1
Consumer Confidence and its effect on Dutch Households’ Stock
Market Participation
Author: Maikel Bijker Student number: S3538605
Supervisor: Dr. C. Laureti
University of Groningen Faculty of Economics and Business
MSc Finance
Abstract
Keywords: Non-participation Puzzle, Consumer Confidence, Stock Market Participation JEL-classification: D10, G10, G40
Word count: 10.979
2
1. Introduction
Since the global financial crisis, stock markets have recovered remarkably well. Some indexes, e.g. the S&P 500 Index and the MSCI World ex USA Index, have reached abnormal numbers1. The S&P 500 Index logged returns at 24% over 2017, which was the index’s best achievement since the year 2009, whereas the MSCI World ex USE Index had a return of 37% over the same year. In earlier years, Mehra and Prescott (2003) state that, by participating in markets through long-term investments, high numbers in the stock market can give individuals an opportunity to make some return. Lee et al. (2015) concur, stating that high return expectations should lead to an increase in stock market participation. In addition, due to interest rates being low, high return expectations should lead to more risk taking (Lian et al. 2019). However, despite recent opportunities, there is an overall limited number of individuals that do partake in the stock market. For example, Beshears et al. (2018) state that many households do not own any kind of stock, directly or indirectly, through mutual funds and pension funds.
This phenomenon is generally referred to as the “non-participation puzzle”, which is first described by Mankiw and Zeldes (1991) and is later discussed in more detail by Haliassos and Bertaut (1995). Mankiw and Zeldes (1991) explore the size of the equity premium and through this, they come across reasons for not holding stocks. Haliassos and Bertaut (1995) study these reasons, such as cultural factors, more specifically and explore irrational behavior. In general, traditional theory states that in perfect market conditions, all individuals are rational and consistent, indicating that they make decisions with reason and based on logic. As a result, these rational individuals obtain information in the most efficient way possible, therefore having perfect knowledge for holding stocks (Sacco 1991). In reality, a large majority of individuals does not participate in the stock market. Becker (1962) states that rationality is often not agreed on, hence individuals show this non-participation behavior. Bertaut (1998) find that individuals do not invest in stocks, since when they invest, they do so in riskless assets. This is later supported by Guiso et al. (2003), who states that regardless of investment opportunities, the number of individuals that do partake in the stock market is limited.
In general, there are various reasons why individuals do not participate in the stock market. For example, Dimmock and Kouwenberg (2010) study whether loss-aversion is a reason for households’ non-participation in the stock market, where loss-aversion is defined as an individual’s preference to avoid losses rather than to gain profits. They find that a higher
3 loss-aversion gives a lower chance of participating in the stock market. In addition, Benartzi & Thaler (1995) study the same relationship through which they also find that higher loss-aversion leads to lower stock market participation.
As with loss-aversion, there are other ‘personality traits’ that affect an individual’s participation in the stock market. For instance, Conlin et al. (2015) focus on personality traits such as extravagance (an individual rather spends money than save it), sentimentality (how an individual is moved over another individuals’ emotional motivation) and dependence (an individuals’ eagerness of another individuals’ approval), finding that all of these traits have a negative effect on stock market participation.
Another reason for individuals’ non-participation in the stock market is its cost (Vissing-Jorgensen 2003). Individuals do not have the required knowledge to overcome these costs. Moreover, knowledge towards the risk and return of an investment is mostly absent, which makes it difficult to acquire an equity premium; the difference between returns on stocks and returns on bonds. This is also what Betraut (1998) tells us: that the costs do not weigh up to the benefits.
Overall, these are merely a few factors out of many examples of why individuals choose not to participate in the stock market. Despite all the previous research, the non-participation puzzle remains. As a result, the puzzle gives opportunities for further research. Hence, this study aims to build on previous research regarding the non-participation puzzle, by adding consumer confidence of Dutch households to the literature, to find a possible relationship towards stock market participation.
In this study, consumer confidence is defined as an individual having a positive expectation towards future events, either towards the financial market or towards their own household. Previous research shows that confidence, or ‘optimism’ as some describe it, has its effect on financial decisions. For example, Brown et al. (2005) find that optimistic financial expectations have a positive effect on the holding of debt. Furthermore, Puri and Robinson (2007) show that optimism correlates with people working harder, expect to retire later and save more. Additionally, they find that optimism increases an individual’s investment in stocks.
4 valuable for policy makers and investors to predict consumer spending behavior and to gain an overall indication of the health of the economy (Kwan and Cotsomitis 2004; Bruno 2014; Lahiri et al. 2015). Additionally, understanding consumers’ financial decision making can be of significance to not only investors, but also to banks. That is, using consumer confidence can help banks to take different measures. For instance, adaptation for lower undertaking of lending when the level of confidence is low.
In general, the main relationship this study explores is whether consumer confidence effects Dutch households’ stock market participation. For this purpose, the following research question is formulated:
What is the effect of consumer confidence of Dutch households on their stock market participation?
Data from the Dutch Household Survey (DHS), which is acquired from CentERdata, is used to find a possible relationship between consumer confidence and stock market participation. The survey, which started in 1993 and conducted annually, contains information of over 2000 households in the Netherlands each year. With regards to the survey, this study makes use of the 2008th wave up to and including the 2018th wave.
5
2. Literature review
The literature review is divided into three subsections. The first subsection explains the introduction of the “non participation puzzle” and theory & empirical evidence behind stock market participation. The second subsection discusses the theory and empirical evidence of consumer confidence.
2.1 “Non participation puzzle” and stock market participation
In the mid eighties, Mehra and Prescott (1985) introduced the equity premium puzzle, a phenomenon that analysis why stocks have abnormal higher historic returns than bonds do. These abnormal returns suggest an extreme amount of risk aversion, which the authors find highly unlikely. As mentioned before, the equity premium is the difference between returns on stocks and returns on bonds. Accordingly, this study is what researchers led to begin addressing non-stock market participation, since, even though stocks have higher returns, most households do not hold any stocks. Ever since the non participation puzzle has been around, research has been done on this subject to try to explain this phenomenon.
