Optimism and limited stock market participation: The effect of positive attitudes towards future events

(1)

Optimism and limited stock market participation:

The effect of positive attitudes towards future events

June 2018

M.B. Diender

Master thesis: MSc. Finance

Supervisor: prof. dr. R.E. Wessels

(2)

Optimism and limited stock market participation

2 Abstract

This thesis explores the effect of dispositional optimism, defined as having an above average positive attitude towards future events, on stock market participation. I assume that the probability of an individual investing in stocks increases with the level of dispositional optimism. This relationship is estimated using a logistic model, employing the data collected in the Survey of Health, Aging and Retirement in Europe (SHARE). I find that the proposed dispositional optimism indicator does not significantly influence stock market participation, after controlling for several well-established factors. Based on the results of my analysis, I conclude that dispositional optimism does not provide a convincing explanation for the low participation puzzle.

Keywords: Dispositional optimism, stock market participation, limited participation puzzle, logistic regression, SHARE.1

1_{This paper uses data from SHARE Wave 5 (DOI: 10.6103/SHARE.w5.610), see Börsch-Supan et al.}

(2013) for methodological details. (1)

(3)

3

1. Introduction

Many papers in the last three decades have studied the financial decision making of households. A prominent and common finding is that very few households participate in financial markets and particularly few participate in the stock market (e.g. Mankiw and Zeldes (1991); Haliassos and Bertaut (1995)). Furthermore, there are large differences in participation rates across countries (Guiso et al. (2003)). Even between countries that are culturally, economically and institutionally similar. As an example, using data from wave 5 of the Survey of Health, Ageing and Retirement in Europe (SHARE, henceforth) there are great differences in participation between Swedish and German households, where in Sweden about 40% of elderly households invest directly in stocks, whereas in Germany only 12% of elderly households hold stocks. Strikingly, even among the very wealthy, participation rates are much lower than what we would expect given the opportunity costs of not investing in the stock market.

Financial markets facilitate the transfer of consumption, income and assets over time, and can serve as a protection against inflation. The low participation rate of households in stock markets is therefore surprising, given the need to smooth consumption of income and assets over time. Moreover, Campbell (2006) observed that low participation rates are evidence against current theories of household financial behaviour that imply, given that there is a positive risk premium, that all households should hold some stocks, regardless of risk aversion. The phenomenon is known as the limited participation puzzle or limited asset market participation puzzle.

Limited stock market participation has significant implications for individual households. For instance, the welfare loss from not participating in the stock market can be large (e.g. Cocco et al (2005)). Furthermore, Mankiw and Zeldes (1991) point out that the limited participation puzzle is also able to provide some explanation to the asset pricing anomaly known as the equity premium puzzle.

The equity premium puzzle was introduced in Mehra and Prescott (1985). They find that the observed equity premium, the return on the market portfolio of stocks minus the return on default free debt, is substantially larger than can be explained by economic models widely used in financial economics, given the observed risk (volatility) of stocks.2 In other words, the observed equity premium is so large that only an extremely risk averse investor would not want to invest in the market portfolio.

Mankiw and Zeldes (1991) examine whether differences in consumption of investors in stocks versus that of non-participants can serve as an explanation for the equity premium puzzle. They find that there are significant differences in the correlation between consumption

2_{Mehra and Prescot (1985) employ variations of the consumption-based asset pricing model of Lucas (1978) set}

(4)

4 and stock returns between stock investors and non-participants, they reason that these differences imply separate degrees of relative risk aversion for these two groups. Stock market participant’s implied relative risk aversion is lower, and therefore providing a piece for solving the equity premium puzzle.3 However, additional reasons why households participate in the stock market remain unexplored.

The work of Mankiw and Zeldes (1991) provides a fertile ground for exploring factors that could possibly explain the limited stock market participation of households. Meaningful factors that have been identified can be categorized by belonging to roughly three types. These types are participation costs; preferences and beliefs. What follows is a brief overview of the three explanations and related factors that have been identified in the literature, starting with participation costs.

First, households might decide not to participate in the stock market, given the costs of participation (Vissing-Jørgensen (2003)). The participation costs households face are those of a monetary or direct nature, such as the costs of trading, and those of a non-monetary nature, i.e. information costs. Understanding what the risks and expected returns of investments in the stock market are, requires considerable knowledge on how financial markets work (see van Rooij et al. (2011)), but at the same time learning costs almost nothing since by investing small amount one can become financially literate at very low cost. The direct costs of participating serve here as a lower bound for participation costs, that is, direct cost of starting a brokerage account and trading are still there. Households are assumed to weigh the costs of participation against the gains of participation in the stock market, i.e. earning the equity risk premium. Many of the factors found that significantly explain, or correlate with stock market participation, can be rationalized by the existence of participation cost (Guiso and Sodini (2013));

Firstly, financial literacy is found to be a substantial predictor of stock market participation (van Rooij et al. (2011)). Rationalizing this result with the existence of participation costs goes as follows: indirect participation costs for financially literate individuals are, ceteris paribus, substantially less than that those of non-literate individuals. These lower costs will lower the necessary return to participate in the stock market. Therefore, a positive correlation between financial literacy and participation is expected.

Furthermore, Christelis et al. (2010) and Cole and Shastry (2009)) find that cognitive ability and level of education are also relevant predictors of stock market participation. Individuals that have more ease in acquiring and analysing information suffer relatively lower indirect cost of participation. Another dominant factor is the relative risk aversion of a household (see Haliassos and Bertaut (1995)), where more risk averse investors require a higher return per unit

3_{The implied relative risk aversion indicator for participants is still implausibly high, however. So, limited}

(5)

5 of risk as a threshold to participate. Extremely risk averse investors might choose not to participate at all with the existence of participation costs. The disparity in country level average participation rates can also be rationalized under this theory, for instance large differences in taxation can explain part of the differences.

A prediction that follows from the participation costs theory, is that stock market participation will increase when the costs of participating become cheaper. Currently, with the existence of budget online brokers and a world of free financial information, participation rates should be expected to increase following this. Guiso and Sodini (2013) note that over the last two decades participation rates have been increasing. However, participation rates are still lower than would be expected. Moreover, the differences in country levels of participation cannot be rationalized by differences in taxations alone.

Another approach that has been taken to give reason to limited stock market participation is to study differences in preferences. Most notably are explanations from an irrational behaviour perspective such as the concept of loss aversion (Tversky and Kahneman (1991); Barberis et al. (2006)).

In loss aversion, possible future regret plays a role. Where regret is defined as the pain that is felt when one realizes that he would be better off is he had taken a different decision in the past. If the decision to participate in the stock market possibly causes future regret, individuals might be likely to not invest at all. As participating in the stock market entails a concrete action, that may cause regret when the return on investing turns out poorly even when participation costs do not exist. This phenomenon can be seen as an aversion to losing, in contrast to solely aversion to risk.

