The healthy appetite for risk: examining the effect of health on portfolio allocation

(1)

The healthy appetite for risk: examining the effect

of health on portfolio allocation

University of Groningen

Faculty of Economics & Business

Msc. Finance & Msc. BA: Health

June 22

nd

2020

By Idsge Pieter van Dijk

Student ID S2915669

Supervisors Charles Adjasi (Finance) Maarten Postma (BA Health)

(2)

Abstract

This thesis examines the effect of the health (SRH) of individuals and their health compared to the previous year (CRH), both measured through subjective measurements, on their financial risk taking behaviour, measured through the fraction of their total wealth invested in risky assets. SRH- and CRH-scores are both scored on a 5-point scale, where a higher score means a better state of perceived health. To this end, data has been obtained from the LISS panel, through the MESS project by CenterData. This data is analysed through Pooled OLS, Fixed Effects, Random Effects and Tobit models. Furthermore, Principal Component Analysis is used to obtain indices for financial literacy for the respondents in the sample. Significant results are found to both, with the fraction of risky assets increasing by up to 6.1% per one-point increase in SRH-score, but decreasing by up to 8.5% per one-point increase in CRH-score. These results seem contradictory, but may be explained by the effect of adverse health shocks. The results imply that the health of investors significantly impact their financial risk taking behaviour. Further research is required to solidify these findings and to explore the exact nature of the relationships found.

Keywords Health, Risk-attitude, Investment, Portfolio allocation

(3)

1. Introduction

Over the past decades, much has been written on risk-taking. Due to the availability of many financial products to individual investors, complicated portfolios can be constructed that match an individuals’ risk-profile. Whereas much has been written on organizational and managerial risk-taking, literature on drivers of personal risk-taking remain scarce. The investors’ risk attitude is most often considered an exogenous variable, a trait which can be measured. A popular measurement of risk-attitudes is the Arrow-Pratt measures of risk-aversion (Arrow, 1971) which is based on expected utility theory where risk-aversion in individuals can be determined through their required risk-premium for certain choices. However, Expected Utility Theory does not account for any outside factors that may influence risk attitudes. This is illustrated by Prospect Theory (Kahneman and Tversky, 1979), who found that risk attitudes may differ based on the framing of the problem and that decision weights do not necessarily correspond to the real probabilities in said problems.

(5)

5

In financial literature, risk-taking is often discussed in a principal-agent context. In the case of household finance, there may be forms of moral hazard connected to decision-making. For example, people who bankrupt themselves on risky investments will not only do so at great personal cost, but they might also incur societal costs (Thompson, Gazel and Rickman, 1997). Examining the effect of health status, this study aims to shed a light on these societal costs and to contribute to the overall debate on the costs and benefits of healthcare. Were this relationship to hold up in practice, a great opportunity cost for investment in healthcare would be revealed. Furthermore, if a link between health and financial risk attitude could be extended to a link with risky behaviour in general, this would be of special interest to insurance firms, who could use such information to more accurately predict future costs based on their client portfolio. Given the research gap regarding the link between physical health and decision making, the following research question arises:

What is the effect of personal health on the financial risk attitude of individual investors?

By answering this research question, I aim to contribute to the existing literature in behavioural finance meant to explain investor behaviour, as well as to contribute to the discussion on the importance of public health.

In order to investigate this relationship, data from the Longitudinal Internet Studies of Social Sciences (LISS) panel will be used. The main explanatory variable is subjective health, the measurement of which will be discussed in the data section of this paper. The research will be structured around the following broad hypotheses:

(1) People with high subjective health scores tend to be less risk averse in their investment behaviour

(2) Changes in subjective health scores tend to go along with changes in risk aversion in investment behaviour over time

Therefore, health is assumed to affect risk aversion as such that healthier people are assumed to be less risk averse. These broad hypotheses will be specified further in the next section, based on a review of existing literature on this topic.

(6)

6

the previous year is related to an 8.5% decrease in the fraction of risky assets. These findings indicate that a relationship between health and investment behaviour does exist, confirming the hypotheses posed above.

This thesis will be structured as follows. Section two will provide a literature review on risk attitude and decision making under risk, linking these concepts to personal health. This section will also further specify the hypotheses. The following section will elaborate on the context of the data, providing a health profile for the Dutch population. Section 4 describes the data, providing some insight into the measurements used in this study. Section 5 covers the methodology used to obtain the results. Section 6 contains the results of the study. Section 7 provides a discussion of the results and discusses potential policy implications. Section 8 discusses the limitations to which this thesis is subjected. The final section of this thesis will provide the conclusion of this thesis.

2. Literature Review

2.1 Defining risk attitude

The human attitude towards risk and uncertainty has long been subject to study. While this has resulted in a rich field of literature, it has also bred ambiguity in concepts and definitions regarding risk behavior. This is illustrated by Bran and Vaidis (2019), who list a number of studies using different terminologies and definitions. These include, among others, risk-taking

tendency, risk-taking propensity, risk appraisal, risk perception, and risk attitudes. While these

(7)

7

2.2 Decision making under risk: Expected Utility Theory and Prospect Theory

Originally introduced by Bernoulli (1738) as an answer to the St. Petersburg Paradox, and later refined by Von Neumann and Morgenstern (1944), Expected Utility Theory has long been the cornerstone of modeling decision making under risk. In Expected Utility Theory, risk attitudes are derived from problems as solved by economic agents; if an agent prefers a certain payoff x over an uncertain payoff equal to x, the agent can be considered risk-averse. An agent who is indifferent between the two would be considered risk-neutral, while an agent who prefers the uncertain payoff would be considered risk-loving. Traditionally, rational agents are considered to be risk-averse, meaning agents should be compensated for any risk they take on, the compensation being a risk premium.

As a reaction to Expected Utility Theory, Prospect Theory was developed by Kahneman and Tversky (1979). This alternative theory was developed due to the inability of Expected Utility

Theory to actively describe decision making under risk in practice (Edwards, 1996). Kahneman

and Tversky described the existence of the Certainty Effect, which entails that individuals tend to overweight certain outcomes relative to uncertain outcomes. Furthermore, they found that the framing of the problem affected the individuals’ preferences, resulting in inconsistent risk preferences dependent on the reference point used in the formulation of the problem. This second finding was dubbed the Isolation Effect. Other salient findings include the overweighting of losses compared to gains, and nonlinear weighting of probabilities. This indicates that, when making decisions under risk, individuals tend to abstract from rationality. Moreover, Tversky and Kahneman added that “Any discussion of the utility function for money

must leave room for the effect of special circumstances on preferences”, further suggesting that

personal characteristics, such as illness, may play a part in the process of risky decision making.