One of the first studies that address this topic, is from Mankiw and Zeldes (1991). During their research regarding the difference of consumption with regards to the amount spend on food by stockholders and non-stockholders, they explore reasons for not holding stocks. One of these reasons is the lack of liquid assets; assets that are simple to turn into cash. More specifically, they indicate that the lack of liquid assets is mainly due to the fact that consumers have a liquidity constraint, which they define as consumers having a limited amount they can borrow. In addition, the relationship between family characteristics and stock market participation is also something they examine. They find that stockholding increases with average labor income. Furthermore, heads of household who are higher educated are more likely to participate in the stock market.
6 a relationship with stock market participation, as mentioned in various studies. For example, Almenberg and Dreber (2015) examine whether the divergence in gender regarding stock market participation can be explained by diversity in financial literacy. They distinguish basic financial literacy and advanced financial literacy, where the difference between basic and advanced financial literacy equals the knowledge an individual has about financial products and concepts. These differences give an explanation for gender gaps in stock market participation.
Besides genetic factors, such as gender, individual “irrational” behavior can explain the low participation rate in the stock market. Irrationality can be defined as behavior where someone is behaving without using common sense and letting their emotions play a determining role. Akhtar and Muhammed (2017) explore this emotional behavior and take a closer look at the relation between emotional intelligence and stock market participation. They define emotional intelligence as an individual’s functional behavior in a professional and structured environment. In this context, motivation and self-awareness is being investigated. Throughout their research they find that motivation has a positive relationship with stock market participation and self-awareness has a negative relation with stock market participation.
7 Another factor that individuals have as a personality trait is awareness. Guiso and Jappelli (2005) do research on awareness and stock market participation, where they define awareness as individuals being aware of stocks, mutual funds and other financial assets. They specifically limit to financial awareness, moreover the lack of financial knowledge towards investment products. They combine cultural & educational factors and personality traits to explain this relationship. For this, they use Italian survey data regarding household income and wealth. Regarding awareness, they find that the education, household resources and long-term bank relations of people are positively correlated towards awareness of stocks, mutual funds and other financial assets. They also find that a lack of financial knowledge has implications on understanding parts of the stock market.
This financial knowledge is also defined as financial literacy in finance literature, a factor which also sees a rise of interest. Research that stands out the most regarding this particular subject is that of Van Rooij et al. (2007). They study the relationship between financial knowledge and its effect on financial decision making. Throughout their study, they find evidence of an independent effect of financial literacy on stock market participation. People that have low financial pre-knowledge are less likely to invest in stocks or the stock market in general.
Another study on the relationship between financial literacy and stock market participation is Xia and Wang (2014). In particular, they study individuals’ overconfidence with regards to their financial literacy, in other words, people overestimating their knowledge of financial assets. Similar to Rao et al. (2016), Xia and Wang (2014) conduct research on Chinese individuals and acquire data regarding consumer finance. Their main finding is that overconfidence has a positive relationship with stock market participation, with overconfident individuals having an increased likelihood of 20% towards underconfident individuals to participate in the stock market. As with individuals having a certain trust in the stock market, overconfident individuals also have a positive feeling towards their investments.
2.2 Consumer confidence
8 returns which he or she anticipated for. Financial awareness and knowledge in the sense that the more an individual possesses these traits, the more confidence an individual might gain. With these factors in mind, one might also say that the more confidence people have towards the future, where confidence is defined as an individuals’ view with regard to the economy as a whole or to their own household getting better, the more this individual might be attracted to the stock market. Due to this confidence, individuals might start to spend more money on the more luxury assets, such as investing spare money in stocks or other financial assets. On the contrary, it can be difficult to dissentangle the effects of different emotions and confidence, since these can relate to one another. Nonetheless, the effect of confidence, or ‘optimism’ as some research would define it, is already explored on a broader perspective according to different literature. For example, household saving and borrowing behaviour. Confidence plays a vital role when it comes to whether an individual invests his or her money or would rather save it. Klopocka (2016) studies this relation and finds that there exists a relationship between confidence and saving & borrowing, which can be used to forecast financial behaviour. Bram and Lydvigson (1998) support this, who suggest that consumer sentiment, which is defined as the opinion of a consumer about the well-being of the economy, can indeed help predict future movements in consumer spending.
Furthermore, confidence can also lead to ‘overconfidence’. Frydman and Camerer (2016) define this excessive amount of confidence as individuals believing they posses more valuable information than they have in reality. They find that overconfidence has a negative effect on financial decision making, in the sense that individuals do not fully use rational thoughts which gives non-optimal decisions. They mention that not only individuals with low education, but also corporate managers who are highly educated, make these overconfident decisions. This is in line with what Camerer and Lovalla (1999) say about overconfidence and its effect on business. In specific, they state that in most cases overconfidence leads to business failures within a few years. Clearly, being overconfident can have a negative impact not only on behaviour, but also on the business side of view. Overall, overconfidence can been as irrational behavior, in contrast with confidence, which Tonkiss (2009) suggests as rational behavior.
9 investment in stocks. As a result, they find significant postive results with regards to optimism and participation in the stock market, specifically in equity. Following Puri and Robinson (2007), Angelini and Cavapozzi (2017) also studied the effect of optimism on stock investment. They use the same approach, but expand it with three measurements. That is, controlling for cognitive skills, additional personality traits and studying the interaction between optimism and different personality traits. Ultimately, they also find a significant positive relationship between optimism and stockholding. Furthermore, Agarwal et al. (2018) argue that confidence can also affect stock market participation through political uncertainty. When a country faces elections and individuals experience uncertainty and are therefore not confident in the future, this affects their stockholding. In the end, the authors find that households significantly lower their stock holding during these stages.
Moreover, confidence does not only show an effect on stockholding, but also on the stock market and returns in general. For example, Jansen and Nahuis (2002), who study the effect of consumer confidence on the stock market and in particular stock returns. They find a positive correlated relationship between the changes in confidence and stock returns. They state that this relationship is established by expectations which include conditions economically wide, and not so much on conditions related to personal finances. One can also see these switches in uncertainties in a broader perspective regarding consumer confidence and the stock market. Karnizova and Khan (2015) find that switches in consumer confidence affect stock markets, in the sense that it affects stock prices and their volatility. For instance, these switches can cause stock prices and stock market volatility to go up or down.
