omitted variables if not added to the equation. It is assumed that parental risky health behavior amplifies the effect of individual risky health behavior on perceived health (and subsequently on stock market participation), because individuals who experienced their parents exhibiting risky health behavior during childhood are more likely to engage in risky health behavior themselves later in life. The rationale behind this is that individuals copy behavior of their parents at least to some extent. My expectation is that parental risky health behavior absorbs some explanatory power of individual risky health behavior and perceived health. The 4th hypothesis of this paper is that parental risky health behavior is negatively associated with stock market participation, and that adding the variables reduces the magnitudes of the relationships of individual risky health behavior and perceived health with stock market participation. Tranche 4 is captured in the equation introduced at the beginning of this chapter, equation (1): 1 !$ = & + ()*+$+ (,-.$+ (/-0$+ (1-2$+ (3*.$+ (4*0$+ 5)61$7+ 5,62$7+ 5/0$7+ 9$ 4.3 Additional Analyses
will be used as dependent variable in the logistic regression. The second analysis replaces ‘perceived health’ as base independent variable with ‘mental health’.
4.3.1 Additional Analysis 1 – Indirect Stock Market Participation
Additional analysis 1 will use the exact same framework as introduced for the core relationship of this paper9. The dependent variable ‘direct stock market participation’ is replaced with its indirect counterpart. The binary variable ‘indirect stock market participation10’ represents individuals that hold stocks via mutual funds or individual retirement accounts (hence, indirectly).
This analysis builds on the core relationship (perceived health vs. stock market participation) of this paper by looking at a dependent variable that reflects a lesser degree of riskiness, in the context of the same set of independent and control variables. It was hypothesized that the core relationship of this paper shows that people in poor health are less likely to participate in the stock market (hence, a negative association). When using a dependent variable that reflects a less risky version of stock market participation, I still hypothesize a negative association, but one that is weaker than with the ‘direct stockholding’ dependent variable. With regard to the effect of the different tranches, I assume the same effects as described in section 4.2, but again with a weaker magnitude due to the fact that indirect stockholding is less risky than direct stockholding. 4.3.2 Additional Analysis 2 – Mental Health
Additional analysis 2 also uses the framework with the four tranches. The core dependent variable ‘direct stock market participation’ is used in this analysis. The core independent variable ‘poor perceived health’ is replaced with ‘poor mental health11’, as sensitivity check. The same equations as presented in section 4.2 are used, with the only difference being the replacement of PH (perceived health) with MH (mental health).
medical expenses or reduced income that might arise from health problems. The same might hold for mental health. Sickness from mental health issues might also lead to costly medicines and treatments, and reduced income. The literature showed that mental health should have roughly the same effect on portfolio choice as perceived health (Bogan and Fertig, 2013). My hypothesis is therefore that mental health status has explanatory power in portfolio choice. People with poor mental health are assumed to be less likely to participate in the stock market. With regard to the effect of the different tranches, I assume the same effects as described in section 4.2.
Table 3: Summary statistics dependent variables
Variable Total Austria Germany Sweden Netherl. Spain
Indirect stockholding 0.272 0.047 0.243 0.737 0.160 0.110 (0.445) (0.213) (0.429) (0.441) (0.367) (0.313) Direct stockholding 0.180 0.040 0.173 0.468 0.186 0.051 (0.385) (0.197) (0.379) (0.499) (0.389) (0.220) Observations 10,728 695 797 1,188 1,081 785
Variable Italy France Denmark Greece Switz. Belgium
5.1 Summary Statistics
Table 3 shows the summary statistics of the dependent variables. 18% of the individuals in the total sample hold stocks directly. 27% participates indirectly in the stock market via individual retirement accounts and mutual funds that have stock holdings. The summary statistics show high heterogeneity across the countries. For example, 47% of Swedish individuals participate directly in the stock market, but only 4% of the Austrian people have direct stockholdings. Atella et al., (2012) explain this heterogeneity nicely by stating that it “stems from several factors, ranging from the development of each country’s financial market to the average level of financial education of households”. The summary statistics of the dependent variables in my sample are roughly in line with previous research that uses household data (e.g. Rosen and Wu, 2004; Atella et al., 2012; Bressan et al., 2014). The summary statistics of the independent variables and control variables are found in table 4. The average individual respondent is 64 years old. There are slightly more females in the sample (53%) than males (47%). 24% attended high to post-secondary education, 67% is together with someone (married or in registered partnership), 88% has children and 45% is employed. Roughly a third of the individuals was better than average at school in mathematics (37%) and language (38%). The average probability of leaving an inheritance of at least 50,000 Euros is 58%. As described in chapter 3, the household resources were transformed for normality purposes. Underlying data shows that the average household has a yearly income of 44 thousand Euros, holds gross financial assets worth of 41 thousand Euros, and has 195 thousand Euros of real assets13.