2.3 Factors affecting decision making under risk

(8)

8

supporting Tversky and Kahneman’s (1979) Prospect Theory. Moreover, Nygren et al. (1996) assert that the increased loss-aversion observed in positive affect people may be because these individuals have “more to lose” than control individuals. The same could be said for people with differing health states, suggesting a link between health status and risk attitude. In existing literature, links between health and decision making are scarce. While many articles are available on risk factors that may affect an individual’s health, not much is written on how health may influence behaviour. Bhatti, Salek, and Finlay (2011) reviewed the literature on the effect of chronic illness on major life-changing decisions. They found that patients suffering from various chronic diseases made major life changing decisions, such as having children or pursuing education, based on their condition, stating that “one negative life event, such as onset

of a chronic disease, may influence decisions relating to several subsequent life events, such as choice over education, career, employment, marriage, housing, having children and moving abroad”.

2.4 Investment behaviour over time: life-cycle models

(9)

9

remaining working life than the old. Given that this human capital is relatively risk-free, this reasoning would lead younger investors to take on larger shares of risky assets in their portfolio than older investors with the same risk-preferences, since the younger investor possesses a relatively risk-free non-financial asset in their human capital. Thus, Bodie et al. (1992) pose an argument for decreasing one’s exposure to risk as their time horizon shrinks. Viceira (2001) tests this argument in the absence of labour income insurance, meaning that labour income and the aforementioned human capital is not considered (near)risk-free. Viceira finds that, as long as labour income is not correlated with risky financial assets, the effect, while reduced, holds.

2.5 Investment behaviour related to health

The leap from arguments based on age to arguments based on health status is not that great in life-cycle models. Where Bodie et al. (1992) and Viceira (2001) consider age to be a depreciating factor to the non-financial human capital, diminishing health should logically considered as such as well. When in poor health, the flexibility of labour supply, on which these life-cycle arguments are at least partly built, decreases, as people in poor health may not be able to work more hours or delay their retirement. While health and age often go hand-in-hand, they are not inextricably linked: young people in poor health experience inflexibility in their labour supply relative to their peers, just as older people experience this inflexibility relative to the young. Furthermore, in the literature on life-cycle investment models mentioned above, different time horizons are considered the consequence of age. Bellante and Green (2004) bridge the gap to health putting forth the argument of diminishing health equalling a shortened time horizon, leading to different risk-preferences. Bellante and Green (2004) find that, in elderly subjects, poor health significantly decreases the fraction of portfolios allocated to risky assets. Therefore, the life-cycle models as presented by Bodie et al. (1992) and Viceira (2001) support the notion that people in poor health should hold a relatively smaller share of risky assets than people in good health, through the mechanisms of labour supply flexibility and differences in time horizons.

(10)

10

financial assets, irrespective of the household’s total wealth. Furthermore, they associate poor health with a smaller share of total wealth in risky assets and a larger share of total wealth in safer assets. They were, however, unable to pinpoint the mechanism through which health affected portfolio allocations in their sample; including variables such as risk preferences, planning horizon, bequest motives, and health insurance did not cause significant changes in the coefficients on the health variable. Fan and Zhao (2009) further support these findings. They examine the relationship between health status and portfolio choices, using four health indices. They find strong correlations between health and both financial and non-financial assets, but also find these results may be largely driven by heterogeneity. Therefore, the causality in their research remains questionable. Their results do however indicate that adverse health shocks significantly shift investors away from risky assets towards safer alternatives. Berkowitz and Qiu (2006) also find that a sudden deterioration of health in households drives a decline in financial wealth larger than the decline in non-financial wealth. They propose that this may be attributed to differences in liquidity and eligibility requirements for insurance in the United States. They further propose that this may be due to a need for liquidity to cover medical expenses. Goldman and Maestas (2013) further support this, finding that households with a higher risk of incurring medical expenses hold lower fractions of risky assets in their portfolios. Edwards (2010) examines a model in which health can change the shape of the utility function. In this model, health shocks again lead investors to choose safer portfolio choices, as this would be optimal if they expect to need more funds in poor health, as the marginal utility of consumption falls. This implies that, if individuals rate their own health as such that they expect their health to deteriorate in the (near) future, they would rationally choose less risky portfolio compositions, securing funds for a higher level of future consumption to maintain a constant level of utility. The notion of health altering the investor’s utility function has also been suggested by Rosen and Wu (2004) and has been argued by Finkelstein et al. (2013), who find that, when an individual’s health declines, the marginal utility of consumption declines along with it. More recently, Crainich et al. (2017) also found support for the adverse relationship between health and financial risk taking, finding that “a deterioration in health reduces

investment in risky financial assets”.

(11)

11

heterogeneity. For married households, they only find small effects (2 to 3 percentage points difference) for households in the lowest tier of self-reported health.

In summary, based on previous literature one may expect people in poor health to hold a smaller proportion of their total wealth in risky assets relative to people in good health, due to (1) a shortened time horizon, (2) a lower flexibility of labour supply, (3) a lower value in human capital and (4) a higher marginal utility of consumption. Therefore, I hypothesize:

H0A: Health does not affect the proportion of risky assets held by individuals relative to their

total wealth

H1A: Health status positively affects the proportion of risky assets held by individuals relative

to their total wealth

Furthermore, due to the forward-looking nature of financial decision making, one may expect an individual’s expectation of their future health to factor into their portfolio allocation as well. Therefore, I also pose the following hypotheses:

H0B: The expectation of future health does not affect the proportion of risky assets held by

individuals relative to their total wealth

H1B: The expectation of future health positively affects the proportion of risky assets held by

individuals relative to their total wealth

3. Dutch health profile

In order to better understand the context of this study, the overall health and health decisions of the Dutch population should be examined more closely. Gaining a deeper understanding of these contextual factors creates a stronger foundation on which the hypotheses as well as the explanation for any findings can be based, contributing to a more meaningful examination of health and risk taking all together.

The resulting profile will consist of three parts. First, the overall health of the Dutch people will be discussed. In the second part, the main challenges to the Dutch population health will be discussed, along with the impact of behaviours that pose a health risk. After this, the implications of the profile in relation to the thesis will be discussed briefly. However, before discussing these topics, the exact definition of health should be determined.

(12)

12

‘a state of complete physical, mental and social wellbeing, not merely the absence of disease or infirmity’ (WHO, 1948)

This definition is commonly referred to as the positive view of health (Taylor and Hawley 2010). This positive view takes into account not just physical health, but also mental and social wellbeing. This definition, as opposed to the negative view of health, which focusses only on the absence of disease or infirmity (Taylor and Hawley, 2010), fits this thesis best, as health in the data used is measured as subjective health. Given that mental and social wellbeing can affect one’s perception of their own health (Taylor and Hawley, 2010), it stands to reason that this positive view of health is most appropriate.

Dutch population health

The population health of the Netherlands is widely regarded as good. While a small decrease in subjective health has been reported amongst the Dutch population (as illustrated in Figure 1), nearly 80% of the mature population experiences their own health as either ‘good’ or ‘very good’ (CBS, 2020).

Figure 1: The percentage of people rating their own health as either "Good" or "Very good" in the Netherlands between 2000 and 2018

Internationally, this can be regarded as a relatively high score; Figure 2 in the appendix shows a comparison with other European Union member states on the same subjective health measure1 (Eurostat, 2019). This graph asserts that the Netherlands scores relatively well on subjective health, leaving only Ireland, Switzerland, Cyprus, Norway, Italy and Sweden with superior 1_{Note that the sample differs slightly from the earlier discussed national figure, as the European comparison}

includes individuals aged 16 and over, whereas the Dutch sample included only individuals aged 18 and over. This accounts for a slight difference in the Dutch results between the European survey and the national survey.