Overall, the various studies mentioned in this paper explain factors that have an effect on stock market participation. With regards to confidence, the field of research does not only show a broader perspective on which consumer confidence has an effect, such as household spending and saving behavior or even corporate behaviour (Camerer and Lovalla, 1999; Brown et al. 2005), it also shows a significant impact on the stock market and on an individuals’ stock market participation (Jansen and Nahuis, 2002; Puri and Robison, 2007; Angelini and Cavapozzi, 2017; Agarwal et al. 2018). Overall, with the various outcomes of these studies and their explanations, there is belief that in this study consumer confidence might also have an positive effect on households stock market participation, specifically on the Dutch population. Therefore, the following hypothesis is formulated:
10
3. Methodology
This section describes the methodology this study uses to find out whether there exists a relationship between consumer confidence and stock market participation. It is divided into two subsections. A subsection explaining the econometric specifications and a subsection dedicated towards explaining the variables of interest.
3.1 Econometric specifications
When looking at stock market participation in this study, there are two possible outcomes. On one hand an individual participates in the stock market and on the other hand an individual does not participate in the stock market. Due to this, stock market participation is a binary variable, which is a variable that can take only values of 0 and 1. To this end, one can either use a linear probability model, a logit model or a probit model. The linear probability model is an OLS regression, where the dependent variable is also a dummy variable. However, for this study, there are two reasons for not initially using this method. First is that of a certain requirement. With the use of a binary variable, the probability of acceptance lies between the interval 0 ≤ 𝑝𝑖 ≤ 1, where 𝑝𝑖 is the probability of an individual. However, this is not always
11 contains observations of multiple individuals which are observed over two or more points in time. The dataset this study uses has some households that can be observed at multiple points in time, hence, this study uses the latter. Furthermore, the dataset is unbalanced, which is defined as a dataset for which some observations are missing.
Furthermore, this study observes data over time. Therefore, it is certain that it deals with unobserved heterogeneity. This is defined as an issue where a correlation exists between the observable variables (variables of interest) and the unobservable variables. In general, there are two models that deal with this issue that are most used in the field of research: the fixed effects model and the random effects model. The fixed effects model apprehends variables that vary over time, which strengthens the accuracy of the model. However, a major disadvantage of the fixed effect model is that it does not capture variables that do not vary over time, e.g. gender. The fixed effects model only controls for these variables. This indicates that these variables are not considered in the estimation and therefore one loses the ability to establish the influence of such variables. Nevertheless, the fixed effects model does control for unobserved individual characteristics, but only time-invariant ones. In contrast to the fixed effects model, the random effects model is a model through which time-invariant variables do not cancel out, hence, these can be used for the estimation analysis. A major drawback of the random effects model is that it assumes that all explanatory variables are uncorrelated with the error term. If this assumption does not hold, it can lead to a bias called the omitted variable bias, which means that the model withdraws relevant variables, leading to the probability of wrong standard errors, but also biased and inconsistent parameter estimates.
Since the main explicative variable consumer confidence is time-variant because it changes over time, this study uses a logit fixed effects model. A Hausman test also supports this model2. The regression this study uses is shown in the formula below:
𝑆𝑀𝑃𝑖𝑡 = 𝛼 + 𝛽1𝐶𝑂𝑁𝐹𝑖𝑡+ 𝑋𝑖𝑡+ 𝑢𝑖+ 𝜀𝑖𝑡 , (1)
where 𝑆𝑀𝑃𝑖𝑡 is the stock market participation of individual i at time t. 𝐶𝑂𝑁𝐹 is the level of confidence towards the future of individual i at time t. The 𝑋′𝑠 represent a vector of control variables for individual i at time t, which include: age, gender, number of children in the
12 household, level of education, marital status, health status and net income. 𝑢𝑖 represents the individual-specified fixed effect. Finally, 𝜀𝑖𝑡 is the error term.
Besides the regression in Eq. (1), this study performs several robustness tests to support the validity of the model. Firstly, the same regression in Eq. (1) is run through a probit model approach. Secondly, this study takes stockholders and non-stockholders to see whether there are any differences between these subgroups. Lastly, it runs Eq. (1) through a linear probability model approach, including a fixed effects model and a random effects model.
3.2 Variables
This subsection explains the main variables this study uses and why it uses these variables, based on the existing literature. These are the dependent variable, the main explicative variable, and several control variables.
3.2.1 Dependent variable
The dependent variable is an individual’s stock market participation (SMP). This variable is a binary variable, taking up a value of 0 for an individual that does not participate in the stock market and 1 for an individual who does participate in the stock market. Following Van Rooij (2011), this study considers stock market participation as having investments in either stocks or mutual funds, or both. For this reason, a dummy variable is created for having investments in either stocks and/or mutual funds, where a value of 0 indicates an individual having no investments in either stocks or mutual funds and a value of 1 where individuals do have an investment in either one of these.
3.2.2 Main explicative variable
13 situation much better than the current situation. Additionally, individuals also respond with “I don’t know”. This is treated as user missing data.
3.2.3 Socio-demographic control variables
The control variables this study uses are based on the existing literature regarding financial decision making and stock market participation. One of these control variables is age. This variable is created through subtracting date of birth on the year of the survey. Agarwal et al. (2009) and Binswanger (2011) both find that stock market participation increases with age to a certain point in the middle ages and then decreases again. This is shown in a U-shaped graph. To find out whether the relationship is linear or U-shaped as described in the literature, the regression includes the square of age.
Additionally, gender is taken as a control variable. As mentioned before, Almenberg and Dreber (2015) address the gender gap in stock market participation, where more men than woman participate in the stock market. A dummy variable explains the difference between men and woman, where a value of 1 represents a male and a value of 0 represents a female.