we go well back into the second part of the 20th century. Back then, the adverse health effects of smoking of were less generally known and accepted. Lastly, 8% of the parents drank heavily during the childhood of the individuals in this sample. These summary statistics are also roughly in line with summary statistics of other papers using SHARE as data source (Atella et
al., 2012; Bressan et al., 2014).
Table 4: Summary statistics independent variables
Variable Mean S.D. Min Max Obs.
Independent variables Poor perceived health 0.330 0.470 0 1 10,728 Poor mental health 0.232 0.422 0 1 10,728 Individual smokes 0.198 0.398 0 1 10,728 Individual drinks (>2 glasses daily for at least 5 days) 0.144 0.351 0 1 10,728 Individual is physically inactive 0.069 0.253 0 1 10,728 Parents smoked during individual's childhood 0.627 0.484 0 1 10,728 Parents drank heavily during individual's childhood 0.076 0.266 0 1 10,728 Control variables Age 64 9 50 90 10,728 Age squared 4,158 1,246 2500 8100 10,728 Male 0.466 0.499 0 1 10,728 Education 0.237 0.425 0 1 10,728 Individual is together with someone 0.665 0.472 0 1 10,728 Individual has children 0.883 0.322 0 1 10,728 Individual is employed 0.448 0.497 0 1 10,728 Relative good performance in mathematics 0.367 0.482 0 1 10,728 Relative good performance in language 0.376 0.484 0 1 10,728 Probability of leaving a >50k inheritance 0.583 0.433 0 1 10,728 Log Household gross income 3.313 1.030 0 8.361 10,728 Log Household gross financial assets 2.383 1.770 0 8.529 10,728 Log Household real assets 4.164 1.986 0 10.133 10,728
Note: The table shows the summary statistics of the independent variables and control variables. The
5.2.1 Dependent Variables
Section 4.3.1 touched upon the difference in interpreting direct stock market participation and indirect stock market participation. In essence, direct stock market participation is a riskier portfolio choice than indirect stock market participation. This is reflected in two t-tests between the dependent variables (see appendix B.1). The t-tests indicate two things. Firstly, if you have direct stockholdings, you are more likely to also hold indirect stockholdings (and vice versa). The correlation of 38% between the two dependent variables also indicates this. Secondly, 42% of the people who hold stocks indirectly, hold stocks directly as well. The other way around shows a higher percentage: 64% of the people who hold stocks directly, also holds stocks indirectly. This result reflects the difference in riskiness between indirect and direct stockholdings: If you hold the riskier form of stocks, you are more likely to hold the less riskier form as well than the other way around.
5.2.2 Independent Variables
The correlations between the independent variables are showed in appendix B.2. As expected, poor perceived health and poor mental health move in the same direction (correlation of 0.279). In section 5.4, perceived health is replaced with mental health in the logistic regression model, as a sensitivity check. This correlation suggests that mental health will show an approximately similar result to perceived health in the logistical regression model, as they move in the same direction.
Surprisingly, the correlations between the risky health behavior variables are all very low. Physical inactivity by an individual even has a negative (though very weak) relationship with the variables ‘individual smokes’ and ‘individual drinks’. The suggested relationship between risky parental health behavior and individual risky health behavior based on copying behavior is not supported by strong positive correlations either.