76 77 78 79 80 81 82 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 % o f p eo p le r ep o rtin g t h eir o w n h ealt h a s 'goo d ' o r 'v ery goo d ' Year

(13)

13

scores in 2017. However, given the influence of cultural factors, objective health measures may be better suited for cross-country comparisons, as highlighted by Meijer, Kapteyn and Andreyeva (2011):

“The distribution of the data on self-reported health is particularly illustrative of the large cross-country differences embedded in self-reports. For example, the percentage of men who rate their health status as poor or fair is more than three times as large in Germany as in Sweden, whereas approximately the same proportion of men in both countries reports having some chronic health condition. Another example is the male population of Denmark whose life expectancy is on average one year less than of French men, but who are 20% less likely than the French to rate their health as poor/fair.”

Therefore, one cannot paint a complete picture of the population’s relative health by only using subjective measures. When taking into account objective measures, the Netherlands again ranks relatively high. Life expectancy at birth, one of the more common objective measures of population health, is measured and ranked by several different organizations, producing slightly different estimates. The table below summarizes some of these rankings.

Table 1: Life expectancy at birth in the Netherlands, measured in various years by various institutions

Organization Year of measurement

Estimate for Dutch life expectancy at birth (male and female combined) Dutch ranking Total number of countries ranked WHO (2016) 2015 81,9 14 183 UNDP (2019) 2018 82,1 17 189 OECD (2020) 2016 81,8 17 44 CIA (2020) 2017 81,4 25 224

WHO = World Health Organization, UNDP = United Nations Development Program, OECD = Organisation for Economic Co-operation and Development, CIA = Central Intelligence Agency (US)

(14)

14

Main challenges and risk behaviours

According to the National Institute of Public Health and the Environment (RIVM, 2018), the main challenges to the Dutch population health, as measured in loss of DALY’s2_{, are cancers}

and cardiovascular diseases. When examining individual ailments, coronary heart disease, stroke, diabetes mellitus and COPD cause the most lost DALY’s amongst the Dutch population. The behaviours that caused the highest disease burden, once again measured in DALY’s, were smoking, unhealthy diets and a lack of movement, accounting for 9.4%, 8.1% and 2.3% of the total disease burden respectively (RIVM, 2018). This is a consistent picture, given the existing links between these behaviours and the aforementioned diseases (U.S. Department of Health and Human Services, 2014;2010;2004; WHO, 2020).

Implications for this thesis

Given the information gathered in the previous section, the overall health of the Dutch population, in subjective as well as objective measures, can be regarded as good. Given the theory built in the earlier section of this paper, this would imply relatively long time horizons, relatively high values in human capital, a relatively flexible labour supply and a relatively low marginal utility of consumption. Following the hypotheses put forward, one thus expects individuals in the sample to hold a relatively large fraction of their wealth in risky assets.

4. Data

For this study, the data collected from the LISS panel (collected by CentERdata through the MESS project) will be used. This data covers a panel of subjects on various topics (including health and personal finance), running from November of 2007 until November of 2019, over 11 so-called “waves”. By analysing the differences in regards to health variables and personal finance variables over time, the study will attempt to examine a possible link between the two. This section will describe the data used in the study in more detail. Furthermore, the section will start with some general information on the LISS panel and CentERdata.

2_{Disability Adjusted Life Years. This is a common measure for losses and gains in population health. It}

(15)

15

4.1 The LISS panel

The LISS panel consists of 4500 Dutch households, comprising 7000 individuals. This panel dataset is collected by CentERdata, a non-profit research institute which settles on the campus of Tilburg University. Subsidized by NWO (Centerdata.nl), CentERdata started the Advanced Multi-Disciplinary Facility for Measurement and Experimentation in the Social Sciences (MESS) project, which encompasses the LISS panel as well as a smaller immigrant panel, consisting of 1600 households, comprising 2400 individuals. In this study however, only the main LISS panel data is used. In order to prevent selection bias, the following two measures were taken in the data collection by CentERdata:

1. The panel is based on a true probability sample of households drawn from the population register by Statistics Netherlands.

2. Households that could not otherwise participate are provided with a computer and Internet connection.

These measures should ensure that the data obtained accurately reflects the Dutch population. In this thesis, different studies of the LISS panel dataset are used. Firstly, the Health dataset is used for any and all variables concerning personal health, subjectively measured as well as objectively measured. This dataset consists of 11 waves, ranging from 2007 until 2018. Second, all data regarding income, investment and wealth have been obtained from the dataset

Economic situation: Assets. This dataset consists of 6 waves, also ranging from 2007 until 2018.

(16)

16

survey subject, as the Health dataset also suffered from this inconsistency. To deal with these differences in timing, the sample was constructed as follows.

First, data was obtained from the Economic Situation: Assets survey for each wave. It is important here to note that the survey asks participants about their situation on the 31st of December of the previous year, effectively eliminating the timing problem in this survey. This means that, for example, the 2018 Economic Situation: Assets dataset (Wave 6) contains data on asset holdings for each participant on December 31st of 2017. Second, given the proposed relationship in this thesis, where asset holdings are affected by the investor’s personal health, it makes little sense to use health data collected after the date for which asset holdings are surveyed. Therefore, data from the Health survey was added from the most recent wave prior to the data covered in the Economic Situation: Assets data on asset holdings. This means that, for example, the 2018 Economic Situation: Assets data (Wave 6) was combined with the Health data from 2017 (Wave 10), linking data on the investors’ health collected between 06-11-2017 and 26-12-2017 to asset holdings surveyed for 31-12-2017. Lastly, background data was added to match the Economic Situation: Assets data (e.g. Background data for December 2017 was matched with Economic Situation: Assets data collected in 2018, which describes the situation on December 31st 2017). This results in a combined dataset which covers bi-yearly data on the three aforementioned surveys. Table 3 below summarizes the datasets which together form the sample used in this paper.

(17)

17

Table 3: Summary of data waves used for the Health, Investment and Background datasets

Investment wave

Investment data period of collection

Health wave

Health data period of collection Background period of collection 6 2018 (describes situation per December 2017) 10 November 2017 – December 2017 December 2017 5 2016 (describes situation per December 2015) 8 July 2015 – August 2015 December 2015 4 2014 (describes situation per December 2013) 7 November 2013 – December 2013 December 2013 3 2012 (describes situation per December 2011) 5 November 2011 – December 2011 December 2011

Moreover, LISS panel data on financial literacy as well as base risk-aversion has been used for control variables. These datasets are derived from one-off experiments, in which panel members participated. Both financial literacy and base risk-aversion are assumed to be constant over time and are therefore held equal over the period of study.

