The effect of political preference on household portfolio decisions regarding the stock market

The effect of political preference on household portfolio decisions regarding the stock market

Martijn van den Bosch ^!,∗ , supervised by Dr. Viola Angelini ^!

! Faculty of Economics and Business, University of Groningen, The Netherlands

ARTICLE INFO

ABSTRACT

JEL classifications:

D14 G11 Keywords:

Household finance Political preference Limited

participation

This paper analyses the role of political preference in household portfolio decisions regarding the stock market, using a Dutch data set from the Longitudinal Internet Studies for the Social sciences (LISS). Results show that political preference is a significant predictor of household portfolio decisions regarding the stock market. Households with leftist political preferences are less likely to invest in the stock market, controlling for various other variables including wealth, trust, financial literacy and sociability. Furthermore, contrary to previous research, leftish political preference is also found to be associated with a smaller portion of the net financial assets being invested in the stock market, conditional on stock market participation. Results are consistent with the idea that resentment against the stock market is the key factor explaining the association between political preference and the stock market allocation and participation decision. Results are inconsistent with the alternative explanation of differences in risk or return expectations.

∗ Corresponding author at: Faculty of Economics and Business, student number s2353385, Nettelbosje 2, 9747 AE Groningen, NL.

E-mail address martijnvdbosch93@gmail.com (G.M. van den Bosch)

1. Introduction

The question why households, and wealthy households in particular, refrain from direct or indirect stock holding has been subject to extensive research over past decades (see Guiso and Sodini, 2013 for an extensive overview). This tendency of non-participation in the stock market by households is particularly interesting, given the historical return premium of equity in comparison to riskless assets (Mehra and Prescott, 2003). The determinants of stock market participation are subject to plentiful theoretical literature. Early research uses static models in which investors maximise expected utility, based on the risk return trade-off of assets (e.g. Arrow 1965).

However, well-established stylised facts (e.g. the absence of universal participation) are difficult to explain with this standard portfolio theory. This has led to empirical research into other household characteristics influencing the stock market participation decision.

Several household characteristics have been found to influence the likelihood of stock market participation. Stock market participation has been found to increase with higher levels of income, education and wealth (Vissing-Jørgensen, 2002; Hong et al., 2004; Calvet and Sodini, 2014). Gender, age and risk aversion are also related to the stock market participation decision (Bertaut, 1998; Hong et al., 2004; Almenberg and Dreber, 2015). Besides these traditional factors, recent studies discovered an interesting set of novel determinants to be of significant importance. These include social interaction, financial literacy, trust, health and IQ. (Hong et al., 2004; Rosen and Wu, 2004; Guiso et al., 2008, Grinblatt et al., 2011; Van Rooij et al., 2011)

Analysing the stock market participation as a consumer choice in addition to an investment decision, could lead to the discovery of further determinants of stock market participation. Fama and French (2007) show that asset holding is not solely based on anticipated monetary payoffs and state that investment assets are similar to consumer goods. Consequently, taste preference could help explain investment decisions.

The aim of the paper is to show that treating the stock market participation decision of households as a consumer decision rather than solely an investor decision could help explain limited participation rate in the stock market among households.

Specifically, this paper employs a model to show that an individuals’ decision to participate and the share invested in the stock market is influenced by personal values, in this paper measured by political preference. Key insight is that people with different political orientations also differ in their opinion regarding the stock market.

Leftist voters would feel that participating in the stock market is not in line with their

personal values. This discrepancy between their personal values and participation in

the stock market, results in utility pay-offs below the monetary pay-offs. Using this line of reasoning, this paper studies the effect of political preference on households’

decisions regarding the stock market. Consequently, the research question is following:

What is the effect of political preference on household decisions regarding the stock market?

This paper expands on the work of Kaustia and Torstila (2011), the first to analyse the relationship between political preference and household decisions regarding the stock market. Using a political preference scale, based on the quantification of party preference, their empirical analysis shows a negative effect of leftist political preference on the likelihood of stock market participation among individuals.

This paper uses longitudinal data of 2,440 Dutch individuals over four waves of the LISS panel to examine the effects of political preference on stock market participation. In line with previous research on household portfolio decisions, two distinct but related questions regarding portfolio decisions are addressed in this research. Firstly, what is the effect of political preference on household participation in the stock market? Secondly, what is the influence of political preference on the portion of the net financial portfolio of the household invested in the stock market, conditional on participation? Additionally, resentment against the stock market as the driver of the relationship between political preference and household decisions regarding the stock market, is explored using data from the World Values Survey and the European Values Study.

This paper finds that leftist political preference is associated with a lower

likelihood of stock market participation, controlling for traditional household

characteristics, trust, risk aversion, financial literacy and sociability. The effect of

political preference on stock market participation is strongly significant, as well as

economically important. Depending on the specifications of the control variables, a

change of one point to the left on a 0-10 left-right scale is associated with a 0.7 to 1.6

percentage point smaller likelihood of stock market participation. Political preference

is also found to influence the household allocation decision. An increase of one point

to the left on the aforementioned scale is associated with a 0.005 to 0.008 smaller

share of net financial assets allocated to the stock market. Additionally, strong

evidence is found for resentment against values, commonly associated with the stock

market among leftist voters. A negative relationship is found between leftist political

preference and trust in the two major groups of organisations active in the stock

market, banks and large companies. Results are therefore consistent with the idea that

resentment against the stock market is the driver of the relationship between political

preference and the stock market allocation and participation decision. The alternative explanation that the results are driven by differences in risk or return expectations is not consistent with the results found.

This paper is able to contribute to the current literature on the effects of political preference on stock market participation and the allocations decision regarding the stock market. This contribution is fourfold. Firstly, it is able to provide additional empirical finding on relationship between political preference and household decisions regarding the stock market, based on a large sample of household data from the Netherlands, a country not previously researched. Secondly, due to the richness of the LISS panel, several important alternative explanations such as risk aversion, trust, financial literacy, wealth, risk aversion and monetary return expectations are controlled for or ruled out. Thirdly, the results are found using a self-reported left- right rating, which could provide a better indicator of the real individuals’ value orientation as compared to party preference. This self-reported left-right rating is not influenced by an individuals’ evaluation of the capabilities of a party and therefore a more direct measure of individuals’ values. Fourthly and most importantly, contrary to previous research, leftist political preference is also found to be associated with a smaller portion of the portfolio being invested in the stock market. This contradicts previous research, which places the effects of political preference on household decisions regarding the stock market among the fixed participation costs. That once overcome, are of no influence on the allocation decision (Kaustia and Torstila, 2011).