(6)

6 Thus, participation costs, differences in preferences and differences in beliefs all play a role in providing an explanation for the limited stock market participation puzzle. Nonetheless, there are more factors at play. For instance, as a prerequisite to participate, individuals must be aware of the stock market. Guiso and Jappelli (2005) show that a lack of awareness might explain why some individuals chose not to invest in stocks. Furthermore, an individual’s trust plays a role (Guiso et al. (2008)). And, institutional quality (Asgharian et al. (2015)), providing a partly an explanation of differences between country level participation. There are numerous other factors that are correlated with stock market participation, such as age and gender, religious upbringing, and social interaction (Campbell (2006); Barber and Odean (2001); Renneboog and Spaenjers (2012); Hong et al. (2004)).

However, the limited participation puzzle still exists, that is, studies yet have not been able to fully rationalize the financial behaviour of individual households within expected utility models. So, there is room left for research in other areas that are relatively unexplored in trying to explain limited stock market participation. One of these areas is the role that personality could play in rationalizing a household’s financial behaviour. If personality plays a substantial role, then proposed solutions to diminish limited stock market participation, such as investing in financial education (Guiso and Jappelli (2008)) are more likely to be proven inefficient. As an example, Brown and Taylor (2014) explore the relationship between 5 personality traits and financial behaviour and find that openness to experience increases the probability to hold stocks.

This thesis will focus on the relationship between dispositional optimism and stock market participation. Dispositional optimism is defined by Scheier and Carver (1985) as having generalized positive expectations about future events. They note that optimism is a stable personality characteristic that could have important implications for the way people make decisions. As Puri and Robinson (2007) put it; optimism tempts individuals to overestimate the chance that favourable events will happen and to underestimate the chance of unfavourable events. The question that will be addressed in this thesis is the following:

What do differences in dispositional optimism contribute towards explaining the puzzle of low participation of individuals in financial markets?

(7)

7 Puri and Robinson (2007) explore this relationship between dispositional optimism and individual economic decision making. To analyse the relationship, they propose a measure of dispositional optimism using the difference between subjective life expectancy, indicated by respondents in the (American) Survey of Consumer Finance, and objective life expectancy implied by statistical tables. They find that this measure is related to having positive beliefs about future events, such as expected income growth. Moreover, they note that it correlates strongly and significantly with psychometric tests of optimism, including the Life Orientation Test that was introduced by Scheier and Carver (1985), see Puri and Robinson (2006). They follow the construction of the measure with an empirical analysis of optimism and economic behaviour. They find that optimists save more and are more likely to hold individual stocks. Additionally, they find that optimists expect to work longer and are more likely to expect that they will never retire. These last findings might be evidence that optimists are individuals that are more aware of longevity risk.

Longevity risk is an important factor in theoretical models of individual financial decision making, including saving and asset allocation. Where individuals exposed to significant longevity risk are expected to save more. Post and Hanewald (2013) investigate the relationship between longevity risk awareness and saving behaviour, using data from SHARE. They observe significant covariance between subjective survival expectations and objective survival expectations and note that there is some awareness of longevity risk. They find however that individuals do not save more when faced with longevity risk, where theory suggests that they should. This results slightly contradicts the findings of Puri and Robinson (2007) who observe that individuals with larger discrepancy between subjective and objective survival expectancy on average save more.

Angelini and Cavapozzi (2017) employ the same approach as Puri and Robinson (2007) to measure dispositional optimism, using wave 2 of SHARE. Subsequently, they use this measure to analyse financial decisions of households. In addition to the analysis performed by Puri and Robinson (2007) they include control variables that should capture cognitive abilities, trust and social interaction of the respondents. They find that for risk tolerant individuals, dispositional optimism is a relevant factor in stock market participation.

There are other reasons why optimism might play a role in explaining why households hold stocks. First, under heterogeneous beliefs, individuals that have an optimistic attitude or nature are probably more likely to be financial optimists as well, expecting that the return on a stock portfolio is higher than pessimistic individuals would believe. For which, the finding of Brown et al. (2005) that there is a positive relationship between financial optimism and unsecured debt holding, might be evidence. Furthermore, optimists might expect the volatility of stock portfolios to be less than pessimists would.

(8)

8 Moreover, an explanation based on differences in preferences, individuals overestimating the possibility that favourable future events will happen and underestimating the possibility of bad events happening, arguably perceive the possibility of regret to be less than more pessimistic individuals. Therefore, if individuals take possible regret into account before deciding to participate, individuals that perceive the chance of regretting a decision to be less are, everything else equal, more likely to participate.

These findings together with the above suggested explanations lead to the following hypothesis:

H1: Dispositional optimism increases the likelihood of stock market participation

The relationship is explored by a logistic model. This estimation method is chosen since it can capture the binary character of the dependent variable, i.e. households either hold stocks (1) or do not hold stocks (0). The construction of the measure of dispositional optimism follows that of Puri and Robinson (2007) and Angelini and Cavapozzi (2017). The estimation is performed controlling for several established factors that influence stock market participation. Most importantly, age, gender, risk aversion, trust, planning horizon and country of residence. Moreover, the model controls for differences in health and cognitive ability that might be captured by the optimism measure when not controlled for.

I find that optimism in terms of a generalized positive attitude towards the future is not able to meaningfully explain differences in households’ stock market participation in Europe, employing a recent dataset of the SHARE.4 Under a minimal specification it is a significant predictor for stock market participation in a logistic model. However, once additional factors that influence stock market participation are accounted for, the influence diminishes. Moreover, this result is robust when analysing participation of individuals that are financial risk tolerant. I conclude that dispositional optimism is not a meaningful predictor of stock market participation. Factors that do play a role are socio economic related, such as income, education and social interaction. Also, cognitive ability, risk aversion, planning horizon, gender and country of residence play a significant role. This result confirms established factors in the literature that influence stock market participation.

