1
Happiness and individuals’ financial decision-making
Author: G.J. Hobma1 Student number: s2354055 Supervisor: Dr. M.M. Kramer Master’s Thesis University of Groningen 9th of January 2017 Abstract
Using the SHARE dataset on individuals of 50 years and older across Europe, which contains questions about the individuals’ self-perceived levels of happiness, as well as detailed information on the individuals’ background and financials, this paper examines the relationship between happiness and individuals’ financial decision-making. This while controlling for a broad range of control variables known to influence the decision-making of individuals in the financial market. The results suggest that happier individuals have a lower chance of owning financial assets and have higher levels of subjective risk aversion. Subjective risk aversion is only weakly mediating in explaining the reduced ownership probabilities, suggesting that happiness influences financial decision-making in a different way. The findings are strengthened by using several robustness tests accounting for endogeneity and using alternative regression specifications, although the effect for some regressions becomes insignificant. This paper does not find any significant effect of happiness on the percentage of gross financial wealth invested in any of the risky assets. Keywords: Household Finance, Happiness, Life Satisfaction, Financial Market Participation, Risk Aversion.
JEL Codes: D14, G11, G31.
2 Acknowledgement:
This paper uses data from SHARE Waves 1, 2, 3 (SHARELIFE), 4, 5 and 6 (DOIs: 10.6103/SHARE.w1.600, 10.6103/SHARE.w2.600, 10.6103/SHARE.w3.600, 10.6103/SHARE.w4.600, 10.6103/SHARE.w5.600, 10.6103/SHARE.w6.600), see Börsch-Supan et al. (2013) for methodological details.
3 1. Introduction
Financial market participation of households is, and has always been, very limited throughout the U.S. and EU (Guiso et al., 2002; Christelis et al., 2010). Understanding the motivations for households to not participate in the financial market is one of the main areas of interest in household finance and is of key importance since households potentially lose utility by not participating. The financial literature recommends individuals to invest a portion of their resources into risky assets, in order to maximize their personal income and therefore their utility. Assuming every individual has resources that can be allocated; everyone should participate in the financial market. This contradicts with the observation that many people do not invest in the financial market, as shown by Georgarakos and Pasini (2011), who observe an average in Europe of only 25%. Consequently, when this non-participation is translated to the potential welfare loss of households, the decision not to participate in the financial market is estimated to correspond to a welfare loss of approximately 1.5%-2% of total consumption in a calibrated life-cycle model (Cocco, Gomes and Maenhout, 2003). The reason for this welfare loss comes, apart from diversification benefits of financial assets, from the equity premium. The corresponding puzzle, the equity premium puzzle, refers to the difference in return between equity and government bonds of approximately 3%-9% on a yearly basis over the long-run (Mehra and Prescott, 2003). This premium partially arises from a lack of household investment in the stock market. Understanding the reasons why households refrain from participating in the financial market therefore becomes very interesting since next to explaining households’ welfare loss it can aid in understanding the equity premium puzzle.
4
The last aspect mentioned, happiness2, has already had vast academic coverage in the
economic world but has only been directly linked to financial market participation once in literature, as done by Delis & Mylonidis (2015). According to prior research, happiness increases personal income and health (Graham et al., 2004) and happiness is highly important in increasing social capital (Guven, 2011). Furthermore, happier people tend to save more and spend less as they take more time for decisions and have more control over expenditures (Guven, 2012). One of the key hypothesized channels through which happiness affects individuals’ decision-making is through differences in risk- and return expectations. However, there are two conflicting views regarding the way in which happiness influences these risk- and return expectations.
On the one hand, Forgas (1995) shows that happier people exhibit more risk-taking behaviour. This relationship is defined as the Affect Infusion Model (AIM). This finding is confirmed by Johnson and Tversky (2003) who find that happier people expect a more favourable gamble outcome. When this theory is tested on financial market decisions, Kaplanski et al. (2015) find that happier people are more optimistic with regards to the domestic Dutch stock market as well as the U.S. stock market compared to unhappier people. Overall, happiness leads to higher expected returns and lower expected risk, even when controlling for individuals’ own optimism. On the other hand there is the Mood-Maintenance Hypothesis by Isen and Patrick (1983), which finds that happier people behave more cautiously in risky situations. This since it is suggested that happier people try to protect their current happy state from real and salient losses. This finding is supported by Drichoutis and Nayga (2013) whom also find increased risk aversion with higher happiness. This psychological theory also apparent in financial market decisions, considering how Delis & Mylonidis (2015) find that happier people are less likely to invest in stocks and are less likely to have insurance. Although not directly tested, they propose the reduced tendency to participate in the financial market is due to an increase in risk aversion for happier people.
These two conflicting theories from the psychological literature both have empirically proven effects on individuals’ financial decision-making. This paper seeks to provide evidence for the two conflicting views by answering the following question:
2 According to Frey and Stutzer (2002) happiness and life satisfaction are part of the psychological term
5
Does happiness affect individual’s financial market participation and financial risk preferences?
This paper utilizes the Survey of Health, Ageing and Retirement in Europe (SHARE). SHARE is a multidisciplinary and cross-national panel database of micro data on health, socio-economic status, and social and family networks of more than 120,000 individuals aged 50 or older in Europe. This database contains information on the ownership of four financial assets (bonds, stocks, mutual funds and individual retirement accounts) and the amounts invested in each of these assets. Furthermore, it contains information on an individual’s willingness to take financial risks. This data is combined with individuals’ data on happiness to test the aforementioned relationships.
This paper finds similar results as Delis & Mylonidis (2015); happier people have a lower tendency to own financial assets, controlling for a large set of relevant variables. Next to a decreased ownership probability, an increase in subjective risk aversion is found for happier people. There is a 0.38 percentage point decrease in the probability to hold any financial asset for a point increase in happiness on a 1-10 scale. This corresponds to a relative 2.5% decrease, since the chance of owning any financial asset equals approximately 15%. Subjective risk aversion, on a 1-4 scale, increases with 0.002 points for every point increase in happiness. These findings hold, although the effects for some regressions become insignificant, in robustness tests such as alternative regression specifications. The problem of endogeneity is accounted for through instrumental variable regression for omitted variable bias, and through regressions using lagged variables for reverse causality bias. No conclusions can be made with regards to the percentage of financial wealth invested in financial assets, which was hypothesized to decrease with an increase in happiness.
6
The remainder of the paper is organized as follows. Section 2 presents an overview of the current literature on financial market participation and a more thorough description of happiness and the effects of happiness on individual’s behaviour. Afterwards, the hypotheses are defined. Section 3 explains the data and the methodology. The results are analysed in Section 4. Section 5 provides an overview of the robustness tests, accounting for endogeneity and using alternative regression specifications. Sections 6 and 7 respectively present the discussion and conclusion.
2. Literature review
First, this section provides a comprehensive overview of why the limited financial market participation is relevant and what explanations currently exist. This is followed by describing the influence of happiness on individuals and therefore on the economy in aggregate. Finally, the two conflicting mechanisms through which happiness affects individuals’ decision-making and financial behaviour is explained before the hypothesis is formulated.