Furthermore, a dummy variable is created for the number of children in the household. According to Van Rooij et al. (2011) this is also one of the determinants of participating in the stock market. The dummy variable has a value of 0 for households with no children and a value of 1 for households having 1 or more children.
Another control variable is the level of education. Hilgerth and Hogarth (2002), Campbell (2006) and Van Rooij et al. (2011) find a positive relationship between a higher level of education and stock market participation, hence, this study also controls for this variable. In the survey, there are two questions regarding education: a question regarding the highest level of education attempted and a question regarding the highest level of education completed. This study uses the latter, since it covers a better representation of an individual’s actual educational level. For the level of education three dummies are created, where the first dummy represents a low educational level, the second represent a mid-level of education and the third represents a high level of education.
14 widowed, or never been married. Therefore, a dummy variable is created, where 1 is equal to being married or having a registered partnership and 0 for being divorced, widowed, or never been married.
A factor that is also being controlled for by various studies is health status. For example, Rosen and Wu (2004) find that households that include individuals with a bad health status tend to participate less in the stock market. The dataset contains a question where one is asked what their general health condition is. Individuals answer these questions with a value of 1 having an excellent health condition, a value of 2 having a good health condition, a value of 3 indicating a fair health condition, a value of 4 representing a not so good health condition and a value of 5 indicating a poor health condition. A dummy variable is created, where a value of 1 indicates having a generally good condition (excellent or good) and a value of 0 having a generally bad condition (fair, not so good or poor).
15
4. Data
This section covers the data that this study uses. It is divided into three separate subsections: data source, sample construction and the descriptive statistics.
4.1 Data source
This study uses data from the DNB Household Survey (DHS). It obtains its data through CentERdata, which is an institution for data collection and research. The DHS contains information of over 2000 households in the Netherlands. The dataset goes all the way back to 1993, in which they started these surveys among the population. In general, such surveys can contain a form of endogeneity. Such a form can cause biased and inconsistent parameter estimates, which can make inferences about these estimates invalid. A common form of endogeneity in surveys is sample selection. Sample selection occurs when individuals in a survey are not selected at random or when certain individuals are excluded from participating in the survey. For this reason, individuals who participate in the DHS are randomly selected each year. These individuals participate online on their computer through the internet. Hence, only households with a computer and internet connection can participate in these surveys, which can cause sample selection bias. However, the DHS prevents this through the availability of TV set-top boxes or computers with an internet connection for the people who have these disadvantages. Not only does the DHS partly deal with endogeneity, this study also adds several control variables to control for endogeneity and therefore getting more precise estimates of the variables of interest.
The DHS contains six main datasets regarding the participating households. These are household information, work & pension data, accommodation data, income data, wealth data and psychological concepts data. In addition, there are two modules which are derived from the six other datasets. These are aggregated income data and aggregated wealth data. For the construction of this study’s dataset, multiple datasets are merged to form one final dataset. This is done through an unique identification number, which is included for every individual and in every dataset.
4.2 Sample construction
16 because they contain key questions. From the household information dataset, general variables such as age, gender, number of children and education are obtained. The wealth dataset contains information regarding whether individuals participate in the stock market, through questions concerning stocks and mutual funds. These are used for the dependent variable. A binary variable is created, which indicates 0 (individuals possess none of these two investments) and 1 (individuals possess at least one of these two investments). The work and pension dataset contains a variable whether an individual is married or not. The income dataset contains information about an individuals’ general health condition. The aggregated income dataset contains information about an individuals’ net income. The psychological concepts dataset contains a key question regarding consumer confidence, the main explicative variable. The question asks individuals how they see the(ir) economic situation in five years from now. The response can vary between a situation that is much worse, worse, (about) the same, better or much better than the current situation. Some individuals also respond with ‘I do not know’, which this study considers as missing values.
In general, this study looks at the period from 2008 to 2018. 2008 is the starting point of this study since the Great Recession might cause individuals to be more conscious about their future. Also, earlier years contain different questions, therefore being another reason for using this starting point.
The merged dataset contains 51,051 observations over the time period 2008 to 2018, relative to 21,336 different households. The dataset has missing values for questions related to whether an individual has investments in either stocks or mutual funds and for the question related to the expected economic situation in the future. The reason for this, is that households simply did not answer the question. After dropping missing values for the questions regarding investments in stocks and mutual funds, the remaining sample contains 23,220 observations, relative to 15,661 different households. Dropping missing values for the question regarding the economic situation in the future gives a final sample of 19,363 observations, relative to 13,984 different households.
4.3 Descriptive statistics.
17 that older individuals invest more in the stock market than younger individuals. In addition, 55% of the individuals are male and 45% of the individuals are female, supporting research of Almenberg and Dreber (2015) regarding gender gap, where more men than woman participate in the stock market. Furthermore, 30% of the households have 1 or more children, 70% of the households are without children. 41% of the households have a high education. Middle and low education are represented by 31% and 28%, respectively. 61% is either married or has a registered partnership. The other 39% has no relationship and is either divorced, widowed or has never been married. Additionally, 70% of the households rate their general health condition as good, while 30% of the households rate their general health condition as bad. With regards to stock market participation, the table reports the difference in holding stocks and mutual funds: 15% of the individuals invest in stocks and 10% invest in mutual funds. The average net income is equal to € 25,165.19.
18
Table 1: Summary statistics
The table represents summary statistics for the original sample and the final sample. It reports statistics for the
dependent variable, independent variable, and socio-demographic control variables. Data is retrieved from the DNB Household Survey. The time-period is between 2008-2018.
Original sample Final sample
Obs. Mean Std. Dev. Obs. Mean Std. Dev.