The contribution to existing literature of this thesis is that it explores the relationship of optimism and stock market participation employing a recent data set. The finding that dispositional optimism is not meaningful in rationalizing limited stock market participation contradicts what is found by Puri and Robinson (2007) who report that the optimism indicator they propose is highly significant in explaining stock market participation. It is found that optimism is also not a meaningful determinant of stock market participation for risk tolerant individuals. This contradicts what is found by Angelini and Cavapozzi (2017) who report that

(9)

9 this dispositional optimism indicator is a relevant predictor of stock market participation for risk tolerant individuals, employing an older dataset of the SHARE that includes less countries. The contradictory findings can be due to several reasons. First, dispositional optimism might correlate with other predictors that influence stock market participation, since it is found that the effect of dispositional optimism diminishes when control variables are added. Puri and Robinson (2007) employ a minimal list of control variables when investigating the level of stock market investments relative to the total amount of wealth. Furthermore, they report that optimism is a relevant predictor for a wealth to income ratio, reporting a large positive coefficient when regressing wealth/income on optimism. Not controlling for this, the optimism indicator might reflect omitted effects of wealth over income on stock market participation. Second, the dataset employed in this study has a significantly larger sample size and includes more countries than Angelini and Cavapozzi (2017). Lastly, the role of dispositional optimism in explaining stock market participation might be time-variant, since this study finds an insignificant effect of optimism employing a more recent dataset of SHARE.5 A possible explanation could be that optimistic beliefs about future events in general are no longer correlated with optimistic beliefs about stock returns or return volatility. Possibly due to a general change in beliefs about the stock market.6 This could be an interesting topic for future research that could be tested by, for instance, employing the longitudinal characteristics of the different SHARE waves.

The rest of this thesis is structured as follows; first, the empirical strategy employed is introduced in the next section, then, the data from the SHARE is presented, subsequently, the results from the empirical analysis are given, finally, the last section provides the conclusion and discussion.

2. Empirical strategy

This section introduces the reader to the logistic model in which direct stock market participation is regressed on a single dispositional optimism indicator. This measure is due to Puri and Robinson (2007) and equals the difference between the subjective probability of living another N years, where N is a standard number of years that depends on the respondent’s current age, and the respondent’s actuarial probability to achieve that age, as observed from expected life tables in the respondent’s country of residence dependent on gender and current age. Optimists are assumed to overestimate, and pessimists to underestimate the probability of

5_{SHARE wave 2, employed by Angelini and Cavapozzi (2017) was collected in 2006, whereas SHARE wave 5,}

employed in this study, was collected in 2013. Although this thesis studies the cross section of dispositional optimism and stock market participation, finding somewhat contradicting results between an older and newer dataset might indicate that there is a time element at play.

6_{For instance, the global financial crisis might have made individuals more pessimistic about the stock market}

(10)

10 survival. The subsequent logistic model regresses direct stock market participation on the dispositional optimism indicator and controls for several established factors that are found to influence stock market participation. Moreover, health and cognitive ability indicators are included in the model to account for variability that could otherwise be captured by the dispositional optimism measure. The following part introduces the model formally and section 3 describes the data and the construction of the control variables. Part 3.8 elaborates on the construction and validity of the dispositional optimism measure.

2.1 Stock market participation and dispositional optimism: the logistic model

To investigate to which degree dispositional optimism is associated with stock market participation, the following multivariate cross-sectional model is estimated:

𝑆𝑀𝑃_𝑟 = 𝛽₁𝑂𝑃_𝑟+ 𝛿₁Χ_𝑟+ 𝜀_𝑟 , (1)

where SMPr is direct stock market participation, OPr is the explanatory variable that represents

the measure of dispositional optimism, Xr is a large set of respondent and household specific

control variables and εr the stochastic disturbance term for each financial respondent r. The

stock market participation measure, SMPr is equal to 1 if the respondent owns stocks or 0 if the

respondent does not own stocks. The dispositional optimism measure is described in depth in the following section, it is scaled to lie between 0 and 1. Angelini and Cavapozzi (2017) propose doing this to provide a clear distinction between the most pessimistic and most optimistic respondents. The model includes a large set of control variables; these variables roughly belong in two groups: variables that reasonably can be assumed to- or variables that have been identified to influence stock market participation and variables that control for effects that otherwise can potentially be captured by the dispositional optimism indicator, if not controlled for. The control variables include demographic, respondent specific and household characteristics. Specifically, the model controls for country of residence, gender, age, number of children of the respondent, and whether the respondent is single, high educated, employed, (the logarithm of) household income, risk aversion of the respondent, a scale for trust, planning horizon of the respondent, several cognitive ability indicators, a measure for social interaction and several individual health indicators. The variables included in the specification of the model are described in the following section.

(11)

11 specifications exist that can take the binary character of dependent variables into account. Most common are the probit and logit (or logistic) regressions. Findings with respect to the coefficients in an OLS specification and probit or logit specification are quite similar. For instance, Angelini and Cavapozzi (2017) report estimates that were regressed using OLS, however they note that they find similar results under a probit regression. Moreover, Hong et al. (2004) note that given the binary nature of stock market participation they estimated their model under logit and probit specifications, finding similar results to an OLS specification. Kaustia and Tortstila (2011) choose to report the coefficients from a logit specification. The advantage of a logit or logistic specification is elaborated on next.

In a logistic model, the dependant variables take the value of 0 or 1. The independent variables are not restricted. The main element of the logistic regression model is an algorithm that restricts the expected value, or fitted value, of the function described by the independent variables to a number to lie also between 0 and 1. This the result can then be interpreted as the conditional probability of observing participation (1) of a given respondent given the observed data from the survey, and 1 minus this probability as the probability of observing non-participation (0) for that same participant. The fitted values for stock market non-participation can be acquired by assuming that the model predicts the respondent to be a stock owner if the predicted probability is larger than 0.5. The degree to which the model is able to correctly predict stock market participation and none participation provides us with a sense for the fit of the model.

The estimated coefficients for the independent variables presented in this study, are referred to as “odds ratios”, representing estimates of how the odds of participating in the stock market change with a 1 unit increase in that independent variable holding all other variables constant. An odds ratio of 2 means that an increment of 1 in the independent variable doubles the predicted odds of observing the stock market participation. That is, an increasing effect on the estimated probability that the respondent holds stocks. An odds ratio of 1 would mean that the independent variable has a 1:1 effect on the predicted probability of holding stocks. In other words, an odds ratio of 1 means that a change in the independent has no meaningful effect on the predicted probability of participating in the stock market.

In this study we are interested in how being optimistic changes the probability of a household participating in the stock market. Therefore, we are interested in the significance and magnitude of β1 of equation (1). If being optimistic increases the odds of participating, one

would expect an odds ratio larger than unit. Formally; the hypothesis that will be tested is as follows:

𝐻₀ ∶ 𝛽₁ = 1, (2)

(12)

12

𝐻_𝑎 ∶ 𝛽₁ > 1. (3)

where β1 is the estimated odds ratio for the dispositional optimism measure. The model is

estimated using robust standard errors, to account for unobserved heteroskedasticity. Furthermore, the sensitivity and specificity of the model will be addressed. Section 4 discusses the results of the empirical analysis. The following section introduces the data employed in this analysis. Furthermore, it addresses the way this study deals with missing data.

