E UROPE . C HILDHOOD SOCIO - ECONOMIC STATUS AND FINANCIAL ASSET OWNERSHIP THROUGHOUT

C

HILDHOOD SOCIO

-

ECONOMIC STATUS AND FINANCIAL

ASSET OWNERSHIP THROUGHOUT

E

UROPE

.

ABSTRACT. Using a European database, I examine the effects of childhood socio-economic status (SES) on household investment behavior. Contrary to previous research I find no evidence of a direct relation between childhood SES and adult financial asset ownership. I do prove that childhood SES has significant predicting power over adult financial asset ownership without controlling for adult variables. The drivers of these effects are risk tolerance, education and wealth. This indirect effect of childhood SES can have serious implications and is likely to contribute to cross-generational inequality. European wide measures would be ineffective and thus country specific approaches are preferred.

Keywords: Household finance, Childhood SES, Financial asset ownership, Risky asset ownership, Risk preferences, Financial upbringing, SES transition, Cross-generational inequality.

Table of contents

1. Introduction ... 1

2. Literature and hypotheses ... 4

2.1. childhood characteristics and financial asset ownership ... 4

2.2. The indirect influence of childhood SES ... 5

2.3. The direct effects of childhood conditions ... 8

3. Data ... 10

3.1. Database ... 10

3.2. Sample selection ... 10

3.3. Dependent variables ... 13

3.4. Independent variables ... 15

3.5. Transforming the data ... 18

4. Empirical methods ... 19

4.1. Hypothesis one and two, the (in)direct effect of childhood SES ... 19

4.2. Additional testing ... 19

5. Empirical results ... 21

5.1. Testing ... 21

5.2. Hypotheses ... 22

5.3. Proxy ... 32

5.4. Summary and implications ... 32

1

1. I

Investors are known to show baffling and incredibly irrational investment behavior (Hirshleifer and Shumway, 2003; Vissing-Jorgensen, 2003). Similar baffling behavior can be observed when examining the household finance literature. Despite recent increases in availability of both investing information and investing platforms, still only a fraction of all households owns financial assets (Bae, Bailey and Mao, 2006; Sprenger, Tumasjan, Sandner and Welpe, 2014; Christelis, Dobrescu and Motta, 2012). Economists have tried to explain this phenomenon for years. Ideally all households would invest in financial assets as rewards greatly outweigh risks (Campbell, 2006). In this research I try to contribute to the puzzle by shining a light on the relation between childhood SES (social economic status) and financial asset ownership.

This area of research is no barren land. Research has been conducted on the impact of childhood SES on the willingness to take financial risk (Christelis et al., 2012). Using data from the Survey of Health, Ageing and Retirement in Europe (SHARE1), they examine the impact of several childhood characteristics, among which SES, on adult financial asset ownership. They find that the amount of books and rooms in the residence when 10 years of age are both positively related to financial asset ownership. Together, these variables are treated as a proxy for childhood SES and thus they conclude that childhood SES directly influences the probabilities of financial asset ownership.

The aim of this thesis is roughly similar to the one of the Christelis et al. (2012) paper, but a few distinct differences make it worthwhile. The Christelis et al. (2012) paper is

1 _{This paper uses data from SHARE Waves 1, 2, 3 (SHARELIFE), 4, 5 and 6} (DOIs: 10.6103/SHARE.w1.600, 10.6103/SHARE.w2.600, 10.6103/SHARE.w3.600, 10.6103/SHARE.w4.600, 10.61 03/SHARE.w5.600, 10.6103/SHARE.w6.600), see Börsch-Supan et al. (2013) for methodological details.

2 based on data that is over 10 years old. In the last decade, the world of finance has been shaken and stirred. Events such as the 2007 crisis and the increasing availability of investment data and platforms impacted investment behavior throughout the world (Malmendier and Nagel, 2011; Sprenger et al., 2014; Bae et al., 2006). Furthermore, Christelis et al. (2012) use a proxy for childhood SES, I use the true variable. The literature shows no consensus on a childhood SES proxy, and thus the conclusions drawn upon the proxy might be inaccurate (Mazzonna, 2014). Lastly, I am able to conduct the research using data on more countries. Culture and origin have well-documented relations with financial asset ownership probabilities (Breuer, Riesener and Salzmann, 2014). Including more countries should give additional insights and create more generalizable results.

In line with Christelis et al. (2012), this thesis uses the SHARE database. This database collects both social and financial information on inhabitants of 15 countries throughout Europe that are at least 50 years of age. The SHARE database is very popular in the household finance literature (Mazzonna, 2014; Christelis et al, 2012; Havari and Mazzonna, 2015)

Recently, additional waves have been added to the SHARE data. The new data contains more observations originating from more countries. The new waves also include new questionnaire modules. Wave five includes an improved childhood questionnaire including a direct question on childhood SES. Using the variable based on this question, it is possible to formally test the effects as inferred by Christelis et al. (2012).

3 relationship with a recreated proxy, the proxy seems to be an inaccurate substitute of real childhood SES.

Alternatively I do find childhood SES to have strong predictive power as to financial asset ownership during adulthood. This relationship holds after controlling for several other childhood variables, birth cohort dummies and country dummies. Running additional testing on the drivers of this relationship yields interesting results. The drivers of this relation are significantly different per country. In general, childhood SES influences financial asset ownership through risk tolerance, education and wealth persistency.

This paper has both economical and empirical relevance. Empirically, I contribute to the puzzle that is household finance behavior. To the best of my knowledge, the only research focusing on childhood SES and financial asset ownership is the Christelis et al. (2012) paper. I prove that the conclusion of the Christelis et al.(2012) paper is wrong. Childhood SES has no direct influence on adult financial asset ownership.

Economically I show that inequalities in financial market participation rates are partly due to SES background. Ideally these inequalities would be non-existent as this can contribute to cross-generational inequalities in wealth. I give handles to reduce the inequalities and note that it is not useful to implement European wide measures. Instead country specific measures must be implemented based on better education, more attention to the possible benefits of taking financial risks and more attention on the benefits of specific asset classes.

4

2. L

In this section, the relevant literature is discussed and the hypotheses are formed.

2.1.

Household finance is a hot topic, but despite the increasing amount of literature, household financial behavior is still very unpredictable (Hirshleifer and Shumway, 2003; Bae et al., 2006; Christelis et al., 2012; Campbell, 2006). Historically, households show suboptimal financial behavior and are severely underrepresented in the financial market. As financial assets ownership comes with a great set of benefits, economists have been trying to explain the lack of household investing for quite some time (Campbell, 2006).

Recently, researchers are trying to explain this phenomenon using seemingly unrelated aspects of life, such as childhood characteristics. As a result, the literature examining the effect of childhood characteristics on household financial behavior is growing rapidly (Christelis et al., 2012; Lusardi and Mitchell, 2014; Palloni, 2006; Case, Fertig and Paxston, 2006; Cole and Shastry, 2009). Existing literature contains several well-documented relations between childhood characteristics and financial asset ownership. The most significant relations can be placed in roughly three childhood characteristics: education, cognition and health.

Education, cognition and health, all show significant direct influences on financial asset ownership. These influences persist after controlling for adult equivalents and have been found in multiple papers using multiple databases (Christelis et al., 2012; Lusardi and Mitchell, 2014; Moschis, 1985; Palloni, 2006; Doyle, Harmon, Heckman and Tremblay, 2009; Brounen, Koedijk, and Pownall, 2016). More recently, research is shifting to examine other childhood characteristics. An example of such research is the paper by Christelis et al. (2012). They focus on the relation between childhood SES and financial asset ownership. To the best of my knowledge, this is the only study that examined this relationship.

5 proxy is a combination of two variables, the amount of rooms in the residence at 10 years of age and the amount of books therein. Together, the variables should approximate actual SES (Christelis et al., 2012). The existing literature however, shows no consensus on such a proxy. Other papers, such as one by Mazzonna (2014), use other variables such as breadwinner occupation to approximate SES.