5.2.3 Core Relationship
The t-tests in table 5 provide a first shot at the core relationship researched in this paper, i.e. perceived health with direct stock market participation, in the context of behavioral health risk variables. The t-statistic of the test between poor perceived health and direct stock market participation is positive and very high (significant at 1% level), indicating that individuals with a poor perceived health (group 1) are less likely to participate in the stock market than individuals with a good perceived health (group 0). This is in line with the literature. Similarly, the t-tests indicate that individuals who smoke, individuals who are physically inactive and individuals whose parents drank heavily are also less likely to have direct stockholdings (all significant at the 1% level). For individuals that (do not) drink heavily or individuals whose parents (did not) smoke, there is no statistically significant difference in explaining direct stockholdings. Overall, these findings are roughly in line with what is hypothesized: that individual and parental risky health behavior is negatively associated with direct stock market participation. The previous chapter explains that it is assumed these variables have their negative effect on stock market participation through perceived health status.
The second wave of t-tests in table 5 digs a bit deeper by looking how the behavioral health risk variables relate to perceived health. Interestingly, the results indicate that individuals who smoke or drink heavily are more likely to indicate their perceived health to be good rather than poor. In line with expectation, the t-tests strongly show that physical inactivity is related to poor perceived health. Individuals whose parents drank heavily are also more likely to rate their perceived health as poor (significant at 1% level). For individuals whose parents smoked, the t-tests show the inverse of what is expected, though the test is not extremely significant. The third wave of t-tests examines how late-in-life individual risky health behavior is related to risky health behavior of parents to which the individuals were exposed during childhood. The t-tests suggest a positive association between parental risky health behavior and late-in-life individual risky health behavior. Only excessive parental drinking behavior does not indicate the copying of drinking behavior of the individual later in life. It could be that the individuals in the sample have had bad experiences with their parents’ drinking behavior, such as aggression, and that instead of copying the behavior it leads to adverse behavior.
Table 5: T-Tests - the core relationship
Dependent variable Independent variable T statistic H0: diff = 0 Sign.
Wave 1
Direct stockholding Poor perceived health 13.45 Rejected ***
Direct stockholding Individual smokes 3.42 Rejected ***
Direct stockholding Individual drinks heavily -0.36 Not Reject
Direct stockholding Individual physically inactive 8.78 Rejected ***
Direct stockholding Parents smoked -1.14 Not Reject
Direct stockholding Parents drank heavily 2.43 Rejected ***
Wave 2
Poor perceived health Individual smokes 3.07 Rejected ***
Poor perceived health Individual drinks heavily 4.80 Rejected *** Poor perceived health Individual physically inactive -22.96 Rejected ***
Poor perceived health Parents smoked 1.72 Rejected **
Poor perceived health Parents drank heavily -5.74 Rejected ***
Wave 3
Individual smokes Parents smoked -10.05 Rejected ***
Individual smokes Parents drank heavily -4.32 Rejected ***
Individual drinks Parents did not smoke -5.77 Rejected ***
Individual drinks Parents did not drink heavily -1.07 Not Reject
Table 6: Logistic regression on dependent variable "Direct stock market participation"
Dependent variable > Direct stock market participation
Tranche 1 Tranche 2 Tranche 3 Tranche 4
Independent variables O.R. S.E. Sign. O.R. S.E. Sign. O.R. S.E. Sign. O.R. S.E. Sign.