3. Data

To construct factors that can explain (limited) stock market participation in Europe. I employ data from the Survey of Health, Ageing and Retirement in Europe (SHARE). SHARE is a representative longitudinal survey of households in Europe.7 There are several waves of the survey available on the website of SHARE. Respondents that participated in previous waves are asked to participate again in new waves, furthermore households, and countries, are added each wave. This results in a growing dataset each wave, with a longitudinal component. SHARE’s goal is to collect data on individuals and their environment as they age to assess the process of individual and population ageing in depth (Malter & Börsch-Supan (2015)). It collects data from individuals aged 50 or older and their spouses from 14 European countries.8 Specifically, these European countries are: Austria, Belgium, Czech Republic, Denmark, Estonia, France, Germany, Italy, Luxembourg, Netherlands, Slovenia, Spain, Sweden and Switzerland. No more than 10% of the sample population lives in the same country, so no country is overrepresented in the data. Data collected by the SHARE spans wide including information on demographics, activities, expectations, employment, assets, social connectedness and health. Additionally, the survey has respondents perform several tests, such as a numeracy, memory and some physical tests. These exercises are then used to develop a scale, index or score that reflects an individual’s health or cognitive ability. The data is employed in numerous disciplines such as psychology, economics, demography and in public health related studies.

To analyse the relationship between dispositional optimism and stock market participation I focus on the data collected in SHARE wave 5. Data collection for wave 5 took place in 2013 and was finished in November of that year. It is desirable to work with the most recent data, however, I choose not to let wave 6 serve as the main dataset as there are, as of writing, not for

7_{For (methodological) details, see Börsch-Supan et al. (2018) available at share-project.org.}

8_{SHARE wave 5, see SHARE compliance profiles – Wave 5. There is also data available for Israel, however}

(13)

13 all the countries life tables available that match the year in which the data in wave 6 was collected (2016). 9

My research focusses on the financial respondents of each household aged 50-90. The financial respondent is the person in charge of taking the financial decisions in the household. The number of financial respondents that answered all the questions is 37,264. Table 2 in the appendix depicts the descriptive statistics of all the variables included in the analysis.

3.1 Missing data

A common characteristic of surveys is that data for a variety of reasons is often missing. The way an empirical study deals with missing data is important as it can introduce biases into the model. This section describes how we deal with missing data.

In the dataset I employ, most missing values occur in the section devoted to the so-called grip strength test. The outcome of the grip test is an important health indicator, which is often used as a proxy for health (Andersen-Randberg et al. (2009)), furthermore it is arguably unrelated to stock market participation. To make sure that the proposed measure for dispositional optimism is not a substitute for healthiness, this study controls for health, and grip strength seems to be an appropriate measure. It is tested as follows: respondents are asked to squeeze as hard as they can into a handle, the number of kilograms they are able to squeeze is the resulting measure. In the survey, around 7% of the respondents were unable to or refused to do this. Of this group, 22% felt it would be unsafe to do so, 19% were unable because of an injury, due to surgery or otherwise, 18% refused and gave no reason, 6% was unable to complete the test, 3% did not understand the instructions and 24% gave another reason.

We deal with missing data by the method of listwise deletion. This method involves omitting respondents in the empirical analysis that did not answer all of the questions. This method does not bias the empirical analysis if the missing values are missing at random, that is, the probability of missing data are random rather than that they have a structural component. From the example of grip strength above, missing values for the grip strength indicator are arguably not missing at random. So, by using listwise deletion, some of the information is lost. Moreover, it might introduce biases into the model as the estimations mainly omit respondents who were not able to, or refused to, squeeze into a handle.

There are other ways to deal with missing data, such as the multiple imputation technique described by Rubin (1987). This method involves using answers to related questions to impute multiple responses to substitute missing answers, using a pre-specified model. This creates a number of datasets, that together are used to analyse the model of interest. Drawbacks of

(14)

14 employing this method are that the imputed responses are dependent on the appropriateness of the model that the researcher proposes to impute them. Moreover, for multiple imputations to be unbiased, the probability of missing data is allowed to correlate with other observed variables, however, the probability that data is missing may not depend on itself. The latter is arguably not the case for the grip strength indicator, as respondents that perceive in unsafe to squeeze into a handle or are not able to so, presumably, all else equal, score lower on the indicator itself than respondents who can squeeze into a handle.

Nevertheless, since squeezing a handle and stock market participation are arguably unrelated and the goal is to analyse the relationship between optimism and participation, the author believes that in this case it is justified for this study to use the method of listwise deletion as a way of dealing with missing data. Readers however should be aware of this.

3.2 Stock market participation

This study assesses stock market participation considering direct stock market participation. Direct stock market participants are defined as the respondents that answered yes to the following question in SHARE:

Do you currently have any money in stocks or shares (listed or unlisted on stock market)?

There are also other ways to participate in the stock market, for instance, via mutual funds or retirements accounts. Indirect stock market participation is defined by the possible ownership of stocks through mutual funds and individual retirement accounts. Indirect stock owners are defined as the households that answered half stocks and half bonds or mostly stocks to the question:

Are these mutual funds, managed investment and/or retirement accounts [that you have] mostly in stocks or mostly in bonds?

Of the 37,264 respondents that answered the question, 13% hold stocks. This percentage increases to 24% if one also considers indirect stock ownership, where the aggregation of direct and indirect stock holding is referred to as total stock ownership. So, only one fourth of the respondents participate in the stock market in some way. There are also significant differences between countries.

(15)

15 have significant indirect ownership, compared to direct ownership. Estonia, Italy, Spain and Slovenia experience low total stock ownership.

One of possible reasons to find such low direct and total levels of stock market participation in certain countries and high in others can be due to differences in institutional quality of the country of residence, for instance Asgharian et al. (2015) link household’s trust to institutional quality find that the part of trust related to institutional quality influences stock market participation. Another explanation might be the wide variety in state pension-systems and/or individual retirement accounts between different countries. Moreover, as predicted by financial theory, differences in personal taxes might explain some of the variability.

This study will not cover the percentage of wealth invested in stocks as Angelini and Cavapozzi (2017) do, since the amount of missing values for this variable is simply too high to draw conclusions with respect to the whole sample (around 70% of observations remain).

In the following part the control variables are introduced, starting with economic attitudes. Next, the social interaction indicator is introduced, then the cognitive ability indicators followed by the health indicators. Finally, the dispositional optimism measure is described, and the validation of this measure is addressed.

3.3 Economic attitudes 3.3.1 Risk aversion

Figure 1: Direct and indirect stock ownership by country

(16)

16 To provide a control for differences in attitudes towards taking of financial risk, respondents were asked the following question:

When people invest their savings, they can choose between assets that give low return with little risk to lose money, for instance a bank account or a safe bond, or assets with a high return but also a higher risk of losing, for instance stocks and shares. Which of the statements on the card comes closest to the amount of financial risk that you are willing to take when you save or make investments?