Existing literature might not reach a consensus on SES proxy, but it does support the reasoning behind the found relationship (Kajonius and Carlander, 2017; Campbell, 2006; Vissing-Jorgensen, 2003; Palloni, 2006). Childhood SES has possible direct and indirect relations with financial asset ownership. Childhood characteristics, such as SES, are known to influence adult characteristics that are related to financial asset ownership (Vissing-Jorgensen, 2003; Shore, 2011; Charles and Hurst, 2003). Childhood characteristics are known to shape personality traits, beliefs and values that can directly and indirectly influence financial asset ownership probabilities (Shore, 2011; Cole and Shastry, 2009; Plomin and Bergeman, 1991). As inferred by Christelis et al.(2012) the direct and indirect relations are likely to exist jointly, but in this thesis, both relations will be tested separately.

2.2. T

SES

6 combination of both nature and nurture and thus both need to be taken into account when examining the possible effect of childhood SES.

One of the most important personality traits regarding financial asset ownership that is known to be shaped during childhood, is financial risk tolerance (Campbell, 2006; Christelis et al., 2012; Bateman and Munro, 2005; Polderman, Benyamin, De Leeuw, Sullivan, Van Bochoven, Visscher and Posthuma, 2015). Financial risk tolerance is a thoroughly researched phenomenon and is known to be shaped by both nature and nurture (Polderman et al., 2015). Financial risk tolerance is strongly related to current SES and SES background (Campbell, 2006; Christelis et al., 2012). A high SES background suggests more risk tolerant parents making it more likely to have inherited a higher level of risk tolerance (Polderman et al., 2015). Children with risk tolerant parents also experience a more risk tolerant upbringing, resulting in additional reinforcement of this relation (Hryshko et al., 2011; Shore, 2011).

Financial asset ownership probabilities encompasses more than just personality traits. Traits such as risk tolerance do have the ability to increase desire for financial risk taking, but this desire can be dimmed as a result of other characteristics and events. Humans are mostly risk averse and are known to weight and compare risks from multiple sources (Campbell, 2006; Heaton and Lucas, 2000). Risks faced in seemingly unrelated areas of life, can have significant resonance on financial risk taking. As a result, actual risk taking is reliant on both risk preference and current risk exposure.

Exposure to risk can come from multiple sources. Some of these sources are known to be influenced by background. Risks such as income risk and health risk have well documented relations with childhood cognition, health and SES (Luo and Waite, 2005; Case et al., 2005; Breuer et al., 2014; Doyle et al., 2009). Childhood SES thus exerts an influence on total risk exposure which in turn is strongly related to financial asset ownership.

7 has a strong and well-documented positive relation with the ownership probabilities of multiple financial assets (Campbell, 2006; Vissing-Jorgensen, 2003; Christelis et al., 2012). Wealth is not only essential to financial asset ownership, it is also strongly correlated to the upside potential of owning financial assets.

Owning financial assets comes with fixed and variable costs, proportionally the total costs will decrease when the amount invested increases (Vissing-Jorgensen, 2003; Campbell, 2006). Furthermore, obtaining financial assets creates additional complexity in the financial environment (Vissing-Jorgensen, 2003). Financial assets need to be understood, picked and monitored. Owning financial assets also creates complexity in seemingly unrelated financial areas such as tax returns and loan applications. To incentivize investing, potential gain, which is highly reliant on portfolio size, must offset the experienced costs and inconveniences (Vissing-Jorgensen, 2003; Campbell, 2006; Grohmann et al., 2015). As a result, the decision to obtain financial assets is highly reliant on household wealth.

Income and wealth are known to be influenced by multiple adult and childhood characteristics such as education, health, cognitive abilities and (childhood)SES (Case, Fertig and Paxson, 2005; Furnham and Cheng, 2013; Kajonius and Carlander, 2017; Campbell, 2006; Charles and Hurst, 2003; Solon, 1992). Income and wealth exert high cross-generational persistency. This cross-generational persistency is mostly due to (financial) support from parents and other family members, which is related to SES background (Charles and Hurst, 2003). Cross-generational persistency of wealth also has more sophisticated aspects (Furnham and Cheng, 2013).

8 As briefly mentioned childhood SES is positively related to human capital during adulthood (Case et al., 2005; Doyle et al., 2009; Esping-Andersen, 2008). Human capital is extremely important in incentivizing financial asset ownership. Human capital helps overcoming the complexities created by financial asset ownership and is positively related to the performance of the financial assets (Gaudecker and Von, 2015; Grohmann et al., 2015; Vissing-Jorgensen, 2003). Past performance is known to be a great incentive or disincentive to pursue further financial asset ownership (Christelis et al., 2012; Grohmann et al., 2015).

It is clear that childhood SES can exert a significant indirect influence on financial asset ownership. Childhood SES is linked to higher human capital and higher levels of wealth, both known to increase upside potential and reduce the proportional costs and complexities. Furthermore, higher childhood SES is linked to lower risk exposure throughout life and higher risk tolerance. This diminishes disincentives due to excess risks.

However, as all argued relations are indirect, they are likely to lose a lot of strength. But, due to the sheer number of relations I do still expect the aggregated effect to be statistically significant. To formally test the relation, the following hypothesis is formed:

H1: Childhood SES possesses a positive indirect relation with financial asset ownership probabilities through relations with several relevant adult variables.

If hypothesis one proves to be supported, additional testing will be done on the individual drivers of the indirect relation.

2.3. T

9 Childhood characteristics, such as SES, can shape an individual’s believes and attitudes without affecting measureable adult variables. SES is positively related to the level of financial socialization experienced during childhood (Lusardi and Mitchell, 2014; Moschis, 1985). Financial socialization is the day to day contact with money and financial information through parents, education, media and other means. Higher levels of financial socialization are known to increase the value attached to financial status later in life (Kajonius and Carlander, 2017). A higher value attached to financial status will increase the desire to obtain wealth. The most well-known instruments used to increase (financial) wealth are financial assets (Campbell, 2006; Ellingsen and Johannesson, 2008). This could ultimately result in a direct relation between childhood SES and financial asset ownership.

Childhood SES can also have a lasting effect on the social circle in adulthood (Judge et al., 2009). A higher SES background is linked to a higher SES social circle. Current SES is positively related to financial asset ownership and as a result higher SES social circles will be more investment heavy. The well-known “keeping up with the Joneses” is very applicable to household finance. This can ultimately result in financial asset ownership probabilities being influenced by social circle which in turn is influenced by childhood SES (Hong, Kubik and Stein, 2004; Ivković and Weisbenner, 2007; Brown, Ivković, Smith and Weisbenner, 2008; Christelis, Jappelli and Padula, 2010; Ellingsen and Johannesson, 2008). As the SES of a social circle is not objectively measurable, this relation will show as “direct”.

As discussed in the previous section, childhood SES is negatively related to multiple risks faced throughout life (Luo and Waite, 2005; Case et al., 2005; Breuer et al., 2014; Doyle et al., 2009). Residual risks that are not included in other variables might show as a “direct” relation between financial asset ownership and childhood SES. Examples of such risks would be future unemployment or health risks. Expectations of future risks can severely decrease current financial risk taking in the form of financial asset ownership (Heaton and Lucas, 2000).

10 (2012) are based on an indirect relation. Furthermore, the proxy used by Christelis et al. (2012) is not unanimously supported by the literature (Mazzonna, 2014). The found relationship very likely to exist, but due to the raised uncertainties it is very interesting to review this. To formally test this, the following hypothesis is formed:

H2: Childhood SES will show a positive direct relation with financial asset ownership probabilities.

3. D

In this section, the collection and treatment of the data is discussed.