2.1. Limited financial market participation
7
2.1.1. Drivers and explanations
A view that has gained considerable coverage and support as for why households refrain from the financial market is due to actual or perceived participation costs. Allen & Gale (1994) show, using a theoretical model, that entry costs limit financial market participation, in a paper focusing on frictions of the financial market itself. Vissing-Jorgensen (2002), building on this theoretical framework, also finds that participation costs is one of the key drivers explaining why households that are able to participate in the financial market, do not actually participate. An individual will invest only if the expected equity premium and the corresponding level of investment is high enough to overcome the fixed and perceived participation costs. Perceived participation costs of $50,- per period could explain approximately 50% of the non-participation decisions, whereas $200,- per year would be sufficient to explain approximately 75%. Important to note here is that participation costs are not merely defined as the costs to set up a brokerage account, but also include mental aspects such as search and information costs. This is confirmed by King and Leape (1998) in a survey-based study on households. Their findings suggest that information costs are one of the key drivers of non-participation in the financial market.
8
bequest motives, planning horizons and health insurance are all rejected in this same paper.
Other factors were found by Guiso et al. (2003), who found a positive correlation of education and stock ownership. Christelis et al. (2010) investigated individuals’ cognitive skills and the decision to directly or indirectly participate in the financial market using a sample of 11 European countries. Their results suggest that, similar to education, high cognitive skills leads to an increased probability of owning assets. Further studies in the realm of cognitive ability show that individuals with low financial literacy are less likely to participate in the financial market (Van Rooij et al., 2005). Similarly, Grinblatt et al. (2011) find that the decision to participate in the financial market is directly related to an individual’s IQ. All these factors can be attributed to an easier overcoming of (perceived) participation costs for those people with higher cognitive skills.
9
One of the most recent papers investigating the novel determinants of financial market participation is by Delis and Mylonidis (2015). Two variables are investigated, trust and happiness, both stand-alone and together, using a representative sample of the Dutch population. They found that the impact of happiness on households’ financial decisions is more economically important than that of trust. Happiness leads to a lower probability of investing into risky financial assets, whereas trust has the expected positive effect.
2.2. The effect of happiness on individuals
That happiness has an effect on individuals’ choices and therefore, on aggregate, on the market itself has long been established by research in the field of economics. As mentioned before, happiness is part of the larger growing concept called subjective well-being, on which already more than 100 papers were written in the period between 2001-2005, whereas this was four in the period 1991-1995 (Kahneman and Krueger, 2006). One of the examples is provided by Graham et al. (2004) who found that higher happiness leads to higher personal income and health. Furthermore, Guven (2011) determined that happiness is an important determinant for social capital. Similarly, Guven (2012) finds that happier people save more and spend less. On an aggregate level, Kaplanski and Levy (2009) find that negative sentiment, driven by low levels of happiness, affects individual’s investment decisions negatively and therefore lowers aggregate asset pricing.
10
There are two theories that explain how happiness affects individuals’ behaviour. The first theory, the Affect-Infusion-Model by Forgas (1995), suggests that higher happiness increases risk-taking behaviour, whereas lower happiness reduces this tendency to take risk. This is supported by, among others, Fehr-Duda et al. (2011), who find that higher happiness is significantly associated with a more positive probability weighting of risky gambles. Johnson and Tversky (1983) also suggest that happy individuals’ higher risk tolerance can be explained by having more optimistic beliefs about a risky gamble. Confirming evidence is provided by Nygren et al. (1996), who found that happier individuals are more optimistic as winning probabilities are overestimated compared to probabilities of losing. Kaplanski et al. (2015) confirm this finding for financial risk-taking, proving that happier individuals have higher stock market return and lower risk expectations. The second theory, the Mood-Maintenance Hypothesis by Isen and Patrick (1983), finds a lower level of risk taking for happier individuals compared to those that are relatively less happy. Drichoutis and Nayga (2013) indicate that higher happiness increases risk aversion and decreases time preference at the same time. This is supported by Loewenstein (2000), who argues that “emotions experienced at the time of making a decision often propel behaviour in directions that are different from that dictated by a weighing of the long-term costs and benefits of actions”. Since investing in the financial market is in essence based on a risk-return trade-off, it is argued that happiness influences individual’s investment decisions. Loewenstein (2000) argues that future risk is discounted disproportionally heavy when low happiness is experienced and therefore more risk is taken. This finding is also found in the financial decision-making of individuals, as Delis & Mylonidis (2015) point to a decrease in probabilities for stock ownership and insurance ownership when individuals are happier. These two conflicting mechanisms and differences in observed financial behaviour provide the starting point for this paper.
2.3. Hypothesis building
11
H1: Higher happiness increases the level of risk aversion.
This paper uses three different proxies for risk aversion. First it is expected that an increase in risk aversion leads to a decreased probability of participating in the financial market, which is measured for four different asset classes. Second, an increase in risk aversion is supposed to affect an individual’s subjective risk aversion. Therefore, it is expected that an individual’s subjective risk aversion increases with higher levels of happiness, which follows as the second proxy. Third, due to the nature of the data used in this paper, it is possible to examine whether the effects of happiness on risk aversion also appear in the percentage of wealth invested in financial assets. Following the reasoning above, an increase in happiness leads to a lower level of risk tolerance. When risk tolerance decreases, so does the optimal share invested into risky assets3. Therefore, it is
expected that happier individuals have a lower percentage of gross financial wealth invested in financial assets, conditional on participation. This is measured for all four financial asset classes and is used as the third proxy for risk aversion, which is also separately tested. In total three proxies are tested: 1) financial asset holding probabilities; 2) subjective risk aversion; and 3) share of gross financial wealth invested in financial assets.
3. Data and Methodology
First in this section, the dataset is introduced and the construction of variables is explained. The descriptive statistics are presented subsequently. Second, the empirical strategy of the baseline regressions is explained.
3.1. Data
This paper uses the Survey of Health, Ageing and Retirement in Europe (SHARE) as data source. SHARE is a multidisciplinary and cross-national panel database of microdata on health, socio-economic status and social networks of more than 120,000 individuals aged 50 or older in 26 European countries. It offers unique combinations which allow to match individual’s specific demographic, social and behavioural information with an individual’s investment portfolio. As of 2017, SHARE has collected six panel waves (2004, 2006, 2011,
3 Consider the standard (quadratic) utility function: 𝑓 = 𝐸[𝑟] − 𝜎2
𝑇. With T being an individual’s risk
tolerance, E[r] is the expected return on a portfolio and σ is the standard deviation. When combining the utility of a risky asset in the utility function with a risk-free asset, yielding a two-asset model, one can calculate the optimal share invested into risky assets. This is based on the following equation: 𝑤∗=
2
𝑇× (
𝐸[𝑟]− 𝑟𝑓
𝜎2 ). With w* being the optimal share invested into the risky asset and 𝑟𝑓being the return on a
12
2013 and 2015). The dataset is representative for the population across the whole of Europe. All the questions used by SHARE are standardized across the participating countries, which allows for direct international comparisons. All core data is collected using face-to-face, computer-aided personal interviews. Some modules are collected using traditional questionnaires4. SHARE has been used in a set of studies analysing
households’ portfolio allocation and their underlying reasons for the allocation decisions (e.g. Georgarakos and Pasini, 2011; Christelis, 2010).
3.1.1. Data set construction
As explained above, SHARE collected six different panel waves to date. This paper’s analysis is based on the fourth and fifth wave which took place in, respectively, 2011 and 2013 in 16 European countries (Austria, Germany, Sweden, the Netherlands, Spain, Italy, France, Denmark, Switzerland, Belgium, Czech Republic, Poland, Hungary, Portugal, Slovenia and Estonia).