Dependent variable
Stock Market Participation 51,051 0.087 0.282 19,363 0.206 0.404
Stock participation 51,051 0.064 0.245 19,363 0.151 0.358
Mutual fund participation 51,051 0.041 0.198 19,363 0.098 0.297
Independent variable Consumer confidence 24,319 2.807 1.015 19,363 2.949 0.769 Control variables Age 51,051 42.59 22.717 19,363 55.56 15.267 Gender Male 51,051 0.494 0.500 19,363 0.546 0.498 Female 51,051 0.506 0.500 19,363 0.454 0.498 Household size Children (1 or more) 51,051 0.531 0.499 19,363 0.301 0.459 No children 51,051 0.469 0.499 19,363 0.699 0.459 Level of education High education 51,051 0.438 0.496 19,363 0.408 0.491 Middle education 51,051 0.267 0.442 19,363 0.309 0.462 Low education 51,051 0.295 0.456 19,363 0.283 0.451 Marital status
Married / registered partnership 51,051 0.291 0.454 19,363 0.610 0.488 Divorced / widowed / never
married 51,051 0.709 0.454 19,363 0.390 0.488
Health status
Good condition (excellent or
good) 51,051 0.344 0.475 19,363 0.702 0.457
Bad condition (fair, not so good,
poor) 51,051 0.656 0.475 19,363 0.298 0.457
19
5. Results
This section covers the results of the Eq. (1) described in section 3. This section is divided into four subsections. The first subsection discusses the possible multicollinearity issue briefly. The second section presents the results of several logistic regressions. The third subsection discusses the results of a comparison between a logistic, fixed and random effects model. Lastly, the fourth subsection discusses various robustness checks.
5.1 Multicollinearity
An issue that might arise in data, is that of multicollinearity. This occurs when the explanatory variables are highly correlated with each other. There are two types of multicollinearity: perfect multicollinearity and near multicollinearity. In the former type an exact relationship between two or more variables exists. This is indicated by a correlation of either 1 or -1. The latter is a non-exact relationship between two or more variables, but still equally important. One can deal with this issue in multiple ways. Firstly, a solution for the issue is to ignore it. Secondly, a solution to deal with the issue is to drop one of the two variables that are highly correlated with each other. Lastly, one can convert the variables that are highly correlated into a ratio and include only the ratio in the regression. For this purpose, this study performs a Pearson correlation coefficient test to see whether this issue is present. Table A1 in Appendix 1 reports the output. Coefficients that yield a correlation above (-) 0.3 are considered to be moderately correlated. Coefficients which yield a correlation above (-) 0.7 are considered to be highly correlated, which is where this issue performs a role. As a results, table A1 shows that no coefficients are higher than (-) 0.7. High and mid education have the highest correlation, (-0.5546). Additionally, one can observe the variance inflation factors (VIF’s), which also measures multicollinearity in the explanatory variables. As a rule of thumb, a value higher than 10 is where multicollinearity arises. None of the explanatory variables has a VIF that is higher than 10. Based on these results, one can conclude that multicollinearity among the independent variables should not be a problem, hence all variables are used in the regression.
5.2 Pooled logit regressions
20 second regression contains all control variables, except for income (columns 3 and 4). The reason for this is that the observations drop significantly when also controlling for income, due to missing values. Finally, the third regression also includes the variable income, to see whether the results change when observations drop significantly (columns 5 and 6).
21
Table 2: Pooled logistic regressions – stock market participation
The table below shows the results of several logistic regressions of consumer confidence on stock market participation (SMP). In regressions 1 (column 1) confidence is regressed on stock market participation, excluding control variables. Regression 2 (column 3) shows results when all socio-demographic control variables, except income, are added. Regression 3 (column 5) shows the fully specified model, also controlling for income. AME indicate the average
marginal effects (column 2,4 & 6). Significance levels of 1%, 5% and 10% are indicated by ***, ** and *, respectively. Standard errors are in the parentheses.
Coeff. AME Coeff. AME Coeff. AME
VARIABLES (1) (2) (3) (4) (5) (6) Consumer confidence -0.048** -0.008 0.054** 0.008 0.064* 0.010 (0.024) (0.027) (0.033) Age 0.105*** 0.015 0.096*** 0.015 (0.010) (0.012) Age squared -0.001*** -0.000 -0.001*** -0.000 (0.000) (0.000) Male 0.818*** 0.119 0.566*** 0.089 (0.041) (0.054) Children -0.175*** -0.025 -0.353*** -0.055 (0.049) (0.061) High education 1.153*** 0.168 0.782*** 0.123 (0.051) (0.062) Mid-education 0.560*** 0.082 0.354*** 0.056 (0.056) (0.067) Marital status -0.197*** -0.029 -0.102** -0.016 (0.041) (0.050) Health status 0.309*** 0.045 0.239*** 0.038 (0.044) (0.057)
Net income (log) 0.575*** 0.090
22 Column 1 shows a coefficient of -0.048, which is statistically significant at the 5% significance level. The average marginal effect (AME) is also reported, which yields an AME of -0.008. This indicates that a marginal change of one unit in the level of consumer confidence decreases the chance that an individual participates in the stock market with 0.8%. Since consumer confidence is an ordinal variable and has values ranging from 1 to 5, the effect multiplies, where an individual who answers ‘much better than the current situation’, with a value of 5, has 3.2% less chance to participate in the stock market than an individual who answers ‘much worse than the current situation’ with a value of 1. Thus, when not controlling for anything, the results show the opposite of the intitial thought.
23 status, a marginal change of one unit decreases the chance of an individual participating in the stock market with 2.9%. Additionally, a marginal change of one unit in health status increases stock market participation with 4.5%. Again, both results are in alignment with findings of Rosen and Wu (2004), Grinblatt et al. (2011) and Rao et al. (2016).
Lastly, column 5 shows the results of a fully specified logistic regression. When observing the main explicative variable, the level of significance drops when controlling for income, but it remains statistically significant at the 10% significance level. Overall, a marginal change of one unit in consumer confidence increases the chance of an individual participating in the stock market with 1.0% to 4.0%. Therefore, these results support the earlier mentioned hypothesis, which states that consumer confidence increases Dutch households’ stock market participation. Furthermore, the results show that income is also an important predictor towards stock market participation, even though the amount of observations drop significantly due to not answering income related questions. Additionally, the pseudo R2 also shows a notable leap when controlling for different variables. Model 1 (excluding control variables) gives a pseudo R2 of 0.0002 and model 3 (including all control variables) gives a pseudo R2 of 0.0950, meaning that the variables included in the model explain 9.50% of the phenomenon. However, these results do not account for possible heterogeneity yet.
