MASTER THESIS
SOCIAL CAPITAL AND THE FUNCTIONING OF
FINANCIAL MARKETS: INTERNET AND STOCK
MARKET PARTICIPATION
BY:LOTTE POOLEN*MSC.ECONOMICS
UNIVERSITY OF GRONINGEN,THE NETHERLANDS
JANUARY 2017
SUPERVISOR: PROF. DR.A.R.SOETEVENT
ABSTRACT
Inspired by the recent discussions around the topics Internet bias, fake news and its unclear consequences on society, this study assessed the effects of Internet use on social capital and stock holding of the Dutch households. By means of the LISS panel, multiple hypothesis are tested, build on a theoretical framework that predicts distinct roles of Internet use, sociability and general trust on portfolio decisions. Various cross-sectional analyses are employed, accompanied with instrumental variables to account for the endogeneity of general trust. The initial results indicated a positive significant effect of Internet use on social capital and stock market participation. However, it is primarily a correlation than a causal consequence. Furthermore, the study showed that trust has a distinct impact on the likelihood of stock market participation in the year 2012. Unfortunately, this study was unable to verify whether the same conclusion holds for the other years, due to weak instruments. Additionally, to test the causal relation, the longitudinal structure of the LISS panel was exploited. By means of various panel estimations to account of unobserved heterogeneity and simultaneity bias the causal effect of Internet use was tested. In contrast of hypothesised predictions, the results indicate that the use of Internet decreases the likelihood of stock market participation and general trust.
1. INTRODUCTION
Trust is an important matter in this world. It keeps the society running and most trivial transactions depend on it. It is regarded as essential for the functioning of market economies (Fukuyama, 1995). Multiple studies find a positive relationship between general trust and economic performance, investments and financial decisions of households (Guiso et al., 2004). In general, sound institutions and well-working law systems could also enforce this, but regions that lack efficient and credible institutions could benefit from general trust. Therefore, it can be concluded that trust is an important factor in a country to strive for economic development and growth. However, the cultivation of trust is not an easy task. It is considered as a part of social capital that is built over time and effort. Regions that enjoy high levels of social capital, trust is rewarded and acts as a substitute for unsound institutions (Guiso et al., 2004). This opposed to places that lack trust and social capital. In general, families and networks seem to be a solution to find ways around the trust barrier, but this is limited. New technologies, such as the Internet could offer solutions to these barriers. It could create possibilities to trust unfamiliar groups, built networks, acquire information and cooperation. It could reduce uncertainty and widen the circle of trust. In other words, it could supplement social capital. However, there are voices that claim that new technologies, as the Internet, alter trust in a negative way. Due to the nature of the Internet there is no need for people to be social or friendly. Hence, it could alter their trust and relation building skills. Additionally, fake news gets spread at a tremendous fast rate and reaches more people than one can account for. This is, among other reasons, a result of the transition in news consumption that has been taken place. Whereas originally one received the news via the traditional media, such as the television, radio and newspapers, now an increasing number of individuals consume the news via new media, such as the Internet and social network sites as Facebook and Twitter. The Internet provides a service whereby everyone can be an editor and add information without a gatekeeper to verify it, hence it is full of bias. For this reasons the Internet has been subject to critique the last years. Unfounded statements are made, even by most influential individuals. The problem is that it is believed by the mass, which could create unfounded fears and could even result in a loss of trust. The Internet is in turn a key-player in spreading this fake news.
economic agents wants to maximize their utility, the number of households participating in the stock market should be higher as currently is (Haliassos & Bertaut, 1995). It is simply beneficial due to the higher expected-return premium on equity compared to riskless assets. Although, the financial markets in the western countries are well developed, it shows that there still remains a barrier to enter. Financial markets are important for the economy, because they provide liquidity: the ease and speed at which agents can convert assets into purchasing power at agreed prices (Levine, 1997). They offer the opportunity to hold liquid assets, which can be bought and sold quickly. If this would not exist, investors would be stuck into long-term illiquid investments, and the problem is that often savers are unwilling to give the control of their savings to investors for a long-term period. Hence, liquidity is important for investments, which in turn is necessary to strive for economic growth. If one could find ways to lower this barrier it would increase the liquidity and the market will function better as a consequence. Therefore, the research question addressed in this study is formulated as follows:
Has Internet lowered the barrier to enter the stock market?
Hence, increases the likelihood of participating. The second channel is an indirect effect going through sociability. The Internet could enlarge the social network of a potential investor and increase the weak ties, which hypothesised as main source to acquire new information. This in turn could lower the participation costs as theorized by Georgarakos and Pasini (2011). The third channel is also an indirect effect going through the social capital proxy general trust. Via this channel the Internet if hypothesised that it could help to trust unfamiliar people, and thus enlarges networks to trust. When this trust cultivates it lowers uncertainty and following the reasoning of Georgarakos and Pasini (2011) this increases the expected returns of an investment and therefore the likelihood of participation.
This study is performed by the use of the LISS panel (longitudinal Internet Studies of Social sciences), which provides extensive information about the Dutch households and their financial decisions. First, by means of a cross-sectional analysis and the use of the instrumental variables to account for endogeneity, the effect social capital on stock market participation is assessed. This is followed by a cross-sectional analysis towards the effect of Internet use on social capital and stock market participation. This cross-sectional analysis is conducted for the three different years. For the reason that cross-sectional analysis brings primarily a correlation than a causal interpretation, the longitudinal format of the LISS panel is exploited. This brings the opportunity to account for several endogeneity concerns and causal effect estimation. Hence, the second analysis is conducted by means of various panel model estimations.
The remainder of this paper is as follows: Section 2 provides a research framework, which contains a literature review about social capital, stock market participation and Internet use. This also provides the theoretical framework and predictions of this study. Section 3 includes the research design, with corresponding models and data retrievement. This is followed by section 4, which presents and discusses the empirical results. The last section provides the conclusion and implications.
2. RESEARCH FRAMEWORK
2.1 LITERATURE REVIEW
This part provides a literate review about social capital and Internet use in relation to stock market participation. Different studies and outcomes will be discussed, which are used to build the underlying framework for this study.