Poor perceived health 0.662 0.046 *** 0.822 0.062 ** 0.827 0.063 ** 0.827 0.063 ** Age 1.060 0.047 1.024 0.050 1.021 0.050 1.020 0.050 Age squared 0.999 0.000 1.000 0.000 1.000 0.000 1.000 0.000 Male 1.543 0.092 *** 1.287 0.083 *** 1.273 0.084 *** 1.272 0.084 *** Education 1.724 0.108 *** 1.272 0.087 *** 1.267 0.087 *** 1.267 0.087 *** Individual is together with someone 1.428 0.101 *** 0.909 0.074 0.906 0.074 0.906 0.074 Individual has children 0.885 0.084 1.012 0.104 1.010 0.104 1.009 0.104 Individual is employed 1.077 0.080 0.960 0.080 0.962 0.080 0.961 0.080 Good in mathematics 1.385 0.087 *** 1.258 0.086 *** 1.260 0.086 *** 1.261 0.086 *** Good in language 1.068 0.068 0.989 0.068 0.988 0.068 0.988 0.068 Prob. of leaving >50k inheritance 3.102 0.234 *** 1.198 0.111 ** 1.199 0.111 ** 1.198 0.111 ** Log Household gross income 1.156 0.047 *** 1.155 0.047 *** 1.155 0.047 *** Log Household gross financial assets 2.110 0.057 *** 2.108 0.057 *** 2.109 0.057 *** Log Household real assets 1.142 0.027 *** 1.140 0.027 *** 1.141 0.027 *** Risky health behavior individual Individual smokes 0.940 0.077 0.936 0.077 Individual drinks heavily 1.093 0.099 1.092 0.099 Individual is physically inactive 0.902 0.168 0.904 0.169 Risky health behavior parents One or two parents smoked 1.053 0.071 One or two parents drank 0.980 0.123
Country dummies included Yes Yes Yes Yes Constant 0.011 0.017 *** 0.002 0.004 *** 0.003 0.005 *** 0.003 0.004 ***
5.3.2 Tranche 2
Tranche 2 builds on the basis set in tranche 1, by adding the resources of the household of which the individual is part. The literature predicted that the magnitude of the base relationship of this paper, poor perceived health versus direct stock market participation, would decrease after controlling for household resources. In the tranche 2 section of table 6 we see that prediction coming through in the results: The odds ratio of poor perceived health remains below 1 but increases to 0.822 (vs. previous 0.662). The main result of tranche 2: “People with poor perceived health are less likely to participate in the stock market, even after controlling for standard demographic variables and household resources. Hypothesis 2 is not rejected.”
The household resources themselves are positively associated (all statistically significant at 1% level) with direct stock market participation. The rationale behind this is as described in section 4.2.2: when you possess more money, you are more likely to put part of your money into stocks. As expected in previous section, the effect of the probability of leaving an inheritance decreases substantially as a result of adding household resources to the equation (from an O.R. of 3.102 to 1.198). The association is still positive at significant at the 5% level. The addition of household resources also decreases the positive effect of education (from an O.R. of 1.724 to 1.272), because individuals with a better education hold more wealth. The results clearly indicate that part of the explanatory power of income and wealth towards stock market participation was hidden within the education component of tranche 1. Lastly, the positive and statistically significant effect of being together with someone fully disappears in tranche 2 due to the addition of household resources. The reason for this might be that household resources are a two-person shared variable for most observations, and that being together with someone is inherent to this.
The results on country dummies, constant term and number of observations remain in line with tranche 1. R-squared slightly increases after controlling for household resources.
5.3.3 Tranche 3
The tranche 3 section of table 6 shows that the variables that were included in the tranche 2 equation are not affected by the addition of late-in-life individual risky health behavior variables. The risky health behavior variables have no indirect effect on stock market participation through the perceived health variable. In addition, the risky health behavior variables themselves have no direct explanatory power in direct stock market participation, since the O.R. of all three variables is close to 1 and none are statistically significant.
The results on country dummies, constant term, R-squared and number of observations remain in line with tranche 2.
5.3.4 Tranche 4
Tranche 4 adds parental risky health behavior to which individuals were exposed during childhood to the equation. The main result of tranche 4: “Parental risky health behavior has no (in)direct explanatory power in the relationship between perceived health and stock market participation. The hypothesis of tranche 4 is rejected”.
The tranche 4 section of table 6 shows that the variables that were included in the tranche 3 equation are not affected by the addition of late-in-life individual risky health behavior variables. The parental risky health behavior variables have no indirect effect on stock market participation through the individual risky health behavior or through the perceived health variable. In addition, the parental risky health behavior variables themselves have no direct explanatory power in direct stock market participation, since the O.R. of all three variables is close to 1 and none are statistically significant.
The results on country dummies, constant term, R-squared and number of observations remain in line with tranche 3.