The results are more in line with the work of Fama and French (2007) that regards financial assets as consumption goods. Subsequently, people can derive or lose direct utility from certain financial assets.

The remainder of the paper is organised as follows. Section 2 presents an overview of current literature on household decisions regarding the stock market and the link between political preference and investment decisions. Afterwards, the hypotheses are stated. Section 3 explains the empirical strategy. Section 4 presents the data sets used and the construction method of measures. Section 5 shows the empirical results. Section 6 discusses the outcomes of this paper and concludes.

2. Literature review and hypothesis 2.1 Stockholding puzzle

Limited stock market participation, especially amongst the wealthy, is one of the

great puzzles for economists today (Cambell, 2006). Although stock market

participation rates have increased steadily over past decades, participation remains

modest in the United States where about 49% of households directly or indirectly own stock. In European countries, the participation rate is even lower. In the Netherlands, this paper’s country of research, the direct stock market participation rate is only around 24% (Guiso et al., 2008).

The motivation of households to invest in risky financial assets has received a lot of academic interest. Researchers are interested in finding the reasons why the majority of households refrain from direct or indirect holdings of risky assets such as equity funds, stocks and long-term bonds; given the substantial equity premium. This phenomenon is known as the stockholding puzzle (Haliassos and Bertaut, 1995). The stockholding puzzle forms a part of the equity premium puzzle which refers to the large difference, around 3-9 % on a yearly basis in the long run, between the return of equity and government bonds (Mehra and Prescott, 2003). As households hold a large part of wealth in the economy, household stock market participation could have a substantial impact on the equity premium (Campbell, 1993; Heaton and Lucas, 1999).

Identification of the drivers of stock market participation could therefore help explain the equity premium puzzle.

Traditional models on stock market participation predict universal participation (e.g. Arrow, 1965). Traditional models assume that investors base their investment decisions on payoffs of these investments with risk preference as only preference parameter. These models predict that all households hold stocks, although they differ in their level of investment, due to differences in risk preference and wealth. This universal participation prediction can simply be rejected by taking into account limited household wealth. Some households are not able to accumulate sufficient wealth to set up a savings account, let alone participate in the stock market. Although this limited wealth can explain the decision of poorer households to stay out of the stock market, limited stock market participation among the wealthy remains unexplained.

2.2 Theory explanations and drivers of limited stock market participation

To address this limited participation in stock markets by households several

theories and determinants of stock market participation have been put forward. A

view that has gained considerable support is that households shy away from the stock

market due to some actual or perceived participation costs. These actual or perceived

participation costs have to be overcome in order to participate. Only if the perceived

equity premium and the planned investment are sufficiently large, relative to these

fixed participation costs, a household will invest. Important to note is that literature

finds that even small participation costs keep many households out of the stock

market. This holds especially if the marginal investor wants to invest a limited amount in the stock market (Haliassos and Michaelides, 2003; Vissing-Jørgensen, 2002).

Allen and Gale (1994) show that in a theoretical framework, entry costs limit participation. When examining this participation costs, Vissing-Jørgensen (2002) estimates that yearly fixed cost for stock market participation of $50 would well explain the decision to stay out of the stock market for half of the non-participants.

Other theoretical explanations are also put forward to explain limited stock market participation among households. King and Leape (1998) find information costs to be among the key drivers. Heaton and Lucas (1997) find human capital uncertainty to affect the participation decision. These theories provide a starting point for the identification of the determinants of participation decisions of households.

The empirical literature on stock market participation seeks to find the observable determinants of this participation decision. Several important variables influencing the participation decision of households have already been identified. Wealth has proven to be an important factor, since relative participation costs decrease as wealth increases. Wealthier households have more financial assets to invest, so fixed costs of participation are spread over a larger sum of investments (Calvet and Sodini, 2014).

Risk aversion also empirically explains participation, as more risk averse individuals are less likely to hold stocks (Bertaut, 1998). In line with information cost, education has a strong positive effect on the probability of stock market participation. This can be explained by arguing that education increases the understanding of the risk-reward trade-offs of the stock market, thereby reducing the fixed costs of participation as shown by Betraut and Starr-McCluer (2002). Christelis et al. (2010) find similar results when analysing the relationship between cognitive skills and direct and indirect stock market participation in 11 European countries. Household characteristics such as gender, race and age also empirically influence the participation decision (Hong et al., 2004; Almenberg and Dreber, 2015).

More recent literature has also focussed on novel determinants of the stock

participation decision, such as sociability, trust, health, cognitive ability and financial

literacy. Hong et al. (2001) find that sociability, increases participation. Individuals

that attend church or interact with their neighbours are found to have a higher

likelihood of participation. Guiso et al. (2008) find that increased levels of trust exert

positive effects on stock market participation. Conditional on participation, the share

of financial assets invested in stocks is also positively associated with trust. Grinblatt

et al. (2011) find that cognitive ability is a key driver of the participation decisions

and share of financial assets invested in the stock market. Individuals with higher IQ

are better at processing information thereby lowering their indirect participation costs.

When specifically examining financial literacy, Van Rooij et al. (2011) find that higher levels of financial literacy are associated with a increased likelihood of stock market participation. Several studies have also investigated the effect of health on financial portfolios. Rosen and Wu (2004) and Edwards (2008) show that self- assessed health status is an important determinant of financial portfolio decisions of households. Households with poor health are less likely to hold risky assets such as equity. Conditional on investing they also invest less in the stock market. Although these researches provide improved insight in the determinants of household participation, a considerable part of participation decisions still remains unexplained.

2.3 Political preference and investment decisions

Recently, the effect of individuals’ values on investment decisions has gained growing academic interest. The effect of individuals’ personal values on decisions regarding consumption and voting has already received interest from the academic community. Doran (2009), among others, shows that personal values, influence consumption decisions. Personal values are found to strongly drive political orientations (Clarke et al., 2004; Green and Hobolt, 2008). In liberal countries like the Netherlands, values even predict political orientation more strongly than demographic variables, making an individuals’ political orientation strongly influenced by value- expressive elements (Schwartz et al., 2011). Previous research has expanded the domain of research to the realm of household finance, partly triggered by recent developments such as the popularity of socially responsible investing.