4.2 Personal health

(18)

18

indicator for the respondent’s expectation of their future health. This CRH-score is obtained from the survey question “Can you indicate whether your health is poorer or better, compared to last year?”. The response was again rated on a 5-point scale, ranging from “considerably poorer” [1] to “considerably better” [5]. Such patient reported measures are not without flaws; Schmidt et al. (1996) pose that self-reported health measures can be biased by the respondent’s mood at the time of surveying, while Caroli and Weber-Baghdiguian (2016) find that self-reported health measures are subjected to gender-based bias, leading women to report poorer health than men due to the influence of social norms. The choice for a subjective measure is, however, not without purpose; personal investors make their choices based on their own beliefs and perceptions. Therefore, it stands to reason that their own perception of their health would drive their investment behaviour in this model, rather than objective (and possibly more accurate) measures.

Unfortunately, the data available did not allow for the construction of conventional patient-reported health indices, such as the EQ-5D or the SF-36. While subjects were surveyed on part of these questionnaires, the majority of data needed to construct either one of these indices was not present in the LISS dataset, rendering the construction of these indices impossible for this sample. Nevertheless, SRH-scores have been found to be a valuable measure, as individuals themselves are thought to be the best judges of their own health. This notion is supported by Idler and Benyamini (1997), Kaplan and Gamacho (1983), Larsson et al. (2002) and Heistaro et al. (2001), who all find that self-reported health is an accurate predictor for mortality.

0,00 10,00 20,00 30,00 40,00 50,00 60,00 70,00 2012 2014 2016 2018

Figure 3: SRH-score distribution per year (%)

(19)

19

Figure 3 shows the distribution of SRH-scores in the sample over time, as a percentage of all respondents for the given year. The scores show relatively little change over time, with the majority of respondents scoring a 3 out of 5 in each period. Table 4 further summarizes this and reveals a slight shift towards higher scores over time.

Table 4: distribution of scores (%), mean scores and standard deviations of SRH-scores per year

1 2 3 4 5 Mean SRH-score Std. Deviation 2012 1,56 15,20 62,21 16,95 4,08 3,07 0,74 2014 1,41 16,77 59,66 17,85 4,32 3,07 0,76 2016 1,42 17,06 59,31 17,94 4,30 3,07 0,76 2018 1,42 16,57 56,08 21,29 4,63 3,11 0,78

4.3 Wealth and asset holdings

The LISS archive contains data on multiple classes of wealth and assets. While there is no catch-all variable for wealth available, a close approximation can be made by adding up multiple significant classes of wealth and subtracting significant debts. More specifically, a figure of total wealth was obtained for each respondent by adding up the value of their total wealth in so-called safe assets, risky assets and real estate and subtracting mortgage debts.

(20)

20

because of this correction, the model is appropriate for examining risk attitudes through risky asset holdings.

Table 5: Summary statistics of the wealth variables in the sample

Variable Number of observations Mean Standard deviation Minimum Maximum Wealth 10,388 40975.19 172846.5 -4647500 8389443

Risky assets divided by wealth 8,907 .0679 .2056548 -1 5 Risky assets 14,319 6.069.31 47543.95 -200 1800000 Safe assets 10,513 26171.06 129689.1 -570000 8389443 Real estate 14,548 9.739.87 77046.84 0 3250000 Mortgage debts 14,636 3.498.17 55056.2 0 5000000

4.4 Background variables

(21)

21

Moreover, an individual investor’s education also affects its financial decision making under risk (Grable, 2000; Bellante and Green, 2004; Guiso and Jappelli, 2005), for which most papers find a positive relationship; that is, individuals who have attained a higher level of education tend to make riskier portfolio choices. In the dataset used, education is included as a categorical variable, scored on a scale of 1 to 6, indicating the highest level of education completed by the respondent. A score of one represents persons having only completed primary school education, whereas a score of 6 represents persons who have completed university education. The categories and exact wording of this survey item can be found in the appendix to this thesis. Respondents who have not finished any education whatsoever have been omitted, due to disparities in measurements for this answer between waves. Furthermore, Berkowitz and Qiu (2006) find that married individuals are more likely to own stock and that married individuals are more willing to accept risks in their portfolios. Barber and Odean (2001) also find that marital status affects portfolio composition. More specifically, they find that being married decreases the differences in risk between genders, theorizing that spouses may influence each other’s investment decisions. In their sample, single men underperform single women by significantly more than married men underperform married women. The role of marital status is further explored by Love (2010), who also finds that family shocks, such as divorces, affect the individuals’ portfolio allocation, with men moving towards riskier investments, while women tend to move to less-risky allocations. Therefore, the variables age, gender, marital status and education from the Background variables dataset have been included in the model put forth in this thesis.

4.5 Financial Literacy

(22)

22

while questions 3 and 4 (the sophisticated questions) test the subjects’ knowledge on diversification and bond prices. To obtain the financial literacy index, they create dummy variables for each question, of which the binary value depends on whether the respondent answered the question correctly. Van Rooij et al. (2011) argue the importance of distinguishing cases in which the wrong answer is given and cases in which the respondents answered “I don’t know”, as these indicate different degrees of financial knowledge. Therefore, an extra set of dummy variables is used in this study; these indicate whether the question was answered with “I don’t know” or not. This results in two dummy variables for each question. The Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy is used to determine whether principal component analysis is appropriate given this data. A KMO score of 0.6837 is obtained, exceeding the threshold value of 0.5, indicating that principal component analysis, while not ideal, is appropriate for this data (Kaiser, 1974). Then, factor loadings are obtained using principal component analysis on the eight dummies. The factor loadings are then used to determine financial literacy scores, employing the methods of Bartlett (1937). Following Kaiser (1960), only components with eigenvalues larger than 1 are retained, resulting in two components, loading on basic and sophisticated financial literacy respectively. Lastly, for ease of interpretation, the scores are standardized, resulting in indices taking values between 0 and 1. The factor loadings can be found in Table 7 in the appendix, along with the screeplot of eigenvalues.

(23)

23

Table 8: summary of responses to financial literacy survey (%)

None One Two Three All Mean

Correct 5,89 13,36 36,84 31,08 12,84 2,32

Incorrect 48,37 36,93 11,65 2,10 0,95 0,70

Do not know 42,09 27,54 23,09 4,85 2,44 0,98

Table 9: summary of the basic and sophisticated financial literacy indices

Mean Std. Deviation Minimum value Maximum value

Basic index 0,531681 0,309598 0 1

Sophisticated index 0,102946 0,206903 0 1

4.6 Data selection

All of the panel members who have not participated in either one of the surveys have been eliminated from the final dataset. Moreover, only members who have participated in at least two waves of both surveys between 2012 and 2018 have been included in the sample, leaving 4735 individuals in the final dataset. Furthermore, the subjective health data has been converted to numerical data, as described in section 5.2. Participants who provided non-sensical or contradictory answers (e.g. investments totalling €9999999 but within the category €2500 - €5000) have also been eliminated from the dataset. For these individuals, one cannot be certain whether the question was misunderstood or if they were not comfortable providing a precise estimation of the value of their asset holdings or if there was another reason why the non-sensical and/or contradictory answers were given. Moreover, participants who either indicated

0 10 20 30 40 50 60 70 80 90 100

Question 1 Question 2 Question 3 Question 4

% o f re sp o n d en ts

Figure 5: distribution of answers on the financial literacy survey

(24)

24

that they are a director and majority shareholder of a private limited company or who failed to answer this question were also eliminated from the sample, since these individuals do not represent the average individual investor. Due to their position in these firms, the asset holdings of these panel members are expected to be linked to their employment status the firm, holding relatively large shares in these firms for the duration of their employment.