5.4 Additional Analyses
of 0.763). As expected, the relationship is less strong than the relationship between perceived health and direct stock market participation (O.R. of 0.662). However, after including the household resources in tranche 2, and adding risky health behavior in tranche 3 and 4, the effect disappears. Compared to the results of the core model, this would indicate that indirect stock market participation does not reflect a strong enough measure of risk taking by the individual. Interestingly, tranche 3 and tranche 4 show that individuals that are physically inactive are less likely to participate indirectly in the stock market, with a strong odds ratio below 1 (0.611 for tranche 3, and 0.613 for tranche 4), statistically significant at the 1% level. 5.4.2 Additional Analysis 2 – Mental Health
8. References Angelini, V., Klijs, B., Smidt, N., Mierau, J., 2016. Associations between childhood parental mental health difficulties and depressive symptoms in late adulthood: the influence of life-course socioeconomic, health and lifestyle factors. PLOS ONE 11, 1-13. Atella, V., Brunetti, M., Maestas, N., 2012. Household portfolio choices, health status and health care systems: A cross-country analysis based on SHARE. Journal of Banking and Finance 36, 1320-1335. Berkowitz, M., Qiu, J., 2006. A further look at household portfolio choice and health status. Journal of Banking and Finance 30, 1201-1217. Bogan, V., Fertig, A., 2013. Portfolio choice and mental health. Review of Finance 17, 955-992. Bertaut, C.C., Starr-McCluer, M., 2002. Household portfolios in the United States. In: Guiso, L., Haliassos, M., Jappelli, T. (Eds.), Household Portfolios. MIT Press, Cambridge, MA, pp. 181– 218. Bressan, S., Pace, N., Pelizzon, L., 2016. Health status and portfolio choice: Is their relationship economically relevant? International Review of Financial Analysis 32, 109-122. Cardak, B., Wilkins, R., 2009. The determinants of household risky asset holdings: Australian evidence on background risk and other factors. Journal of Banking and Finance 33, 850-860. Carroll, C., 2000. Portfolios of the Rich. In: Guiso, L., Haliassos, M., Jappelli, T., Household Portfolios. MIT Press, Cambridge, MA, pp. 389–430. Christelis, D., Jappelli, T., Padula, M., 2010. Cognitive abilities and portfolio choice. European Economic Review 54, 18-38.
Christiansen, C., Joensen, J., Rangvid, J., 2015. Understanding the effects of marriage and divorce on financial investments: The role of background risk sharing. Economic Inquiry 53, 431-447. Coile, C., Milligan, K., 2009. How household portfolios evolve after retirement: The effect of aging and health shocks. Review of Income and Wealth 55, 226-248. Dohmen, T., Falk, A., Huffman, D., Sunde, U., Schupp, J., Wagner, G.G., 2011. Individual risk attitudes: Measurement, determinants and behavioral consequences. Journal of the European Economic Association 9, 522-550.
Edwards, R., 2008. Health risk and portfolio choice. Journal of Business and Economic Statistics 26, 472-485.
Fan, E., Zhao, R., 2009. Health status and portfolio choice: Causality or heterogeneity? Journal of Banking and Finance 33, 1079-1088.
Goldman, D., Maestas, N., 2007. Medical expenditure risk and household portfolio choice. Rand Working Paper 325. Guiso, L., Haliassos, M., Jappelli, T., 2002. Household Portfolios: An International Comparison. Household Portfolios, 1-24. Guiso, L., Paiella, M., 2008. Risk aversion, wealth and background risk. Journal of the European Economic Association 6, 1109-1150. Gupta, V., 2007. Wealth shock and impact of health on risk aversion and savings. Towers Watson Technical Paper 7, 1-15.
Heaton, J., Lucas, D., 2000. Portfolio choice and asset prices: The importance of entrepreneurial risk. Journal of Finance 55, 1163–1198. Hurd, M., 2002. Portfolio holdings of the elderly. MIT Press, Cambridge. Kimball, M., 1990. Precautionary saving in the small and in the large. Econometrica 58, 1-53. Love, D., Smith, P., 2010. Does health affect portfolio choice? Health Economics 19, 1441– 1460. Markowitz, H., 1952. Portfolio selection. The Journal of Finance 7, 77-91. McGinnis, M., Foege, W., 1993. Actual Causes of Death in the United States. The Journal of the American Medical Association 18, 2207-2212. Palumbo, M., 1999. Uncertain medical expenses and precautionary saving near the end of the life cycle. Review of Economic Studies 66, 395-421. Prince, M., Beekman, A., Deeg, D., Fuhrer, R., Kivela, S., Lawlor, A., Lobo, A., 1999. Depression symptoms in late life assessed during the EURO-D scale. Effect of age, gender and marital status in 14 European centres. British Journal of Psychiatry 174, 339–345. Rosen, H., Wu, S., 2004. Portfolio choice and health status. Journal of Financial Economics 72, 457–484. Rubin, D., 1976. Inference and missing data. Biometrika 63, 581-592.