Hong and Kostovtsky (2012) show that even for professional investors, personal values influence their investment decisions. They find that fund managers making campaign donations to the Democratic Party, hold less of their portfolio in socially irresponsible companies compared to fund managers who make contributions to the Republican Party or do not make contribution to political parties at all. This clearly shows the effect of individuals’ values, measured by political preference on investment decisions among sophisticated investors.

Kaustia and Torstila (2011) are the first to examine the effects specifically for household investment decisions, using four Finnish data sets on politicians and households at individual or zip code level. For their analysis they construct a right-left axis based on the quantification of party preference. Their research finds that value- expressive reasons measured by political preference, influence stock market participation among households and members of parliament. Their analysis finds leftist voters and politicians to be less likely to participate in the stock market.

Conditional on participation, no relationship is found between political preference and

the share of wealth allocated to the stock market. They therefore conclude that the effects of political preference are similar to variables linked to fixed costs. These variables influence the participation decision, but not the allocation decision. In their empirical analysis, they are able to control for education and several other relevant factors. However, possibly due to data limitations, they are not able to control for wealth, trust, financial literacy and sociability. This could be problematic, since financial literacy and wealth are also found to influence political preference (Powdthavee and Oswald, 2014; Montagnoli et al., 2016). Changwony et al. (2015) find similar results for the relationship between political preference and stock market participation in the United Kingdom. The abovementioned papers find a strong and significant relationship between the personal values of investors and their respective investment decisions.

The negative effect of leftist political preference on stock market participation of households is explained by a discrepancy between personal values and the stock market. Kaustia and Torstila (2011) find that Finish leftist voters perceive the stock market to be more harmful to society than their rightist counterparts. Furthermore, leftist Finnish voters view the stock market to be excessively worshipped and that Finnish society has succumbed too much to market forces and selfish profit seeking.

Leftist voters do not see investments in shares as a way to increase national wealth.

Hence, their analysis shows that there is antipathy among voters with a leftist view agianst the stock market, a so-called stock market aversion.

2.4 Political preference as a measure of personal values

Previous research has used party preference as a measure of personal values (Kaustia and Trostila, 2011; Changwony et al., 2015). However, this measure of personal values is not solely influenced by one’s political ideology. Research into political preference in the United Kingdom suggests that the expression of political preference in elections is in general determined by two sets of influences: political identification based on ideology and consumer voting based on individuals’

evaluation of the competence of a party (Clarke et al., 2004; Green and Hobolt, 2008).

This would mean that an individual with an outspoken rightist political ideology

could still prefer a more centrist party if they find the competences of right-wing

parties to be insufficient. Although political ideology remains an important

determinant, the importance of consumer voting has increased over the past decades

(Clarke et al, 2004). Another problem arises from the quantification of party

preference (Janda, 1982). The diversification of parties may not be sufficient to

ensure that one’s values closely match party preference. A measure that more closely resembles individuals’ political preference has not been employed so far.

2.5 Institutional background and stock market participation in the Netherlands

The Netherlands form an interesting country with respect to studying limited stock market participation of households, since the Dutch household portfolios have changed rapidly during the past decades (Alessie et al., 2000). In the 1980s, stocks and bonds were seen as exclusively for the rich. In general, households would buy a house and take a mortgage. When these households would acquire wealth, money would be used to pay-off mortgage. During the 1990s, this pattern changed; stocks, bonds and mutual funds were no longer seen as exclusively for the rich and the variety of investment products increased. These developments however still surpassed the majority of Dutch households that stuck to traditional saving instruments. This happened despite high returns, tax advantages and great availability of information on investing.

This decision is even more interesting in the light of the Dutch institutional setting. Financial markets in the Netherlands are well developed and households can acquire information on the possibilities of the stock market through an extensive number of channels. In general, financial institutions in the Netherlands offer a wide range of products together with free advice (Alessie et al., 2000). Also, the education system and average IQ in the Netherlands is amongst the highest in Europe. Dutch individuals should therefore be well able to process and evaluate the opportunities of investing in the stock market (Christelis, 2010; Marks, 2013).

2.6 Hypothesis stock market participation and allocation decision

Previous literature on political preference has shown that leftist political orientation reduces the likelihood of a household participating in the stock market.

The effects of political preference are expected to be similar to those found in previous research (Kaustia and Torstila, 2011; Changwony et al., 2015). Therefore the first hypothesis is formulated as follows:

H1: Leftist voters are less likely to participate in the stock market.

This paper is able to further examine the effects of political preference on

portfolio allocation over time on a large sample, which is especially interesting for

investigating the mechanisms through which political preference influences the share

of assets allocated to stocks. This enables to test whether the effect of political preference is limited to the participation decision or also influences the share invested in the stock market.

The effect of personal values on stock holdings could be categorical as found by Kaustia and Torstila (2011). They were the first and to the knowledge of the author, the only ones to look at the effect of political preference on the share invested in the stock market. This would mean that the participation decision is influenced by individuals’ values while the share invested in the stock market is not. Making the effect of individuals’ values on decisions regarding the stock market comparable to that the dietary choice of vegetarians who also base their choice on value expressive reasoning. Resentment against consumption of meat would predict a higher likelihood of being a vegetarian, but not influence the amount of meat consumed, conditional on being a meat eating person.

The other explanation is more in line with the theoretical framework of Fama and French (2006), which would explain both a lower likelihood for leftst voters to participate in the stock market and a lower share of the portfolio invested in the stock market, conditional on participation. Fama and French (2006) state that households may experience a taste for assets as for consumption goods. Consider two groups:

rightist investors and leftist investors. Rightist investors would evaluate assets solely on the monetary pay-offs. The taste preference of leftist investors on the other hand, would lower the utility of holding the assets below the monetary pay-offs. This results in a higher threshold of equity premium and planned investments for leftist voters.

Furthermore, the share of the portfolio invested would also be lower compared to rightist voters, due to differences in the marginal utility of stocks.