5. Methodology

5.1 The base model

After combining the variables, the following model is obtained:

𝑦_𝑖𝑡 = ∝ + 𝛽₁𝑥₁+ 𝛽₂𝑥₂+ 𝛽₃𝑥₃+ 𝛽₄𝑥₄+ 𝛽₅𝑥₅+ 𝛽₆𝑥₆+ 𝛽₇𝑥₇+ 𝜀_𝑖𝑡 (1)

𝑦𝑖𝑡 = Wealth in risky assets

divided by total wealth

𝑥5 = Gender

𝑥₁ = SRH-score 𝑥₆ = Marital status

𝑥2 = CRH-score 𝑥7 = Financial literacy score (basic)

𝑥3 = Education 𝑥8 = Financial literacy score (sophisticated)

(25)

25

Note here that the variables Gender and Marital status are dummy variables, where 𝑥₅ = 0 or 1 for male and female subjects respectively4_{, and 𝑥}

6 = 0 or 1 for unmarried and married subjects

respectively.

5.2 Panel data techniques

With the final dataset, the hypotheses can be tested using statistical analysis. Because this thesis uses panel data, regular regression analysis is not appropriate. Instead, panel data techniques are required for this study. This section will discuss three such techniques, which are often used in similar studies: (1) Pooled OLS, (2) Random effects and (3) Fixed effects.

1.2.1 Pooled OLS

Estimating a pooled ordinary least-squares regression (Pooled OLS) is arguably the simplest way of dealing with panel data. With this technique, a regression is estimated on all observations pooled together, using the following regression model:

𝑦_𝑖𝑡 = 𝛼 + 𝛽_𝑖𝑥_𝑖𝑡 + 𝑢_𝑖𝑡 (2) Where 𝑦_𝑖𝑡 is the dependent variable, 𝛼 is the constant, 𝛽𝑥_𝑖𝑡 represents the independent variable and its respective coefficient and 𝑢𝑖𝑡 represents the error term. While this technique is attractive

due to its simplicity, it assumes no heterogeneity in the data, which is unlikely when dealing with many different individuals over time.

1.2.2 Fixed effects models

Simple fixed effects models allow the intercept in the regressions to differ cross-sectionally, but not over time. In other words, the use of a fixed effects model allows for differentiation between different individuals in the dataset. A fixed effects model is structured as follows:

𝑦_𝑖𝑡 = 𝛼 + 𝛽𝑥_𝑖𝑡 + µ_𝑖 + 𝜈_𝑖𝑡 (3) Where once again, 𝑦_𝑖𝑡 , 𝛼 and 𝛽𝑥_𝑖𝑡 represent the dependent variable, constant and the combination of the independent variable and its coefficient respectively. µ_𝑖 is the entity fixed effect in this model, while 𝜈𝑖𝑡 is the error term. The major drawback of such fixed effects

4_{The survey only allowed for dichotomous answers for gender. Furthermore, no changes in gender in individuals}

(26)

26

models is that variables which do not vary over time will cancel out, rendering the model incapable of displaying the influence of independent variables which affect 𝑦_𝑖𝑡 but which do not vary over time. For this study, this means that a fixed effects model will not include the effects of financial literacy, since this variable is taken from a single-wave study and is assumed to remain constant over time.

1.2.3 Random effects models

Random effects models assume that the variation across entities (individuals in this study) is assumed to be random and uncorrelated with the independent variable. In other words, the entity effect, as described in the fixed effects model, is assumed to be random, producing the following model:

𝑦_𝑖𝑡 = 𝛼 + 𝛽𝑥_𝑖𝑡 + 𝑢_𝑖𝑡 4.1

Where

𝑢𝑖𝑡 = µ𝑖 + 𝜈𝑖𝑡 4.2

Where, like in the fixed effects model, 𝑦𝑖𝑡 , 𝛼 and 𝛽𝑥𝑖𝑡 represent the dependent variable,

constant and the combination of the independent variable and its coefficient respectively. The difference here lies in the error term, as µ_𝑖 is now seen as part of the error term, due to the assumption that µ_𝑖 is random. Therefore, the random effects model is most appropriate when static differences across individuals are expected to influence the dependent variable. This means that the effect of variables such as financial literacy, which is assumed to remain constant in this study, can be examined with a random effects model, as opposed to fixed effect models.

1.2.3 Tobit estimation

(27)

27

choice is made because, like the case was in Rosen and Wu (2004), the data clusters only at zero and not at any other value.

Figure 6: Histogram of the value of risky assets divided by the total wealth of individuals in the sample

5.3 Model selection

Generally, the fixed and random effects models are best fitted to deal with panel datasets. The choice between these models is usually made using the Hausman specification test. The Hausmann test tests whether the error term 𝜇𝑖 in the model is correlated with the explanatory

(28)

28

6. Results

6.1 Estimation of the full model

Table 13 presents the estimation results of the full model. It displays significant coefficients at the one percent level for both SRH and CRH scores in the pooled OLS (0.01 and -0.013 respectively) and random effects (0.011 and -0.016 respectively) models. In the fixed effects model, the coefficients for SRH (0.013) and CRH (-0.019) scores are significant at the five percent and one percent level respectively. There results indicate that both the current health status as well as the health status compared to the previous year significantly impacts the share of risky assets as a proportion of total wealth in the individuals in the sample. Interestingly, the SRH score yields positive signs, whereas the CRH score yields negative signs in all models, revealing contradicting effects. Overall, greater health seems to lead to a greater proportion of wealth invested in risky assets, but an improvement in health compared to the previous year leads to a decrease in the proportion of wealth invested in risky assets. More specifically, a one-point increase in SRH-score seems to cause an increase of 1% to 1.3% in the fraction of risky assets held, while a one-point increase in the CRH-score lowers this fraction by about 1.3% to 1.9%, dependent on the model used. Given that the fixed effects model has been identified by the Hausman test as the most appropriate, the fixed effects results are expected to be the most accurate. The fixed effects model does not yield any further significant results for the other variables. Note here that, due to the nature of the fixed effects estimation technique, no coefficients are obtained for any variables that do not show any changes over time. In this model, this results in no coefficients for the variables Gender and Financial literacy (both basic and sophisticated). The pooled OLS and random effects models both yield significant results for the variables Education (0.013 for both), Age (0.001 for both), Sophisticated financial

literacy (-0.096 and -0.093 respectively) and Net income (<0.001 for both). Most of these

(29)

29

Table 13: Estimation results of the full model, using the Pooled OLS, Fixed Effects and Random Effects estimation techniques

6.2 Estimations with SRH and CRH in isolation

(30)

30

Tables 14 and 15 present the estimation results for the models where CRH-score and SRH-score are omitted respectively.