The possible answers are: 1. Take substantial financial risks expecting to earn substantial returns, 2. Take above average financial risks expecting to earn above average returns, 3. Take average financial risks expecting to earn average returns and 4. Not willing to take any financial risks. 73% of financial respondents chose option 4, saying that they are not willing to take any financial risks. A dummy variable is constructed that is equal to 1 for respondents that marked option 4, that is, the respondents that are not willing to take any financial risks. Furthermore, only 1% of the respondents chose option 1. declaring that they are willing to take substantial financial risks expecting to earn substantial returns, 3% chose option 2. saying that they are willing to take above average financial risks and 23% chose option 3. declaring to be willing to take average financial risks.

3.3.2 Trust

As Angelini and Cavapozzi (2017) correctly suggest, it is important to distinguish optimism from trust. That is, one needs to make sure that the dispositional optimism variable is not a substitute for trust. As Guiso et al. (2008) find that trust is an important factor influencing stock market participation. They find that less trusting respondents are less likely to hold stocks, moreover, conditional on stock holding they find that less trusting individuals hold less stocks as a percentage of wealth. The trust variable is constructed using the following question:

Finally, I would like to ask a question about how you view other people. Generally speaking, would you say that most people can be trusted or that you can’t be too careful in dealing with people?

Respondents are asked to rate this on a scale from 0-10, with 0 meaning you cannot be too careful and 10 meaning that most people can be trusted. Around 4% of the respondents indicated that one cannot be too careful (score 0) and 5% indicated that most people can be trusted. The average of the trust indicator is 6.0.

(17)

17 Another economic attitude of interest is respondents planning horizon. Investing in the stock market may have to do with the willingness and ability of respondents to plan consumption and saving ahead. Moreover, respondents with longer planning horizons might be more willing to participate in the stock market as the volatility in stocks is smoothened over time, therefore the disutility, or participation costs for investing in stocks might be lower for respondents with longer planning horizons. Renneboog and Spaenjers (2012) find planning horizon to be a relevant factor in stock market participation. The proposed dispositional optimism indicator basically measures the overprediction of the survival probability. Respondents that expect to live longer arguably might be expected to assume longer planning horizons in planning consumption and saving of income. To take this into account, a control for planning horizon is included in the estimation. It comprises a dummy variable that is equal to 1 for respondents that indicated that planning of saving and spending was most important for them for a time-period of at least one year, and 0 otherwise. Around 62% of the respondents indicated that a time-period of at least one year was most important in planning their saving and spending.

3.4 Social interaction

Previous research has suggested that social interaction influences stock market participation, e.g. Hong et al. (2004) construct a model that predicts higher participation rates among social individuals than under non-social individuals, employing the US Health and Retirement Study, they find that social households are more likely to invest in the stock market. Social households are defined by 3 dummy variables, knowing your neighbours, visiting neighbours or attending church. Furthermore, Bailey et al. (2017) suggest that social connectedness at the county-level plays a role in facilitating economic and social interactions, they find this by constructing the social connectedness index using links between Facebook users.

This study also employs an indicator of social activities, in line with Angelini and Cavapozzi (2017) a variable is constructed that reflects whether respondents engage in social activities

In wave 2, the dataset Angelini and Cavapozzi (2017) employ, the question: Have you done any of these activities in the last month?

Is asked. They construct a social activities dummy that equals one if the respondents engaged in any of the activities mentioned. Unfortunately, this question is no longer asked in wave 5 of SHARE. However, the question can be found somewhat restated as:

(18)

18 The activities listed are: 1. Done voluntary or charity work, 4. Attended an educational or training course, 5. Gone to a sport, social or other kind of club, 6. Taken part in activities of a religious organization, 7. Taken part in a political or community-related organization, 8. Read books, magazines or newspapers, 9. Did word or number games such as crosswords puzzles or Sudoku, 10. Played cards or games such as chess, 96. None of these. In the last twelve months 29.7% of respondents played card games or games such as chess, 29.5% went to a sport/social/other kind of club and 18.3% did voluntary/charity work. Then, the next question asks:

How often in the past twelve months did you do this activity?

The options are: 1. Almost daily, 2. Almost every week, 3. Almost every month and 4. Less often. Fortunately, this question thus provides some measure of frequency of the social activities.

The social interaction dummy created now takes value 1 if the respondent engaged at least almost every month in any of the listed activities, excluding reading books, magazines or newspapers and doing word or number games such as crosswords puzzles or Sudoku, since these activities are assumed to on average not constitute a social activity. Of the respondents 54% reported that they attended at least one of these activities almost every month. This is around the involvement in social activities that Angelini and Cavapozzi (2017) find.

3.5 Cognitive skills

Serving many disciplines, SHARE constructs several (psychological) scales and multi-item indicators. These scales and multi-item indicators are measures and/or proxies of certain abilities and (psychological, medical) characteristics. These measures are derived from answers to multiple questions or are scores from tests performed in the questionnaire. SHARE for instance has respondents conduct several tests to measure cognitive abilities.

(19)

information-Optimism and limited stock market participation

19 intensive) and cognitive skills is less apparent. In what follows, the method of testing and scoring the measures in described.

The numeracy indicator is constructed by scoring the following questions:

1. If the chance of getting a disease in 10 per cent, how many people out of 1,000 would be expected to get the disease?

2. In a sale, a shop is selling all items at half price. Before the sale, a sofa costs 300 [country currency]. How much will it cost in the sale?

3. A second-hand car dealer is selling a car for 6,000 [country currency]. This is two-thirds of what it costs new. How much did the car cost new?

4. Let’s say you have 2,000 [country currency] in a savings account. The account earns ten per cent interest each year. How much would you have in the account at the end of two years?

The numeracy score, measuring mathematical ability is then scored 1-5. With 5 indicating good mathematical ability (4 correct answers) and 1 indicating bad mathematical performance (0 correct answers). Only new respondents are asked to perform this test; therefore, SHARE fills in the scores of respondents that participated in other waves. The score is 0 (bad) for 4% of the respondents, 20% score a 5 (good), 13%, 30% and 34% of the respondents score 2, 3 and 4 respectively.

All respondents in wave 5 are asked to participate in a second numeracy test (subtraction test, henceforth), this test is designed to measure if participants can correctly perform a simple subtraction multiple times. The question reads:

1. Now let’s try some subtraction of numbers. One hundred minus 7 equals what? 2. – 5. And 7 from that?

The subtraction-indicator is than simply the number of correct answers, ranging from 0 (bad) to 5 (good). The relative plainness of this test is reflected in the scores, as 67% of the respondents score a 5. Moreover, it increases to 83% if a score of 4 is included.