Empirical research so far finds that political preference does not influence the share of financial assets allocated to stocks (Kaustia and Torstila, 2011). Hence, the second hypothesis is formulated as follows:

H2: Conditional on stock market participation, financial asset allocation to stocks is not influenced by political preference.

3. Methodology

To analyse the mechanism of how political preference influences stock market

participation, this paper will examine both participation and allocation decision

regarding the stock market. This is done in line with the majority of empirical

research on portfolio choice of households (e.g. Rosen and Wu, 2004; Arrondel et al.,

2014). In this section firstly the choice for random effects models is explained.

Secondly, the random effects probit regression that is used to examine the effects of political preference on this participation decision, is presented. Thirdly, the random effects tobit model employed in this paper to examine the effect on the share of the asset within the portfolio of financial assets is shown.

3.1 Random effects model

Given the availability of longitudinal data, the sample has observations of the individuals over time, one could argue to include fixed effects, in order to control for possible unobserved heterogeneity. As a fixed effects model captures time variation in the variables, which could make the model more accurate.

The disadvantage of a fixed effects model is that it cannot estimate the effect of time-invariant characteristics. The result of using a fixed effects model is that individuals, whose characteristics do not change during the panel period, will not contribute to the estimation. In this case the within-variance in the dependent variable ¹ is not sufficient for a fixed-effects approach and numerous of the control variables are time-invariant. Additionally, in short panels with a large number of variables, as employed in this research, the estimates of the fixed effects model may become biased due to the incidental parameter problem. Therefore random effects models will be employed in this paper.

3.2 Methodology participation decision

Firstly, the association between political preference and stock market participation is analysed. Since stock market participation, is binary, linear regression analysis is unfit to examine the effects of political preference on the likelihood of a household participating in the stock market. In line with the vast majority of literature, this paper employs a probit model in order to overcome this problem (e.g. Rosen and Wu, 2004; Guiso et al., 2008; Van Rooij et al., 2011). The estimates of the probit models are obtained using the standard maximum likelihood procedure. In the random effects probit model employed in this paper, the dependent variable is binary and takes on a value of one for stock market participation and zero otherwise. Henceforth, stock market participation is examined by the following random effects probit model:

𝑆𝑀𝑃 ^∗ _!,! = 𝛼 + 𝛽 _! 𝑝𝑜𝑙𝑖𝑡𝑖𝑐𝑎𝑙 𝑝𝑟𝑒𝑓𝑒𝑟𝑒𝑛𝑐𝑒 _!" + 𝛽 _! 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠 _!" + 𝑢 _! + 𝜀 _!" (1)

1 See Table A1 in the Appendix for the within and between variation of the dependent and

independent variables.

where 𝑆𝑀𝑃 ^∗ _!" ,of individual i at time t, is the probability of an individual participating in the stock market, political preference is the variable measuring political preference on a left-right 0-10 scale, controls is the vector of control variables including the year dummies of which one is dropped to prevent perfect multicollinearity, 𝑢 _! is the individual-specific random effect and 𝜀 _!" is the idiosyncratic error term.

𝑆𝑀𝑃 ^∗ _!" is a latent variable that is not directly observed. The observed stock market

participation, 𝑆𝑀𝑃 _!" , is defined as follows:

𝑆𝑀𝑃 _!" = 1 𝑖𝑓 𝑆𝑀𝑃 ^∗ _!" > 0

0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒 (2)

where for the participation dummy 𝑆𝑀𝑃 _!" , one is observed as the individual participates and zero otherwise. One of the advantages of such a binary measure is that it does not suffers from a self-reporting bias (Campbell, 2006).

3.3 Methodology allocation decision

Subsequently, influence of political preference on the share of liquid financial assets invested in the stock market, is examined. A statistical problem arises from the fact that portfolio shares are between zero and one and there is a substantial concentration of the portfolio shares at zero, as approximately 78% of the respondents hold no investments in the stock market at all. Hence, the share measure is a limited dependent variable. Research in household finance has used a variety of econometric methods to overcome this problem. Heaton and Lucas (2000) employ a two-part model. In their research, ordinary least squared is used after dropping the sample individuals who fall below a certain level of asset holding. Bertraut and Starr- McCluer (2002) use the Heckman selection model in order to overcome this problem while Rosen and Wu (2003), Arondel et al. (2014) employ a tobit model.

When comparing the empirical options the Heckman model is assumed to be less suited for this research since there is no valid exclusion restriction in the data.

Due to fixed costs of participation the amount invested in the stock market is assumed to dependent on the optimal amount invested as a threshold. Under this threshold, zero amounts of the share of financial assets invested in the stock market are unobserved. Consequently, the tobit model is preferred over the two-part model, since the tobit model is able to deal with these unobserved latent observations of the share of net financial assets invested in the stock market. A robustness check compares both options.

Thus, although each approach has advantages and disadvantages, this paper

employs the random effects tobit model with a truncation at zero to overcome the

statistical issue mentioned above. This results in following equation for the random effects tobit model used to examine the relationship between and the share of net financial assets invested in the stock market and independent variables:

𝑆ℎ𝑎𝑟𝑒 ^∗ _!" = 𝛼 + 𝛽 _! 𝑝𝑜𝑙𝑖𝑡𝑖𝑐𝑎𝑙 𝑝𝑟𝑒𝑓𝑒𝑟𝑒𝑛𝑐𝑒 _!" + 𝛽 _! 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠 _!" + 𝑢 _! + 𝜀 _!" (3) where 𝑆ℎ𝑎𝑟𝑒 ^∗ _!" is the latent variable of share of net financial assets invested in the stock market, again 𝑝𝑜𝑙𝑖𝑡𝑖𝑐𝑎𝑙 𝑝𝑟𝑒𝑓𝑒𝑟𝑒𝑛𝑐𝑒 _!" is the variable measuring political preference on a left-right 0-10 scale , 𝑐𝑜𝑛𝑡𝑟𝑜𝑙𝑠 _!" is the vector of control variables including the year dummy of which one is dropped to prevent perfect multicollinearity and 𝜀 _!" is idiosyncratic error term, 𝑢 _!" represents the individual- specific random effect. 𝑆ℎ𝑎𝑟𝑒 ^∗ _!" is the latent variable that can be written as:

𝑆ℎ𝑎𝑟𝑒 _!" = 𝑆ℎ𝑎𝑟𝑒 ^∗ _!" 𝑖𝑓 𝑆ℎ𝑎𝑟𝑒 ^∗ _!" > 0

0 𝑖𝑓 𝑆ℎ𝑎𝑟𝑒 ^∗ _!" ≤ 0 _!" (4)

were the observable variable 𝑆ℎ𝑎𝑟𝑒 _!" , representing the observed share of net household wealth invested in the stock market, is equal to the latent variable if the latent variable is above zero and zero for the other observations.