Table 14: Estimation results of the SRH-isolated model, using the Pooled OLS, Fixed Effects and Random Effects estimation techniques

(31)

31

Table 15: Estimation results of the CRH-isolated model, using the Pooled OLS, Fixed Effects and Random Effects estimation techniques

(32)

32

the results for the other variables do not show any meaningful differences from those in the full model.

6.3 Estimation of the tobit models

Lastly, Table 16shows the estimation results for the tobit estimations of the full model and both SRH-score and CRH-score in isolation. The results for the SRH-score and CRH-score variables are similar to those of the other estimation, the most notable difference being the lack of statistical significance in the isolated SRH-model. Furthermore, the coefficient sizes are larger in these tobit models, suggesting stronger effects. The estimation results of the full model suggest a 6,1% incease in the fraction of wealth invested in risky assets per one-point increase in SRH-score and an 8,5% decrease in the fraction of wealth invested in risky assets per one-point increase in CRH-score, all significant at the 1% level. In isolation, the SRH-score does not yield a significant coefficient, while the CRH-score coefficient drops, now reflecting a 6,8% decrease per one-point increase in CRH-score, again significant at the 1% level. Furthermore, the results for the control variables also change in the tobit model. As shown in Table 16, the coefficients for Education, Age and Sophisticated financial literacy are still significant and have significantly larger coefficients in terms of absolute value. Gender and Marital status remain non-significant. Basic financial literacy is now significant at the one-percent level, with again a greater coefficient than in previous estimations. Interestingly, the sign has changed for Basic

financial literacy, as it is now significant and negative, whereas in the other estimations it is

(33)

33

(34)

34

7. Discussion

The main aim of this thesis has been to closely examine the relationship between an investor’s health and their financial decision making. The results presented in the previous section strongly hint at the existence and nature of this relationship. More specifically, the results indicate a weakly significant relationship between investors’ current health status and their portfolio choices and a strongly significant relationship between the investors’ health as compared to the previous year and their portfolio choices. This separation indicates that not only the current health status, as described in the first hypothesis of this thesis, but also the direction in which their health is moving, factors in to financial decision making, as described in the second hypothesis. First, an investors health, as measured by their SRH-score is positively related to their exposure to risky assets, rejecting H0A in favour of H1A. This finding is in line

with previous literature, as it supports the life cycle models and the general notion that a shortened time horizon would lead investors to move towards less risky portfolio allocations. The results indicate that in better health, and therefore a longer time-horizon, invest a higher proportion of their wealth in risky assets. While this finding is significant, the differences may be marginal: at most a 1.3% increase in the proportion of risky assets held per one-point increase in SRH-score in the regular models. However, in the tobit model this increase jumps to a 6.1% increase per one-point increase in SRH-score, indicating a much larger impact. Furthermore, the results show a significant negative relationship between the fraction of risky assets held and the investor’s health compared to the previous year. This reveals that, aside from the current status of one’s health, the general trend of the investor’s health factors into this relationship as well, rejecting hypothesis H0B. Interestingly, this relationship is, as opposed to the relationship

with current health status, negative, contradicting hypothesis H1B. Again, the effect is, while

(35)

35

enjoy a remarkable improvement in health, the base health status has to be low to begin with. This would mean that the CRH-score variable actually, at least partly, captures the effect of the score variable. However, given that a significant relationship is found when the SRH-score is controlled for in the model, this is unlikely to be the sole explanation. Increases in health are expected to come after adverse health shocks, not when health has been stable over time or is already at its highest. When this increase occurs, the individual may still incur medical costs due to this health shock, as theorized by Fan and Zhao (2009) and Berkowitz and Qiu (2006). Goldman and Maestas (2013) further confirm that households with a higher risk of incurring medical costs hold smaller fractions of risky assets. Perhaps health shocks may even cause individuals to be permanently more conservative in their financial decision making, because they anticipate the possibility of further health shocks in the future. These effects might explain the findings, as this would be consistent with both a positive relationship between SRH-score and risky investment as well as a negative relationship between CRH-SRH-score and risky investment. More research on the longer-term behavioural effects of health shock is needed to explore the exact nature of this relationship. Building on this argument, an adverse health shock may permanently affect an individual, effectively diminishing their human capital. Following Bodie et al. (1992), this would then drive these individuals to less risky investment decisions. If the human capital is permanently diminished, this effect would remain even if the individual’s health improves following the adverse health shock. Moreover, Nygren, Isen, Taylor and Dulin (1996) find higher loss-aversion for positive affect people. An upwards-trending health status might cause a stronger positive affect than a constant state of good health. This would mean that, even if good health is related to more financial risk taking, recent improvements in health would still cause a decrease in financial risk taking.

7.1 Implications for health policy

(36)

36

(37)

37

8. Limitations

This thesis is subject to several limitations. First, Baron-Epel and Kaplan (2001) find that age and education affect how accurate and consistent subjects answer questions meant to measure subjective health. The risk of these discrepancies in the measurements of variables is, unfortunately, inherent to subjective self-reported measures. Due to a limited availability of data, more standard measures, such as the EQ-5D or SF-36 surveys, could not be used in this thesis. Furthermore, these indices may be more complete, painting a more accurate picture of an individual’s health than solely a subjective score scaled from one to five. A follow-up study using such measures could eliminate these concerns. Second, the total wealth of individuals in the sample is not measured directly, but rather derived as the sum of the value of all major asset classes, subtracted by all significant debts. While this should yield a close approximation of total wealth, large portions of wealth in niche classes may be overlooked. Furthermore, no exact data is available for many individuals in the sample. These individuals provided only a range wherein the value of their assets in a given category lies. By taking the midpoint values for these individuals, while it may be a close approximation, the values obtained for their wealth in different asset classes are not completely accurate. Third, the indices measuring financial literacy are based on partial information; due to a limited availability of data, only four survey questions were used to construct these indices, rather than the full set of thirteen questions included in Lusardi and Mitchell (2009). Had the LISS panel dataset included the remaining questions, the resulting financial literacy indices would have been more accurate. Fourth, given that a sample of only Dutch households is used in this research, the study is inherently limited in its explanatory power for non-Dutch investors. Follow-up studies using international samples would be required to examine situations outside of the Dutch context.