Each wave consists of several modules, which collect information on a different subject. So, as a first step, all the relevant questions which are provided in separate datasets have to be matched for each person within a single wave. Subsequently, the matched datasets for the two waves are appended into one dataset for both waves which is done using a fixed unique identification number per individual. Most of the questions asked by SHARE are on the individual level, such as questions on demographics, cognitive abilities, family network, social activities and health status. However, there are also questions referring to the household. This is the case for questions regarding the ownership of bonds, stocks and mutual funds. These questions are therefore aggregated over the individuals who have indicated to be a couple, using the unique household identifier, following Christelis et al. (2010).
As a final step, observations reporting a negative amount for any of the financial assets are dropped from the dataset. Furthermore, observations reporting that individuals own one of the financial assets but denoting a financial asset amount of zero are corrected by replacing the binary ownership variable by zero. Lastly, observations with missing values for any of the dependent or independent variables are dropped. This yields a total dataset covering 48,082 individuals of which 19,595 have answered both questionnaires in 2011 and 2013. This leads to a total number of observations of 67,677.
3.1.2. Variable construction
3.1.2.1. Dependent variable construction
13
There are in total three proxies for risk aversion, with different dependent variables, to be tested. The first proxy describes the probabilities that individuals own any of the financial assets; the second describes an individual’s subjective risk aversion; and the third describes the relative share of gross financial wealth invested in any of the financial assets.
The first proxy analyses the impact of happiness on financial market participation. Four financial assets are studied: bonds, stocks, mutual funds and individual retirement accounts. In addition, another variable is created that indicates whether the individual owns one or more of the aforementioned financial asset classes. These five variables are all binary, taking a value of one if the person holds any of the assets and zero otherwise. The second proxy analyses the effect of happiness on subjective risk aversion. This variables is based on a question in the survey which asks to what degree individuals are willing to take financial risks with respect to their investments. There are four possible answers, in order of increasing risk aversion: 1) take substantial financial risks expecting to earn substantial returns; 2) take above average financial risks expecting to earn above average returns; 3) take average financial risks expecting to earn average returns and 4) not willing to take any financial risk.
The third proxy analyses the effect of happiness on the relative share of the household gross financial wealth invested in risky assets, conditional on ownership. As a first step, the share measure is constructed using the pre-defined variable of household gross financial assets. This variable contains the sum of 1) bank accounts; 2) bonds, stocks and mutual funds; and 3) savings for long-term investments. Then, dividing the amount of each of the financial assets (bonds, stocks, mutual funds or IRAs) by the gross financial assets yields the share measure. In total, five share measures are used: the share of bonds, stocks, mutual funds, IRAs and total financial market investment respectively. This approach is similar to the one employed by Rosen and Wu (2004) and only employs the financial assets of households and therefore excludes assets such as houses, human capital and cars. However, considering that the amounts of financial assets are registered manually on individuals’ own indication, there is a relatively high chance of errors. When examining the distribution of the newly constructed wealth measures, there is a number of observations that display figures higher than one. Since it is evident that these observations have wrong values, they are dropped from the analysis. In total, 163 observations are dropped, which has already been included in the observation figures in the above paragraph.
14
In order to analyse the hypothesis with the three proxies, two independent variables have been created. These are based on the independent variables constructed by Delis & Mylonidis (2015) in their study of happiness using the Dutch LISS database. In their paper, three measures are used, which all yielded similar statistical significance. These three measures are based on the following questions: 1) How happy would you say you are?; 2) In general, how satisfied are you with your life?; and 3) How satisfied are you with the life that you lead at the moment? Since this paper uses the SHARE database, which is slightly different from the LISS database in the questions asked, the measures cannot be replicated perfectly. The measures used in this paper are elaborated upon now.
One of the questions in the SHARE database reads: “On a scale from 0 to 10, how satisfied are you with your life?”. This question takes a stance between the second and third measure used by Delis & Mylonidis (2015) and does not make a distinction between life satisfaction in general and current life satisfaction. However, it is very similar to the two above and therefore this question is the basis for the first independent variable: Happiness (satisfaction).
The second independent variable is constructed differently. Since there is no question asking respondents’ happiness directly in SHARE, a factor has been created based on questions related to happiness using Principal Component Factoring (PCF). The specific notation and questions used can be found in Table 1. As can be seen in Table A1 in the Appendix, only one factor is retained. This confirms the fact that all four variables measure one underlying concept. This factor has subsequently been renamed as Happiness (factor) and is used as the second independent variable. Therefore this paper uses two measures of happiness, capturing both forms of happiness as defined by Frey and Stutzer (2002): evaluated happiness and experienced happiness.
Table 1. Independent variable Happiness (factor) definitions and sources.
Notation Type of variable Question
Satisfaction activities Interval variable On a scale from 0 to 10, how satisfied are you with the activities that you mentioned?
Looking forward day Categorical variable How often do you look forward to each day?
Life meaning Categorical variable How often do you feel that your life has meaning?
Look back life happiness Categorical variable How often, on balance, do you look back on your life with a sense of happiness? Note: The activities mentioned refer to an open question about activities undertaken over the past year. The categorical variables have four answering possibilities: 1) Often; 2) Sometimes; 3) Rarely; and 4) Never.
3.1.2.3. Control variable construction
15
financial assets’ holding probabilities in previous literature on household finance. Several types of control variables can be distinguished: demographic, socio-economic, financial and more value-based variables.
Demographic and socio-economic variables, such as respondents’ age, gender, relationship status, employment status, education and whether or not the respondent has children have strong theoretical and empirical backing in the literature. Furthermore, measures of wealth and income have been used as control variables (e.g. Calvet and Sodini, 2014; Guiso et al., 2003; Heaton and Lucas, 2000). The financial variables for income and wealth have been transformed since they are highly skewed to the right of the distribution curve. For these variables, inverse hyperbolic sine transformation has been performed to account for the zero and negative values5. The variables before and after
transformation can be found in Table A2 in the Appendix. The skewness and kurtosis values normalize after the hyperbolic sine transformation. The new variables have been included in the regressions as control variables.
Self-assessed health status is another important determinant of financial risk-taking at an older age, since a higher health risk automatically implies more uncertainty regarding medical expenditures. Edwards (2005) and Rosen and Wu (2004) investigate the role of self-assessed health status on financial risk-taking and find that it significantly influences the holding probabilities of financial assets and amounts held conditional on ownership. To account for this relationship between financial risk-taking and self-assessed health status a new control variable is used based on a question how individuals rank their own health on a 1-5 Likert scale.
Furthermore, it has been shown that financial literacy is an important determinant, next to an individual’s education level (Van Rooij et al., 2011). There is a positive relationship; higher financial literacy is positively related with the probability of direct and indirect stockholdings. A dummy variable has been created which equals one if the respondent correctly answered the question on interest accumulation on savings accounts and zero otherwise. Additionally, following the approach of Hong et al. (2004), a sociability dummy is constructed. It takes on a value of one if respondents indicate to at least have partaken once over the past year in voluntary, educational, social club, religious or political activities and zero otherwise.