2.1.1 DEFINING SOCIAL CAPITAL
The term social capital exists already since 1890, but became widely used in the 1990s by social sciences (Coleman, 1988; Putnam 1995). Over the years, the term has grown towards one of the most actively debated terms in social sciences and it is used to explain fluctuations in various subjects as voting, economic performance, health, job-finding and crime (Putnam, 1995; Granovetter, 1973; Islam et al., 2006). Despite its well-studied outcomes, the definition of the term remains vague. Some characterize it as the relationships among individuals, while others define it as the positive externalities it creates, such as economic growth. Let’s start with Coleman (1988) who defines social capital as: ‘’social organisation constitutes social capital, facilitating the achievement of goals that could not be achieved in its absence or could be achieved only at a higher cost’’. In turn, Robert Putman (1995), a strong advocate of this phenomenon, defines it as: ‘’the connections among individuals, social networks and the norms of reciprocity and trustworthiness that arise from them’’. Fukuyama (2001) defines social capital as: ‘’an informal norm that promotes cooperation between individuals’’. These different definitions indicate a non-consensus between researchers about this phenomenon. It is evident that it is hard to reduce social capital to a single definition and therefore it is often subject to critique (Durlauf, 2002). It is hard to measure something if you cannot define it. Durlauf and Fafchamps (2004) analysed the different definitions and reduced social capital into three underlying common ideas (p. 1644):
1. Social capital generates positive externalities for members of a group;
2. These externalities are achieved through shared trust, norms and values and their consequent effects on expectations and behaviour;
3. Shared trust, norms, and values arise from informal forms of organisations based on social networks and associations.
information and a sense of belonging. The second one is participatory capital, which is characterized with involvement in politics and voluntary organisations that offers an opportunity to build networks. Second, there is a distinction made between cognitive and structural social capital (Islam et al., 2006). The former includes values, norms, attitudes and beliefs and is operationalized by individuals’ perceptions about the levels of interpersonal trust, sharing and reciprocity, while the latter consists out of externality aspects of social organization as social networks, or civic participation. Furthermore, within social capital there exists bonding and bridging social capital and additionally, weak ties and strong ties, which are in turn linked together. Putnam (1995) defines bonding social capital as the value assigned to social networks between homogeneous groups of individuals and bridging as that of social networks between socially heterogeneous groups of individuals. The distinction between strong ties and weak ties is that the former is defined by close social relations, which is characterized by trustworthiness such as friends and family, while the latter one is defined by the more superficial relationships as friends-of-friends (Granovetter, 1973). As can be seen, weak ties are linked to bridging capital, whereas strong ties are linked to bonding capital. Former research has concluded that weak ties have a strength that strong ties do not have. This is reflected in the novelty of information sharing. Most novel sharing flows trough weak ties, because the information of friends often overlaps with that what we already know. Conversations with people outside of our own network cycle provide use more likely new information and better sources as job opportunities (Granovetter, 2005).
To continue this reasoning, it indicates that there could be positive externalities from social capital. In economic terms, social capital could increase Pareto efficiency: a state in which the allocation of resources are most efficient and in which it is impossible to make any individual better off without making one individual worse off (Durlauf & Fafchamps, 2000). Inefficiencies are caused by factors as negative externalities, free riding, imperfect information, imperfect competition and so on. Hence, for social capital to create positive externalities, it should help to solve one of these inefficiencies. To reflect this on the topic of this study, social capital and Internet could eliminate inefficiencies caused by imperfect information about the stock market. Due to asymmetric information stock market trade is hindered and doesn’t take place, because people don’t have the information, can’t find each other or do not trust each other. Hence, if Internet facilitates trust, networking and search, social exchange can be improved and Pareto efficiency could be increased.
2.1.2 SOCIAL CAPITAL: TRUST AND NETWORKS
efficiently. Trust can be defined as an optimistic expectation or belief of another persons’ behavior (Fafchamps, 2004). According to Knack and Keefer (1997), who define trust in terms of beliefs in a game theoretic framework, trust is a fraction of the people in a society who expect that most others will act cooperatively in a prisoner’s dilemma context. Trust is an aspect of a community in which people life and determined by the aggregated perception of trust off all members of that community, thus it is not a personality trait. This in turn means that individuals are trust-takers and no one can individually change the overall level off trust. Consequently, it is an endogenous belief about others, created by interaction between people, the society, institutions and the economy and takes time and effort to built. This type of trust is also often referred to as general trust, the general knowledge about the population and often seen as a proxy for social capital (Guiso et al., 2004). This explanation of trust also indicates there could arise some concerns with the measurement of general trust. First, it is an endogenous belief, influenced by the aggregated perception of the society, whereas many econometric models ask strict exogeneity to measure the impact of a variable. Second, many scholars used the self-indicated trust question to measure this concept. This question asks whether the respondent beliefs that in general, most people can be trusted or one cannot be too careful. Glaeser et al. (2000) have shown by means of an experiment that this question about trust predicts primarily trustworthy behaviour, than it measures trusting behaviour. Hence, it is a challenging job to assess this right.
Networks, mentioned as the third factor of social capital in the definition of Durlauf and Fafchamps (2004), play also an important role in this matter. It could offer a substitute for low general trust. If there would exist perfect generalized trust, everyone would belong to the same network, which is in turn most efficient for trade and therefore economic growth. On the other hand, if people only trust only certain groups, there exist many small networks, which in turn is inefficient because they could benefit from trading with each other. Networks create trust, norms and values. Thus if one could be find ways to enlarge the networks, it could lead to cooperation and positive externalities
2.1.3 SOCIAL CAPITAL & THE NETHERLANDS
In a country as the Netherlands, which enjoys sound and credible institutions, well-organized law systems and high generalized trust, one would assume that informal organization and higher levels of social capital do contribute significantly more to economic growth. Still, there remains one topic where social capital could play a role and be beneficial. This is the stock market. As stated in the introduction, the question remains why so many households do not participate in the stock market. Much research has been done towards this question (Bertaut & Starr-McCluer, 2000; Georgarakos & Inderst 2014). Certain determinants1 of stock market participation are already known. First, participation is increasing in wealth, because the fixed costs are less of an obstacle to invest. Another one is education, which decreases the fixed participation costs due to easier learning and understanding the risk-rewards trade-offs, executing trade and to deal with trading mechanisms. Other determinants are found in personality traits as risk attitude. Hurd et al. (2009) analysed the Stock market expectations of Dutch Households and found evidence that expectations of individuals are correlated with stock holding. When the expected return is low, there is less incentive to invest. Also, the imperfect information about the stock market and the corresponding high transaction cost to acquire this information has a negative impact on the likelihood of stock holding. Trust plays an important role in this, because financial contracts are primarily trust-intensive contracts. In turn, Guiso et al. (2008) studied the effect of low levels of trust on stock market participation in the Netherland. They found evidence that it is an important factor to explain financial decisions of households, even in a developed country such as the Netherlands. Thus in conclusion, trust it could contribute to the development of financial markets.