The choice of control variables in the regressions is motivated by previous literature on stock market participation and political preference. To control for time effects, time dummies are employed. Included demographic factors are age, education and gender (Bertaut 1998; Hong et al., 2004) In addition, economical characteristics such as income and wealth are employed (Vissing-Jørgensen, 2002; Hong et al., 2004;

Calvet and Sodini, 2014). Also several other determinants of stock market participation and allocation are included such as risk aversion, financial literacy, health trust and sociability (Hong et al., 2001; Rosen and Wu, 2004; Guiso et al., 2008; Van Rooij et al., 2011). Employing these control variables could increase the validity of the model, while several of these control variables such as wealth and financial literacy are found to be related to both political preference and financial decisions of households (Powdthavee and Oswald, 2014; Montagnoli et al., 2016).

3.4 Marginal effects

The results of the probit model and tobit model only indicates whether a variable

has a positive or negative effect on the dependent variable and the respective

significance of the correlation. Therefore, marginal effects will be estimated in order

to determine the magnitude. The marginal effect will therefore also be estimated after

the probit and tobit regressions and can be found in the Appendix in table A2, A3 and

A4.

4. Data and measures 4.1 Dataset

This study uses data from the LISS panel over the time period of 2008-2014. The LISS household panel is managed by CentERdata, affiliated with Tilburg University.

The panel consists of 7,000 individuals spread over 4,500 households. The questionnaires of the LISS Core Study are conducted annually or biannually. The LISS panel is based on a true probability sample of Dutch households. CentERdata and Centraal Bureau voor de Statistiek (CBS) invites participants to create a representative sample of the Dutch population.

Members of the panel complete a questionnaire over the Internet each month. For individuals without computer or Internet access, a device with Internet access is made available without charge. Participants in the LISS panel are reimbursed for their contribution. Households are followed across time through the LISS Core Study.

The LISS Core Study has questionnaires on 1) Background variables, 2) Health, 3) Religion and Ethnicity, 4) Social Integration and Leisure, 5) Family and Household, 6) Work and Schooling, 7) Personality, 8) Politics and Values, 9) Assets, 10) Income, and 11) Housing.

Of particular interest for this research is that the LISS Core Study provides information on both political preferences as well as a comprehensive set of information on financial assets, including investments in the stock market. Next to the LISS Core Study there is room for collection of data for research purposes. The data of these studies are particularly useful for this research to construct multiple control variables and examine risk and return expectations as an alternative explanation for the results found.

The data on risk aversion is based on research into the higher order risk attitudes of the general population of Noussair et al. (2014). This research was completed by 3,457 individuals of the LISS panel. In order to analyse the relationship between political preference and risk and return expectations, a study conducted in 2010 among the members of the LISS panel that participate in the stock market, is used. A total of 761 members of the LISS panel participated in this study. The data on financial literacy is obtained from a study conducted in 2011 among 4,860 members of the LISS panes. The relationship between political preference and stock market aversion is studied, using the European Values Study and the World Values Survey.

The European Values Study, conducted in 2008, is a cross-national research exploring

individuals’ values. This papers sample of 548 individuals comes from the LISS panel

version of this survey and is specifically used to examine the relationship between political orientation and values. The Dutch version of the World Values Survey, completed by 1,884 individuals of the LISS panel in 2012, is used for the measures regarding trust in banks and large companies.

4.1.1 Data set construction

The LISS panel data is provided in separate data sets. Each individual has a unique identifier, which allows tracking individuals over time. The data sets of each wave are matched, based on the unique individual indicator. The data sets covering each wave are combined in order to create a data set covering the waves of 2008, 2010, 2012 and 2014. The use of the LISS panel data set is limited to those waves, as these are the only years in which questions regarding assets are included. Afterwards, all waves are appended into a single data set. Subsequently, the unique identification number and wave year are used to sort the data set, which results in a total of 23,107 observations with corresponding 7,502 individuals.

The data of the LISS panel is collected on the individual level. However, the financial decision of the different household members is probably not independent. It can be assumed that the household head makes the majority of financial investment decisions. Hence, in order to remove double data entries for one household and base the analysis on the characteristics of the decision maker, all other participants except the household head are dropped. Afterwards, all individuals with missing values (including those that were unable, or preferred not to answer questions) regarding the dependent and independent variables except wealth, financial literacy and risk aversion are also dropped from the analysis. To correct for differences in price level between the years, all monetary values are corrected for inflation with 2008 as the base year. This leads to an unbalanced data set covering four waves with 2,440 individuals and 7,487 observations.

As often with data based on questionnaires, the data on wealth is incomplete. The data for the measures of financial literacy and risk aversion is also limited, while the measures were only collected once. Therefore the models are estimated, including and excluding wealth, financial literacy and risk aversion as control variables.

4.2 Measures

In the following, the constructions of the variables employed in this paper will be

described. The first part defines the dependent variables. The second part shows the

measures of political preference and several of the control variables.

4.2.1 Stock market participation and allocation measures

The stock market participation variable is based on a question in the questionnaire regarding assets. Participants are asked whether they possess growth funds, share funds, debentures, stocks, options or warrants. If they indicate they own these assets, the stock market participation variable takes on a value of one and zero otherwise.

The share measure is constructed using the sum of positive savings (in the Netherlands it is also possible to have a small amount of debt on a savings account;

these values are replaced with a zero for the calculation of the financial assets) and investment in the stock market as total net financial assets. Subsequently, dividing investments in the stock markets by the net total financial assets, provides the share measure. When studying portfolio share, defining the wealth measure in the denominator of the portfolio share measure, is of significant importance. Different results may arise if homes and cars are included in the household wealth measure.