3.1. D

In this thesis, the SHARE database is used. The SHARE database includes variables on a broad spectrum of categories, among which financial, cognitive and social characteristics. The data is based on an extensive questionnaire. SHARE collects data on households from 15 European countries (Austria, Belgium, Czech, Switzerland, Germany, Denmark, Estonia, Spain, France, Italy, Israel, Luxembourg, Netherlands, Sweden and Slovenia). All participants of SHARE are at least 50 years of age. SHARE is a recurring survey, every three to four years a new wave is released. This creates an opportunity to follow participants over time.

The SHARE database is unique in its size and multinational nature. As a result, SHARE is very suitable for socio-economic research on Europe.

SHARE contains survey based retrospective data which could result in biased and unreliable variables. Havari and Mazzonna (2015) investigated this issue and concluded that the data is of sufficient quality to be used for research purposes. A more thorough exposition on the biases and disadvantages of the SHARE database and how these are safeguarded will be discussed later on.

3.2. S

11 3.2.1. CHARACTERISTICS OF THE SAMPLE

This thesis aims to test the relation between childhood SES and financial asset ownership. The sample selection is tailored to obtain as many quality entries of the childhood SES variable while creating an unbiased and representative sample. Childhood SES is based on the following question found in the childhood questionnaire:

“Think about your family when you were growing up, from birth to age 15 included. Would you say your family during that time was pretty well off financially, about

average, or poor?”.

The sample contains a fair and even distribution of childhood SES backgrounds. Of the final sample, 18.2% of the observations experienced a high SES childhood, 64.2% of the observations experienced an average SES childhood and 16.7% of the observations experienced a low SES childhood. The remainder experienced varying childhood SES.

This thesis is based on the groundwork done by Christelis et al. (2012) who use a proxy for childhood SES. It is interesting to compare the proxy to the actual childhood variable. To do this, the proxy is recreated. The questions used to obtain the proxy are:

“Approximately how many books were there in the place you lived in when you were 10? Do not count magazines, newspapers, or your school books”

and

“We would like to find out more about the accommodation where you lived when you were 10 years old. How many rooms did your household occupy in this

accommodation, including bedrooms but excluding kitchen, bathrooms, and hallways?”

12 some question marks are raised. The proxy dummy on books seems to grossly overestimate the amount of observations that experienced a high childhood SES. Additionally, the correlation between both proxy variables and actual childhood SES is 0.27. The relatively low correlation, the lack of consensus in the literature and the apparent overestimation of high childhood SES observations raises some serious doubts on the quality of the used proxy. To further elaborate on the proxy effects and because the variables have previously shown a significant relation with financial asset ownership, both variables have been included in the regressions (Christelis et al., 2012).

More detailed specifications on actual childhood SES and the proxy variables can be found in Table 1.

3.2.2. TREATMENT OF THE SAMPLE

Due to several characteristics of the SHARE data, transformations have to be done before it is suitable for research purposes. SHARE presents financial data on household levels, based on one financial respondent. This requires the testing to be carried out on the household level as well. As a result, individual variables have to be aggregated over households. The aggregating is done similar to Christelis et al. (2012), the entries are aggregated per household, maximizing childhood SES, (childhood) cognition, education, (childhood) health and risk tolerance.

Blindly aggregating households has distinct disadvantages. The “cocktail” of values is in essence random and does not take into account financial decision hierarchy. An additional approach would be to use the characteristics of the main respondent, assuming that the financial respondent is more likely to have high-ground in the financial decision making hierarchy. However, most households make joint decisions on financial matters. And thus, blindly omitting non-financial respondents is also unlikely to approximate true financial decision hierarchy (Bateman and Munro, 2005; Breuer et al., 2014). Additionally, the dropped observations will induce possible bias as they cannot be considered random.

13 The questions in SHARE are plain in language and easy to comprehend. Missing entries can be seen as random and should be unrelated to personal characteristics such as intelligence and education2. The observations are independent and the dropped observations due to missing variables are considered random. In total 14,293 households are dropped from the sample, most of which due to missing childhood SES entries (9,000+).

No evidence is found of trends in the dropped observations. Similarly, examining the final sample (Table 1), no evidence is found of a biased or unrepresentative sample. The financial sample is balanced with roughly similar amount of observations per country and a fair representation of all SES backgrounds.

3.3. D

SHARE collects data on five financial asset classes: stocks, mutual funds, bonds, individual retirement accounts (IRA) and life insurance. In this thesis, a distinction is made between liquid assets (stocks, mutual funds and bonds) and illiquid assets (IRAs and life insurance). The respective ownership ratios of the liquid financial assets are 11.9% for mutual funds, 11.1% for stocks and 4.8% for bonds. The illiquid financial assets are more popular with ownership ratios of 24.6% for IRAs and 24.5% for life insurance. In total about 46.1% of the households hold any form of financial asset.

The focus of this thesis will lie on the liquid financial assets. Obtaining liquid financial assets is a conscious decision, obtaining illiquid financial assets might not be as they are known to be part of other formal agreements (Kitao, 2010). As the liquid asset classes might contain cross-contamination3, an aggregated liquid asset class is created (Campbell, 2006; Christelis et al., 2010). In the final sample, 20.8% of the households owns at least one type of liquid financial asset. In line with expectations, we see huge fluctuations in ownership percentages between countries (Breuer et al., 2014).

2

Some variables require extra care. These variables are discussed in more depth later on.

14

Table 1: Descriptives wave 5 (mean per household)

Austria Germany Sweden Netherlands Spain Italy France Denmark Switzerland Belgium Israel Czech Luxembourg Slovenia Estonia Stock ownership 6.7% 11.1% 37.5% 9.8% 5.2% 4.2% 10.4% 31.6% 21.0% 14.9% 10.0% 4.4% 10.4% 8.2% 1.8%

Mutual fund ownership 8.2% 14.2% 39.0% 11.9% 4.4% 7.3% 12.0% 15.5% 19.9% 20.4% 14.9% 6.5% 18.2% 2.7% 3.0%

Bonds ownership 2.7% 9.2% 8.9% 2.5% 0.9% 13.9% 1.4% 5.5% 10.3% 9.8% 9.8% 1.8% 4.5% 0.8% 0.2%

IRA ownership 10.1% 25.9% 50.8% 10.2% 14.9% 3.0% 29.2% 52.1% 36.5% 39.5% 22.8% 50.7% 19.4% 5.7% 9.4%

Life insurance ownership 27.7% 33.1% 44.7% 26.9% 12.7% 7.7% 41.6% 33.5% 18.3% 32.1% 23.2% 22.9% 20.9% 32.9% 5.2%

Liquid asset ownership 14.3% 24.8% 58.1% 17.0% 8.8% 20.0% 19.1% 40.0% 34.7% 31.5% 21.4% 10.1% 23.4% 9.8% 4.3%

Is an entrepeneur 4.0% 8.4% 19.4% 9.8% 11.1% 6.2% 5.2% 14.6% 10.9% 7.1% 6.9% 3.8% 4.3% 3.4% 4.4% Risk appetite (1-4) 1.31 1.36 1.78 1.36 1.18 1.34 1.40 1.79 1.39 1.37 1.51 1.41 1.32 1.19 1.16 Age 67.91 65.04 67.72 65.30 66.39 66.31 67.33 63.70 66.65 64.03 69.32 67.69 65.35 67.44 69.84 Is in a couple 53.4% 70.1% 68.9% 69.6% 75.1% 69.1% 59.1% 71.6% 69.2% 62.8% 67.7% 56.1% 69.6% 67.3% 51.5% Amount of children 2.00 1.92 2.30 2.25 2.05 1.78 2.17 2.22 2.00 2.11 2.99 2.14 1.96 1.93 1.99 ISCED level (1-6) 3.48 3.69 3.48 3.22 2.41 2.44 2.97 3.81 3.47 3.47 3.54 2.95 2.79 3.17 3.42 Health (1-4) 2.23 2.01 2.61 2.29 2.13 2.08 2.06 2.84 2.51 2.27 2.13 1.83 2.10 1.92 1.29 Verbal (1-5) 3.54 3.33 3.44 3.14 2.37 2.08 2.65 3.79 3.21 3.13 2.44 3.35 2.33 3.45 3.00 Numeracy (1-5) 4.05 3.80 3.87 3.99 3.10 3.35 3.55 3.98 4.06 3.72 3.84 3.71 3.58 3.41 3.55