2.1.4 SOCIAL CAPITAL & STOCK MARKET PARTICIPATION
Several studies analysed the effect of social capital on financial decisions of households. Guiso, Sapienza and Zingales (2004), did research towards the role of social capital in financial development by exploiting the differences in social capital across different regions of Italy. They analysed the effect of social capital on different forms of financial decisions, which are in turn subject to different risk-taking levels. The higher the risk, the more trust is needed. They found indeed significant effects of social capital on financial decisions of households. In areas with high levels of social trust, households make different financial decisions in terms of a higher likelihood of risky investments, holding less cash and make more often use of checks. Additionally, they also found evidence that in regions where legal enforcement is weaker and education is low, the effect of trust is stronger. This implies that general trust can work as a substitute for weak institutions.
contribute to information sharing, reciprocity and trust. They studied whether socials, defined as people who know their neighbours, spend time with them, and join religious services, are more likely to invest in the stock market than their counterpart non-socials. This prediction is based on two theories, which involves information sharing. The first is word-of-mouth or observational learning in which potential investors may learn from one another either about the stock market. The second one is that stock market participants like talking about the market with friends who are also participants, but also with friends who are not. In their theoretical model, both type of investors face fixed participation costs, but these are lower for the ‘socials’ when the participation rate is high among their peers, caused to the possible information sharing in their network. The results showed indeed that socials are more likely to invest in the stock market. Additionally, they tested the hypothesis that if socials are more likely to invest in the stock market than their non social counterparts, the marginal effect should be higher in areas where there is a high density of stock market participants. They found indeed significant impact of household sociability on stock market participation in areas where those rates are higher.
Georgarakos and Pasini (2011) combined the studies of Hong et al. (2004) and Guiso et al. (2004), and studied the effect of sociability and trust on stock market participation. They used internationally comparable household data and supplemented this with regional information on generalised trust and on specific trust to financial institutions. They developed a theoretical model to predict the effect of trust and sociability on stock market participation, which affects the likelihood by two separate channels. The first channel hypothesises that sociability, in term of social interactions, reduces fixed participation cost through cheaper information sharing, just like in the model of Hong et al. (2001). The second channel implies that trust has an effect on the expected return of an investment, because an investor has to take into account the possibility that the financee will not respect a contract. The higher the trust, the higher the expected return of an investment, hence the higher the likelihood of an investment. They showed that trust and sociability have significant effect on stock market participation and that sociability is most likely to act as a substitute effect to low generalized trust on stock market participation. The effect of sociability is stronger in areas with higher stock market participation rates, as theorized by the model. The effect of generalized trust is mainly significant in regions with low stock market participation and its effect is strong for the wealthy.
2.1.5 SOCIAL CAPITAL & THE INTERNET
and advantageous for people who find it difficult to maintain those ties. Also, the Internet makes it possible to learn detailed information about persons and topics of interest. Furthermore, it offers a forum to acquire cheap and easy information about the different topics as the stock markets and thus could reduce the transaction costs. This forum of information can reduce uncertainty, which is in turn a necessity for developing norms of trust and reciprocity (Berger & Calabrese, 1975). However, others argue that the Internet could create narcissism and reduces sociability because there is no need for social face-to-face interactions (Fernback, 1997). Olken (2009) confirms that TV consumption crowds out social participation in the context of a developing country. Furthermore, it is argued that it could lead to self-selection and even ideological segregation (Mullainathan & Shleifer, 2004; Sunstein, 2009).
Bauernschuster, Falck, Woessmann (2011) analysed how broadband Internet affects several dimensions of social capital by exploiting the GSOEP panel data. In this study they measure social capital based on the definition of Putnam (1995) and made a distinction between formal and informal social capital. By means a value-added model and instrumental variables to account for the endogeneity, they showed that Internet does not decrease social capital, but has a positive effect on it.
Valenzuela, Park and Kee (2009) examined whether the social network site ‘Facebook’ is related to attitudes and behaviours that enhance individuals’ social capital. Their study showed a small, but positive significant relation between intensity of Facebook use and students’ social capital defined as life satisfaction, social trust, civic engagement and political participation. Additionally, Facebook also enhances bridging capital, (Ellison, Steinfield and Lampe, 2007) and provides a medium in which weak ties are easily to form and maintain, because the technology is well suited for this.
Bogan (2008) studied the relation between the availability of Internet and the possible implications for the stock market participation puzzle. She theorized that due to the rise of Internet, the information costs and transaction costs of participation decreased significantly. By analysing panel data she found convincing evidence of an overall increase in stock market participation by Internet using households compared to household that do not use the Internet. In conclusion, this means that that these types of costs do significantly affect stock market participation and are affected by the Internet.
2.2 THEORETICAL FRAMEWORK
study is defined by the general trust of a household. The latter is defined by informal social participation, which involves weak ties due to the hypothesised likelihood of acquiring new information. Second, this is followed by a theoretical framework on how Internet could affect trust, sociability and stock market participation.
2.2.1THEORETICAL FRAMEWORK:SOCIAL CAPITAL & STOCK MARKET PARTICIPATION
The theoretical framework of Georgarakos and Pasini (2011) predicts that social capital affects stock market participation by two separate channels. This will be explained by means of a household portfolio model.
The first theory in the analytical framework of Georgarakos and Pasini (2011) is related to trust and its effect on stockholding. Financial agreements, such as stocks, require a sufficient level of transparency and trust, because the buyer of the agreement must belief that the issuer keeps its agreement and will be able to repay the investment in the future. Also, he must belief in the case of litigation the attorney costs are bearable and the juridical process is fair. Often, these conditions can be achieved with sound institutions and law, but when a country or a region lacks efficient working institutions and law systems, generalized trust could be a substitute (Guiso et al., 2004). In this theoretical framework, trust acts as a channel, which can increase the likelihood of participation by increasing the expected return of an investment.
In the model there are two investment options available, a safe asset with a risk-free return for simplicity, and a risky investment option with an uncertain distribution. The investor, as a homo economicus, wants to maximize its utility. Therefore he will participate if the expected return of the investment in the risky asset is larger or equal2 to the expected return of the safe asset. The difference between the return of the risk-free asset and the risky asset is the return of bearing the risk, also known as the risk premium. Within this premium the investor may face other risks, which are not accounted for, like a not enforceable contract or high costs in case of the previous mentioned lawsuit. This is measured with mistrust, the probability of other risks, which is in turn independently of the probability distribution of the risky investment return that investors assign to the event that the value of the investment goes to zero. This assumption implies that mistrust works as a discount function of the expected return of the investment. Hence, trust acts as a channel that impacts the likelihood of stock market participation: the lower the mistrust, the higher the expected return, the higher the expected utility of stock market participation.