One could also define household assets even more broadly by incorporating human capital. This paper uses an approach similar to other research on individuals’ asset decisions and employs the net financial assets of households (Rosen and Wu, 2004).

Therefore, human capital, cars, houses, and associated debts are excluded when calculating the share of household assets invested in the stock market.

4.2.2 Measures of independent variables

In line with the research of Kaustia and Torstila (2011) and Changwony et al.

(2015), this paper measures values by political preference. Contrary to previous research, this paper employs a more direct measure of values, namely; self-assessed political preference on a left-right scale. Political preference is based on a single-item measure where participants report their left-right political orientation using an eleven- point scale ranging from left to right, where zero is left and ten is right. Although this is a simple scale, it has proven to accurately predict voting behaviour of individuals and even used to determine genetic and brain structure contributions to political orientation (Kanai et al., 2011). This axis value is used as the measure of political preference. The main advantage of this measure is that it is not influenced by individuals’ assessment of party capabilities.

The trust dummy is based on a question from the personality survey and

constructed in line with the work of Guiso et al. (2008). Respondents are asked to

zero indicates that you cannot be too careful in trusting people and ten would indicate that most people can be trusted. The trust dummy takes on a value of one for respondents with a six or higher on the scale.

The sociability dummy is based on the question ‘‘How often do you do the following? Spend an evening with someone from the neighbourhood?’’ those who indicate to never do this take on an value of zero for the social interaction dummy and one otherwise. This question and the construction approach of the measure is similar to the one Hong, Kubik and Stein (2004) use to construct one of their social interaction measures.

The bad health dummy is based on the self–assessed health scale, ranging from one to five with one being unhealthy and five being perfectly healthy. The bad health dummy is assigned to respondents who report poor (1) or moderate (2) health status.

The wealth measure is constructed in line with previous research that uses LISS data to examine household portfolios (Estrada-Mejia et al., 2016). The wealth variable was constructed as the sum of savings balance, long-term insurance balance, investments and real estate, reduced by mortgage liabilities and other loans. The distribution of the wealth variable is skewed to the right with a long thick tail and a proportion of the households even report wealth above 500,000 euros. To overcome this skewed distribution one could take the natural logarithm. However, taking the natural logarithm is not preferable since large portions of the households hold no or negative amounts of wealth and for the natural logarithm zero and negative values are undefined. To overcome this problem, the inverse hyperbolic sign transformation which is an alternative to natural logarithm transformations if the variable takes on negative or zero values, is applied. This results in 𝑤𝑒𝑎𝑙𝑡ℎ (𝐼𝐻𝑆), which is defined as:

𝑤𝑒𝑎𝑙𝑡ℎ (𝐼𝐻𝑆) _!" = 𝑙𝑜𝑔 (𝑤𝑒𝑎𝑙𝑡ℎ _!" + 𝑤𝑒𝑎𝑙𝑡ℎ _!" ^! + 1) (5)

The risk aversion variable is based on a single wave study conducted in 2010. In this wave an questionnaire was added in order to measure the risk attitudes of respondents. Part of this questionnaire allows to construct a measure of risk aversion.

This measure is constructed in line with Noussair et al. (2014). The first part of the questionnaire consisted of five questions in which the participant was given the choice between a sure pay-off and an uncertain lottery pay-off. The number of certain choices is used to construct the measure of risk aversion, which ranges from zero to five. Five indicating the highest level of risk aversion. Since Chiappori and Paiella (2011) and Sahm (2012) find that relative risk aversion is constant over time, risk aversion is assumed to be constant.

The measure for financial literacy is based on a study of financial literacy

conducted in 2012 among the members of the LISS panel. The measure is based on

the first three questions of the study. The questions regard respondents’ understanding of basic numeracy, compounding of interest and the difference in risk profile between funds and stocks. The first two questions correspond to the first two basic financial literacy questions of Van Rooij et al. (2011), while the other question focusses on the respondents’ knowledge of the characteristics of investment options. If the three questions are answered correctly, the financial literacy dummy takes on a value of one and zero otherwise. Since the time period of the panel is relatively short and little information is known about the changes of financial literacy over time, financial literacy is also assumed to be constant.

5. Analysis

In this section, the results of the empirical analysis are reported. The objective is to present an overview of the impact of political preference on stock market participation and allocation. Firstly, the descriptive statistics are presented. Secondly, simple cross tabulations are shown. Thirdly, the random effect probit regression results are presented in order to show the effects of political preference on the likelihood of a household participation in the stock market. Fourthly, the results of the random effects tobit model are presented to show the effects of political preference on the allocation of assets to the stock market. Finally, stock market aversion as driver of the relationship between political preference and household decisions regarding the stock market, is explored.

5.1 Descriptive statistics

Table 1 shows the summary statistics of the variables employed in this paper,

including demographic characteristics and political preference. The average net

income of households is 20,500 euros. The average age is 54 years. Approximately

46% of the respondents is female. Around 22% of households report that they have

stocks, bonds, mutual fund holdings or options in line with previous research on stock

market participation in the Netherlands (Guiso et al., 2008). Conditional on

participating in the stock market, the portion of net financial assets invested in

investment assets makes up around 44%. The left-right scale value ranges from 0 to

10 and has a mean of 5.165. Around 27% of the sample holds a degree from an

university of applied sciences (UAS) and approximately 10% holds an university

degree.

Table 1

Summary statistics

This table reports the summary statistic for the variables used in the empirical analysis.

Data is obtained from the 2008, 2010, 2012 and 2014 waves of the LISS panel.