It is assumed that macro-economic experiences also significantly contribute to an individual’s willingness to take financial risk. Malmendier and Nagel (2011) investigated
5 𝐼𝑛𝑣𝑒𝑟𝑠𝑒 ℎ𝑦𝑝𝑒𝑟𝑏𝑜𝑙𝑖𝑐 𝑠𝑖𝑛𝑒 (𝑣𝑎𝑟𝑖𝑎𝑏𝑙𝑒
16
this and found that individuals are less willing to take financial risk if they experienced low stock market returns in the past. To account for this relationship, two new dummy variables are introduced. These equal one if individuals were born during the Great Depression in the 1930s or during the second World War respectively and zero otherwise. Based on Bucciol and Zarri (2015), these two dummy variables sufficiently account for the macro-economic experiences.
Two extra, more value-based, control variables have been included. The first variable is trust which is based on respondents indicating their trust in others on a Likert scale ranging from 0-10. This follows the approach by Guiso et al. (2008) who finds a positive relationship between trust and stock-holding probabilities. Based on research by Kaustia and Torstila (2012), a variable representing individuals’ self-assessed political values by a left-right scale has been included. In their paper, the authors find that individuals with a left-wing political preference have a lower tendency to hold financial assets. Finally, to account for possible time-variance and heterogeneity between countries (Christelis et al., 2012), dummy variables have been created for the different waves and countries used in this analysis.
3.1.3. Descriptive statistics
Table 2 shows the descriptive statistics of the variables used in this paper, as the total dataset and per wave. The dependent, independent and control variables are presented. With regards to the dependent variables, it can be seen that the holding probabilities for all the individual asset classes lie between 3% and 7%. The holding probability for any of the four financial assets equals 14.7%. Furthermore, it can be observed that the probability of holding (at least one) financial asset(s) increases over the course of time. The biggest increase is seen in the probability of holding stocks which increases from 5.7% to 8.2%, an increase of 2.5 percentage points. Whereas the probability of holding any financial asset is 11.7% in 2011, this increases to approximately 17.8% in 2013. The risk aversion variable equals 3.76 on a 1-4 scale, indicating a very high level of risk aversion. Although the risk aversion variable drops slightly in the 2013 as compared to 2011, this is not in the same magnitude as the increased holding probabilities for participating in the financial market. The share invested in each of the financial assets lies anywhere between 3.3% for IRA’s to 35.1% for stocks, and does not change in an easily observable direction over time.
17
variable with mean zero and standard deviation equalling one. In 2011 the average lies at -0.07 in 2011 whereas this is 0.06 in 2013, also indicating a slight increase in happiness. The control variables provide an idea of the dataset that is employed in this paper. Firstly, the average age of the respondents is high at 66.17. This is not surprising however, since SHARE only surveys persons aged 50 and over and their partners. The same holds for the relatively high number of war and depression babies, 27.5% and 6.8% respectively. Lastly, the positive skewness in terms of age displays itself in the amount of respondents that are currently unemployed, which lies at approximately 68%. The number of respondents indicating to have a self-assessed good health or better equals 57.1% for the total dataset, but notably increases from 54.4% in 2011 to 59.7% in 2013. With regards to the more value-based control variables, the averages for trust and sociability equal 5.79 (1-10 scale) and 51.1% (dummy variable) respectively, but increase notably in 2013 compared to 2011. More elaborate descriptive statistics can be found in Appendix A3.
3.2. Empirical strategy
In this section, the methodology for the baseline regressions is elaborated on. These regressions cover all three proxies for risk aversion using both independent variables.
3.2.1. Baseline regressions
The goal for the baseline regression is to estimate the impact of happiness on the probability of holding financial assets, on individuals’ subjective risk aversion, and on the relative share invested conditional on ownership. The methods used are explained subsequently.
For the first analysis Ordinary Least Squares (OLS) is used as the regression to analyse the impact of Happiness (satisfaction) and Happiness (factor) on two proxies: 1) the holding probabilities of the four financial assets and the holding probabilities for at least one of the four financial assets; and 2) subjective risk aversion. Both independent variables are used separately in order to assess their respective impact. Whereas the first five regressions use a dummy variable as dependent variable; the sixth and final regression used is a categorical variable on a scale from 1-4. The following regression specification is used:
𝑌𝑖 = 𝛼 + 𝛽𝐻𝑖 + 𝛿𝑋𝑖′+ 𝜃1𝐷1+ . . + 𝜃𝑛𝐷𝑛 + 𝜀𝑖 (1) Where Y can be either one of the dependent variables (holding probabilities and subjective risk aversion). With H as either one of the independent variables Happiness (satisfaction) and Happiness (factor). The vector of control variables is denoted by 𝑋𝑖 and
18
Table 2. Descriptive statistics for dependent, independent and control variables.
Dependent variables: Total Wave 4 (2011) Wave 5 (2013)
Holding prob. bonds 0.0304 (0.1718) 0.0274 (0.1631) 0.0333 (0.1794)
Share bonds 0.3430 (0.2997) 0.3332 (0.2980) 0.3505 (0.3010)
Holding prob. stocks 0.0699 (0.2549) 0.0573 (0.2323) 0.0817 (0.2739)
Share stocks 0.2705 (0.2690) 0.2819 (0.2767) 0.2630( 0.2636)
Holding prob. mutual funds 0.0636 (0.2440) 0.0433 (0.1990) 0.0845 (0.2781)
Share mutual funds 0.3705 (0.2819) 0.3284 (0.2792) 0.3899 (0.2811)
Holding prob. IRA’s 0.0337 (0.1804) 0.0277 (0.1642) 0.0329 (0.1943)
Share IRA’s 0.2791 (0.1978) 0.2764 (0.1951) 0.2809 (0.1996)
Holding any financial asset 0.1466 (0.3537) 0.1154 (0.3195) 0.1761 (0.3809)
Risk aversion 3.7599 (0.5216) 3.7687 (0.5189) 3.7517 (0.5239)
Independent variables: Total Wave 4 (2011) Wave 5 (2013)
Happiness (satisfaction) 7.3919 (1.9260) 7.3488 (1.9632) 7.4325 (1.8893)
Happiness (factor) 0.0000 (1.0000) -0.0682 (1.0368) 0.0622 (0.9610)
Control variables: Total Wave 4 (2011) Wave 5 (2013)
Age 66.168 (10.248) 64.542 (10.401) 67.700 (98.584) Gender 0.4259 (0.4945) 0.4261 (0.4945) 0.4260 (0.4944) Couple 0.6860 (0.4641) 0.6890 (0.4629) 0.6837 (0.4652) Having Children 0.9056 (0.2924) 0.9046 (0.2938) 0.9063 (0.2911) Not working 0.6444 (0.4787) 0.6182 (0.4859) 0.6693 (0.4706) Post-secondary education 0.2531 (0.4348) 0.2519 (0.4341) 0.2551 (0.4354)
Self-reported good health 0.5707 (0.4950) 0.5438 (0.4981) 0.5967 (0.4907)
Financial literacy 0.0696 (0.2544) 0.1425 (0.3496) 0.1209 (0.3095)
War baby 0.2749 (0.4465) 0.2544 (0.4356) 0.2944 (0.4557)
Depression baby 0.0681 (0.2519) 0.0672 (0.2503) 0.0689 (0.2534)
Sociability 0.5109 (0.4999) 0.4142 (0.4926) 0.6471 (0.4781)
Trust 5.7895 (2.3718) 5.6392 (2.4122) 5.9320 (2.3242)
Left Right Politics 4.9720 (2.2723) 4.9649 (2.2724) 4.9832 (2.2723)
Gross hh income (‘ 000s) 3.5104 (1.1520) 3.3760 (1.1244) 3.6394 (1.0747)
Gross hh fin assets (‘ 000s) 2.3684 (2.0339) 1.9994 (1.9732) 2.7230 (2.0290)
Real hh assets (‘ 000s) 4.5095 (2.2467) 4.3558 (2.2220) 4.6588 (2.2611)
Note: Reported are the means of the variables and the standard deviation between parentheses. The “share” variables are conditional on participating. Happiness (factor) is created using Principal Component Factoring and therefore has a zero mean and standard deviation of one.