The second theory is based on structural social capital and its hypothesised implications. This part of social capital plays its role by acquiring cheap and easy information about the stock market, retrieved by social interactions. As stated earlier, social interactions can serve as word-of-mouth information or observational learning, which in turn reduces information costs. Again, this effect is mainly created by the weak ties, which are important for acquiring ‘new’ information
(Granovetter, 1973). This is explained in the portfolio model with fixed participation costs, which is in turn an important determinant that influences the likelihood of stock market participation. These costs consist out of tangible costs such as brokerage fees, and intangible costs such as time, acquiring information and keeping up with market developments. In the model, this is incorporated by the introduction of a fixed cost of participation, which has to be paid when entering the market. Hence, the investor has the choice of not participating in the stock market and safes the fixed costs or he can invest, has to pay the fixed costs and has to allocate its disposable income (Wealth minus fixed participation cost) between the safe and risky asset. These fixed costs can be reduced by acquiring information in which sociability plays a major role. Sociability allows individuals to reduce the fixed participation costs by cheap information sharing. By continuing this reasoning it therefore also affects disposable wealth by augmenting it and thus increases the likelihood of participation. In turn, the fixed cost of participation depends on the number of individuals that participate in the stock market in the social network of that particular investor. The more participate within that network, the higher the likelihood to gain information about the stocks. Hence, the fixed costs are decreasing in sociability. The network works as a word-of-mouth information acquiring. Thus, by having a larger social network, the likelihood to meet people who invested in the stock market increases. Hence, the likelihood of participation increases.
In conclusion, the investor chooses the option, investing or not investing, in order to maximize his expected utility, which depends on trust, sociability, wealth and expected return. This theoretical framework leads to the following hypotheses:
H1: Trusting households are more likely to invest in the stock market than their counterpart non-trusting households, given the level of sociability.
H2: Social household are more likely to participate in the stock market than their counterpart on-social households, given the level of trust.
H3: The effect of trust of a household on stock market participation should be higher for the wealthy, than for their less wealthy counterparts.
2.2.2 THEORETICAL FRAMEWORK: INTERNET USE, SOCIAL CAPITAL & STOCK MARKET
PARTICIPATIONThe economic rational behind the Internet use and its effect on social capital and stock market participation is in threefold. First, there could exist a direct effect on the likelihood of stock market participation, because it reduces the transaction costs of buying, holding, following and maintaining stocks (Bogan, 2008). The second channel is that the Internet could increase sociability. It provides a forum to meet and maintain relations. Additionally, it also provides a way to gain information about organizations and clubs of interest and meeting times. Hence, it could enlarge the social network of the potential investor and thus the weak ties. As explained in the theoretical model of Georgarakos and Pasini (2011), this enlargement of network could in turn decrease the fixed participation costs and therefore adds to the likelihood of stock market participation. The third channel in which Internet could enlarge the likelihood of stockholding is through the channel of trust. Internet could help to trust unfamiliar, enlarge networks, and reduce uncertainty by the availability of easy information. This in turn is a necessity for building trust (Berger & Calabrese, 1975). According to the previous specified model, an increase in trust, leads to a higher expected return of an investment, therefore enhances the likelihood of stock market participation. This leads to the following hypotheses:
H5: Households that make use of the Internet are more likely to be trusting. H6: Households that make use of the Internet are more likely to be social.
H7: Household that make use of the Internet have a higher likelihood to participate in the stock market.
Again, these hypotheses represent the alternative hypotheses. The corresponding null hypothesis will be tested, which assumes that the corresponding parameter is equal to zero. That is, there is no significant effect of Internet use on the likelihood social capital, defined as sociability and trust, and stock market participation.
3. RESEARCH DESIGN
SOCIABILITY
GENERAL TRUST
INTERNET STOCK MARKET
PARTICIPATION
cross-sectional analysis towards the effect of the Internet on social capital and stock market participation. Within cross-sectional analysis it is not possible to give causal interpretations about the results, because the model suffers from multiple endogeneity concerns. Therefore the structure of the dataset is exploited and different types of panel models are used to account for possible heterogeneity bias and causal interpretation. Figure 1 presents the conceptual model.
FIGURE 1:CONCEPTUAL MODEL
3.1 RESEARCH METHOD
The aim of this study is to estimate the causal effect of Internet use on social capital and stock market participation. To determine causality it is important that the model has to comply two factors (Verbeek, 2012). First, there has to exist a correlation between the independent variable and the dependent variable. Second, and most importantly, the independent variable has to be exogenous. While the first factor is relatively easy to estimate, the second factor is harder to account for. If the predictor not exogenous, it is so called endogenous. The problem of an endogenous regressor is that it is correlated with the error term. This will lead to inconsistent and biased estimators and therefore not possible to account for causality. Endogeneity bias could be caused by multiple factors as: measurement error, such as misreporting in the surveys; omitted variables, which are correlated with the predictor and dependent variable; reverse causality, in which the predictor not only has an effect on the dependent variable, but also the other way around. While it is hard to determine causality this study, because the variables are subject to various endogeneity concerns, there are some models that could offer possibilities, which will be discussed elaborately.
solutions by controlling for variables, which are not observed or cannot be measured. This in turn reduces identification problems of the endogenous regressors, caused by omitted variable bias. This creates the advantage to estimate more precise and realistic models under certain assumptions and brings the study a step closer to determine causality. Therefore, in the second part of the analysis, multiple panel models are employed to determine the causal relationship between Internet use, social capital and stock market participation. Below, the different models are explained elaborately.
3.1.1 CROSS-SECTIONAL ANALYSIS: SOCIAL CAPITAL & STOCK MARKET PARTICIPATION
To test whether social capital has an effect on stock market participation a cross-sectional analysis is conducted, based on the study of Georgarakos and Pasini (2011). Out of the literature review, sociability and general trust could explain the variation in stock market participation, hence will be accounted for as the main predictors in this model. This effect will be estimated by a binary choice model, due the nature of the outcome variable investment in the stock market. This leads to the following econometric specification:Yi =Xi'β+γ1Generaltrustj +γ2Sociabilityi +ui (1)
1 if U(A)>U(B)
Where Yi = , and ui ~ N(0,1)} 0 otherwise
By means of a probit model the underlying latent variable will be estimated by the maximum likelihood. Yi is the binary dependent variable, which equals 1 if the household, i, invested in the stock market and zero otherwise. Both the main independent variables general trust and sociability are binary in nature and equals 1 if at least one of the household members is trusting or social. The model controls also for a vector of demographics, Xi that are discussed elaborately in the part data collection. These variables include among others: wealth, education and personality traits, which in turn could affect the dependent variable according to the literature review. It should be noticed that sociability is a proxy for the unknown ‘the mount of peers that participate in the stock market around the household’. By means of regional data it could be tested whether the effect of sociability is higher in regions with high density of stock market participation. Unfortunately, this was not possible, because regional data was not available. In this study, sociability is assumed to be exogenous if one takes the assumption that no one of the households participates in one of the social organization to learn about stock holding. Furthermore, to account for endogeneity caused by omitted variable bias, certain personality traits are added into the model, which are discussed more in detail later on.
individually affect the aggregated trust-level (Knack & Keefer, 1977). General trust is thus an endogenous belief about the behavior of others, in other words, individuals are trust-takers. The question used in this study is the reported trust-level. As mentioned earlier, also this self-reported trust-level is subject to some measurement issues as that it primarily predicts trustworthy behavior than it predicts trusting behavior (Glaeser et al, 2000). Georgarakos and Pasini (2011) handled this endogeneity problem by assigning each individual the regional average of trust. Again, since there is no regional data available, this is unfortunately not possible. Therefore, to account for the endogeneity of this variable an instrumental variable method is employed. The reasoning behind an instrumental variable is that this instrument will isolate the part of the endogenous independent regressor that is uncorrelated with the error term, the exogenous part (Verbeek, 2012). It is possible to use multiple instruments. There are two conditions needed to account for a valid instrument.