Variable Mean Std dev Min Max N

Dependent variables

Stock market participation 0.221 0.415 0 1 7,487

Share of financial assets in the stock market 0. 094 0.219 0 1 3,275 Share of financial assets in the stock market,

conditional on stock market participation

0.442 0.297 0 1 694

Independent variables

Left-right scale 5.165 2.135 0 10 7,487

Age 54.08 14.92 18 92 7,487

Income/€1,000 20,474 0.0123 0 180.000 7,487

Female 0.464 0.499 0 1 7,487

Have kids 0.748 0.434 0 1 7,487

UAS 0.268 0.443 0 1 7,487

University 0.102 0.303 0 1 7,487

Bad health 0.162 0.369 0 1 7,487

Trust 0.715 0.452 0 1 7,487

Social interaction 0.767 0.423 0 1 7,487

Wealth (IHS) 8.775 7.047 -17.010 16.636 3,296

Financial literacy 0.471 0.499 0 1 2,230

Risk aversion 3.435 1.685 0 5 2,230

Year 2008 0.251 0.442 0 1 7,487

Year 2010 0.267 0.442 0 1 7,487

Year 2012 0.235 0.424 0 1 7,487

Year 2014 0.247 0.431 0 1 7,487

5.2 Some cross tabulations

Table 2 provides a first glimpse at the effect of political preference on stock

market participation. The sample is divided into a left-wing voters group, which

consists of households with a self-assessed left-right rating below six and a right-wing

voter group which consists of households with a rating of six or higher. Of the left-

wing voters, 19.4% participate in the stock market while of right-wing voters 25.3%

participate in the stock market. Conditional on participation, left-wing voters have on average 43.0% of their net financial assets invested in the stock market. Right-wing on the on other hand right-wing voters allocate 45.3% of their financial assets to the stock market, conditional on participation. Although these results are no substitute for the more extensive analysis that follows, it indicates the basic pattern.

The cross tabulation of means is also executed for a number of variables not controlled for in previous literature. The cross tabulations especially show the importance of controlling for financial literacy, as it seems to be over ten percentage point higher among right-wing voters. The same holds for wealth, where the right- wing voters are found to have a higher level of wealth. While both variables are also related to the stock market participation decision, not controlling for wealth and financial literacy could result in omitted variable bias. The political preference measure could also partially capture the effect of wealth and financial literacy, if not controlled for. Risk aversion, trust, health and sociability measure are not found to differ significantly between the left-wing and right-wing voters.

Table 2

Cross tabulation dependent and various independent variables

An individual is classified as ‘‘left’’ if (s)he reports a political preference below six on the left-right scale ranging from 0-10 ‘‘right’’ otherwise. Means of variables are reported.

Left Right Dependent variables

Stock market participation 0.194 0.253

Share of financial assets in the stock market 0. 080 0.111 Share of financial assets in the stock market, conditional on stock

market participation

0.430 0.453

Independent variables

Wealth (IHS) 8.217 9.381

Bad health Trust

Social interaction

0.178 0.721 0.751

0.143 0.707 0.787 Financial literacy

Risk aversion

0.416 3.398

0.523 3.465

N 4,085 3,402

5.3.1 The effect of political preference on stock market participation

The first step of the analysis is to examine the effect of individuals’ political preference on the probability of stock market participation. The goal is to determine whether political preference exerts an independent effect on the probability that a household participates in the stock market. The dependent variable is stock market participation which takes a value of one when the household holds stocks either directly or through equity mutual funds and zero otherwise. On the right side of the equation, the measures for political preference and controls for income, wealth and other demographic variables are used. A probit random effects estimator is employed in order to examine this relationship. To control for year effects, a dummy variable is included for the waves of 2010, 2012 and 2014.

Table 3 shows the results of the random effects probit regression estimates. The first column presents the results for the basic specification of control variables. The second column shows the basic specification of the sample that has data on wealth.

The third column presents the results controlled for wealth.

Left-wing political preferences are significantly negatively related to the likelihood of household stock market participation. The right-left axis value is significant for all combinations of the explanatory variables and takes on a t-value from 2.5 to 5.0 depending on the combination of independent variables. The effects remain strong after adding the control variable for wealth, although due to incomplete data, adding this wealth control variable reduces the number of observations.

The difference in stock market participation could be influenced by a difference in risk aversion or financial literacy between leftist and rightist voters, as risk preference could also influence political preference. Individuals with a leftist political preference could be risk averse and therefore support an ideology that emphasises risk sharing and more government intervention. More rightist voters on the other hand, might favour competition, minimal government intervention and a system in which individuals are more self-reliant. Furthermore, as shown in the cross tabulation rightist voters have a higher level of financial literacy. Therefore adding risk aversion and financial literacy could improve the reliability of relationships found. Both risk aversion measure and financial literacy measure are based on a single wave study of the LISS panel. This results in a significant drop in the number of observations.

However, since previous studies find both factors to be of significant influence on

stock market participation and the share of assets allocated to stock incorporating the

measures could provide more reliable results.

Table 3

The effect of political preference on stock market participation

This table reports the results of the probit regression. Estimation is by random effects.

The dependent variable is a dummy equal to one if the household participates in the stock market, i.e. holds shares, mutual bonds, corporate bonds and/or options. Sample size differs between the first columns and second and third column due to missing information on the wealth measure, Standard errors are reported in parentheses. *** indicates the coefficient is different from zero at the 1% level, ** at the 5% level and * at the 10% level.

Dependent variable: Stock market participation

(1) (2) (3)

Left-right scale value 0.110*** 0.113*** 0.096**

(4.97) (3.06) (2.47)

Income/€1,000 0.026*** 0.041*** 0.033***

(5.23) (4.00) (3.57)

Female -0.557*** -0.834*** -0.793***

(-4.79) (-4.41) (-3.98)

Have kids 0.100 -0.605*** -0.535**

(0.77) (-2.72) (-2.31)

UAS 0.914*** 1.032*** 0.940***

(6.85) (4.48) (3.92)

University 1.691*** 2.286*** 1.964***

(8.52) (5.14) (4.75)

Age 0.095*** 0.063 0.024

(3.92) (1.61) (0.62)

Age squared -0.001*** -0.000 -0.000

(-3.23) (-0.82) (-0.16)

Trust 0.391*** 0.331** 0.303*

(4.34) (2.11) (1.89)

Social interaction 0.054 -0.133 -0.150

(0.57) (-0.81) (-0.94)

Bad health -0.208* -0.236 -0.149

(-1.84) (-1.22) (-0.77)

Wealth (IHS) 0.129***

(5.10)

Year 2010 -0.361*** -0.097 -0.087

(-4.24) (-0.58) (-0.51)

Year 2012 -0.616*** -0.597*** -0.573***

(-6.56) (-3.22) (-3.09)

Year 2014 -0.857*** -0.899*** -0.784***

(-8.36) (-4.43) (-4.05)

Constant -6.135*** -6.400*** -5.821***

(-8.84) (-4.97) (-4.66)

N 7,487 3,296 3,296

Log likelihood -2918.706 -1227.23 -1188.22

The results of the random effects probit regression, with risk aversion and financial literacy as control variables are presented in table 4. Column one gives the basic regression results for the sample, which includes measures on financial literacy and risk aversion. Column two presents the results as risk aversion and financial literacy are added as control variables. In the fourth column, the basic results are presented for the sample that contains measures of risk aversion, financial literacy and wealth. Column five shows the results when controlling for wealth, risk aversion and financial literacy. This is the most extensive controlled specification in this paper.