Real wealth (median) 82250 83000 152265 118800 163974 178350 184428 147493 219673 213000 187864 39970 550000 124435 50274

Financial wealth (median) 6807 18000 46447 16518 4000 3000 16500 49620 90290 43379 6471 4146 25000 500 650

Household income (median) 25485 30680 47440 37294 18480 18596 27353 44254 73289 34924 24222 8299 76350 12476 7452

In financial distress 19.5% 24.8% 13.3% 19.1% 39.1% 59.9% 31.5% 11.2% 12.2% 29.0% 53.5% 43.0% 17.4% 57.0% 59.9%

10+ books at age 10 69.6% 76.9% 86.8% 78.2% 62.2% 49.1% 65.1% 88.3% 79.8% 75.4% 81.5% 86.7% 67.1% 56.2% 73.6%

Rooms at age 10 3.69 4.37 4.19 5.12 4.25 3.54 4.55 5.27 5.30 5.66 3.21 2.85 5.25 2.71 2.70

Math at age 10 (1-4) 2.42 2.45 2.68 2.47 2.24 2.35 2.34 2.63 2.47 2.58 2.58 2.48 2.45 2.37 2.51

Language at age 10 (1-4) 2.52 2.47 2.66 2.41 2.24 2.32 2.43 2.65 2.52 2.59 2.69 2.54 2.41 2.32 2.54

Had childhood health issues 20.4% 25.0% 13.5% 21.8% 12.5% 13.0% 21.2% 14.1% 19.8% 22.9% 20.7% 24.7% 18.0% 18.3% 27.5%

High childhood SES 9.4% 20.8% 25.8% 27.7% 12.4% 11.9% 22.0% 14.4% 14.9% 30.4% 17.8% 10.5% 27.2% 5.3% 12.2%

Average childhood SES 66.1% 61.6% 61.8% 58.5% 68.0% 62.5% 59.6% 75.8% 67.5% 56.8% 62.9% 65.5% 53.9% 64.0% 57.4%

Low childhood SES 23.2% 16.6% 9.5% 11.1% 18.9% 24.9% 17.5% 9.0% 17.0% 11.8% 18.5% 23.0% 16.8% 30.3% 29.4%

Varied childhood SES 1.3% 1.0% 2.9% 2.7% 0.7% 0.7% 0.9% 0.8% 0.6% 1.0% 0.8% 1.0% 2.1% 0.4% 1.0%

Observations 4,382 5,752 4,556 4,168 6,708 4,750 4,506 4,146 3,051 5,640 2,599 5,643 1,610 2,958 5,752

15

3.4. I

To make sure the found relations between childhood SES and financial asset ownership are correct, it is of utmost importance to include sufficient independent variables. This section discusses the vector of control variables used in the specification of the latent variables.

3.4.1. HUMAN CAPITAL

Human capital is the combination of cognitive abilities and education. Research has shown two types of cognitive abilities to be positively related to financial asset ownership. These cognitive abilities are language proficiency and numerical abilities (Christelis et al., 2012; Christelis et al., 2010; Hanushek and Woessmann, 2008). Education has well-documented relations with risk tolerance, risk exposure and financial asset ownership (Cole and Shastry, 2009; Hryshko et al., 2011).

SHARE contains variables on both language proficiency and numerical abilities. The numerical variable is based on objective testing and compares results within the sample. The results are reported on a five point scale that ranges from bad to good. SHARE contains two variables on language proficiency, both of the variables have significant drawbacks. In existing literature, preference is given to a reading score (Christelis et al., 2012; Hanushek and Woessmann, 2008). This reading score is a self-reported variable. Self-reported variables are known to possess unknown biases that are hard to control post hoc (Conway and Lance, 2010). Alternatively, SHARE contains a verbal fluency score. This is an objective test, but the test is done in native languages. Not all languages are equally complex and thus a bias could be reflected in the fluency score. As the nature of the fluency test bias will be known, and partly refuted by the inclusion of country dummies, this variable is preferred. The variable originally is based on a 50 point scale. This is transformed to a five point scale, the scale is based on results within the sample and thus reported in the same manner as numeracy test.

Education is measured using the ISCED97 standard. ISCED97 is a uniform measure of education levels maintained by UNESCO4 and can be seen as objective. ISCED97 measures the highest achieved level of education and is uniform across countries.

16 3.4.2. ADULT SES

Adult SES is split into three variables collected on household level: income, financial wealth and real wealth. Financial wealth is the net amount of financial assets or debt. Real wealth is the net amount of non-financial assets and non-financial possessions. Real wealth, financial wealth and income are all important predictors of financial asset ownership and financial risk appetite (Campbell, 2006; Vissing-Jorgensen, 2003; Christelis et al., 2012).

SHARE contains data on all three SES variables and contains multiple variables on household income. In this thesis, the more sophisticated, calculation based measure of income is preferred. This income is based on multiple specific questions that have been combined by computations. The alternative income variable is a self-reported monthly income estimate and might include more bias (Conway and Lance, 2010). In practice the variables are remarkably similar.

3.4.3. FINANCIAL RISK TOLERANCE

Financial risk tolerance is known to be a strong predictor of financial asset ownership (Christelis et al., 2012; Breuer et al., 2014; Campbell, 2006). SHARE contains a question dedicated to financial risk tolerance:

“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 being:

“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 4. Not willing to take any

financial risks”

This question is easy to understand, yet it gives a thorough image of actual financial risk tolerance. As a result, the answers can be deemed unbiased and accurate.

17 substantial financial risk. Given the high average age of the sample, this level of risk-aversion is in line with expectations (Hryshko et al., 2011; Ameriks and Zeldes, 2004; Campbell, 2006).

3.4.4. UNRELATED RISK

Actual financial risk taking is reliant on more than preference only. Actual risk taking is highly reliant on seemingly unrelated risk exposure and risk altering conditions. An example of a risk altering condition is being in a relationship. Jointly made financial decisions are known to be significantly more risk-averse than individual risk preferences would suggest (Bateman and Munro, 2005). As a result, couples take on fewer financial risk. A similar relation is found between having children and financial risk taking (Christelis et al., 2012).

Several other characteristics are known to increase risk exposure throughout life. As financial asset ownership is negatively related to total risk exposure, it is important to include sufficient documented sources of risk exposure in the vector of control variables (Campbell, 2006; Heaton and Lucas, 2000).

Firm-ownership induces significant exposure to income risk (Campbell, 2006). Age also creates significant exposure to health and income risk (Hryshko et al., 2011; Ameriks and Zeldes, 2004). Furthermore, as age increases and lifespan decreases, time to enjoy the merits of financial risk taking will decrease and thus risk taking incentive will decrease. As the sample includes a relatively wide age range, the effects are likely to be non-linear. To safeguard this, the square of age is added to the vector of control variables.

18 risk (Luo and Waite, 2005; Case et al., 2005; Breuer et al., 2014; Doyle et al., 2009). These variables are already included in the vector of control variables, however, human capital is also known possess a direct relation with financial asset ownership (Hanushek and Woessmann, 2008; Christelis et al., 2012).

In an effort to include residual financial risk exposure, a dummy variable on financial-distress has been created. This variable takes the value of one if a household had some or significant trouble meeting financial obligations any time in the last 12 months.

3.4.5. CHILDHOOD VARIABLES

Multiple childhood variables are known to have a direct effect on financial asset ownership probabilities.

Childhood cognitive abilities and childhood health both impact financial asset ownership probabilities after controlling for adult equivalents (Hanushek and Woessmann, 2008; Christelis et al., 2012).

Furthermore, Christelis et al. (2012), found significant relations between the number of rooms in the residence at age 10 and the books therein. Although I do not deem this a suitable proxy for childhood SES, the variables did show a direct relation with financial asset ownership and are thus included.