(20)

20 recall any words the first time and 8.8% the second time, 0.5% were able to recall all the words the first time and 0.6% were able to do it the second time. The average of the recall indicator is 9.6 so on average respondents were able to recall 9.6 words when we sum the two attempts.

According to Rosen (1980), particularly for the elderly, verbal fluency is another indicator of cognitive impairment. The verbal fluency indicator in SHARE is designed to test executive function. Respondents are asked to name as many words as possible from a semantic category in one minute. The verbal fluency score is simply the number of correct and unique words. The semantic category in SHARE is animals, since this category is assumed to be clear across languages and cultures.10 Respondents were on average able to name 21 animals. The maximum that respondents were allowed to name is 100 animals. There was one respondent who was able to do so.

The relationship with cognitive ability and these measures is assumed to be positive, i.e. the scores are assumed to be caused by a respondent’s cognitive ability. Where a higher score indicates a higher cognitive ability. A principal-component analysis, not reported here, that was performed to explore the possibility of modelling cognitive ability as a latent variable, showed that the cognitive ability indicators load substantially on the same component. Moreover, being highly educated also loads substantially on this component. Indicating that indeed these measures share a common factor. The following section introduces the socio-economic related factors.

3.6 Socio-economic status

The goal here is to describe the control variables for socio economic status. I employ three indicators of socio-economic status. These indicators are: education, employment and household income. It is assumed that socio-economic status is an important predictor of stock market participation. For instance, household income could work as a necessary condition for stock market participation. Also, being highly educated and being employed might work through socio-economic status as a predictor of stock market participation.

The household income measure is constructed by taking the logarithm of yearly household income reported in SHARE, who annualize the answer to a question about monthly income. The average yearly income of the households is 44,528 euro and the highest household income reported was 1.2 million euros.

The education measure is a dummy variable that is equal to 1 is the education that the respondent received was at least labelled a 4 in the International Standard Classification of Education (ISCED, 1997), which corresponds to having received at least post-secondary,

(21)

21 tertiary education. In other words, respondents who received a high education. Of the respondents 29% are highly educated.

The employment measure is a dummy that is equal to 1 if the respondent indicated to be employed. In this set, 28% of the respondents are employed. This is, however, not surprising since the survey is held under respondents aged 50 or above.

3.7 Health

The employed dispositional optimism indicator is measured as the discrepancy between subjective and objective survival probabilities. Therefore, if not controlled for, the indicator could be a substitute for a respondent’s health status, for instance, if on average more healthy respondents estimate the chance of survival to be higher, the measure could capture health status. To make sure this optimism indicator is not partly a substitute for healthiness, this study controls for a respondent’s health status by including the following control variables:

Grip strength: This is the maximum grip strength of the right or left hand of two attempts, measured in KG’s. Respondents are asked to squeeze as hard as they can into a handle and the resulting measure is grip strength in KG’s. This variable is included since it is assumed to be independent of stock market participation and is found to be a strong and cheap estimator of healthiness, and an appropriate indicator of mortality (Leong et al. (2015)).11 The average maximum grip strength of the respondents is 33.8 kg’s.

Limitations indices: Two limitation indices are included. The ADL index of Katz et al. (1963) is included to assess the number of limitations with activities of daily living that are essential for an individual’s independence, such as dressing, walking and eating. The more limitations, the higher the index on a scale from 0-6. Furthermore, the IADL index of Lawton and Brody (1969) is included. This index measures the number of limitations with instrumental activities of everyday life, such as making phone calls, managing money and taking medication. It is measured on a scale from 0-7, the higher the index the more limitations the respondent’s experiences. The average for ADL is 0.25 and the average for IADL is 0.40. The reason for finding reasonably low indexes for a survey under respondents aged 50-90 might be that SHARE aims to survey under none-institutionalized individuals. The measure that is included in the analysis is the aggregation of the two indices. This is done to avoid possible multicollinearity problems, moreover it decreases the number of coefficients to be estimated. The next part introduces the dispositional optimism indicator and addresses the validity of this measure.

(22)

22 3.8 Dispositional Optimism

The measure of dispositional this study employs is a variant of the measure used in previous research (see Puri and Robinson (2007) or Angelini and Cavapozzi (2017)). It is defined as the difference between self-perceived chances of survival until a certain age and the chances according to life tables. Life tables are tables constructed by demographers that incorporate birth and mortality rates among cohorts of people, these table are constructed per country and gender. Dispositional optimism/pessimism is then assumed to be increasing in the discrepancy between the two chances.

Self-perceived chances of survival are obtained from the following question in SHARE: What are the chances that you will live to be [TARGET AGE] or older?

This question is then answered on a scale from 0 to 100 % chance. The [Target age] differs for individual respondents, dependent on their age. Specifically, the age group 50-65 is asked for the chances of living to be 75 or older, ‘the 5-year’-goup aged 66-70 is asked for the chances of living to be 80 or older, ‘the 5-year’-goup aged 71-75 is asked for the chances of living to be 85 or older, this goes on until the 86-90 group is asked for the chances of living to be 100 or older. Mainly, the question is constructed in such a way that it asks respondents to estimate their chances of survival for at least ten years into the future.

Subsequently, the life tables provided by the Human Mortality Database12 are used to compute actuarial probabilities of surviving to be the target age or older, conditional on current age, gender and country of residence. The measure for optimism of respondent r is the discrepancy between self-perceived chances of survival and that computed from the life tables:

𝑂𝑃_𝑟 = 𝑆𝑢𝑏𝑗𝑒𝑐𝑡𝑖𝑣𝑒 𝑠𝑢𝑟𝑣𝑖𝑣𝑎𝑙 𝑝𝑟𝑜𝑏._𝑟− 𝐴𝑐𝑡𝑢𝑎𝑟𝑖𝑎𝑙 𝑠𝑢𝑟𝑣𝑖𝑣𝑎𝑙 𝑝𝑟𝑜𝑏._𝑟. (4)

The measure is then demeaned per age cohort and scaled such that it lies between 0 and 1. The demeaning per age cohort is done to control for potential biases. Since, life cycle trends in subjective survival probabilities exist; as Grevenbrock et al. (2018) note that younger individuals tend to underestimate, and older individuals tend to overestimate survival chances, where the turning point seems to be around 70 years. Scaling between 0 and 1 is proposed by Angelini and Cavapozzi (2017) to clearly define benchmarks that thus identify the most optimistic and pessimistic respondents in the sample. This measure then reflects the assumed

(23)

23 relative optimism of respondents relative to their peers of the same age.13_{The following section} elaborates on the validity of this measure.