1. Instrument relevance: Cov(z,x)≠0 2. Instrument exogeneity: Cov(z,ε)=0
The first one is related to relevance and states that the variation in the instrument should be correlated with the variation of the endogenous regressor. The second one states that is that the instrument has to be exogenous. In other words, the instrument should not be correlated to the error term. If the instrument fulfills these two assumptions, then by means of a Two Stage Least Square estimator the coefficient can be estimated and will be consistent. This method is, as the name already indicates, performed in two steps. In the first step, the endogenous variable is regressed against the instrument and decomposed into two elements: a component that may be correlated with the error term and a problem-free component uncorrelated with the error term. In the second stage, the problem-free component is used to estimate the coefficient. Hence, the two-stage model can be described as follows:
Investmenti=Xi'β+γ1Generaltrustj +γ2sociabilityi +ui (2) Generaltrustj =Xi'β+ γ1Z +γ2sociabilityi +ui (3)
3.1.2
CROSS-SECTIONAL ANALYSIS:
INTERNET USE, SOCIAL CAPITAL
& STOCK MARKET
PARTICIPATIONY i =Xi'β+γTotalinternetusej +ui (4) 1 if U(A)>U(B)
Where Yi = , and ui ~ N(0,1)} 0 otherwise
Also these models are binary of nature. Yi represents the dependent variables sociability, general trust and stock market participation. This equals 1 if at least one of the household members is social, trusting, or participates in the stock market. The predictor total Internet use is a continuous variable. Furthermore, a vector of demographics is added in this model, which are the same controls as the model above.
Also in this model causal interpretations of the predicted estimators could be hindered. The model could suffer from multiple endogeneity problems, caused by omitted variables and reverse causality. It is possible that a social interested individual uses the Internet to gain more information about the social activities Likewise, trusting individuals are more likely to use the Internet because they are more open to it than their non-trusting counterparts. Furthermore, stock market interested individuals make more use of the Internet to find more information about the stock market. Additionally, there could exist unobserved heterogeneity that the model does not control for, which could affect both Internet use and the dependent variable. These problems could lead to upward or downward biased estimators. Hence, the results of this cross-sectional analysis will give primarily a correlation, than it provides a causal interpretation. To account for these problems also a panel analysis is conducted after the cross-sectional analysis and discussed in the next part.
3.1.3 PANEL ANALYSIS: SOCIAL CAPITAL & STOCK MARKET PARTICIPATION
As stated before, with cross-sectional analysis it is hard to give a causal interpretation of the predicted estimators, because there are multiple endogeneity concerns. Therefore, various non-linear panel methods are employed. It will start with the analysis of to the effect of social capital on stock market participation. The following econometric specification for the binary panel analysis is formulated in terms of the underlying latent model:
y*it = Xit'β + αi+ uit (5) 1 if y*>0
Where y*i =
Again, both the independent variables general trust and sociability are binary in nature and equals 1 if at least one of the household members is trusting or social. Furthermore, this panel model controls for the same vector of demographics, X, also used on the cross-sectional analysis. First, the analysis will starts with a pooled logit regression. This will not measure causality, but an average correlation corrected for correlation over time for a given individual, therefore likely more precise than cross-sectional (Verbeek, 2012). As stated before, the model suffers from endogeneity concerns, which have to be accounted for. There could exist some unobserved heterogeneity within individuals, which causes individuals to be more social or trusting and at the same time to be more likely to invest in the stock market as. Whereas it is in cross-sectional estimation common to make use of instrumental variables to account for endogenous regressors caused by omitted variables, panel analysis can account for this by the use of fixed effects estimation. However due to the binary nature of the dependent variable this has to be estimated by means of a conditional fixed effects model in which the estimation is executed by the conditional maximum likelihood. This fixed effects model concentrates on within individual differences and accounts for unobserved heterogeneity. It assumes that if there exist certain unobserved characteristics, it is correlated with the dependent variable and independent variable, which in turn could bias the estimators if this is not controlled for. The fixed effects model removes this individual specific effect, denoted by αi. Additionally, this study made use time fixed effect, which controls for variables, which are constant across households, but change over time. A small note on the fixed effects model is that this model only accounts for the within change of households (Verbeek, 2012). When this is non-existing, the variable will be dropped out of the model and the observations will decrease significantly. This also means that this model can’t estimate the effect of time-invariant predictors, as gender. Additionally, if the chosen variables show little within variation it will lead to large standard errors. In this study, one of the main predictors general trust is most likely subject to slow within change. For the reason that it takes time and effort to build trust it is rather time-invariant. This could mean that the fixed effects model is not an efficient model to estimate the effect of social capital on investment. Therefore, also a random effects model is used to measure the effect of general trust and sociability on stock market participation. This model assumes that the unobserved characteristics are not correlated with the predictor, but are random and varies between individuals. In other words, the independent variables are exogenous. This is rather a strong assumption and unlikely to hold within this study. A big advantage of this model is that it can predict the time-invariant estimators.
reasoning behind this test is that one model is consistent under both the null-hypothesis and the alternative hypothesis and one only under the null-hypothesis. Hence, when the estimates differ significantly, the Hausman test will reject the null-hypothesis. Often, when the null hypothesis is rejected, it is assumed that the fixed effects model is the preferred model. Rejection can be caused by the likelihood of correlation between the omitted variables and the regressors. The interpretation of this test should be taken with caution, because there could be other reasons why this test is rejected or accepted (Verbeek, 2012). Therefore, the choice of the right model should be built an underlying economic theory.