19
𝑌𝑖 = { 1 𝑌𝑖∗ = 𝛼 + 𝛽𝐻𝑖+ 𝛿𝑋𝑖′+ 𝜃1𝐷1+ . . + 𝜃𝑛𝐷𝑛+ 𝑣𝑖 > 0
0 else (2) Where the variables are similar as the ones in Equation (1), only the error term changes to 𝑣𝑖, which represents the zero-mean residual. In the second step of the two-part model OLS is used with the following specification:
𝑌𝑖 = 𝛼 + 𝛽𝐻𝑖 + 𝛿𝑋𝑖′+ 𝜃1𝐷1+ . . + 𝜃𝑛𝐷𝑛+ 𝑣𝑖 (3) Where 𝑌𝑖 is the relative share invested in the asset of interest. Other variables are similar to the ones in Equations (1) and (2). One control variable is dropped from this regression, the variable for gross financial wealth. It is dropped since this variable is used as denominator to calculate the independent variable share of gross financial wealth invested in any of the financial assets.
4. Analysis and results
In this section, the results of the empirical analysis are reported. The goal of this section is to provide a comprehensive overview of the effects of happiness on risk aversion. Firstly, a correlation matrix and univariate cross-tabulations are reported and elaborated on. In the second part, the multivariate results from the baseline regressions are reported and elaborated on.
4.1. Univariate results
Table 3 reports the correlation matrix between the independent variables used in the baseline regressions. As can be seen, Happiness (satisfaction) and Happiness (factor) have a high correlation equalling 0.490. The variables (3) to (6) are used to create Happiness (factor). Almost all variables show at least a weak positive relationship with each other.
Table 3. Correlation matrix between independent variables.
(1) (2) (3) (4) (5) (6)
(1) Happiness (satisfaction) 1
(2) Happiness (factor) 0.490 1
(3) Satisfaction activities 0.425 0.514 1
(4) Looking forward day 0.251 0.700 0.168 1
(5) Life meaning 0.371 0.794 0.238 0.419 1
(6) Look back life happiness 0.326 0.697 0.202 0.272 0.394 1
Note: Happiness (factor) is the factor extracted from a PCF analysis using variables (3) – (6): Satisfaction activities; Looking forward day; Life meaning; and Look back life happiness.
20
Table 4. Cross-tabulations of Happiness (satisfaction) and Happiness (factor) on three proxies for risk aversion.
Panel A: Happiness (satisfaction) on three proxies for risk aversion: 1) Has any financial asset; 2) Subjective risk aversion; and 3) Portfolio share.
Has any financial asset Subjective risk aversion Portfolio share (in quartiles)
No Yes N 1 (low) 2 3 4 (high) N 1 (low) 2 3 4 (high) N
H appine ss (sa tisfa ct ion) 0 97.29% 2.71% 480 1.25% 1.67% 4.79% 92.29% 480 23.08% 15.38% 30.77% 30.77% 13 1 96.30% 3.70% 243 1.23% 1.65% 8.64% 88.48% 243 33.33% 11.11% 22.22% 33.33% 9 2 96.29% 3.71% 512 1.17% 0.59% 5.27% 92.97% 512 31.58% 5.26% 10.53% 52.63% 19 3 96.50% 3.50% 1,258 0.64% 0.79% 7.47% 91.10% 1,258 31.82% 25.00% 22.73% 20.45% 44 4 96.12% 3.88% 1,496 0.74% 1.20% 9.36% 88.70% 1,496 25.86% 15.52% 29.31% 29.31% 58 5 94.08% 5.92% 8,688 0.55% 1.06% 10.17% 88.21% 8,688 27.43% 25.49% 25.10% 21.98% 514 6 91.94% 8.06% 5,658 0.65% 1.43% 13.11% 84.80% 5,658 28.95% 24.56% 22.37% 24.12% 456 7 86.41% 13.59% 11,188 0.88% 1.97% 18.02% 79.13% 11,188 23.09% 26.18% 26.12% 24.61% 1,520 8 81.95% 18.05% 19,508 0.76% 2.44% 20.67% 76.13% 19,508 26.26% 25.13% 24.96% 23.74% 3,521 9 78.13% 21.87% 8,994 0.76% 3.24% 23.54% 72.47% 8,994 23.79% 25.01% 26.08% 25.11% 1,967 10 81.32% 18.68% 9,652 1.04% 2.36% 17.46% 79.14% 9,652 23.68% 24.35% 28.34% 23.63% 1,803 Total 85.34% 14.66% 67,677 0.79% 2.11% 17.41% 79.69% 67,677 25.00% 25.00% 25.86% 24.14% 9,924
Pearson χ2 (10) 1800.00 Pearson χ2 (10) 1300.00 Pearson χ2 (10) 36.69
p-value 0.000 p-value 0.000 p-value 0.111
Panel B: Happiness (factor) on three proxies for risk aversion: 1) Has any financial asset; 2) Subjective risk aversion; and 3) Portfolio share.
Has any financial asset Subjective risk aversion Portfolio share (in quartiles)
No Yes N 1 (low) 2 3 4 (high) N 1 (low) 2 3 4 (high) N
H appine ss (fa ct or ) I n qua rt ile s 1 (low) 89.72% 10.28% 14,425 0.70% 1.59% 14.20% 83.51% 14,425 23.06% 25.29% 25.83% 25.83% 1,483 2 84.71% 15.29% 15,320 0.86% 1.98% 19.15% 78.02% 15,320 26.22% 23.06% 26.13% 24.59% 2,342 3 80.42% 19.58% 15,723 0.76% 2.58% 21.97% 74.69% 15,723 25.79% 26.24% 26.18% 21.79% 3,079 4 (high) 78.26% 21.74% 11,665 0.94% 2.96% 22.12% 73.98% 11,665 24.76% 25.43% 26.38% 23.42% 2,536 Total 83.48% 16.52% 57,133 0.81% 2.25% 19.28% 77.66% 57,133 25.19% 25.08% 26.17% 23.56% 9,440
Pearson χ2 (10) 761.04 Pearson χ2 (10) 476.00 Pearson χ2 (10) 184.06
p-value 0.000 p-value 0.000 p-value 0.031
21
It can be observed in Table 4 that the chance of owning any financial asset increases from approximately 3% to 20% between the highest and lowest scores of Happiness (satisfaction). The same relationship can be seen with Happiness (factor). Unsurprisingly, both Pearson χ2 statistics are highly significant. Regarding the cross-tabulations of both
measures for Happiness with subjective risk aversion, a similar relationship can be found, namely that the level of subjective risk aversion decreases with higher levels of happiness. There is a decreased probability of almost 15 percentage points (p.p.) in having the highest level of risk aversion between a 0 and a 10 for Happiness (satisfaction). This equals approximately 7% between the lowest and highest quartiles for Happiness (factor). Similarly, the Pearson χ2 statistic is highly significant. No conclusions can be
drawn for the last proxy, which is portfolio share.