3.5. T

To make sure the data is suitable to use in a logistic regression and to obtain accurate results, some transformations have been done that have not been mentioned yet.

All dichotomous variables have been transformed to dummy coding. Childhood SES and country variables have also been transformed to dummy coding as the effects are not expected to be proportional. Based on the age variable, five birth cohort dummies are created to test possible childhood macro-economic shocks.

19 preferred over a natural logarithm as both variables can have negative entries. The income variable cannot have negative entries and is transformed using the natural logarithm. The amount of rooms is transformed into a five-point ordinal scale.

4. E

In this section, the model choice, the weaknesses and strength of the models and the assumptions that need to be met to obtain an accurate conclusion will be discussed.

4.1. H

,

(

)

SES

The aim of this thesis is to test the effects of childhood SES on the ownership probabilities of several financial asset classes. The decision making process is a latent variable and has to be judged by the outcome, financial asset ownership. Due to the dichotomous nature of financial asset ownership, it is not possible to use a linear regression. It is however, possible to specify a linear latent variable 𝑦_𝑖∗ to approximate the decision making process. The latent variable 𝑦_𝑖∗ _{can be specified as} follows

𝑦_𝑖∗ _{= 𝛼 + 𝛽𝑐ℎ𝑆𝐸𝑆 + 𝛽𝑋𝑖 + 𝜖 (1)}

where 𝛼 is a constant, 𝑋𝑖 is a vector of control variables discussed in the data section and 𝜖 is the error term.

The latent variable 𝑦_𝑖∗ can be used to estimate probabilities on the outcomes of the observable variable 𝑦𝑖, actual financial asset ownership. The observable variable 𝑦𝑖 can be seen as a product of the latent variable 𝑦_𝑖∗ taking the following values

𝑦𝑖 = {1, 𝑖𝑓 𝑦𝑖 ∗_{> 0} 0, 𝑖𝑓 𝑦_𝑖∗≤ 0

(2)

To estimate 𝑦_𝑖∗ a logistic regression model is used. This method is preferred over a probit model based on a better goodness of fit5_.

20 When employing a logistic regression, some assumptions need to be met to assure the accuracy of the results. As my sample is large and the variables are independent, most assumptions are met by definition. However, assumptions as linearity with log-odds and absence of multicollinearity need to be formally tested. Using the Box-Tidwell test, several variables that exhibited non-linear relations with the log-odds have been identified and subsequently these variables are transformed as described earlier. No evidence of multicollinearity6 is found. All assumptions are met and the conclusion can be deemed accurate.

4.2. A

To elaborate on some interesting findings in the hypothesis testing, additional testing has been done. These additional tests require different testing methods. The additional testing aims to find relations between personal characteristics and is thus a different, non-aggregated, sample is used. Similar to the original testing I will focus on the nature and significance and not on the marginal effect.

4.2.1. ORDINAL VARIABLES

Several of the additional tests have been done on ordinal dependent variables. To test the ordinal dependent variables, ordered logistic regressions are employed. It is not possible to observe the processes that determine the ordinal variables. These processes will be approximated using a latent variable 𝑦_𝑖∗. The specification of the latent variable 𝑦_𝑖∗ _is

𝑦_𝑖∗= 𝛽𝑐ℎ𝑆𝐸𝑆 + 𝛽𝑋𝑖 + 𝜖 (3)

where 𝑋𝑖 is a vector of control variables and 𝜖 is the error term. The used method does not estimate a constant and constructing a constant post hoc is not necessary to obtain the desired conclusions.

The observable variable 𝑦𝑖 is ordinal and can be seen as a product of 𝑦𝑖∗ taking the following values

6

21 𝑦_𝑖 { 1, 𝑖𝑓 𝑦_𝑖∗ ≤ 𝑢1, 2, 𝑖𝑓 𝑢1 < 𝑦𝑖∗ ≤ 𝑢2, ⋮ 𝑁, 𝑖𝑓 𝑢𝑁< 𝑦𝑖∗ (4)

where 𝑢𝑁 is the 𝑁𝑡ℎ cutoff point. The cutoff points are not important to the conclusion and thus not reported.

The advantage of the ordered logistic regression is the simplicity of the assumptions that are in line with a normal logistic regression described above. The ordered logistic regression however, has an additional proportionality assumption that does not uphold7. This assumption is important to discuss marginal effects, in thesis I do not look at the marginal effects and thus this violation poses no problem.

4.2.2. CONTINUOUS VARIABLES

Lastly, testing has been done on continuous dependent variables. These tests are done on a linear regression model using OLS.

The continuous dependent variables 𝑦𝑖 have been tested using the following equation

𝑦𝑖 = 𝛼 + 𝛽𝑐ℎ𝑆𝐸𝑆 + 𝛽𝑋𝑖 + 𝜖 (5)

where 𝛼 is a constant, 𝑋𝑖 is a vector of control variables and 𝜖 is the error term.

Due to heteroscedasticity, the linear regressions have been done using robust standard errors. Other Gauss-Markov assumptions are not violated and thus the results can be deemed accurate.

5. E

5.1. T

In this section, the testing of the hypothesis, the results and the interpretations will be examined. Relations between childhood SES, a vector of control variables and multiple financial asset classes will be discussed. This thesis focusses on the nature

22 and significance of the found relationships, for that matter coefficients will be reported. This thesis does not discuss marginal effects.

The multinational nature of SHARE will result in country based biases. To safeguard this and to examine country specific problems, a vector of country dummies has been included in all regressions. Sweden is the base country as Sweden is present in all waves and thus can be used in both hypothesis- and robustness testing.

5.2. H

5.2.1. THE INDIRECT INFLUENCE OF CHILDHOOD SES

The indirect relation between childhood SES and financial asset ownership has been tested controlling for childhood variables on cognition, health and SES, birth cohort effects and country based effects. The relations have been tested using a logistic regression and the regression coefficients on both the individual and aggregated financial asset classes can be found in Table 2a and Table 2b.

Taking a look at the vector of control variables, we can establish that they are mostly in line with existing literature. Cognition is positively related to all types of financial asset ownership. The vector of country dummies show significant relations with all financial asset classes. However, the nature of the relationships differ per financial asset class. For example, Germany and Italy show significant positive relations with bond ownership, but negative relationship with all other asset classes. This implicates that the problem on household financial asset ownership as proposed in the literature needs to be tackled based on country specific basis.

23

Table 2a: Regression results: the indirect influence of childhood SES on household financial asset ownership

Stocks Mutual funds IRA

Coeff. Std. Err. Coeff. Std. Err. Coeff. Std. Err.