3.2.2 PANEL ANALYSIS: INTERNET, SOCIAL CAPITAL & STOCK MARKET PARTICIPATION
Based on the literature review and theoretical framework, Internet use is hypothesised to have a causal effect on social capital and stock market participation This part of study assesses this prediction. Initially, conditional fixed effects estimation seemed to be most appropriate, due to the plausible unobserved and unmeasured heterogeneity within households for which the fixed effects estimator can account (Verbeek, 2012). Unfortunately, the conditional fixed effects estimator was not possible to conduct, due to the low levels of within variation of the subjects. Therefore, the method of first differencing is employed by which the estimators are predicted by means of a multinomial logit regression. This method enjoys the same advantages as fixed effects model does. First, by means of first differencing, the individual unobserved time-invariant effects are removed and are not longer an unmeasured problem. This estimation would in turn lead to the same unbiased results as the fixed effects estimator would. Normally in a linear regression, the first differences are regressed by applying OLS to the equation. In this case, this is not possible, because to model is binary of nature. This also means that a normal logit regression is not suitable, because the first differences of the dependent variables are not bounded anymore between 0 and 1. The first differences will result in -1,0 and 1, a negative change, no change and a positive change, respectively. Therefore, a multinomial logit model is employed in which the household has a choice between M (in this case 3) alternatives. The estimators are estimated by the conditional maximum likelihood. The chosen alternative contains the highest utility. In this model a reference category is chosen and the predicted estimators holds the probability of one alternative compared to the probability of the reference category.As mentioned earlier, the Internet suffers also from endogeneity bias caused by reverse causality. To account for this, a lagged predictor of Internet is added in the model. This lead to the following model specification:
Specification 6 shows that this model needs three time periods. For the dependent variable and control variables, the first difference represent the difference between the years 2012 and 2010. For the independent endogenous variable Internet use, the first difference represents the difference of Internet use between the year 2010 and 2008. The reasoning behind a first differenced lagged predictor is that this predicted estimator could be correlated with future independent variable Internet use, but it cannot be correlated with the future error term (Vaisey and Miles, 2014). Thus Internet use predicts, general trust, sociability and stock market participation. Not the other way around. Hence, this addresses the endogeneity concern of reverse causality.
3.3 DATA COLLECTION
To study whether the Internet has lowered the barrier to participate in the stock market, the LISS panel is used. This panel is a longitudinal data study, which consists out of more than 4500 households and 7000 individuals. It is a representative sample of the Dutch population in which the selected panel members complete every month online questionnaires. For the cross-sectional analysis and the panel study the years 2008, 2010 and 2012 years are employed3.
The main variables retrieved from the LISS-panel are based on the study of and Guiso et al. (2008) and Georgarakos and Pasini (2011). These are elaborately explained in the appendix, table 1. The main dependent variable of this study is stock market participation. It should be noted that within the LISS survey stocks are grouped with other investments4 with various risk-classes, which made it not possible to distinguish between them. To measure whether a household participates in the stock market, a dummy is created, which equals 1 if the household invested in the stock market. The main independent variable, Internet use, is measured as the total amount of hours a household spend per week on average on the Internet. To measure generalized trust this study uses a question that is used by many other studies as a proxy for social capital (Glaeser et al., 2000; Guiso et al., 2004; Georgarakos & Pasini, 2011):
In general, would you say that most people can be trusted, or that you can’t be too careful in dealing with people? This question is measured on a Likert scale that ranges from 0 to 10. Where 0 equals: you can’t be too careful enough, whereas 10 equals: most people can be trusted. The variable general trust is converted to a dummy variable, which equals 1 if one of the household members is trusting5.
3 The years 2008 2010 and 2012 are used, because the main explanatory- and dependent variables are only measured in these particular years.
4 The LISS panel asks the respondent the question whether he or she possesses one or more of the following assets. The corresponding answer that measures stock market participation is investments. Investments in this study consists out of growth funds, share funds, bonds, debentures, stocks, options, warrants and so on.
As stated before, general trust suffers from endogeneity bias. To solve this an instrumental variable method is employed and two instruments are used in this study. The first instrument, neighborhood, is computed out of questions based on neighborhood characteristics. The second instrument is measured by questions about perceived safeness of the neighborhood. Table 1 contains the questions used to compute the instruments.
As stated earlier, there are two assumptions at which the instruments must comply, but more importantly, one must underpin the chosen instruments with an economic theory. The first assumption includes the relevance of the instruments. Theoretically, the chosen instruments are perceptions about the local neighborhood and corresponding safeness, which are given and could alter the trust-level of the respondents, but are not directly related to stock market participation. It is possible to perform an empirical test to measure the relevance of the variables by checking for possible correlation. Table 2 shows the correlation between the endogenous variable general trust and the chosen instruments.
TABLE 1
DESCRIPTION INSTRUMENTS: RETRIEVED FROM LISS-PANEL: CONVENTIONAL AND COMPUTER CRIME VICTIMIZATION
Neighborhood Possible answers Ac12c015: People in this neighborhood are willing to help each other.
Ac12c018: People in this neighborhood don’t get along so well.
1. Disagree entirely 2. Disagree
3. Neither agree, nor
disagree
4. Agree 5. Agree entirely
Safe Possible answers
Ac12c013: How often does it happen that you leave valuable items at
home to avoid theft or robbery in the street?
Ac12c014: How often does it happen that you make a detour, by car
or on foot, to avoid unsafe areas?
1. (Almost) never 2. Sometimes 3. Often
4. I don’t know/prefer not
to say
TABLE 2
CORRELATION MATRIX ENDOGENOUS VARIABLES AND INSTRUMENTS
2008 2010 2012
[1] [2] [3] [1] [2] [3] [1] [2] [3]
[1] General trust 1 1 1
[2] Neighborhood 0.13 1 0.14 1 0.20 1
[3] Safe 0.16 0.17 1 0.15 0.15 1 0.13 0.18 1
The second assumption involves the required exogeneity of the instruments, which implies that the instruments are not correlated with the error term. In this case neighborhood characteristics do not have a direct obvious relation with stock market participation. It is possible to test for exogeneity of the instruments, because the number of instruments exceeds the number of endogenous regressors, that is, it is over identified. This is done by Newey's minimum chi-squared test. The null-hypothesis implies that the instruments are valid, in other words, uncorrelated with the error term.
Additionally, it is also possible to test for endogeneity of the instrumented variables, in this case general trust. This is done by a Wald-test, which is provided by Stata automatically when modeling the ivprobit. The null hypothesis tests for the imaginable exogeneity of the instrumented variable. Under the null-hypothesis, the ivprobit and probit are both consistent and should only differ by sampling error. Under the alternative hypothesis, which rejects the exogeneity assumption, the regressors are endogenous and an instrumental variable method is in place. Even if the null-hypothesis is rejected, one could look at the relevance of the instruments, and keep in mind that the economic theory behind it is the most important part.
person to be less social, but also to a lower likelihood of a risky investment. This is measured by a question out of the LISS Health study in which is asked to the level of own risk the household takes on for a health insurance. This equals 1 if at least one of the household members is risk seeking7. Sociability may also be related to pessimism. If a person is pessimistic it is most likely less social and in also less likely to participate in the stock market, because of the pessimistic expectations of the returns. In this study this is measured by self-reported depression by the question:
‘’In general, how to you feel?’’
This is again converted into a dummy variable, which equals 1 if one of the household members is depressed8. Also subjective health indicators are taken into account. The health of a household member could lead to a less social household, because it is more enthralled at home. In turn, it leads to a lower likelihood of stock market participation because households with health problems are discouraged by financials to invest in the stock market (Rosen and Wu, 2004).This is accounted for by the self-indicated question about the respondents’ health:
How would you indicate your health, generally speaking?