5.3 Fixed effects model and random effects model
24
Table 3: Comparison between logistic model, FE logistic model and RE logistic model
The table below reports the results of the three models. The first column shows the results of the fully specified logistic regression of Eq. (1). The second column shows the results of a fixed effects logistic regression. The third column shows the results of a random effects logistic regression. Significance levels of 1%, 5% and 10% are indicated by ***, ** and *, respectively. Standard errors are in the parentheses.
(1) (2) (3)
VARIABLES Simple
(pooled) Logit
Fixed effects Random effects
Consumer confidence 0.064* -0.074 -0.098 (0.033) (0.106) (0.093) Age 0.096*** 0.060 0.437*** (0.012) (0.097) (0.055) Age squared -0.001*** -0.003*** -0.004*** (0.000) (0.001) (0.001) Male 0.566*** 2.674*** (0.054) (0.282) Children -0.353*** -0.027 -0.666*** (0.061) (0.401) (0.239) High education 0.782*** 31.603 3.387*** (0.062) (1365.437) (0.375) Mid-education 0.354*** 14.712 1.275*** (0.067) (842.749) (0.354) Marital status -0.102** 0.099 0.106 (0.050) (0.313) (0.214) Health status 0.239*** -0.096 0.045 (0.057) (0.193) (0.168)
Net income (log) 0.575*** -0.066 0.306***
(0.040) (0.116) (0.103)
Constant -11.207*** -24.570***
(0.521) (1.882)
Observations 12,813 1,984 12,813
Number of households 3,653 318 3,653
25 variable male, since gender does not vary over time. The model still controls for this variable, but it is not estimated for. In addition, point estimates show that most control variables are not statistically significant in the fixed effects model and are therefore not in line with the earlier discussed literature. However, point estimates in the random effects model show that most of these control variables are statistically significant. Despite the statistical significance of these variables, marital status and health status are not statistically significant, hence one cannot make any inferences about these variables. A possible explanation for this might be that marital status and health status are reasonably fixed over time. The only variable that yields significance across the three models is age squared, indicating that the older an individual gets, the more he or she participates in the stock market and at a point in time decreases its participation again. A possible reason for the insignificant results in the fixed effects model is the significant drop in observations. The fixed effects regression only uses individuals who change their behaviour over time. In this case, individuals who do not change their level of confidence over time are automatically dropped.
5.4 Robustness checks
To see whether the results are valid, this study performs various robustness tests to strengthen the model that it uses. First, to see whether the pooled logistic regressions are valid, a pooled probit model is regressed to see whether it results in the same outcomes. The same regression is run as in the logistic model. The results are shown in table A2 in appendix A. Point estimates show that results of the logistic regression also hold in the probit regression. More specifically, consumer confidence still has a positive statistically significant effect on stock market participation, where a marginal change in consumer confidence increases the chance of an individual participating in the stock market with 0.9% to 3.6%, depending on the level of confidence. Besides marital status, which is significant at the 10% level with respect to a 5% significance level in the logistic model, overall, the results are the same.
26
Table 4: Panel logistic regressions – comparison between direct and indirect stock market participation.
The table reports the results of two fully specified logistic regressions of consumer confidence on direct and indirect stock market participation (SMP). The first column shows the results of the regression on direct stock market participation. The second column shows results of the regression on indirect stock market participation. AME indicate the average marginal effects. Significance levels of 1%, 5% and 10% are indicated by ***, ** and *, respectively. Standard errors are in the parentheses.
(1) AME (2) AME
VARIABLES Direct SMP Indirect SMP
Consumer confidence 0.040 0.005 0.102** 0.009 (0.037) (0.044) Age 0.130*** 0.017 0.055*** 0.005 (0.013) (0.015) Age squared -0.001*** -0.000 -0.000** -0.000 (0.000) (0.000) Male 0.507*** 0.066 0.704*** 0.065 (0.060) (0.076) Children -0.488*** -0.063 -0.105 -0.010 (0.070) (0.080) High education 0.745*** 0.096 0.680*** 0.063 (0.069) (0.085) Mid-education 0.175** 0.023 0.426*** 0.039 (0.075) (0.091) Marital status -0.184*** -0.024 0.087 0.008 (0.055) (0.066) Health status 0.220*** 0.028 0.209*** 0.019 (0.063) (0.076)
Net income (log) 0.596*** 0.077 0.592*** 0.055
(0.045) (0.054) Constant -12.357*** -11.575*** (0.592) (0.693) Observations Number of households Pseudo R2 12,813 3,653 0.0944 12,813 3,653 0.0797
27 table A3 in Appendix A does yield significant results for consumer confidence at the 5% significance level. However, this only yields for indirect stock market participation. Moreover, the coefficient has a negative sign. This is contrary to the described literature regarding confidence & optimism and stock market participation. A possible explanation for this is that over time, Dutch households have a cynical view on the stock market itself (Hurd et al. 2011).
28
6. Conclusion
To this date the non-participation puzzle still remains a mystery. Since the introduction of this phenomenom by Mankiw and Zeldes (1991), researchers still study this behavior through different kind of aspects, for example loss-aversion or happiness. This study aims to explain this phenomenon through consumer confidence, specifically for Dutch households. The study observes panel data over the time-period 2008-2018 to try and explore the relationship between consumer confidence and stock market participation. The results show some significant results that are in line with the literature, in which they find that confidence has a positive effect on the stock market and on stockholding (Jansen and Nahuis, 2002; Puri and Robison, 2007; Angelini and Cavapozzi, 2017; Agarwal et al. 2018).
This study obtains its results through various models. The main model is a logistic model, which shows the results of a regression of consumer confidence on stock market participation, controlling for several variables which are mentioned in the field of research. The model shows different signs of consumer confidence. A regression excluding control variables gives a negative coefficient. A possible reason for a negative relationship might be that Dutch households are rather pessimistic on the stock market (Hurd et al. 2011). Although the effect of consumer confidence diminishes when controlling for several variables (and the sign changes from a negative coefficient to a positive coefficient), the results remain significant. Therefore, when an individuals’ level of consumer confidence increases, the chance of them participating in the stock market increases. However, these results can be biased through correlation between the explanatory variables and omitted variables. For this reason, this study uses both a fixed effects model and a random effects model. Consequently, the results show that consumer confidence is insignificant. One of the reasons could be the significant drop in observations.