3.8.1 Validation of the dispositional optimism indicator

The purpose of this part is to provide insight into the validation of the optimism indicator. There are several reasons why the discrepancy between self-perceived chances of survival and those obtained from life tables might be unrelated to optimism. For instance, due to an information discrepancy between demographers and individuals or to differences in cognitive abilities. That is, individuals have private information concerning their behaviours, life-style and health, and might be therefore better suited to assess the probability of survival. Consequently, it is important to investigate the validity of the optimism measure.

Therefore, Puri and Robinson (2007) carry out several tests to examine the validity of the measure. As touched upon in the introduction, Scheier and Carver (1985) define dispositional optimism as having generalized positive expectations regarding future events. Puri and Robinson (2007) relate the measure they propose to having positive beliefs about future (economic) conditions. Moreover, they find that this measure strongly and significantly correlates with psychrometric tests of optimism, such as the LOT-R of Scheier et al. (1994) (see Puri and Robinson (2006)).14 Angelini and Cavapozzi (2017) validate the indicator by comparing it to other items that are arguably related to optimism or to a positive attitude towards life. Such as, perceiving the chances that one’s standard of living will improve in the next five years. They find a strong correlation between the dispositional optimism measure and the perceived chances that the standards of living will improve.

Noteworthy, measuring an unobservable variable, like dispositional optimism, empirically with indicators is associated with measurement error. Because of measurement error, estimated parameters can possibly suffer from inconsistency. I have explored the possibility of modelling the measurement errors, however, due to the characteristics of the data this did not result in a feasible model.

13_{Another reason for demeaning the measure per age cohort is that it now measures the relative discrepancy in}

responses to the same question; e.g. respondents aged 55 all got asked to estimate the chance to live to be 75, whereas respondents aged 75 were asked to estimate the chance to live to be 85, which is in fact not the same question. The measure now reflects the discrepancy in the questions asked by measuring the relative

discrepancy between respondents of the same age.

(24)

24

4. Results

In this section, the results of the empirical analysis described in section 2 is presented. First the results from the main logistic regression are presented, then the results from the model considering only respondents that are willing to take financial risk are given.

Table 6 in the appendix shows the pairwise correlation matrix of the independent variables. High correlation can indicate multicollinearity. I assume the common threshold that a correlation of below 0.7 is acceptable to be able to correctly estimate the model. From this table we see that the highest correlation is between the recall and fluency measures and is equal to 0.5. This is deemed an acceptable correlation between explanatory variables, and thus, requires no changes in the specification of the proposed model.

Table 1 presents the results of the logit regression of equation (1) under different specifications. The first column represents the estimated odds ratios using a minimal set of (demographic) control variables, these are country of residence, age, gender, whether a respondent is single and the number of children the respondent has. In column 2, control variables for education, occupation and income are included. Column 3 also accounts for economic attitudes, which are risk aversion, planning horizon and trust. Column 4 includes indicators for cognitive ability, column 5 controls for the respondent’s health status. Finally, column 6 includes a fully specified model that also takes social interaction into account.

Table 1: Direct stock market participation and dispositional optimism

This table depicts the results of 6 logistic regressions of equation (1) in section 3; 𝑆𝑀𝑃𝑟 = 𝛽1𝑂𝑃𝑟+ 𝛿1Χ𝑟+ 𝜀𝑟,

where SMPr is the stock market participation variable of interest, OPr is the explanatory variable that

represents the indicator of dispositional optimism, Xr is a large set of respondent and household specific

control variables and εr the stochastic disturbance term for each financial respondent r. From left to right

(25)

25 Direct stock market participation

(Odds ratios) 1 2 3 4 5 6 Optimism 1.445*** 1.278*** 1.131 1.071 1.033 1.002 (0.115) (0.104) (0.0978) (0.0944) (0.0916) (0.0891) Male 1.504*** 1.439*** 1.205*** 1.171*** 1.169*** 1.174*** (0.0516) (0.0504) (0.0441) (0.0445) (0.0444) (0.0447) Age 1.000 1.011*** 1.017*** 1.025*** 1.027*** 1.027*** (0.00180) (0.00231) (0.00244) (0.00262) (0.00277) (0.00279) Single 0.557*** 0.622*** 0.644*** 0.667*** 0.672*** 0.671*** (0.0220) (0.0253) (0.0273) (0.0285) (0.0288) (0.0288) Number of children 0.915*** 0.913*** 0.917*** 0.919*** 0.919*** 0.916*** (0.0116) (0.0120) (0.0124) (0.0127) (0.0127) (0.0127) High education 1.982*** 1.616*** 1.397*** 1.399*** 1.382*** (0.0703) (0.0603) (0.0539) (0.0541) (0.0536) Employed 1.094* 0.987 0.970 0.966 0.983 (0.0517) (0.0487) (0.0482) (0.0482) (0.0493) Income (log) 1.244*** 1.191*** 1.159*** 1.155*** 1.150*** (0.0292) (0.0278) (0.0275) (0.0279) (0.0285) Risk averse 0.270*** 0.283*** 0.283*** 0.286*** (0.0101) (0.0106) (0.0106) (0.0107) Trust 1.025*** 1.015* 1.014 1.011 (0.00846) (0.00852) (0.00853) (0.00853)

Planning horizon at least a year 1.387*** 1.326*** 1.316*** 1.303***

(0.0590) (0.0566) (0.0563) (0.0558) Numeracy 1.175*** 1.172*** 1.166*** (0.0247) (0.0246) (0.0246) Subtraction 1.111*** 1.107*** 1.103*** (0.0239) (0.0239) (0.0238) Recall 1.024*** 1.023*** 1.021*** (0.00630) (0.00631) (0.00632) Fluency 1.018*** 1.018*** 1.016*** (0.00271) (0.00272) (0.00274)

Limitation with activities 0.940*** 0.946**

(0.0205) (0.0205) Max handgrip 1.003 1.003 (0.00257) (0.00258) Social activities 1.332*** (0.0522) Constant 0.423*** 0.0160*** 0.0246*** 0.00395*** 0.00374*** 0.00359*** (0.0618) (0.00497) (0.00793) (0.00140) (0.00141) (0.00137) Observations 37,264 37,264 37,264 37,264 37,264 37,264 Pseudo R-squared 0.144 0.164 0.215 0.224 0.224 0.226

Robust SE YES YES YES YES YES YES

Country effects YES YES YES YES YES YES

(26)