Appendix A: Overview of variables
SHARE Code Variables Description
Dependent variables
as002d3 Direct stock market participation Dummy variable that takes value 1 if the individual holds stocks (as002d3 = 1) as002d4
as002d5 Indirect stock market participation Dummy variable that takes value 1 if the individual holds mutual funds (as002d4 = 1) and/or individual retirement accounts (as002d5 = 1).
Independent variables
spheu Poor perceived health Dummy variable that takes value 1 if the individual perceives his or her own wealth to be poor (spheu = 3, 4, or 5).
eurodcat Poor mental health Dummy variable that takes value 1 if the individual is depressed. This is measured by the EURO-D scale (eurodcat > 3).
cusmoke Individual smokes Dummy variable that takes value 1 if the individual currently smokes (cusmoke = 1). drinking2 Individual drinks heavily Dummy variable that takes value 1 if the
individual drinks more than 2 glasses per day for at least 5 days a week (drinkin2 = 1). phactiv Individual is physically inactive Dummy variable that takes value 1 if the
individual says to be physically inactive (phactiv = 1) sl_hs045d1 One or two parents smoked Dummy variable that takes value 1 if one or two of the parents smoked during the individual's childhood (sl_hs045d1 = 1). sl_hs045d2 One or two parents drank Dummy variable that takes value 1 if one or two of the parents drank heavily the individual's childhood (sl_hs045d2= 1). Control variables
age Age Age of the individual
gender Male Dummy variable that takes value 1 if the
individual is male and value 0 if the individual is female.
isced Education Dummy variable that takes value 1 if the
individual has attented post secondary education. This is based on the ISCED scale (isced = 4,5 or 6).
nchild Individual has children Dummy variable that takes value 1 if the individual has at least one child (nchild > 0). cjs Individual is employed Dummy variable that takes value 1 if the individual is employed (cjs = 2). sl_cs010_ Good in mathematics Dummy variable that takes value 1 if the individual was relatively good in mathematics during childhood (sl_cs010_ = 1 or 2). sl_cs0101_ Good in language Dummy variable that takes value 1 if the individual was relatively good in language during childhood (sl_cs010_ = 1 or 2). ex004_ Prob. Of leaving >50k inheritance Variable that takes a value between 0 and 1
based on the probability of leaving an inheritance of more than 50k.
thinc Log Household gross income Natural logarithm of the amount of household gross income.
hgfass Log Household gross financial assets Natural logarithm of the amount of household gross financial assets hrass Log Household real assets Natural logarithm of the amount of
household real assets.
country Austria Country dummy for Austria (country = 11)
country Germany Country dummy for Germany (country = 12)
country Sweden Country dummy for Sweden (country = 13)
country The Netherlands Country dummy for The Netherlands
(country = 14)
country Spain Country dummy for Spain (country = 15)
country Italy Country dummy for Italy (country = 16)
country France Country dummy for France (country = 17)
country Denmark Country dummy for Denmark (country = 18)
country Greece Country dummy for Greece (country = 19)
country Switzerland Country dummy for Switzerland (country =
20)
country Belgium Country dummy for Belgium (country = 23)
Appendix B: Bivariate Sample Statistics
Appendix B.1: T-tests - dependent variables
Dependent: Indirect stockholdings Dependent: Direct stockholdings
stockholding Direct stockholding No direct stockholding Indirect stockholding No indirect
Observ. 2,917 7,811 Observ. 1,935 8,793 Mean 0.422 0.090 Mean 0.636 0.192 T-statistic -43.0 T-statistic -43.0 H0 Rejected H0 Rejected
Significance 1% level Significance 1% level
Appendix B.2: Correlation matrices - independent variables
Poor health perceived Poor health mental
Poor perceived health 1.000 0.279
Poor mental health 0.279 1.000
Individual smokes Individual drinks Physically inactive smoked Parents Parents drank
Appendix C: Logistic regression on dependent variable "Indirect stock market participation"
Dependent variable > Indirect stock market participation
Tranche 1 Tranche 2 Tranche 3 Tranche 4
Independent variables O.R. S.E. Sign. O.R. S.E. Sign. O.R. S.E. Sign. O.R. S.E. Sign.