30
Reference
Agarwal, S., Driscoll, J.C., Gabaix, X., Laibson, D., 2009. The age of reason: financial decisions over the life cycle and implications for regulation. Brookings Papers on Economic Activity, 2009(2), 51-117.
Agarwal, V., Aslan, H., Huang, L., Ren, H., 2019. Political uncertainty and household stock market participation. AFA 2019 Annual Meeting Paper, 1-57.
Akhtar, M., Muhammed, F., 2017. Financial sophistication, emotional quotient, and stock market participation: theory and evidence. Journal of Islamic Business and Management, 7(1), 87-99
Almenberg, J., Dreber, A., 2015. Gender, stock market participation and financial literacy. Economics Letters, 137, 140-142
Angelini, V., Cavapozzi, D., 2017. Dispositional optimism and stock investments. Journal of Economic Psychology, 59, 113-128.
Becker, G.S., 1962. Irrational behaviour and economic theory. Journal of Political Economy, 70(1), 1-13.
Benartzi, S., Thaler, R.H., 1995. Myopic loss aversion and the equity premium puzzle. The Quarterly Journal of Economics, 110(1), 73-92.
Bertaut, C.C., 1998. Stockholding behaviour of U.S. households: evidence from the 1983-1989 survey of consumer finances. Review of Economics and Statistics, 80(1), 263-275. Beshears, J., Choi, J.J., Laibson, D., Madrian, C.B., 2018. Behavioural household finance. Handbook of Behavioural Economics, (1), 178-235.
Binswanger, J., 2011. Dynamic decision making with feasibility goals: a procedural-rationality approach. Journal of Economic Behavior & Organization, 78(3), 219-228.
Bram, J., Ludvigson, J., 1998. Does consumer confidence forecast household expenditure? A sentiment index horse race. FRBNY Economic Policy Review, 59-78.
Brigss, J.S., Cesarini, D., Lindqvist, E., Östling, R., 2015. Windfall gains and stock market participation. Working Paper No. 21673.
Brown, S., Garino, G., Taylor, K., Price, S.W., 2005. Debt and financial expectations: an individual and household-level analysis. Economic Inquiry, 43(1), 100-120.
Bruno, G., 2014. Consumer confidence and consumption forecast: A non-parametric approach. Empirica, 41(1), 37–52.
Calvet, L.E., Sodini, P., 2014. Twin picks: disentangling the determinants of risk-taking in household portfolios. The Journal of Finance, 69(2), 867-906.
31 Campbell, J.Y., 2006. Household finance. The Journal of Finance, 61(4), 1553-1604. Christelis, D., Jappelli, T., Padula, M., 2010. Cognitive abilities and portfolio choice. European Economic Review, 54(1), 18-38.
Conlin, A., Kyrolainen, P., Kaakinen, M., Jarvelin, M.R., Perttunen, J., Svento, R., 2015. Personality traits and stock market participation. Journal of Empirical Finance, 33, 34-50.
Dimmock, S.G., Kouwenberg, R., 2010. Loss-aversion and household portfolio choice. Journal of Empirical Finance, 17, 441-459.
Frydman, C., Camerer, C.F., 2016. The psychology and neuroscience of financial decision making. Trends in Cognitive Sciences, 20(9), 661-675.
Grinblatt, M., Keloharju, M., Linnainmaa, J., 2011. IQ and stock market participation. The Journal of Finance, 66(6), 2121-2164.
Guiso, L., Haliassos, M., Jappelli, T., 2003. Household stockholding in Europe: where do we stand and where do we go? Economic Policy, 18(36), 123-170.
Guiso, L., Jappelli, T., 2005. Awareness and stock market participation. Review of Finance, 9(4), 537-567.
Guiso, L., Sapienza, P., Zingales, L., 2008. Trusting the stock market. Journal of Finance, 61(6), 2557-2600.
Haliassos, M., Bertaut, C.C., 1995. Why do so few hold stocks? Economic Journal, 105, 1110-1129.
Hogarth, J.M., Hilgert, M.A., 2002. Financial knowlegde, experience and learning preferences: preliminary results from a new survey on financial literacy. Consumer Interest Annual, 48(1), 1-7.
Hurd, M.D., Van Rooij, M., Winter, J., 2011. Stock market expectations of Dutch households. Journal of Applied Econometrics, 26(3), 416-436.
Jansen, W.J., Nahuis, N.J., 2002. The stock market and consumer confidence: european evidence. Economics Letters, 79(1), 89-98.
Karnizova, L., Khan, H., 2015. The stock market and the consumer confidence channel: evidence from Canada. Empirical Economics, 49, 551-573.
Kaustia, M., Torstila, S., 2011. Stock market aversion? Political preferences and stock market participation. Journal of Financial Economics, 100(1), 98-112.
32 Kwan, A.C.C., Cotsomitis, J.A., 2004. Can consumer attitude forecast household spending in the United States? Further evidence from the Michigan survey of consumers. Southern Economic Journal, 71(1), 136–144.
Lachowska, M., 2013. Expenditure, confidence, and uncertainty: identifying shocks to consumer confidence using daily data. Upjohn Institute working paper, 13(197).
Lian C., Ma, Y., Wang, C., 2018. Low interest rates and risk-taking: evidence from individual investment decisions. The Review of Financial Studies, 32(6), 2107-2148.
Mankiw, G.N., Zeldes, S.P., 1991. The consumption of stockholders and nonstockholders. Journal of Financial Economics, 29(1), 97-112.
Mehra, R., Prescott, E.C., 1985. The equity premium: a puzzle. Journal of Monetary Economics, 15(2), 145-161.