Childhood healthissues 0.028 0.048 -0.009 0.046 0.069 0.038 *

Childhood math 0.333 0.024 *** 0.261 0.023 *** 0.166 0.019 ***

Childhood language 0.042 0.025 * 0.064 0.024 *** 0.115 0.020 ***

High books 0.567 0.058 *** 0.632 0.056 *** 0.497 0.044 ***

Rooms 0.093 0.018 *** 0.063 0.017 *** 0.054 0.014 ***

High childhood SES 0.201 0.048 *** 0.185 0.045 *** 0.103 0.040 ***

Low childhood SES -0.211 0.064 *** -0.158 0.060 *** -0.241 0.047 ***

Varied childhood SES -0.196 0.183 0.003 0.163 0.044 0.141

Austria -1.912 0.093 *** -1.764 0.087 *** -2.293 0.084 *** Belgium -1.257 0.077 *** -0.881 0.072 *** -0.810 0.068 *** Czech -2.410 0.104 *** -2.099 0.091 *** 0.139 0.065 ** Switzerland -0.754 0.083 *** -0.861 0.084 *** -0.724 0.078 *** Germany -1.466 0.076 *** -1.228 0.071 *** -1.353 0.067 *** Denmark -0.281 0.074 *** -1.274 0.085 *** -0.263 0.073 *** Estonia -3.271 0.135 *** -2.812 0.108 *** -2.252 0.077 *** France -1.469 0.092 *** -1.349 0.088 *** -0.975 0.075 *** Spain -2.117 0.107 *** -2.351 0.113 *** -1.897 0.079 *** Italy -2.268 0.134 *** -1.749 0.108 *** -3.629 0.156 *** Israel -1.589 0.121 *** -1.200 0.105 *** -1.261 0.097 *** Netherlands -1.704 0.096 *** -1.510 0.089 *** -2.615 0.096 *** Luxembourg -1.575 0.113 *** -0.930 0.094 *** -1.730 0.095 *** Slovenia -1.484 0.098 *** -2.757 0.149 *** -2.832 0.113 *** Generation1 0.083 0.065 0.074 0.063 2.076 0.057 *** Generation2 0.237 0.063 *** 0.189 0.061 *** 1.930 0.056 *** Generation3 0.251 0.063 *** 0.266 0.060 *** 1.342 0.056 *** Generation4 0.333 0.064 *** 0.327 0.061 *** 0.608 0.060 *** Constant -2.510 0.111 *** -2.273 0.107 *** -2.409 0.095 *** Observations 30,934 30,934 30,934 Log-likelihood -9,381 -10,074 -13,421 Pseudo R2 0.146 0.122 0.232

***, ** and * coefficients are significant at the 1%, 5% and 10% level respectively

A significant and positive relation between childhood SES and financial asset ownership is found. The aggregated indirect effects of childhood SES on risk exposure, higher education levels, better cognitive abilities, better health and higher financial risk tolerance, are strong enough to significantly influence financial asset ownership probabilities in adulthood. Hypothesis one is supported.

24

Table 2b: Regression results: the indirect influence of childhood SES on household financial asset ownership

Liquid assets Bonds Life insurance

Coeff. Std. Err. Coeff. Std. Err. Coeff. Std. Err.

Childhood healthissues 0.030 0.038 0.114 0.067 * 0.168 0.035 ***

Childhood math 0.292 0.019 *** 0.233 0.035 *** 0.133 0.018 ***

Childhood language 0.073 0.020 *** 0.108 0.036 *** 0.080 0.019 ***

High books 0.608 0.044 *** 0.460 0.079 *** 0.374 0.040 ***

Rooms 0.089 0.014 *** 0.093 0.025 *** 0.056 0.013 ***

High childhood SES 0.188 0.039 *** 0.170 0.067 *** 0.018 0.038

Low childhood SES -0.216 0.048 *** -0.273 0.091 *** -0.174 0.043 ***

Varied childhood SES -0.097 0.144 -0.194 0.281 -0.172 0.137

Austria -1.929 0.074 *** -1.030 0.147 *** -0.692 0.067 *** Belgium -1.105 0.066 *** 0.202 0.108 * -0.867 0.068 *** Czech -2.391 0.079 *** -1.474 0.163 *** -1.011 0.068 *** Switzerland -0.904 0.075 *** 0.272 0.121 ** -1.485 0.087 *** Germany -1.334 0.064 *** 0.232 0.102 ** -0.673 0.063 *** Denmark -0.762 0.071 *** -0.447 0.137 *** -0.823 0.074 *** Estonia -3.219 0.094 *** -3.709 0.388 *** -2.618 0.090 *** France -1.608 0.077 *** -1.727 0.217 *** -0.085 0.070 Spain -2.422 0.089 *** -2.064 0.239 *** -1.818 0.081 *** Italy -1.356 0.080 *** 0.932 0.111 *** -2.319 0.108 *** Israel -1.550 0.095 *** 0.227 0.138 * -0.937 0.095 *** Netherlands -1.917 0.081 *** -1.247 0.175 *** -1.029 0.075 *** Luxembourg -1.430 0.088 *** -0.566 0.168 *** -1.325 0.092 *** Slovenia -2.153 0.091 *** -2.092 0.273 *** -0.310 0.072 *** Generation1 -0.002 0.051 -0.530 0.093 *** 1.837 0.053 *** Generation2 0.172 0.050 *** -0.312 0.089 *** 1.657 0.052 *** Generation3 0.247 0.049 *** -0.058 0.084 1.086 0.053 *** Generation4 0.310 0.050 *** 0.193 0.083 ** 0.577 0.057 *** Constant -1.593 0.090 *** -3.901 0.163 *** -2.239 0.090 *** Observations 30,934 30,934 30,934 Log-likelihood -13,497 -5,249 -14,794 Pseudo R2 0.147 0.132 0.149

***, ** and * coefficients are significant at the 1%, 5% and 10% level respectively

The additional regressions contain a vector of childhood and adult control variables that has been discussed previously. All regressions include country dummies to examine cross-country differences.

5.2.1.1. RISK TOLERANCE

25

Table 3: Regression results: risk tolerance

Risk loving Risk loving

Coeff. Std. Err. Coeff. Std. Err.

Male 0.613 0.024 *** 0.618 0.024 *** Verbal skills 0.094 0.010 *** 0.089 0.010 *** Numeracy test 0.144 0.013 *** 0.143 0.013 *** Education level 0.213 0.010 *** 0.212 0.010 *** Squared Age 0.000 0.000 * Age 0.003 0.014 Has children -0.067 0.041 * -0.066 0.041 In a couple 0.127 0.030 *** 0.111 0.030 ***

Self reported health 0.167 0.012 *** 0.165 0.012 ***

Childhood healthissues 0.041 0.033 0.038 0.033

Childhood math 0.109 0.015 *** 0.111 0.015 ***

Childhood language 0.021 0.015 0.023 0.015

High books 0.235 0.030 *** 0.235 0.030 ***

Rooms 0.029 0.006 *** 0.029 0.006 ***

High childhood SES 0.114 0.034 *** 0.117 0.034 ***

Low childhood SES -0.169 0.031 *** -0.171 0.031 ***

Varied childhood SES 0.071 0.083 0.069 0.083

Austria -1.336 0.056 *** -1.330 0.056 *** Belgium -1.232 0.056 *** -1.228 0.056 *** Czech -0.714 0.051 *** -0.719 0.051 *** Switzerland -1.080 0.062 *** -1.075 0.062 *** Germany -1.268 0.052 *** -1.266 0.052 *** Denmark -0.242 0.055 *** -0.237 0.055 *** Estonia -2.056 0.062 *** -2.057 0.062 *** France -0.765 0.059 *** -0.765 0.059 *** Spain -1.809 0.068 *** -1.808 0.068 *** Italy -0.907 0.066 *** -0.915 0.066 *** Israel -0.671 0.064 *** -0.670 0.064 *** Netherlands -1.218 0.060 *** -1.215 0.060 *** Luxembourg -1.241 0.075 *** -1.240 0.075 *** Slovenia -1.714 0.069 *** -1.709 0.069 *** Generation1 0.604 0.041 *** Generation2 0.466 0.041 *** Generation3 0.360 0.039 *** Generation4 0.228 0.040 *** Observations 45,771 45,771 Log-likelihood -28,348 -28,337 Pseudo R2 0.124 0.124

***, ** and * coefficients are significant at the 1%, 5% and 10% level respectively

26 The regression shows a significant and positive relation between childhood SES and financial risk tolerance. The effect of childhood SES on risk tolerance can be seen as an important driver of the aggregated individual effect.

5.2.1.2. ECONOMIC MEANS

The cross-generational persistency of wealth and income has been tested using linear regressions (Table 4). Note that SHARE collects wealth and income on household levels. It is therefore extremely important to include a couple dummy, this also explains the very high coefficient of the dummy.

Table 4: Regression results: economic means

Real wealth Financial wealth Income

Coeff. Std. Err. Coeff. Std. Err. Coeff. Std. Err.