At first glance, the univariate tabulation suggest that the hypothesis that an increase in happiness leads to a decrease in risk aversion can be firmly contested. The probability of owning a financial asset increases drastically and the level of risk aversion decreases. This is contrary to the hypothesis of this paper. Although these results are no substitute for the more extensive multivariate analysis, a certain pattern can be implied.
4.1.1. Multivariate results
In this paragraph, the results of all the baseline regression analyses are presented. In Section V the results of the robustness tests are presented and explained.
22
Table 5. OLS regression results of Happiness (satisfaction) on the probabilities to hold financial assets and individuals’ risk aversion levels.
% chance Bonds % chance Stocks % chance Mutual Funds % chance IRAs % chance Has any financial asset Subjective risk aversion
Happiness (satisfaction) -0.0021*** (0.0004) -0.0016*** (0.0006) -0.0024*** (0.0005) -0.0007* (0.0004) -0.0038*** (0.0008) 0.0023* (0.0013) Age 0.0003*** (0.0001) -0.0002 (0.0001) 0.0002 (0.0001) -0.0009*** (0.0001) -0.0005** (0.0002) 0.0050*** (0.0003) Gender -0.0021 (0.0017) 0.0070*** (0.0026) 0.0033 (0.0024) 0.0127*** (0.0019) 0.0110*** (0.0032) -0.1218*** (0.0051) Couple 0.0021 (0.0019) -0.0008 (0.0028) -0.0070*** (0.0027) 0.0270*** (0.0017) 0.0145*** (0.0035) 0.0469*** (0.0060) Having children -0.0084*** (0.0033) -0.0103** (0.0045) -0.0003 (0.0041) 0.0081*** (0.0026) -0.0034 (0.0054) 0.0073 (0.0089) Not working 0.0076*** (0.0020) 0.0082*** (0.0030) 0.0112*** (0.0029) -0.0095*** (0.0022) 0.0129*** (0.0038) 0.0337*** (0.0062) Post-secondary education 0.0108*** (0.0021) 0.0197*** (0.0032) 0.0183*** (0.0031) 0.0053** (0.0023) 0.0346*** (0.0040) -0.0971*** (0.0062)
Self-reported good health 0.0015 (0.0016) 0.0010 (0.0024) 0.0055** (0.0023) 0.0005 (0.0017) 0.0067** (0.0032) -0.0417*** (0.0054)
Financial literacy 0.0067** (0.0034) 0.0209*** (0.0046) 0.0023 (0.0041) 0.0050 (0.0035) 0.0169*** (0.0055) -0.0642*** (0.0097) Sociability 0.0014 (0.0016) 0.0083*** (0.0024) 0.0070*** (0.0022) 0.0031* (0.0018) 0.0156*** (0.0032) -0.0430*** (0.0054) War baby 0.0037* (0.0020) 0.0047* (0.0028) 0.0083*** (0.0028) -0.0104*** (0.0018) 0.0066* (0.0036) 0.0000 (0.0054) Depression baby -0.0001 (0.0036) -0.0098** (0.0046) -0.0181*** (0.0044) 0.0027 (0.0023) -0.0163*** (0.0060) -0.0181** (0.0088) Trust 0.0002 (0.0003) 0.0003 (0.0005) 0.0022*** (0.0005) 0.0013*** (0.0004) 0.0026*** (0.0006) -0.0069*** (0.0011) Left-right politics 0.0000 (0.0003) 0.0013*** (0.0005) 0.0001 (0.0005) 0.0006* (0.0004) 0.0005 (0.0007) -0.0110*** (0.0011)
Hyp. household income 0.0017** (0.0008) 0.0061*** (0.0012) 0.0066*** (0.0011) 0.0054*** (0.0009) 0.0116*** (0.0016) -0.0201*** (0.0030)
Hyp. gross financial assets 0.0214*** (0.0007) 0.0380*** (0.0009) 0.0405*** (0.0009) 0.0178*** (0.0006) 0.0797*** (0.0011) -0.0408*** (0.0017)
Hyp. real assets 0.0007* (0.0004) 0.0063*** (0.0005) 0.0020*** (0.0005) 0.0001 (0.0004) 0.0061*** (0.0007) -0.0077*** (0.0012)
Dummy: Wave 4 0.0127*** (0.0017) 0.0166*** (0.0024) -0.0116*** (0.0024) -0.0047*** (0.0018) 0.0047 (0.0031) -0.0067 (0.0050)
Constant -0.0468*** (0.0080) -0.0697*** (0.0114) -0.0744*** (0.0106) 0.0060 (0.0071) -0.1105*** (0.0145) 3.8182*** (0.0247)
Observations 48,780 48,780 48,780 48,780 48,780 48,780
Adjusted R2 0.084 0.170 0.150 0.088 0.292 0.145
23
Table 6. Baseline regression results of both measures of Happiness on the three proxies for risk aversion.
Panel A: OLS regression with Happiness (satisfaction) and Happiness (factor) on the ownership probabilities of financial assets and on the level of risk aversion. % chance Bonds % chance Stocks % chance Mutual Funds % chance IRAs % chance Has any financial asset Subjective risk aversion
Happiness (satisfaction) -0.0021*** (0.0004) -0.0016*** (0.0006) -0.0024*** (0.0005) -0.0007* (0.0004) -0.0038*** (0.0008) 0.0023* (0.0013)
Controls? Yes Yes Yes Yes Yes Yes
Observations 48,780 48,780 48,780 48,780 48,780 48,780
Adjusted R2 0.084 0.170 0.150 0.088 0.292 0.145
Happiness (factor) -0.0015* (0.0009) -0.0029** (0.0013) -0.0026** (0.0012) -0.0001 (0.0009) -0.0053*** (0.0016) 0.0082*** (0.0027)
Controls? Yes Yes Yes Yes Yes Yes
Observations 42,972 42,972 42,972 42,972 42,972 42,972
Adjusted R2 0.085 0.170 0.151 0.090 0.291 0.144
Note: Reported are the regression coefficients and standard errors in parentheses. ***, **, * denote significance at 1%, 5% and 10% respectively. Standard errors are clustered at the individual level.
Panel B: Two-part model Happiness (satisfaction) and Happiness (factor) on the relative share of gross financial assets invested in the financial asset classes.
