An Inquiry into the Effects of
Personality on Household Finance
M.Sc. Finance Thesis June 8, 2017 Remco Pander rpander93@gmail.com s2150018 Supervisor: dhr. A. A. Tsvetkov a.tsvetkov@rug.nl
Keywords: big five traits, personality, household finance, financial assets, unsecured debt JEL classification: C24, D03, D14, H31
ABSTRACT
Introduction
For long, economic and financial theory assumed that man based his decision-making exclusively on rationale and self-interest, so as to optimize the expected value of the choice under investigation. In essence, it was presumed that there was no limit on the rational capacity of man and neither was it possible that (un)subconscious influences could alter the decision-making process. In the second half of the twentieth century, resistance against this proposition grew from multiple disciplines of science. Researches started to examine the multitude of biases inherent in every human. These biases, often also referred to as heuristics, allow for quick decision-making using various shortcuts in the thought process. They are often inherited from our ancestors, who had to make snap decisions in situations where mere seconds could mean the difference between life and death. The findings from these researches has spawned the field of behavioral economics, combining psychology, sociology and economics. It has attracted a lot of attention since then, and has largely falsified the concept of man as a purely rational decision maker. Moreover, recent evidence has shown that non-cognitive skills have a strong effect on a large number of socio-economic outcomes. In particular, a person’s personality plays a large part in the decisions and outcomes obtained in many situations. This research zooms in on these outcomes, and focuses specifically on financial investment decisions. We will research the link between personality traits as measured by the “Big Five” framework and personal finance decisions. These incorporate the amount of financial wealth and debt held by individuals. Some earlier research has been performed on this subject, but evidence is still scarce in this specific area. Furthermore, personality is partially dependent on culture (Diener, Oishi, and Lucas, 2003), which of course varies from country to country. If consistent results can be found originating from multiple countries and continents, the evidence becomes substantially stronger. This research focuses on The Netherlands, using data available from the Dutch Central Bank.
holding. Only a limited number of authors have looked at the possible interaction effects between personality. While more evidence is required to strengthen the findings, it does provide an interesting starting point for future research.
The rest of this paper is divided as follows. Section 2 will present an overview of the origins and existing literature in behavioral economics. It mentions several papers specifically relevant to this research. Section 3 gives the research questions and hypotheses based on the existing literature. Section 4 describes the dataset used and gives an explanation of the methodology. Section 5 presents the results. Finally, section 6 discusses the results and limitations, and finally gives a conclusion.
Literature review
From Rationality Toward Behavioral Economics
The concept of man as a utility maximizer, acting solely out of self-interest, is well-embedded within most economic theories. It posits man as rational agents, entirely aware of its environment and capable of interpreting all available relevant information, using logic to come to an optimum. This concept is dubbed “homo economicus”, having its roots in the nineteenth century (Persky, 1995). It has transformed economics, a social science, into one of a highly mathematical nature where the goal often is to maximize the utility function. The foundations for expected utility theory is based on the works of Neumann and Morgenstern (1947) who proved that, under four axioms of rational behavior, a decision maker will have a utility function that represents his preferences. Furthermore, the decision maker will act to maximize the expected value of this utility function, defined over the set of possible outcomes. It is referred to as the von Neumann-Morgenstern utility theorem. Self-interested, rational consumers are of course at the core of capitalism as defined by Adam Smith (1776). The concept is also applied in much of financial theory. As an example, modern portfolio theory (MPT) dictates that the portfolios of all investors should be based on mean-variance analysis, maximizing the return for any given level of risk (Markowitz, 1952). Inputs required for the theory are based on expected values for all relevant variables, which the investor is assumed to be able to calculate correctly based on information available.
Tversky, 1979), which is an alternative to expected utility theory in decision-making. It posits that rather than weighting available options based on expected utility, man looks at potential gains and losses, where equivalent gains and losses are weighted differently. Kahneman and Tversky estimated that losses carry about double the weight of gains. Moreover, probabilities are not uniformly assessed. Low probability events are overstated, while high probability events are understated. The resulting effects on the decision-making process are markedly different than what would be expected from utility theory. Behavioral economics has since then revealed much more evidence that man does not base his decision-making solely on rationale, but rather often employs rules of thumb to assess situations. Multiple researchers have proposed the dual process theory (Evans, 1984; Kahneman, 2003), explaining that the thinking process is dictated by two separate “systems”, one automatic uncontrolled system, basing its decisions on rules of thumb and experience, and one system basing its decisions on logic and conscious reasoning. Bounded rationality is at the source of this distinction, a concept describing that the human mind is incapable of effectively processing all available information and analyzing complex situations due to limited time and limited cognitive capability (Gigerenzer and Selten, 2002).