This is converted into a dummy variable, which equals 1 if one of the household members indicates its health with ‘very poor’.
The LISS panel contains also a broad set of background variables and demographics of the household members. These are included as control variables. Previous studies concluded that stock market participation is increasing in wealth, income, education and age, which are included in the model (Bertaut and Starr-McCluer, 2000). The continuous variables net income and wealth are converted by an inverse hyperbolic sine transformation to account for the non-positive values and to convey the expected nonlinear relationship with investment. Education is measured on grounds of the highest achieved level of education with diploma of the household. This equals 1 if the highest education is high school or university9. Furthermore, age is measured as the average age of the household and used as a second order polynomial due to the expected concave relationship with stock market participation. The model also considers gender, marital status, employment status and retirement. All of them are based on a household level.
7 A household is defined as risk-seeking when at least one of the household members has indicated a higher level of own risk insurance than the median level of €200.
8 A household is defined depressed when at least one of the members answered below median depressed of 5. This was measured on a Likert-scale, which ranged from 0-10
4. EMPIRICAL RESULTS
This part contains the empirical results of the different models. First, the cross-sectional results over the year 2012 will be elaborately discussed in this chapter. The same analysis is conducted over the other years and added in the appendix. The cross-sectional analysis is conducted by means of a probit model. This because this particular model allows to account for the endogenous variable general trust by means to a two stage least square estimation, whereas logit model does not. Second, the results of the panel analysis will be discussed. This is in turn estimated by a logit model, because this model allows to conduct a conditional fixed effects estimation, whereas the probit does not. In general, the results of probit and logit should similar. Both models have zero mean, however the variance of the standard normal distribution in case of the probit model is equal to 1, while the variance of the standard logistic distribution is equal to π2/3 (Verbeek, 2012). The outcomes resulting out these nonlinear models are not directly interpretable. Therefore they are converted into marginal effects and odds ratios.
4.1EMPIRICAL RESULTS: CROSS-SECTIONAL ANALYSIS
The following provides the results retrieved from the various cross-sectional analyses toward the effect of social capital on stock market participation and the effect of Internet use on social capital and in turn stock market participation.
4.1.1SOCIABILITY AND TRUST
minimum chi-square test is conducted to test for the validity of the instruments, that is, whether they are exogenous and not correlated with the error term. This test-statistic results in a number of 0.445 (P=05049), which cannot be rejected, and hence it can be concluded that the instruments are exogenous. In conclusion, the chosen variables comply the two assumptions of exogeneity and relevance and therefore are valid instruments in the sample of 2012.
By means of a probit regression the likelihood of stock market participation is regressed on different household characteristics and the main predictors sociability and general trust separately. Table 4 shows the average marginal effects and standard errors resulted from the regressions. Model 1 and model 3 controls for the main predictors general trust and sociability separately, without controlling for personality traits. These controls are accounted for in model 2 and 4. Model 5 includes both predictors and controls variables. For the models without the estimation of instrumental variables, McFadden’s pseudo R2 is denoted. Because it is a nonlinear regression, it is gives not the same interpretation as the R2 in the nonlinear regression (Verbeek, 2012). It mainly gives a comparison between a model that only has a constant as explanatory variable and a model with the chosen predictors. The larger the difference, the more the full model adds to the restrictive model.
As the table shows, there are 1086 observations left when only controlling for the variable sociability, whereas 1050 observations are left in the model when controlling for general trust. In each of the models, the likelihood ratio – and the Wald chi-square test are rejected, which indicates that at least one of the regressors in the model is not equal to zero10.
Model 1 shows a significant positive marginal effect of sociability on stock market participation. This effect remains strong when one controls for the different personality traits. Social households have a 0.0797 higher probability to invest in the stock market than their non-social counterparts, ceteris paribus. This effect slightly decreases to a 0.0678 probability when the personality controls are added, but remains statistically significant. Also, general trust appears to have a strong independent positive effect on stock market participation and this remains when the controls for the personality traits are added. Without the controls for personality traits, trusting households have a 0.841 higher probability to invest in the stock market, than non-trusting households. When the controls are added, this effect also drops slightly to 0.709, but again remains statistically significant.
In model 5, both predictors and controls are added in one model. Again, the likelihood decreases slightly for the trusting households to 0.633, but remains statistically significant. In turn, sociability loses its significance when it is estimated in the same model as general trust. This means that in this test, the null hypothesis of sociability as represented in section 2, cannot be rejected. In turn, the null hypothesis of trust, represented in table two can be rejected in this model, which means that trust has a significant independent impact on the likelihood of stock
market participation. Moreover, the control variables contain the expected sign, similar with results of other studies, but are not all statistically significant as they were expected.
TABLE 4
THE EFFECT OF SOCIABILITY AND GENERAL TRUST ON STOCK MARKET PARTICIPATION (YEAR: 2012)
(1) (2) (3) (4) (5) VARIABLES Marg. Eff. Marg. Eff. Marg. Eff. Marg. Eff. Marg. Eff. Sociability 0.0797*** 0.0678** 0.134 (0.0284) (0.0291) (0.0993) General trust 0.841** 0.709** 0.633* (0.336) (0.332) (0.351) Net income 0.0508** 0.0468** 0.108 0.105 0.105 (0.0207) (0.0212) (0.0666) (0.0672) (0.0670) Wealth 0.00693*** 0.00581*** 0.0178*** 0.0159*** 0.0161*** (0.00194) (0.00198) (0.00616) (0.00616) (0.00617) Age -0.0131 -0.0139 -0.0326 -0.0382 -0.0393 (0.00889) (0.00903) (0.0275) (0.0279) (0.0278) Age2 0.000132 0.000147* 0.000300 0.000382 0.000394 (8.69e-05) (8.84e-05) (0.000271) (0.000274) (0.000273) Couple -0.0617 -0.0587 -0.238 -0.212 -0.218 (0.0472) (0.0487) (0.149) (0.152) (0.152) Single male 0.0287 0.0140 0.106 0.0767 0.0895 (0.0687) (0.0706) (0.213) (0.218) (0.218) Nr. children 0.00310 0.00242 0.00423 -0.000884 -0.00604 (0.0175) (0.0181) (0.0547) (0.0565) (0.0565) High school -0.0425 -0.0502 -0.0928 -0.115 -0.113 (0.0360) (0.0364) (0.113) (0.114) (0.114) University 0.174*** 0.180*** 0.461*** 0.483*** 0.478*** (0.0383) (0.0390) (0.125) (0.129) (0.129) Employed 0.0843* 0.0783 0.0930 0.103 0.109 (0.0508) (0.0514) (0.171) (0.172) (0.173) Retired 0.0397 0.0306 -0.00325 -0.0128 -0.0158 (0.0509) (0.0511) (0.169) (0.169) (0.168) Depressed -0.0672 -0.0899 -0.0744 (0.0607) (0.199) (0.198) Poor health -0.0682 -0.160 -0.152 (0.144) (0.435) (0.433) Risk seeking 0.0648* 0.192* 0.194* (0.0334) (0.104) (0.104) Observations 1,086 1,050 1,082 1,050 1,050 Log Likelihood -607.91482 -585.12179 -1290.5838 -1290.5838 -1241.5185 LR (Wald)Chi2* 68.12 69.25 63.87 65.05 69.04 Prob>Chi2 0.0000 0.0000 0.0000 0.0000 0.0000 Pseudo R2 0.0531 0.0559
*Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
** This accounts for the Wald chi2 in the ivprobit. and is referred to in the other models F-statistic= 11.29
The proxies for poor health and depression are not significant (but contain the expected sign). In turn, the proxy for risk attitude has a significant positive impact on stock market participation. Risk seeking households have a 0.194 higher probability to invest in the stock market than their non-risk seeking counterparts, ceteris paribus. Additionally, holding a university diploma is strongly significant and also wealth explains a significant part of the variation.