% Share Bonds % Share Stocks % Share Mutual Funds % Share IRAs % Share portfolio
Happiness (satisfaction) -0.0005* (0.0003) 0.0007* (0.0004) 0.0005 (0.0004) 0.0003 (0.0003) 0.0011 (0.0007)
Controls? Yes Yes Yes Yes Yes
Observations 45,363 45,363 45,363 45,363 45,363
Adjusted R2 0.254 0.066 0.121 0.075 0.103
Happiness (factor) -0.0005 (0.0005) 0.0004 (0.0007) 0.0007 (0.0008) 0.0006 (0.0005) 0.0011 (0.0013)
Controls? Yes Yes Yes Yes Yes
Observations 40,582 40,582 40,582 40,582 40,582
Adjusted R2 0.249 0.063 0.118 0.077 0.101
24
As can be seen in Table 6 – Panel A, Happiness (satisfaction) has a strongly significant negative effect on all the financial asset ownership probabilities. Specifically, a one-point increase in Happiness (satisfaction) leads to an approximate 0.2 p.p. decrease in the probability of holding bonds, stocks and mutual funds. For IRAs, this effect is only marginal at -0.075 percentage points. As can be expected, the total chance of holding any of the four financial assets is reduced by 0.38 p.p. for every one-point increase in Happiness (satisfaction). These findings confirm the first proxy of the hypothesis. Corresponding to the reduced holding probabilities, there is a statistically significant positive relationship between Happiness (satisfaction) and the subjective risk aversion level of respondents. Specifically, a one-point increase in Happiness (satisfaction) leads to a 0.0023 increase in the level of subjective risk aversion. This suggests that the theory by Delis & Mylonidis (2015) holds, where it is reasoned that reduced ownership probabilities of stocks are channelled through individuals’ reduced willingness to take financial risk. Section V will further elaborate on the expected mediating effect of subjective risk aversion. The regression with the second independent variable, Happiness (factor), finds the same negative effects for ownership probabilities. However, now there is no more statistical significance for IRAs and the general strength of significance decreases slightly. When analysing the effects on subjective risk aversion, one sees a statistically significant positive effect, meaning that for every one-point increase in Happiness (factor), the level of subjective risk aversion increases by 0.008 points. It is evident that the effects of Happiness on the ownership probabilities are negative and, with one exception, statistically significant. The level of subjective risk aversion increases as the level of Happiness increases. This implies that the first two proxies for risk aversion confirm the hypothesis.
25 5. Robustness tests
In this section, several robustness tests are presented to strengthen the validity of the baseline regression results. As a first remark, endogeneity has not been accounted for so far. This not only includes measurement errors, which is discussed in Section 6, but also includes omitted variable bias and reverse causality bias. These biases are accounted for in this section. Further in this section, alternative regression specifications are used to test the hypothesis; as well as in- or exclusion of other control variables in the regressions.
5.1. Endogeneity
5.1.1. Omitted variable bias
Happiness is known to be a concept for which it is difficult to establish a relationship stronger than a simple correlation due to the omitted variable bias that often exists (Guven & Hoxha, 2015). The effect of happiness on financial market decisions could for instance stem from the fact that there are well-functioning (governmental or financial) institutions in a certain country, which makes it easier for individuals to invest in the financial market, but simultaneously makes people happier. Many examples exist, which is why happiness is often instrumented. Guven (2009, 2015) uses weather (sunshine) as an instrument for happiness in financial market decisions whereas Delis & Mylonidis (2015) use family relationships as an instrument. Family relations is used as an instrument since the relation between family relations and happiness is positive. Diener & Seligman (2002) compared the upper 10% of consistently very happy people with the average and the most unhappy 10%. It was found that very happy people have stronger social relationships than less happy people. Among a wide list of variables concerning both personality and behavioural traits, ‘social relationships’ was the only variable to sufficiently predict happiness of individuals. Apart from this paper, there are several other papers indicating that social relationships play an important role in predicting happiness (Helliwell and Putnam, 2005; Goswami, 2012). The variable social relationships by Delis & Mylonidis (2015) is approximated by using the question how satisfied individuals are with their family relations, not focusing on other relatives. In this paper however, a closer approximation to social relations can be made by using the question: “Overall, on a scale
from 0 to 10, where 0 means completely dissatisfied and 10 means completely satisfied, how satisfied are you with the [ relationship that you have with the person/ relationships that you have with all the people] we have just talked about?”. The individuals first had to
indicate a number of people they felt close to, before this question was asked. This variable constructed from the above question is called network satisfaction.
26
the error term (Verbeek, 2012). Therefore, there can be no effect of an omitted variable. In a two-stage least squares models’ first stage, a new variable is created using an instrument variable. In the second stage, the estimated values from stage one are used in place of the actual values of the potentially problematic indicator. The regression to be run is based on Equation (1), the same as in the baseline regressions, except for the independent variable being estimated with an instrumental variable. See Appendix C for more information on the equations and assumptions for using instrumental variables. In Table 7 – Panel A, the estimates of the regressions using two-stage least squares with network satisfaction for Happiness can be found. As can be seen, the expected negative relationship between Happiness and the holding probabilities of financial assets is found, similar to the baseline regression. While there is no statistical significance for the holding probability of bonds, there is a statistically significant 0.94, 1.57 and 0.49 p.p. decrease in the holding probabilities for stocks, mutual funds and IRAs respectively for a one-point increase in Happiness. For any of the four financial assets, there is a decrease of 1.98 percentage points. Similarly, the relationship between Happiness and subjective risk aversion is statistically significant and positive with a coefficient of 0.054. Since the same findings are found when using the instrumented variables directly, there is extra confirmation for the first and second proxies of risk aversion.
5.1.2. Reverse causality bias
As mentioned earlier, happiness can also suffer from endogeneity bias through reverse causality. Simply think of the following situation: would happier persons not invest in the stock market because they are happy, or would it be that they are happy because they do not invest in the stock market? In order to account for this possible reverse causality, a lagged predictor for Happiness (satisfaction) and Happiness (factor) is added to the model. The independent variables from 2011, the first time period in this dataset, have been used as part of the regression with the other variables from 2013. The same baseline Equation (1) is used, only now both independent variables have been lagged. This follows the approach of Delis & Mylonidis (2015) to account for reverse causality bias.
27
of second independent variable, Happiness (factor), on the binary holding probabilities has the same signs as for the regressions using Happiness (satisfaction). However, now there is strong statistical significance for the holding probabilities of stocks, mutual funds and at least one of the four financial assets. No statistical significance is found for the amount of bonds.
Concluding for lagged Happiness, there is not such consistent statistical significance as when regressing current Happiness. However, significance is consequently found for the holding probabilities of any of the four financial assets. Inconsistent but some statistical significance is found for the four individual asset classes. With regards to subjective risk aversion, no statistical significance is found for either independent variable. Full regression specifications for both Panels of Table 7 can be found in Appendix – D1-3.
Table 7. Results for the robustness tests of omitted variable and reverse causality bias.
Panel A: Two-stage least squares Instrumental Variables regression results using network satisfaction as an instrument for Happiness on the probabilities to hold financial assets and individuals’ risk aversion levels.
% chance
Bonds % chance Stocks
% chance Mutual Funds % chance IRAs % chance Has any financial asset Subjective risk aversion Happiness (instrument) (0.0025) -0.0034 -0.0094** (0.0037) -0.0157*** (0.0037) -0.0049* (0.0029) -0.0198*** (0.0049) 0.0543*** (0.0084)
Controls? Yes Yes Yes Yes Yes Yes
Observations 48,780 48,780 48,780 48,780 48,780 48,780
Adjusted R2 0.084 0.170 0.145 0.087 0.289 0.123
Panel B: OLS regression with lagged Happiness (satisfaction) and lagged Happiness (factor) OLS regression results using lagged happiness factor on the probabilities to hold financial assets and individuals’ risk aversion levels.