Male -9217 4165 ** 890 1711 1039 764 Verbal skills 12005 1784 *** 4079 693 *** -589 327 * Numeracy test 13913 2182 *** 5978 843 *** 2257 416 *** Education level 28575 1891 *** 11834 740 *** 2116 325 *** Squared Age -110 15 *** -21 6 *** 12 3 *** Age 17123 2098 *** 3506 764 *** -2035 415 *** Has children 22099 6899 *** -8194 2615 *** 1224 1441 *** In a couple 111583 4118 *** 31067 1730 *** 11219 849 ***

Self reported health 26261 2226 *** 9850 841 *** 2540 384 ***

Childhood health issues -4716 5761 -824 2247 39 1028

Childhood math 15298 2805 *** 4198 1116 *** -156 490

Childhood language -13713 2979 *** -1328 1270 779 482

High books 23422 4466 *** 6238 1573 *** 2842 876 ***

Rooms 12455 1400 *** 3931 560 *** 708 224 ***

High childhood SES 65582 7963 *** 18765 3253 *** 2303 1321 *

Low childhood SES -13417 4382 *** -1808 1612 -1225 836

Varied childhood SES 19579 16870 -2054 9362 1860 3001

Austria -72892 8306 *** -65212 2967 *** 7194 1481 *** Belgium 39981 10713 *** 3085 3837 6197 1625 *** Czech -151541 7354 *** -66929 2818 *** -29688 914 *** Switzerland 243777 22612 *** 126956 9690 *** 76001 4131 *** Germany -93088 8378 *** -41631 3064 *** 8471 1505 *** Denmark 26137 17297 45823 5446 *** 4960 1407 *** Estonia -111153 7706 *** -73521 2906 *** -29289 952 *** France 21894 10959 ** -3284 5857 2771 1693 Spain 88338 12105 *** -45781 3037 *** -10230 1139 *** Italy 51356 9129 *** -51127 3069 *** 849 1648 Israel 134859 14297 *** 69141 7811 *** -11678 1014 *** Netherlands -89111 8842 *** -26008 4081 *** -4241 1467 *** Luxembourg 499970 19656 *** 34866 8971 *** 111903 5763 *** Slovenia -63692 7821 *** -77132 2929 *** -21999 984 *** Constant -779027 74652 *** -179497 26692 *** 91334 14167 *** Observations 45,771 45,771 45,354 R2 0.145 0.146 0.508

27 Examining the results, the vector of control variables show relationships as found in the existing literature. Cognition, education and health are positively related to all economic performance measures. Interestingly childhood language shows a negative relation to real wealth. Childhood cognitive abilities are known to be related to adult occupation and thus earnings in two ways. Desiring a certain career can increase drive to do well at a subject in school and alternatively doing well at a subject in school can increase appeal to a certain career (Howard and Walsh, 2011). This will find resonance in income and considering the age of the sample, ultimately wealth (Huang and Pearce, 2013).

The results indicate significant cross-generational SES persistency especially among high SES participants. The results show a significant and positive relation between childhood SES and all three measures of economic performance. The found wealth-persistency is known to influence financial asset ownership and thus can be seen as a driver of the indirect relation of high childhood SES.

5.2.1.3. HUMAN CAPITAL

The relations between childhood SES, cognitive abilities and education have been tested using ordered logistic regressions (Table 5).

Taking a look at the vector of control variables we find results as suggested by the existing literature. Interestingly, a negative relationship is found between high childhood SES and numerical abilities. This will weaken the indirect effect of high childhood SES. No relationship is found between childhood SES and verbal abilities. Childhood SES is positively related to level of obtained education after controlling for cognitive abilities. Education is linked to better economic performance, less risk exposure and higher levels of financial asset ownership. This relation is likely to be an important drive of the aggregated indirect relation between childhood SES and financial asset ownership. Education is also strongly related to economic and thus the effect might be greater than expected at first sight.

28

Table 5: Regression results: human capital

Verbal skills Numeracy test Education

Coeff. Std. Err. Coeff. Std. Err. Coeff. Std. Err.

Male -0.142 0.018 *** 0.569 0.018 *** 0.280 0.018 *** Verbal skills 0.271 0.008 *** 0.204 0.008 *** Numeracy test 0.346 0.010 *** 0.362 0.010 *** Education level 0.200 0.008 *** 0.282 0.008 *** Squared Age -0.001 0.000 *** -0.001 0.000 *** 0.000 0.000 *** Age 0.127 0.010 *** 0.074 0.010 *** 0.012 0.010 Has children 0.126 0.030 *** -0.007 0.031 -0.131 0.031 *** In a couple 0.197 0.021 *** 0.111 0.022 *** 0.043 0.022 **

Self reported health 0.237 0.009 *** 0.141 0.009 *** 0.187 0.009 ***

Childhood healthissues 0.138 0.024 *** -0.041 0.025 * 0.135 0.025 ***

Childhood math 0.078 0.011 *** 0.445 0.011 *** 0.279 0.012 ***

Childhood language 0.117 0.012 *** -0.001 0.012 0.375 0.012 ***

High books 0.373 0.021 *** 0.247 0.021 *** 0.830 0.021 ***

Rooms 0.017 0.005 *** 0.011 0.005 ** 0.032 0.005 ***

High childhood SES -0.012 0.027 -0.097 0.028 *** 0.454 0.028 ***

Low childhood SES -0.018 0.021 0.022 0.022 -0.158 0.022 ***

Varied childhood SES 0.039 0.064 -0.040 0.065 0.227 0.067 ***

Austria 0.365 0.046 *** 0.714 0.047 *** 0.410 0.048 *** Belgium -0.425 0.046 *** -0.134 0.047 *** 0.226 0.050 *** Czech 0.200 0.044 *** 0.124 0.045 *** -0.201 0.046 *** Switzerland -0.413 0.052 *** 0.537 0.053 *** 0.337 0.053 *** Germany -0.145 0.043 *** 0.014 0.044 0.830 0.045 *** Denmark 0.177 0.051 *** -0.086 0.052 * 0.485 0.053 *** Estonia -0.110 0.043 *** -0.146 0.044 *** 0.858 0.045 *** France -0.651 0.049 *** -0.068 0.050 0.014 0.053 Spain -0.986 0.047 *** -0.707 0.046 *** -0.818 0.050 *** Italy -1.536 0.052 *** -0.202 0.053 *** -0.518 0.055 *** Israel -1.313 0.055 *** 0.304 0.057 *** 0.775 0.060 *** Netherlands -0.349 0.049 *** 0.551 0.052 *** -0.189 0.052 *** Luxembourg -1.229 0.057 *** 0.133 0.060 ** -0.556 0.063 *** Slovenia 0.646 0.049 *** -0.542 0.050 *** 0.629 0.051 *** Observations 45,769 45,769 45,769 Log-likelihood -64,208 -57,286 -59,757 Pseudo R2 0.119 0.111 0.136

***, ** and * coefficients are significant at the 1%, 5% and 10% level respectively

5.2.1.4. TOTAL INDIRECT RELATION

The drivers of the indirect influences seem to be as proposed in the literature. The drivers vary drastically per country and thus country specific measures would be more suitable than uniform Europe wide measures.

5.2.2. THE DIRECT INFLUENCE OF CHILDHOOD SES

29 regression and the regression coefficients on both the individual and aggregated financial asset classes can be found in Table 6a and 6b.

Taking a look at the vector of control variables, we can establish that the found relationships are mostly in line with existing literature. Significant positive relations are found between financial asset ownership and current SES, human capital, health and childhood cognition. Significant negative relations are found between financial asset ownership and the risk exposure variables: having children, being in a couple and experiencing financial distress. The relationships mostly hold for all asset classes but are the weakest for bond ownership. More information demanding asset classes (i.e. stocks and mutual funds) show the strongest relations. In contrast to the other asset classes, life insurance ownership shows a positive relation with the risk exposure variables. In contrast to the other financial assets, life insurance is risk reducing.

One control variable shows a rather surprising significant relationship with financial asset ownership. Higher levels of language proficiency during childhood is negatively related to stock ownership and aggregated liquid asset ownership during adulthood. As showed earlier, childhood cognitive abilities could be a predictor of career choice and occupation is known to influence financial asset ownership (Howard and Walsh, 2011; Xiao, 1996).