Risk aversion is found to be insignificant, while financial literacy is found to be significant and positively related to stock market participation. The effect of the left- right scale value remains highly significant and positive, even though the number of observations drops substantially from 3,660 to 2,442. Thus, H1 is corroborated.

The economical interpretation is that a one point move to the left on the political preference scale is associated with a 0.7 to 1.6 percentage point decrease of the likelihood of a household participating depending on the combination of control variables.

The outcomes of the other variables in table 3 and table 4 are discussed briefly.

The findings are broadly consistent with previous research. Income and wealth are found to be statically significant and positively related to stock market participation.

Higher levels of education are also associated with a higher likelihood of participation. A college education seems to have more influence on the participation decisions compared a degree from a university of applied sciences. The female dummy is negatively related with participation and significant for al specifications.

The trust dummy is significant for the five combinations of control variables and

positively related to stock market participation. However, when the number of

observations is significantly reduced and more controls are employed, as with the

sample that includes data on risk aversion and financial literacy, the results become

insignificant. The wealth variable is significant and positively related to stock market

participation. The income measure is positively related and significant for all

specifications of the control variables except the third and fourth of table 4. The year

effects are also significant in the majority of the specifications.

Table 4

The effect of political preference on stock market participation: extended analysis

The dependent variable is a binary holding variable equal to one if the household participates in the stock market i.e. holds shares, mutual bonds, corporate bonds and/or options. This table reports the random effects probit regression estimates. The t-statistics are below coefficients in parentheses. All significance tests are two-sided, *, ** and *** indicate significance at the 10%, 5% and 1% levels, respectively.

Dependent variable: Stock market participation

(1) (2) (3) (4)

Left-right scale value 0.158*** 0.142*** 0.168*** 0.124**

(3.61) (3.21) (2.89) (2.14)

Income/€1,000 0.022** 0.018* 0.020 0.009

(2.32) (1.96) (1.54) (0.83)

Female -0.828*** -0.523** -1.294*** -0.850***

(-3.66) (-2.32) (-4.29) (-2.80)

Children -0.456* -0.500* -1.054*** -1.011***

(-1.71) (-1.84) (-2.83) (-2.77)

UAS 1.467*** 1.158*** 1.657*** 1.208***

(5.32) (4.32) (4.28) (3.24)

University 1.890*** 1.479*** 2.155*** 1.472***

(4.81) (3.91) (3.81) (2.72)

Age 0.113** 0.104** 0.140** 0.055

(2.29) (2.11) (2.16) (0.89)

Age ^! -0.001* -0.001 -0.001 -0.000

(-1.79) (-1.52) (-1.48) (-0.31)

Trust 0.469*** 0.399** 0.277 0.157

(2.64) (2.26) (1.23) (0.69)

Social interaction -0.197 -0.188 -0.265 -0.225

(-1.14) (-1.07) (-1.16) (-1.05)

Bad health -0.249 -0.197 -0.255 -0.151

(-1.22) (-0.93) (-0.95) (-0.55)

Wealth (IHS) 0.139***

(3.38)

Risk aversion 0.011 0.032

(0.19) (0.41)

Financial literacy 1.479*** 1.558***

(5.97) (4.81)

Year 2010 -0.456*** -0.434** -0.266 -0.207

(-2.59) (-2.46) (-1.02) (-0.81)

Year 2012 -0.667*** -0.661*** -0.738*** -0.744***

(-3.48) (-3.44) (-2.68) (-2.74)

Year 2014 -1.189*** -1.215*** -1.198*** -1.149***

(-5.61) (-5.66) (-4.00) (-4.01)

Constant -6.679*** -7.089*** -8.039*** -7.103***

(-4.62) (-4.80) (-3.96) (-3.50)

N 2,230 2,230 1,566 1,566

Log likelihood -827.15 -807.46 -554.51 -520.21

5.4 The effect of political preference on allocation decision

Furthermore, it is interesting to examine whether variations in political preference exert an independent effect on the share of net financial assets invested in the stock market or the effect of political preference is merely categorical. A random effects tobit model is employed to analyse this relationship.

Table 4 reports the outcomes of the random effects tobit model. The first column shows the results of the basic regression. The second column provides the result of the basic regression on the sample that contains measures of risk aversion and financial literacy. The fifth column displays the most extensive random effects tobit regression specification which includes controls for financial literacy, wealth and risk aversion.

The left-right scale value is significant and has a positive impact on the share of assets invested in the stock market for all specifications of the control variables. The left- right scale value is significant for all combinations of the variables and has a t-value of 3.9 to 2.8 depending on the specification of the model.

Based on the marginal effects, a one point move to the left of the scale is associated with a 0.005 to 0.008 decrease of share invested in the stock market depending on the combination of control variables.

This leads to rejection of H2 that expects political preference to have no influences on the share allocated to the stock market. This means that the effect of political preference on household decisions regarding the stock market is not merely categorical, but also influences the share allocated to stocks. These results are in line with the insight that consumers gain or lose direct utility by holding certain investments, as is the case with consumer goods. The result differs from the results found by Kaustia and Torstila (2011). This difference could be the results of their severe data limitation, since their two cross-sectional tobit model estimates were based on a low number of observations (210 and 65 respectively).

The results for coefficients on the other variables in table 5 are in general in line

with previous research. Education, wealth and financial literacy are significant in all

specifications of the model and positively associated with the share of financial assets

invested in the stock market. Females allocate less of their financial assets to the stock

market. Contrary to previous research health, trust, sociability and risk aversion are

found to be insignificant with respect to the allocation decision.