“Big Five” Personality Traits
Linking Personality to Economic and Financial Concepts
Studies linking personality traits with a multitude of socio-economic variables are abundant. Uysal and Pohlmeier (2011) find that conscientiousness and neuroticism have a strong impact on the duration of unemployment for a person. There is a positive relationship between conscientiousness and the probability of finding a job, while neuroticism has a negative relationship. Heineck and Anger (2010) conduct an interesting study, examining the effect of cognitive ability and personal traits on job performance as measured by hourly wage. They find mixed results for the five-factor model personality traits, differing by gender. For example, the authors find an inverse u-shaped relationship between conscientiousnes and hourly wages for females, while they find a regular u-shaped relationship between these same variables for males. Cognitive ability is only positively related with hourly wages for men. One personality trait not included in the five-factor model is found to have the most significant effect, namely “locus of control”, which measured the extent to which persons attribute positive and negative outcomes to themselfves or to the environment. An external locus of control is associated negatively with hourly wage. Another study questions the standard economic assumption that marginal utility is the same for all individuals, being independent of personality (Boyce and Wood, 2011). They find that individuals with high levels of conscientiousness or extraversion get more satisfaction from income increases, while individuals with high levels of openness, friendliness or neuroticism tend to get lower satisfaction from income increases. That the effect of emotions and personality on (financial) decisions is not limited to non-professionals can be confirmed by the works of Fenton-O'Creevy, Soane, Nicholson and Willman (2011) and Lo, Repin and Steenbarger (2005) who both find that emotions have a significant effect on the profits made by professional financial traders. Sadi et al. (2011) look at the relationship between each of the five-factor model personality traits and observed financial biases while selling and buying risky assets, and find a significant correlation for four of the five personality traits. For example, extraversion is positively related with hindsight bias.
Household Finance
research. Second, the two personality traits under research, openness to experience and conscientiousness did not predict short-term investing using structured equation modelling. Openness to experience did however predict long-term investing in the model. Moreover, the authors found that neuroticism and extraversion, had a significant relationship with short-term investing. Neuroticism reportedly decreased the intentionality to conduct short-term investing, while extraversion increased the intentionality of short-term investing. However, they had no impact on long-term investing. Conlin et al. (2015) analyze the relationship between personality and stock market participation using the “Temperament and Character Inventory” (TCI) model, which is an alternative to the Big Five personality traits. In summary, using logistic regression analysis the authors find economically significant effects for several personality traits. Focusing on savings, Cobb-Clark, Kassenböhmer and Sinning (2016) look specifically at the effect of locus of control and various aspects of savings. Locus of control captures personal beliefs about the controllability of events and is most closely related to the neuroticism trait of the Big Five model. They find households with a high locus of control to save more, in percentage of household income and in absolute monetary levels. Most closely related to this research are the papers by Brown and Taylor (2014) and Bucciol and Luca (2015). The first pair of authors, Brown and Taylor (2014) research the effect of the Big Five personality traits on individual and household finance using a censored regression approach and data of British households. They find that some personality significanly affect the amount of unsecured debt and (risky) financial assets held. Extraversion and openness to experience show a large positive effect on debt, while extraversion simultaneously shows a large negative effect on financial assets. Interestingly, neuroticism and conscientiousness show no consistent effect on any of the holdings. Finally, openness to experience increases the probability of holding risky assets. Bucciol and Zarri (2015) use an extension of the Big Five model to research further on stock market participation and risk-taking. Employing a logistic model, they find that individualistic personalities show a greater dispense to risk taking. Additionally, the authors also have information on the assessed chance that the market goes up in the next year. Similarly, this estimation goes up with traits related to individualism. This research will refer back to the conclusions of the two sets of authors throughout this paper.
Research questions and hypotheses
Big Five model and the amount of financial assets, unsecured debt and proportion of risky assets over total assets will be investigated in an empirical setting. Albeit earlier research has looked into these subjects, this research intends to deepen the evidence and look for possible interaction effects between personality aspects.
Firstly, concerning the effect of personality on financial assets and unsecured debt, behavioral finance and personality research suggests that extraversion and openness to experience asserts a positive effect on the amount of financial assets and unsecured debt an individual holds. Both of these traits relate to the willingness of a person to engage with the world. This makes it more likely that persons scoring high on these traits will more easily make decisions increasing financial assets or debt compared to individuals scoring low on these traits. Additionally, low openness to experience indicates a sense of restraint, which can easily be related to a restraint to engage with risky financing such as unsecured debt. Conscientiousness, relating to the degree to which a person is willing to stick to standards and norms, will have a positive effect on financial assets and a negative effect on debt. These individuals are more disciplined than individuals scoring low on this trait, making it probable they can adhere to proficient financial standards increasing financial assets while simultaneously refraining from taking unnecessary spending and increasing debt.