The vector of country dummies show significant relations with all financial asset classes. However, the nature of the relationships differ per financial asset class. The country dummies signal significant direct cultural after controlling for a broad scale of adult characteristics (Breuer et al., 2014). It shows that countries as the Netherlands and Austria experience a relatively high rate of underinvestment when taking into account the characteristics of its inhabitants.

30

Table 6a: Regression results: the direct effect of childhood SES on household financial asset ownership

Stocks Mutual funds IRA

Coeff. Std. Err. Coeff. Std. Err. Coeff. Std. Err.

Verbal skills 0.082 0.019 *** 0.074 0.018 *** 0.070 0.015 *** Numeracy test 0.132 0.027 *** 0.101 0.025 *** 0.022 0.021 Firm ownership 0.142 0.063 ** -0.175 0.064 *** 0.382 0.056 *** Education level 0.132 0.018 *** 0.097 0.018 *** 0.070 0.015 *** Squared Age 0.000 0.000 * -0.001 0.000 *** 0.001 0.000 *** Age 0.064 0.027 ** 0.106 0.026 *** -0.171 0.025 *** Has children -0.173 0.068 ** -0.155 0.064 ** 0.035 0.056 Financial distress -0.408 0.066 *** -0.463 0.062 *** -0.303 0.044 *** In a couple -0.283 0.057 *** -0.332 0.053 *** 0.018 0.044 Riskloving 0.715 0.029 *** 0.646 0.028 *** 0.266 0.026 **

Self reported health 0.041 0.023 * 0.025 0.022 0.046 0.018 ***

Income 0.204 0.033 *** 0.216 0.031 *** 0.184 0.024 *** Real wealth 0.097 0.009 *** 0.069 0.008 *** 0.019 0.005 *** Financial wealth 0.159 0.009 *** 0.205 0.010 *** 0.172 0.005 *** Childhood healthissues 0.051 0.052 0.008 0.050 0.091 0.041 ** Childhood math 0.116 0.027 *** 0.062 0.026 ** -0.010 0.022 Childhood language -0.067 0.028 ** -0.037 0.027 0.042 0.022 * High books 0.112 0.064 * 0.217 0.061 *** 0.128 0.049 *** Rooms 0.003 0.019 -0.017 0.018 -0.006 0.015

High childhood SES 0.021 0.052 0.036 0.049 -0.018 0.044

Low childhood SES 0.023 0.069 0.080 0.065 -0.036 0.051

Varied childhood SES -0.097 0.197 0.091 0.175 0.177 0.156

Austria -1.616 0.102 *** -1.466 0.095 *** -2.162 0.091 *** Belgium -0.998 0.086 *** -0.645 0.080 *** -0.706 0.077 *** Czech -1.548 0.121 *** -1.194 0.109 *** 1.197 0.083 *** Switzerland -0.922 0.093 *** -1.131 0.093 *** -1.065 0.086 *** Germany -1.142 0.085 *** -0.914 0.080 *** -1.304 0.075 *** Denmark -0.349 0.083 *** -1.539 0.093 *** -0.328 0.083 *** Estonia -2.119 0.150 *** -1.528 0.125 *** -1.247 0.092 *** France -1.199 0.101 *** -1.119 0.096 *** -0.759 0.083 *** Spain -1.291 0.117 *** -1.542 0.122 *** -1.416 0.090 *** Italy -1.586 0.145 *** -0.982 0.120 *** -3.280 0.163 *** Israel -1.315 0.136 *** -0.821 0.121 *** -0.809 0.111 *** Netherlands -1.389 0.104 *** -1.207 0.097 *** -2.706 0.104 *** Luxembourg -1.506 0.127 *** -0.852 0.108 *** -1.801 0.107 *** Slovenia -0.456 0.115 *** -1.728 0.160 *** -2.110 0.123 *** Constant -10.857 0.998 *** -11.706 0.963 *** 3.314 0.857 *** Observations 30,933 30,933 30,933 Log-likelihood -7,990 -8,639 -11,564 Pseudo R2 0.273 0.247 0.338

***, ** and * coefficients are significant at the 1%, 5% and 10% level respectively

31

Table 6b: Regression results: the direct effect of childhood SES on household financial asset ownership

Liquid assets Bonds Life insurance

Coeff. Std. Err. Coeff. Std. Err. Coeff. Std. Err.

Verbal skills 0.077 0.016 *** 0.006 0.027 0.044 0.014 *** Numeracy test 0.130 0.022 *** 0.086 0.037 ** 0.030 0.019 Firm ownership 0.068 0.058 0.072 0.089 0.322 0.051 *** Education level 0.124 0.015 *** 0.074 0.026 *** 0.044 0.014 *** Squared Age -0.001 0.000 *** -0.001 0.000 ** 0.001 0.000 *** Age 0.114 0.022 *** 0.123 0.037 *** -0.148 0.022 *** Has children -0.232 0.055 *** -0.246 0.087 *** 0.204 0.051 *** Financial distress -0.428 0.049 *** -0.293 0.092 *** -0.066 0.039 * In a couple -0.334 0.046 *** -0.283 0.078 *** 0.148 0.040 *** Riskloving 0.793 0.026 *** 0.247 0.042 *** 0.142 0.024 ***

Self reported health 0.034 0.019 * 0.014 0.032 -0.006 0.016

Income 0.209 0.026 *** 0.032 0.040 0.236 0.021 *** Real wealth 0.083 0.006 *** 0.045 0.011 *** 0.024 0.005 *** Financial wealth 0.204 0.008 *** 0.540 0.024 *** 0.069 0.003 *** Childhood healthissues 0.049 0.043 0.152 0.071 ** 0.175 0.037 *** Childhood math 0.079 0.023 *** 0.052 0.038 0.003 0.020 Childhood language -0.039 0.023 * 0.023 0.039 0.027 0.020 High books 0.182 0.049 *** 0.051 0.087 0.157 0.042 *** Rooms 0.002 0.016 -0.002 0.027 0.008 0.013

High childhood SES 0.017 0.044 -0.028 0.071 -0.069 0.040 *

Low childhood SES 0.028 0.054 -0.039 0.097 -0.042 0.045

Varied childhood SES -0.025 0.160 -0.206 0.297 -0.041 0.143

Austria -1.796 0.086 *** -0.366 0.156 ** -0.437 0.071 *** Belgium -0.954 0.078 *** 0.427 0.117 *** -0.755 0.072 *** Czech -1.685 0.097 *** -0.157 0.181 -0.396 0.078 *** Switzerland -1.284 0.087 *** -0.106 0.131 -1.699 0.091 *** Germany -1.129 0.075 *** 0.759 0.113 *** -0.532 0.068 *** Denmark -1.032 0.083 *** -0.580 0.142 *** -0.918 0.078 *** Estonia -2.161 0.111 *** -1.804 0.399 *** -1.957 0.098 *** France -1.523 0.088 *** -1.535 0.224 *** 0.162 0.075 ** Spain -1.752 0.101 *** -1.119 0.246 *** -1.465 0.087 *** Italy -0.602 0.096 *** 2.222 0.133 *** -2.001 0.113 *** Israel -1.290 0.115 *** 0.453 0.154 *** -0.587 0.101 *** Netherlands -1.793 0.092 *** -0.821 0.181 *** -0.921 0.079 *** Luxembourg -1.481 0.104 *** -0.491 0.182 *** -1.364 0.099 *** Slovenia -1.230 0.107 *** -0.420 0.285 0.273 0.082 *** Constant -11.752 0.821 *** -15.806 1.364 *** 2.312 0.756 *** Observations 30,933 30,933 30,933 Log-likelihood -10,975 -4,459 -13,957 Pseudo R2 0.306 0.263 0.197

***, ** and * coefficients are significant at the 1%, 5% and 10% level respectively