It can be concluded that in the cross-sectional analysis of 2012 general trust has a significant effect on stock market participation. To paraphrase, trusting households are more likely to invest in the stock market than their non-trusting counterparts. To test for potential heterogeneity, the sample is grouped in ‘below medium wealth’ and ‘above medium wealth’. According to the predictions of Georgarakos and Pasini (2011), the effect of general trust on stock market participation should be mainly strong for the wealthy, while the effect of sociability on stock market participation should be more significant in places where there is a high-density of stock market participation. As stated before, this last statement is unfortunately not testable in this study.
TABLE 5
THE EFFECT OF GENERAL TRUST AND SOCIABILITY BY BELOW AND ABOVE MEDIUM WEALTH
Below median wealth
(Obs. 514)
Above median wealth
(Obs. 528)
Full sample
(Obs. 1050)
Variable Marg.Eff. (Std. Error) Marg.Eff. (Std. Error) Marg. Eff. (Std.Error)
General
trust 0.902*** (0.503) 0.249 (0.474) 0.633* (0.351) Sociability 0.090 (0.147) 0.213 (0.140) 0.134 (0.099)
* Medium net housing wealth in 2014 is €120,000
this case, this effect is significantly stronger for the group above medium wealth, than the group below medium wealth (insignificant effect of trust) as predicted by the theoretical model11. In the year 2010 the opposite applies. The effect of trust is not significant, but sociability in this case is. A social household has a 0.261 higher probability to invest in the stock market. This is effect is stronger for the less wealthy (marg. Eff.: 0.345).
The different cross-sectional analyses presents contradicting results for the predictors between the different years. A potential reason could be that the years 2008 and 2010 suffer from weak instruments. While the chosen instruments resulted to be relevant and exogenous for 2012, for the years 2008 and 2010 this is not the case (2008: F=6.07<10; 2010: F=5.14<10). This could severely bias the estimators and is thus not useful to account for the endogeneity of the variable general trust. Additionally, there could be other unobserved characteristics, which are not included in this model. Another problem arises if the self-indicated proxies for personality traits are not measuring the real personality trait. These problems could hamper the results. Hence, a panel analysis would probably result in more precise and consistent estimators.
4.1.2
INTERNET, SOCIAL CAPITAL & STOCK MARKET PARTICIPATION
The next part of the cross-sectional analysis tested the effect of Internet use on sociability, general trust and stock market participation. Table 6 presents the different estimated models in which the estimators are converted into average marginal effects. Model 1 specifies the effect of Internet use on sociability. Model 2 specifies the effect of Internet use on general trust. Model 3 accounts for the effect of Internet use on stock market participation.
All three models account for the demographic controls and chosen personality traits, which were also added as controls in the previous cross-sectional analysis. As model 1 presents, it appears that no relationship exists between Internet use and sociability. That is, if people are increasing their Internet use, there does not exist a positive nor a negative significant impact on being social, classified as participation or performing voluntary work in one of the social organisations. Interestingly is that the chosen personality traits depressed and poor health represents the expected sign in relation to sociability (except for risk seeking).
Model 2 shows the effect of Internet use on the trust level of the household. There appears to be a positive significant effect of Internet use on the trust level of a household. Because Internet use is a continuous variable, the marginal effect of Internet use is the instantaneous rate of change, whereas discrete variable measures the discrete change. If one would assume that the assumptions comply within this model, that is, the predictors are exogenous, it means that an instant change increase in Internet use, leads to significant increase in the probability of being a trusting household by 0.00332. Moreover, this model indicates that there are positive significant
effects of net income, holding a university diploma, being employed and retired on general trust, as expected. Fascinating, although beyond the scope of this study, is the result for depression on general trust, which is significantly negative. To paraphrase, households who report to be depressed are more likely to have lower trust levels.
TABLE 6
THE EFFECT OF INTERNET USE ON SOCIABILITY, GENERAL TRUST & STOCK MARKET PARTICIPATION (2012)
(1) (2) (3) Dependent Variable Sociability General trust Investment VARIABLES Marg. Eff. Marg. Eff. Marg. Eff. Internet use -6.94e-05 0.00332*** 0.00447*** (0.000671) (0.000716) (0.000627) Net income 0.0492** 0.0571** 0.0371* (0.0232) (0.0238) (0.0213) Wealth 0.000532 0.00167 0.00464** (0.00215) (0.00219) (0.00202) Age 0.00141 0.00166 -0.000596 (0.0104) (0.0107) (0.00941)
Age2 4.79e-06 6.18e-05 6.00e-05 (0.000101) (0.000104) (9.06e-05) Couple 0.0573 0.0717 -0.125** (0.0618) (0.0640) (0.0546) Single male -0.123 -0.0705 0.0195 (0.0807) (0.0834) (0.0703) Divorced -0.0899 0.0189 -0.137** (0.0741) (0.0760) (0.0669) Nr. children 0.0540*** 0.0142 -0.0155 (0.0205) (0.0207) (0.0190) High school -0.0687* -0.0593 -0.0416 (0.0396) (0.0406) (0.0364) University 0.119** 0.132*** 0.164*** (0.0477) (0.0475) (0.0393) Employed 0.0895 0.237*** 0.0732 (0.0576) (0.0599) (0.0512) Retired 0.130** 0.193*** 0.0228 (0.0577) (0.0593) (0.0505) Depressed -0.232*** -0.265*** -0.0975 (0.0644) (0.0698) (0.0603) Poor health -0.0870 -0.0380 -0.0843 (0.138) (0.138) (0.140) Risk seeking -0.00278 -0.0171 0.0411 (0.0385) (0.0392) (0.0342) Observations 1,051 1,051 1,050 Log Likelihood -684.63881 -667.05422 -558.26149 LR Chi2 75.11 114.29 122.97 Prob>Chi2 0.0000 0.0000 0.0000 Pseudo R2 0.0520 0.0789 0.0992