Mehra, R., Prescott, E.C., 2003. The equity premium puzzle in retrospect. Handbook of the Economics of Finance, 889-938.
Puri, M., Robinson, D.T., 2007. Optimism and economic choice. Journal of Financial Economics, 86(1), 71-99.
Rao, Y.R., Mei, L., Zhu, R., 2016. Happiness and stock-market participation: empirical evidence from China. Journal of Happiness Studies, 17, 271-293.
Rosen, H.S., Wu, S., 2004. Portfolio choice and health status. Journal of Financial Economics, 72(3), 457-484.
Sacco, P.L., 1991. Rationality and stock market behavior: what theoretical framework (if any?). International Economic Journal, 5(4), 17-41.
Tonkiss, F., 2009. Trust, confidence and economic crisis. Intereconomics. Review of European Economic Policy, 44(4), 196–202.
Van Rooij, M., Lusardi, A., Alessie, R., 2007. Financial literacy and stock market participation. Journal of Financial Economics, 101(2), 449-472.
Vissing-Jorgensen, A., 2003. Perspectives on behavioral finance: does “irrationality” dissapear with wealth? Evidence from expectations and actions. NBER Macroeconomics Annual, 18, 139-194.
33 Appendix A: Results
Table A1: Pearson correlation coefficient matrix
The table below shows the correlation coefficients between the main explicative variable and other explanatory variables. Correlation coefficients above 0.3 are moderately correlated. Correlation coefficients above 0.7 are considered highly correlated. ***, ** and * indicate that the coefficient is significant at respectively the
1%, 5% or 10% significance level. Data is retrieved from the DNB Household Survey.
Consumer High Mid
Confidence Age Male Children Education Education MS HS Net income
34
Table A2: Panel probit regressions – stock market participation.
The table below shows the results of several probit regressions of consumer confidence on stock market participation (SMP). In the first column, confidence is regressed on stock market participation, excluding control variables. Regression 2 shows results when all socio-demographic control variables, except income, are added. Regression 3 shows the fully specified model, including controlling for income. AME indicate the average marginal effects. Significance levels of 1%, 5% and 10% are indicated by ***, ** and * respectively. Standard errors are in the parentheses.
(1) AME (2) AME (3) AME
VARIABLES SMP SMP SMP Consumer confidence -0.028** -0.008 0.028* 0.007 0.034* 0.009 (0.013) (0.016) (0.019) Age 0.059*** 0.015 0.055*** 0.015 (0.005) (0.007) Age squared -0.000*** -0.000 -0.000*** -0.000 (0.000) (0.000) Male 0.461*** 0.118 0.332*** 0.090 (0.023) (0.030) Children -0.097*** -0.025 -0.207*** -0.056 (0.028) (0.035) High education 0.652*** 0.167 0.464*** 0.126 (0.028) (0.035) Mid-education 0.308*** 0.079 0.202*** 0.055 (0.031) (0.037) Marital status -0.112*** -0.029 -0.053* -0.014 (0.024) (0.029) Health status 0.169*** 0.043 0.138*** 0.037 (0.024) (0.032)
Net income (log) 0.298*** 0.081
35
Table 4: Comparison fixed effects and random effects between direct and indirect stock market participation
The table reports the results of fixed effects and random effects models on direct and indirect stock market participation. The first and third column show the results of fixed effects models on direct and indirect stock market participation. The second and fourth column show results of the random effects models on direct and indirect stock market participation. Significance levels of 1%, 5% and 10% are indicated by ***, ** and * respectively. Standard errors are in the parentheses.
(1) Direct SMP (2) Direct SMP (3) Indirect SMP (4) Indirect SMP
VARIABLES Fixed effects Random
effects
Fixed effects Random effects Consumer confidence -0.089 -0.106 -0.287** -0.171 (0.113) (0.092) (0.136) (0.106) Age -0.129 0.385*** 0.219* 0.323*** (0.109) (0.049) (0.123) (0.052) Age squared -0.001 -0.004*** -0.004*** -0.003*** (0.001) (0.000) (0.001) (0.000) Male 2.002*** 2.316*** (0.246) (0.257) Children -0.363 -0.840*** 0.093 0.022 (0.412) (0.227) (0.406) (0.235) High education 29.751 2.773*** 15.299 2.359*** (1,064.752) (0.310) (497.754) (0.312) Mid-education 14.016 0.748** 13.582 1.271*** (657.946) (0.307) (497.752) (0.318) Marital status -0.052 -0.023 -0.167 0.105 (0.327) (0.196) (0.356) (0.204) Health status -0.356* -0.144 -0.027 0.154 (0.210) (0.166) (0.245) (0.186)
Net income (log) -0.065 0.375*** -0.049 0.405***
(0.127) (0.106) (0.163) (0.122)
Constant -22.744*** -23.453***
(1.762) (1.963)
Observations 1,848 12,813 1,311 12,813
36
Table A4: Comparison between a simple pooled OLS, FE and RE model
The table below reports the comparison of the results between three models. The first column shows the results of a simple pooled OLS from table 2 column 3. The second column shows the results of a fixed effects model. The third column shows the results of a random effects model. Significance levels of 1%, 5% and 10% are indicated by ***, ** and * respectively. Standard errors are in the parentheses.
(1) (2) (3)
VARIABLES Pooled OLS Fixed effects Random effects
Consumer confidence 0.009* -0.001 -0.002 (0.005) (0.004) (0.004) Age 0.014*** -0.000 0.017*** (0.002) (0.004) (0.002) Age squared -0.000*** -0.000*** -0.000*** (0.000) (0.000) (0.000) Male 0.095*** 0.119*** (0.008) (0.012) Children -0.052*** 0.008 -0.028*** (0.009) (0.014) (0.010) High education 0.141*** 0.172*** 0.149*** (0.009) (0.047) (0.015) Mid-education 0.058*** 0.105*** 0.057*** (0.010) (0.038) (0.015) Marital status -0.012 0.006 0.000 (0.008) (0.010) (0.008) Health status 0.041*** -0.010 0.003 (0.009) (0.008) (0.007)
Net income (log) 0.055*** 0.001 0.014***