Poor perceived health 0.763 0.047 *** 1.003 0.071 1.034 0.074 1.030 0.074 Age 1.051 0.044 1.045 0.049 1.036 0.049 1.036 0.049 Age squared 0.999 0.000 ** 0.999 0.000 * 0.999 0.000 0.999 0.000 Male 1.428 0.080 *** 1.156 0.073 ** 1.129 0.073 * 1.125 0.073 * Education 1.934 0.117 *** 1.459 0.101 *** 1.456 0.101 *** 1.459 0.101 *** Individual is together with someone 1.423 0.092 *** 0.931 0.072 0.931 0.072 0.931 0.072 Individual has children 0.782 0.067 *** 0.933 0.090 0.928 0.089 0.926 0.089 Individual is employed 1.289 0.088 *** 1.212 0.096 ** 1.214 0.096 ** 1.214 0.096 ** Good in mathematics 1.073 0.064 0.948 0.064 0.951 0.064 0.951 0.064 Good in language 1.233 0.073 *** 1.154 0.078 ** 1.147 0.077 ** 1.146 0.077 ** Prob. of leaving >50k inheritance 2.809 0.190 *** 1.257 0.110 *** 1.257 0.110 *** 1.256 0.110 *** Log Household gross income 1.135 0.043 *** 1.133 0.043 *** 1.133 0.043 *** Log Household gross financial assets 2.447 0.063 *** 2.445 0.063 *** 2.447 0.063 *** Log Household real assets 0.994 0.021 0.992 0.021 0.993 0.021 Individual smokes 1.026 0.081 1.014 0.080 Individual drinks heavily 1.088 0.094 1.083 0.094 Individual is physically inactive 0.611 0.103 *** 0.613 0.104 *** One or two parents smoked 1.134 0.074 One or two parents drank 1.049 0.123
Appendix D: Logistic regression on dependent variable "Direct stock market participation"
Dependent variable > Direct stock market participation
Tranche 1 Tranche 2 Tranche 3 Tranche 4
Independent variables O.R. S.E. Sign. O.R. S.E. Sign. O.R. S.E. Sign. O.R. S.E. Sign.
Poor mental health 0.843 0.065 ** 0.883 0.074 0.888 0.075 0.888 0.075 Age 1.064 0.047 1.024 0.050 1.020 0.050 1.020 0.050 Age squared 0.999 0.000 * 1.000 0.000 1.000 0.000 1.000 0.000 Male 1.543 0.093 *** 1.277 0.084 *** 1.263 0.085 *** 1.261 0.084 *** Education 1.768 0.110 *** 1.284 0.088 *** 1.278 0.087 *** 1.278 0.087 *** Individual is together with someone 1.418 0.100 *** 0.900 0.073 0.898 0.073 0.898 0.073 Individual has children 0.886 0.084 1.015 0.105 1.013 0.105 1.012 0.104 Individual is employed 1.126 0.083 0.977 0.080 0.978 0.081 0.977 0.081 Good in mathematics 1.389 0.087 *** 1.257 0.086 *** 1.259 0.086 *** 1.260 0.086 *** Good in language 1.076 0.068 0.992 0.068 0.990 0.068 0.989 0.068 Prob. of leaving >50k inheritance 3.170 0.238 *** 1.200 0.111 ** 1.200 0.111 ** 1.199 0.111 ** Log Household gross income 1.157 0.047 *** 1.156 0.047 *** 1.156 0.047 *** Log Household gross financial assets 2.120 0.057 *** 2.117 0.057 *** 2.117 0.057 *** Log Household real assets 1.143 0.027 *** 1.142 0.027 *** 1.142 0.027 *** Individual smokes 0.940 0.077 0.936 0.077 Individual drinks heavily 1.096 0.099 1.095 0.099 Individual is physically inactive 0.867 0.161 0.869 0.161 One or two parents smoked 1.052 0.071 One or two parents drank 0.968 0.122