% chance
Bonds % chance Stocks
% chance Mutual Funds % chance IRAs % chance Has any financial asset Subjective risk aversion Happiness-1 (satisfaction) -0.0013** (0.0006) (0.0010) -0.0014 (0.0011) -0.0013 (0.0007) 0.0007 -0.0023* (0.0013) (0.0021) 0.0033
Controls? Yes Yes Yes Yes Yes Yes
Observations 20,392 20,392 20,392 20,392 20,392 20,392
Adjusted R2 0.084 0.173 0.161 0.100 0.305 0.167
Happiness-1
(factor) (0.0014) -0.0018 -0.0049** (0.0021) -0.0049** (0.0022) (0.0014) 0.0009 -0.0088*** (0.0027) (0.0039) 0.0038
Controls? Yes Yes Yes Yes Yes Yes
Observations 18,378 18,378 18,378 18,378 18,378 18,378
Adjusted R2 0.083 0.173 0.161 0.100 0.303 0.165
28
5.1.3. Mediating effect of subjective risk aversion
Whereas so far the level of subjective risk aversion has been used as dependent variable, many other papers in the household finance literature use it as a control variable in the regression specification (Rosen and Wu, 2003; Van Rooij et al., 2011). To test whether the decreased financial asset holding probabilities found in this paper and the paper by Delis & Mylonidis (2015) stem from an increase in subjective risk aversion, a mediation analysis is conducted. For this reason, also following Van Rooij et al. (2011), an OLS regression using subjective risk aversion as an extra control variable is estimated to examine if the found decreased financial asset holding probabilities come from an increase in subjective risk aversion or have another reason. The regression specification equals Equation (1) except for including subjective risk aversion as extra control variable. Full regression specifications for both panels of Table 8 can be found in Appendix D4-5
Table 8. Mediation analysis of subjective risk aversion on the holding probabilities of assets. Panel A: Coefficients of regressions with Happiness (satisfaction) on the holding probabilities of financial assets, with and without subjective risk aversion as extra control variable.
% chance Bonds % chance Stocks
% chance Mutual Funds % chance IRAs % chance Has any financial asset Subjective risk aversion Happiness (satisfaction) with risk aversion as control -0.0020*** (0.0004) -0.0016*** (0.0006) -0.0023*** (0.0005) -0.0007* (0.0004) -0.0037*** (0.0008) Happiness (satisfaction)
without risk aversion as control
-0.0021***
(0.0004) -0.0016*** (0.0006) -0.0024*** (0.0005) -0.0007* (0.0004) -0.0038*** (0.0008) (0.0013) 0.0023* % effect subjective
risk aversion 4.8% 0.0% 4.2% 0.0% 2.6%
Panel B: Coefficients of regressions with Happiness (factor) on the holding probabilities of financial assets, with and without subjective risk aversion as extra control variable.
% chance Bonds % chance Stocks
% chance Mutual Funds % chance IRAs % chance Has any financial asset Subjective risk aversion Happiness (factor)
coefficient with risk aversion as control -0.0009 (0.0009) -0.0023* (0.0012) (0.0012) -0.0016 (0.0009) -0.0001 -0.0038** (0.0016) Happiness (factor) coefficient without risk aversion as control -0.0015* (0.0009) -0.0029** (0.0013) -0.0026** (0.0012) (0.0009) -0.0001 -0.0053*** (0.0016) 0.0082*** (0.0027) % effect subjective risk aversion 40.0% 20.7% 38.5% 0.0% 28.3%
Note: Reported are regression coefficients and standard errors in parentheses. ***, **, * denote significance at 1%, 5% and 10% respectively. Standard errors are clustered at the individual level. The % effect subjective risk aversion is the relative difference between the two coefficients.
29
anywhere between 20% and 40% for Happiness (factor), except for IRAs. This suggests that there is a (very) weak mediating effect of subjective risk aversion on the financial holding probabilities, and that the largest effect comes from different factors.
5.2. Alternative regression specifications 5.2.1. Logistic regressions
For the first step, the holding probabilities, logistic regressions are used to account for the fact that the dependent variable is binary in nature and therefore OLS might misfit the regression. This leads to the same model specification as the first step in the two-part model, Equation (2). Since this model is for binary variables, subjective risk aversion as a variable with a 1-4 scale is not regressed. Therefore, only the first proxy of risk aversion is tested with this logistic regression.
To ease interpretation, marginal effects are reported. These reflect the change in the holding probability of financial assets when the variables of interest change by one unit. For dummy variables, this is the change from zero to one. Table E1 in the Appendix provides the results for the logistic regression using Happiness (satisfaction) and Table E2 in the Appendix provides results for Happiness (factor). In Table E1, only the effects for the holding probabilities of bonds are strongly significant. Specifically, a one-point increase in Happiness (satisfaction) corresponds to a 0.17 p.p. decrease in the holding probability of bonds. There is no statistical significance for any of the other three financial assets nor for the four financial assets combined. When glancing at Table E2 similar results are seen. This time however, neither of the five dependent variables has statistical significance. All the coefficients in both Tables E1 & E2, except for IRAs, are negative, so the direction implied is equal to the baseline regression. Since there is no statistical significance, no conclusions can be made.
5.2.2. Tobit censored regressions on relative share invested
30
model, as employed by Rosen and Wu (2004) and Bertaut and Starr-Mccluer (2002), accounts for these unobserved latent observations6.
The marginal effects for the tobit regressions are specified in Table E3 and Table E4 in the Appendix respectively for Happiness (satisfaction) and Happiness (factor). The results are somewhat different to the ones when using the standard two-part model. As a first observation in Table E3, all coefficients except for bonds are positive and the coefficients for the percentage invested in stocks, mutual funds and the portfolio as a whole are all strongly statistically significant. More specifically, a one-point increase in Happiness (satisfaction) leads to a 0.97 p.p. increase in the share of individuals’ gross financial wealth invested in financial assets. With regards to the relative share of stocks and mutual funds, there is an approximate 1.1 p.p. increase in the percentage of wealth invested. Comparing to the two-part model in the baseline regression, where a negative effect on the relative share of bonds was found, the tobit regression finds no statistical significance. The same finding as the baseline regressions are found for the percentage invested in stocks, they increase with happiness. Observing Happiness (factor) in Table E4, there are again mostly positive effects, but only the percentage invested in IRAs is weakly significant and increases with 1.72 percentage points. The other regressions with Happiness (factor) do not yield any statistical significance.
The contrary effects within the regressions and between the regressions lead to uncertainty concerning the third risk aversion proxy. On the basis of the robustness test, as well as the baseline regressions, no conclusions can be drawn on the third proxy for risk aversion, the percentage of wealth invested in financial assets.
5.2.3. OLS regression controlling for optimism
As a final step, a strongly reduced regression is run by controlling for self-assessed level optimism. For instance, Puri and Robinson (2007) find that (excessive) optimistic investors invest more heavily in stocks, which suggests controlling is necessary. Self-assessed optimism however has only been asked in the second wave of the SHARE questionnaire (2006). In this wave, respondents had to respond to the following question: “I’m always optimistic about my future.” on a 1-5 Likert scale, with 1 corresponding to strongly agree and 5 corresponding to strongly disagree. The method is identical to Delis
6 The following specification is used for the tobit model: 𝑌
𝑖∗ = 𝛼 + 𝛽𝐻𝑖+ 𝛿𝑋𝑖′+ 𝜃1𝐷1+ . . + 𝜃𝑛𝐷𝑛+ 𝜀𝑖.
𝑌𝑖∗ is the relative share invested, and is a latent variable which can be written as: 𝑌𝑖= {
𝑌𝑖∗ if 𝑌𝑖∗ > 0
0 if 𝑌𝑖∗ ≤ 0