(38)

38

9. Conclusion

(39)

39

References

Agnew, J., Balduzzi, P., & Sundén, A. (2003). Portfolio choice and trading in a large 401 (k) plan. American Economic Review. https://doi.org/10.1257/000282803321455223

Almenberg, J., & Dreber, A. (2015). Gender, stock market participation and financial literacy. Economics Letters. https://doi.org/10.1016/j.econlet.2015.10.009

Anderson, D. (1990). Investment and economic growth. World Development. https://doi.org/10.1016/0305-750X(90)90088-F

Barber, B. M., & Odean, T. (2001). Boys will be boys: Gender, overconfidence, and common stock investment. Quarterly Journal of Economics. https://doi.org/10.1162/003355301556400

Baron-Epel, O., & Kaplan, G. (2001). General subjective health status or age-related subjective health status: Does it make a difference? Social Science and Medicine. https://doi.org/10.1016/S0277-9536(00)00426-3

BARTLETT, M. S. (1937). THE STATISTICAL CONCEPTION OF MENTAL FACTORS. British Journal of Psychology. General Section. https://doi.org/10.1111/j.2044-8295.1937.tb00863.x

Bellante, D., & Green, C. A. (2004). Relative risk aversion among the elderly. Review of Financial Economics. https://doi.org/10.1016/j.rfe.2003.09.010

Berkowitz, M. K., & Qiu, J. (2006). A further look at household portfolio choice and health status. Journal of Banking and Finance. https://doi.org/10.1016/j.jbankfin.2005.05.006

Bernoulli, D. (1954). Exposition of a New Theory on the Measurement of Risk. Econometrica. https://doi.org/10.2307/1909829

Bhatti, Z. U., Salek, M. S., & Finlay, A. Y. (2012). Chronic diseases influence major life changing decisions: A new domain in quality of life research. Journal of the Royal Society of Medicine. https://doi.org/10.1258/jrsm.2011.110010

Blais, A.-R., & Weber, E. U. (2006). A Domain-Specific Risk-Taking (DOSPERT) scale for adult populations. Judgment and Decision Making.

Bodie, Z., Merton, R. C., & Samuelson, W. F. (1992). Labor supply flexibility and portfolio choice in a life cycle model. Journal of Economic Dynamics and Control. https://doi.org/10.1016/0165-1889(92)90044-F

(40)

40

Bran, A., & Vaidis, D. C. (2019). Assessing risk-taking: what to measure and how to measure it. Journal of Risk Research. https://doi.org/10.1080/13669877.2019.1591489

Cardak, B. A., & Wilkins, R. (2009). The determinants of household risky asset holdings: Australian evidence on background risk and other factors. Journal of Banking and Finance. https://doi.org/10.1016/j.jbankfin.2008.09.021 Caroli, E., & Weber-Baghdiguian, L. (2016). Self-reported health and gender: The role of social norms. Social Science and Medicine. https://doi.org/10.1016/j.socscimed.2016.02.023

CBS. (2020). Gezondheid, leefstijl, zorggebruik en -aanbod, doodsoorzaken; vanaf 1900. Retrieved April 20, 2020, from https://opendata.cbs.nl/statline/#/CBS/nl/dataset/37852/table?ts=1582718436427

CenterData. (2020). MESS project. Retrieved May 15, 2020, from https://www.centerdata.nl/nl/nieuws/mess-project-gaat-nieuwe-fase-in

CIA. (2020). The world factbook. Retrieved April 20, 2020, from https://www.cia.gov/library/publications/the-world-factbook/rankorder/2102rank.html

Cohn, R. A., Lewellen, W. G., Lease, R. C., & Schlarbaum, G. G. (1975). INDIVIDUAL INVESTOR RISK AVERSION AND INVESTMENT PORTFOLIO COMPOSITION. The Journal of Finance. https://doi.org/10.1111/j.1540-6261.1975.tb01834.x

Crainich, D., Eeckhoudt, L., & Le Courtois, O. (2017). Health and portfolio choices: A diffidence approach. European Journal of Operational Research. https://doi.org/10.1016/j.ejor.2016.10.022

Edwards, K. D. (1996). Prospect theory: A literature review. International Review of Financial Analysis. https://doi.org/10.1016/S1057-5219(96)90004-6

Edwards, R. D. (2010). Optimal portfolio choice when utility depends on health. International Journal of Economic Theory. https://doi.org/10.1111/j.1742-7363.2010.00131.x

Eurostat. (2019). Health variables of EU-SILC. Retrieved April 20, 2020, from https://ec.europa.eu/eurostat/cache/metadata/en/hlth_silc_01_esms.htm

Finkelstein, A., Luttmer, E. F. P., & Notowidigdo, M. J. (2013). What good is wealth without health? The effect of health on the marginal utility of consumption. Journal of the European Economic Association. https://doi.org/10.1111/j.1542-4774.2012.01101.x

Gao, M., Meng, J., & Zhao, L. (2019). Income and social communication: The demographics of stock market participation. World Economy. https://doi.org/10.1111/twec.12777

(41)

41

Grable, J. E. (2000). Financial Risk Tolerance and Additional Factors that Affect Risk Taking in Everyday Money Matters. Journal of Business and Psychology. https://doi.org/10.1023/A:1022994314982

Guiso, L., & Jappelli, T. (2005). Awareness and stock market participation. Review of Finance. https://doi.org/10.1007/s10679-005-5000-8

Heistaro, S., Jousilahti, P., Lahelma, E., Vartiainen, E., & Puska, P. (2001). Self rated health and mortality: A long term prospective study in eastern Finland. Journal of Epidemiology and Community Health. https://doi.org/10.1136/jech.55.4.227

Idler, E. L., & Benyamini, Y. (1997). Self-Rated Health and Mortality: A Review of Twenty-Seven Community Studies. Journal of Health and Social Behavior. https://doi.org/10.2307/2955359

Jianakoplos, N. A., & Bernasek, A. (1998). Are women more risk averse? Economic Inquiry. https://doi.org/10.1111/j.1465-7295.1998.tb01740.x

Jianakoplos, N. A., & Bernasek, A. (2006). Financial Risk Taking by Age and Birth Cohort. Southern Economic Journal. https://doi.org/10.2307/20111864

Jones, R., & Bradley, E. (2007). Health issues for adolescents. Paediatrics and Child Health. https://doi.org/10.1016/j.paed.2007.08.002

Kahneman, D., & Tversky, A. (1979). Prospect Theory: An Analysis of Decision under Risk. Econometrica. https://doi.org/10.2307/1914185

Kaiser, H. F. (1974). An index of factorial simplicity. Psychometrika. https://doi.org/10.1007/BF02291575 Kaiser, H. F. (1960). The Application of Electronic Computers to Factor Analysis. Educational and Psychological Measurement. https://doi.org/10.1177/001316446002000116

Kaplan, G. A., & Camacho, T. (1983). Perceived health and mortality: A nine-year follow-up of the human population laboratory cohort. American Journal of Epidemiology. https://doi.org/10.1093/oxfordjournals.aje.a113541

Larsson, D., Hemmingsson, T., Lundberg, I., & Allebeck, P. (2002). Self-rated health and mortality among young men: What is the relation and how may it be explained? Scandinavian Journal of Public Health. https://doi.org/10.1080/14034940210133997

Love, D. A. (2010). The effects of marital status and children on savings and portfolio choice. Review of Financial Studies. https://doi.org/10.1093/rfs/hhp020

(42)

42

Meijer, E., Kapteyn, A., & Andreyeva, T. (2011). Internationally comparable health indices. Health Economics. https://doi.org/10.1002/hec.1620

Menezes, C. F. and D. L. H. (1970). On the Theory of Risk Aversion. In International Economic Review (Vol. 11, pp. 481–487).