26 The first regression suggests that dispositional optimism has a positive effect on the likelihood, or on the predicted probability, that a respondent participates in the stock market since the odds ratio is larger than 1 and is statistically significant. It is also significant when control variables for education, occupation and income are introduced, see column 2. However, this effect seems to disappear when other control variables are introduced. A logistic regression including a specification with economic attitudes results in an insignificant estimated odds ratio for the dispositional optimism indicator, see column 3. Seemingly, the other factors are better able to explain the variety in stock market participation. For instance, the level of education is a well-suited characteristic that can explain stock market participation, that is, highly educated respondents are, ceteris paribus, a predicted 38% more likely to participate in the stock market than respondents that are not highly educated, in the fully specified model. Being employed does not seem to significantly influence stock market participation, nor does the respondent’s trust in other people. Income however does significantly predict stock market participation. This regression also confirms that well established economic attitudes such as risk aversion and planning horizon are relevant estimators of participation. There is also a noteworthy jump in pseudo R-squared once economic attitudes are incorporated, see column 2 and 3. Cognitive ability indicators are also positive and strong indicators for stock market participation as is social interaction, which confirms the findings of Christellis et al. (2010) and Hong et al. (2004). The limitations index, indicating health impairment is also a strong estimator of stock market participation, where health impairment has a negative impact on the predicted probability that a household holds stocks. This is in line with the sign Angelini and Cavapozzi (2017) find for the respective health indices.

The finding that dispositional optimism is not a meaningful estimator of stock market participation contradicts what is found by Puri and Robinson (2007) who find optimism to be a strong and statistically significant predictor of stock market participation. The contrasting finding can be due to the fact that I specify a larger list of control variables, whereas Puri and Robinson (2007) employ a smaller set.

(27)

27 Table 4 in the appendix addresses the sensitivity and specificity of the logistic model under full specification (column 6 in table 1). As can be seen the sensitivity of the model is poor, since if this were a medical test only 1 out of 5 people with a disease would be identified as having the disease on average. The model thus correctly predicts 21.6% of stock owners to be stock owners based on the observed variables and the estimated coefficients. The specificity of 97.5% is good, in medical terms, 97.5% of the population that does not suffer from the disease would not be identified in this test as having the disease. On average 87.6% of the respondents are correctly categorized by the model.

5. Conclusion and discussion

Substantial research in the last few decades has analysed the financial behaviour of households. The rationalization in financial models of low stock market participation among households proves to be a puzzle that is still incomplete. In this thesis I focus on analysing the effects of dispositional optimism on stock market participation. I do this using a representative dataset from the fifth wave of SHARE that consists of more than 37,000 households from 14 European countries. Dispositional optimism is measured by the discrepancy between subjective and objective survival expectancies, as proposed by Puri and Robinson (2007). Moreover, following Angelini and Cavapozzi (2017) the effect of dispositional optimism on stock market participation is also estimated separately for respondents that are willing to take financial risk. The relationship between dispositional optimism and direct stock market participation is estimated controlling for several factors that have been identified to influence stock market participation, or factors that could otherwise be captured by the dispositional optimism measure.

I find that dispositional optimism is not a meaningful factor in explaining the limited stock market participation of households. This result is robust when conditioning for financial risk tolerant individuals. This result contradicts what is found by Angelini and Cavapozzi (2017), who find that dispositional optimism is relevant for the stock market participation of financial risk tolerant individuals. Moreover, this finding opposes what is found by Puri and Robinson (2007) who note that dispositional optimism is a strong predictor of stock market participation. It appears that other household and individual characteristics are better suited to explain the limited stock market participation of households. These characteristics are country of residence, risk aversion, planning horizon, social interaction, gender, age, relation status, number of children, cognitive ability, education and income of the financial decision maker of the household. These findings reconfirm previously identified predictors of stock market participation.

(28)

28 It might for instance be that the prediction that follows from a participation costs explanation; that the availability of free financial information on the internet and the existence of online budget brokers should increase stock market participation, has more impact on younger persons than on the elderly. Furthermore, SHARE aims to survey under non-institutionalized individuals. The probability of being institutionalized, e.g. living in a nursing home, is arguably increasing with age. This may lead to sample selection bias. In the empirical analysis I try control for this effect by not including respondents older than 90 in the analysis.

Another limitation of this study is the presence of missing values, and the way they are treated here. Since this study uses listwise deletion as a way of dealing with missing data, possible biases might arise if missing values are not missing at random. The main concern here is the grip strength indicator, where some respondents indicated not to be able to participate because of an injury or because they perceived it not to be save. Therefore, caution is needed when generalizing the results.

(29)

29

6. Literature

Andersen-Ranberg, K., Petersen, I., Frederiksen, H., Mackenbach, J. P., & Christensen, K. (2009). Cross-national differences in grip strength among 50+ year-old Europeans: results from the SHARE study. European Journal of Ageing, 6(3), 227-236.

Angelini, Viola, and Danilo Cavapozzi, 2017, Dispositional optimism and stock investments, Journal of Economic Psychology 59, 113-128.

Ardila, A., Ostrosky‐Solís, F., & Bernal, B. (2006). Cognitive testing toward the future: The example of Semantic Verbal Fluency (ANIMALS). International Journal of Psychology, 41(5), 324-332.

Asgharian, Hossein, Lu Liu, and Frederik Lundtofte, 2015, Institutional Quality, Trust

and Stock-Market Participation: Learning to Forget,

(https://ssrn.com/abstract=2369732).

Bailey, M., Cao, R. R., Kuchler, T., Stroebel, J., & Wong, A. (2017). Measuring social connectedness (No. w23608). National Bureau of Economic Research.

Barber, B. M., & Odean, T. (2001). Boys will be boys: gender, overconfidence, and common stock investment. Quarterly Journal of Economics, 116, 261-292.

Barberis, N., Huang, M., & Thaler, R. H. (2006). Individual Preferences, Monetary Gambles, and Stock Market Participation: A Case for Narrow Framing. American Economic Review, 96 (4), 1069-1090.

Börsch-Supan, A. (2018). Survey of Health, Ageing and Retirement in Europe (SHARE) Wave 5. Release version: 6.1.0. SHARE-ERIC. Data set. DOI: 10.6103/SHARE.w5.610

Brown, S., Garino, G., Taylor, K., & Price, S. W. (2005). Debt and Financial Expectations: An Individual‐and Household‐Level Analysis. Economic Inquiry, 43(1), 100-120.

Brown, S., & Taylor, K. (2014). Household finances and the “Big Five” personality traits. Journal of Economic Psychology. https://doi.org/10.1016/j.joep.2014.10.006 Campbell, J. Y. (2006). Household finance. The journal of finance, 61(4), 1553-1604. Christelis, D., Jappelli, T., & Padula, M. (2010). Cognitive abilities and portfolio choice. European Economic Review, 54(1), 18-38.