Nygren, T. E., Isen, A. M., Taylor, P. J., & Dulin, J. (1996). The influence of positive affect on the decision rule in risk situations: Focus on outcome (and especially avoidance of loss) rather than probability. Organizational Behavior and Human Decision Processes. https://doi.org/10.1006/obhd.1996.0038

OECD. (2020). Health status: life expectancy at birth. Retrieved from https://data.oecd.org/healthstat/life-expectancy-at-birth.htm

RIVM. (2018). Volksgezondheid Toekomst Verkenning. Retrieved from https://www.vtv2018.nl/

RIVM. (2020). Ziektelast in DALY’s. Retrieved from https://www.volksgezondheidenzorg.info/ziektelast-nederland

Schmidt, N. B., Telch, M. J., & Joiner, T. E. (1996). Factors influencing health perceptions in patients with panic disorder. Comprehensive Psychiatry. https://doi.org/10.1016/S0010-440X(96)90004-8

Smith, C. A. B., Neumann, J. Von, & Morgenstern, O. (2007). Theory of Games and Economic Behaviour. The Mathematical Gazette. https://doi.org/10.2307/3610940

Sundén, A. E., & Surette, B. J. (1998). Gender Differences in the Allocation of Assets in Retirement Savings Plans. American Economic Review. https://doi.org/10.2307/116920

Taylor, G., & Hawley, H. (2010). Key debates in health care. Maidenhead: Open University Press ; Retrieved from http://site.ebrary.com/id/10404008

Thompson, W. N., Gazel, R., & Rickman, D. (1997). Social and Legal Costs of Compulsive Gambling. Gaming Law Review. https://doi.org/10.1089/glr.1997.1.81

Topolski, T. D., Patrick, D. L., Edwards, T. C., Huebner, C. E., Connell, F. A., & Mount, K. K. (2001). Quality of life and health-risk behaviors among adolescents. Journal of Adolescent Health. https://doi.org/10.1016/S1054-139X(01)00305-6

U.S. Department of Health and Human Services. (2014). Tobacco Reports And Publications. Retrieved from https://www.hhs.gov/surgeongeneral/reports-and-publications/tobacco/index.html

(43)

43

U.S. Department of Health and Human Services. (2004). The health consequences of smoking: a report of the

Surgeon General. Retrieved from

https://www.ncbi.nlm.nih.gov/books/NBK44695/pdf/Bookshelf_NBK44695.pdf

UNDP. (2019). Human Development Index Ranking. Retrieved April 20, 2020, from http://hdr.undp.org/en/content/2019-human-development-index-ranking

van Rooij, M., Lusardi, A., & Alessie, R. (2011). Financial literacy and stock market participation. Journal of Financial Economics. https://doi.org/10.1016/j.jfineco.2011.03.006

Viceira, L. M. (2001). Optimal portfolio choice for long-horizon investors with nontradable labor income. Journal of Finance. https://doi.org/10.1111/0022-1082.00333

Wachter, J. A., & Yogo, M. (2010). Why do household portfolio shares rise in wealth? Review of Financial Studies. https://doi.org/10.1093/rfs/hhq092

Watson, J., & McNaughton, M. (2007). Gender differences in risk aversion and expected retirement benefits. Financial Analysts Journal. https://doi.org/10.2469/faj.v63.n4.4749

WHO. (1948). Preamble to the Constitution of the World Health Organization. International Health Conference, New York, 19-22 June, 1946; Signed on 22 July 1946 by the Representatives of 61 States. https://doi.org/10.1016/B978-012466044-1/50296-9

WHO. (2016). Global Health Observatory (GHO) data, Annex B: t. Retrieved from https://www.who.int/gho/publications/world_health_statistics/2016/Annex_B/en/

(44)

44

Appendix

A. Survey questions

Table 2: An overview of the survey questions used in the variables pertaining to either health or wealth.

Question Obtained from

dataset

Corresponding variable

How would you describe your health, generally speaking?

Health SRH-score

Can you indicate whether your health is poorer or better, compared to last year?

Health CRH-score

What was the total balance of your current accounts, savings accounts, term deposit accounts, savings bonds or savings certificates and bank savings schemes on 31 December [YEAR]? In case of a negative balance, please add a minus sign before the amount.

Economic situation: Assets

Wealth in risky assets divided by total wealth

What was the total value of your investments (growth funds, share funds, bonds, debentures, stocks, options, warrants) on 31 December [YEAR]? (Please note: written options represent a debt, and in case of a negative balance, please add a minus sign before the amount.)

Wealth in risky assets divided by total wealth

What was the total value of the real estate (not used as personal home / second home / holiday home) on 31 December [YEAR]?

(45)

45 How high was the remaining mortgage debt on 31 December [YEAR]? Economic situation: Assets Wealth in risky assets divided by total wealth Were you a majority-shareholder director (Dutch:

DGA) of a private limited company (Dutch: BV) in [YEAR]? Economic situation: Assets Wealth in risky assets divided by total wealth

Financial Literacy questions:

1. Suppose you have 100 euros on a savings account and the interest is 2 percent per year. How much do you think you will have on the savings account after five years, assuming that you leave all your money on this savings account: more than 102 euros, exactly 102 euros, less than 102 euros?

(a) more than 102 euros (b) exactly 102 euros (c) less than 102 euros (d) I don’t know

(e) I would rather not say

2. Suppose that the interest on your savings account is 1 percent per year and that inflation amounts to 2 percent per year. After 1 year, would you be able to buy more, exactly the same, or less than you could today with the money on that account?

(a) more than today

(b) exactly the same as today (c) less than today

(d) I don’t know

(46)

46

3. A share in a company usually offers a more certain return than an investment fund that only invests in shares.

(a) True (b) Not true (c) I don’t know

(d) I would rather not say

4. If the interest rate goes up, what should happen to bond prices? (a) They should increase

(b) They should decrease (c) They should stay the same (d) None of the above

(e) I don’t know

(47)

47

B. Dutch health statistics

C. LISS panel data descriptive statistics

Table 6: Number of individuals per education level (highest completed) per year in the sample

2012 2014 2016 2018 Total Primary school 267 299 277 221 1064 VMBO (intermediate secondary education) 863 1003 984 800 3650 HAVO/VWO (higher secondary education) 343 418 455 341 1557 MBO (intermediate vocational education) 761 929 1013 867 3570 HBO (Higher vocational education 738 907 1000 887 3532 WO (University) 238 333 434 402 1407 Total 3210 3889 4163 3518 14780 0 10 20 30 40 50 60 70 80 90 Lithuania Portugal Poland Croatia Slovenia Bulgaria France Austria Luxembourg Spain United Kingdom The Netherlands Italy Cyprus Ireland

(48)

48

D. Financial literacy indices

Table 7: Factor loadings

Factor 1 Factor 2

Question Type Correct Don’t know Correct Don’t know

Interest compounding Basic -.3287 .3556 .4021 -.4009

Money illusion Basic -.3915 .4067 .2427 -.287

Diversification Advanced -.3719 .3771 -.3606 .3915

Bond pricing Advanced -.258 .3158 -.3416 .3693

These factor loadings are obtained using principal component analysis.

The healthy appetite for risk: examining the effect of health on portfolio allocation