Summarizing, this results in the following testable questions and hypotheses:
1. What is the effect of personality traits on financial asset and unsecured debt holding of an individual?
Hypothesis: extraversion, openness and conscientiousness have a positive relationship with financial assets.
Hypothesis: extraversion and openness have a positive relationship with unsecured debt. Conscientiousness will have a negative relationship with unsecured debt.
2. What is the effect of personality traits on the proportion of risky assets over total financial assets of an individual?
Hypothesis: openness and conscientiousness will have a positive relationship with the proportion of risky assets. Neuroticism will have a negative relationship with the proportion if risky assets.
Data and Methodology
Data Gathering
To conduct a proper empirical analysis, data is required on two distinct subjects. First, the dataset should contain information on psychological questions from which a reliable personality trait score can be derived. Second, data on aggregate wealth and debt is required on an individual level. There is a limited availability of datasets that contain both these subjects. Several research h projects that did contain this information have stopped in the past decade. Fortunately, the Dutch Central
Bank still conducts an annual household survey (DHS) that is sufficient to conduct the analysis. The household survey is conducted by CentERdata since 1993 (CentERdata, 2016), which is in itself a collaboration of several Dutch institutions.
are not available in the data, so this variable is not included in our analysis. The latest wave available is from 2016, which has around 2,200 observations.
In the context of this research, it is however appropriate to opt for a different approach and combine the datasets of 2013, 2014 and 2015, for two reasons. Firstly, the questionnaires in these years is identical with respect to the questions relevant for this research, meaning that additional observations can be extracted by merging the datasets. This yields an additional 1,000 observations compared to the 2016 dataset. Secondly, the 2016 dataset contains just 10 questions measuring personality (2 for each Big Five trait), whereas the 2013 – 2015 questionnaires include 50 questions (10 for each Big Five trait) on this subject, improving the reliability of the personality trait indexes derived from these questions. Since the 2012 and earlier editions of the questionnaire again has just 10 questions on personality, they are excluded.
Results from the survey are distributed in seven separate data files for each year, containing each sub-section of the survey. Three of these seven are relevant, containing general household information, aggregated wealth and debt data, and psychological concepts. The 2015 dataset contains 2,400 observations, the 2014 dataset holds 2,155 observations and lastly the 2013 dataset has 2,088 observations. Each observation in the dataset contains a unique household identification number and person in household number. Combining these using the following formula yields a unique person identification number that can be used to identify individuals across years.
assets and unsecured debt) are constructed from the individual components of financial assets and debt available as variables in the dataset. A list of included components can be found in the footnotes of table 1. From categorical variables on employment status, highest education completed and self-assessed health status, dummy variables are extracted.
Table 1 shows descriptive statistics for the relevant variables, except the personality traits which will be handled subsequently. Control variables are taken from existing literature in behavioral economics (see for example Dohmen et al., 2011; Guiso and Paiella, 2008; Guiso and Sodini, 2013) and are similar to those used in Brown and Taylor (2014) and Bucciol and Zarri (2015).
Table 1: Descriptive statistics for the dataset derived from the DNB Household Survey compiled using the results from 2013 - 2015. The exhibit illustrates the mean (and standard deviation) for each variable. Additionally, it indicates whether the variable is a dummy variable extracted from a categorical variable in the source data. The sample contains 2,703 unique observations across 2,116 households. Logarithms have been calculated as ln(variable + 1) to account for zero-values. For household income, the omitted category is <10k. For education, the omitted category is no high school or college education. For health, the omitted category is poor health.
Dummy Mean (Standard Deviation) Dependent variables
Total financial assets1 No 35,391 (96,090)
→ ln(Total financial assets) No 7.98 (3.64)
Total risky assets2 No 6,692 (38,423)
→ ln(Total risky assets) No 1.43 (3.45)
Total unsecured debt3 No 4,604 (37,793)
→ ln(Total unsecured debt) No 1.77 (3.56)
Control variables Male Yes .53 (0.49) Age No 56.4 (16.3) Adults in household No 1.82 (.47) Children in household No .69 (1.05) Employed Yes .47 Self-Employed Yes .13 Partner Yes .78
Home owner Yes .76
Household income: poor (10k – 14k) Yes .06 (.23) Household income: lower-class (14-22k) Yes .14 (.35) Household income: middle-class (22k – 40k) Yes .45 (.49) Household income: upper-class (>40k) Yes .32 (.47)
1 Includes: checking accounts, savings accounts, savings letters, life insurance policies, growth funds, investment funds, bonds, stocks, options purchased, receivables and substantial interests
Table 1 continued on next page
Education: high school Yes .96
Education: college Yes .61
Health: fair Yes .19
Health: good Yes .62
Health: excellent Yes .13
For each of the five-factor model personality traits, ten questions are asked in the survey that are related to it. Each of these questions is answered on a 5-point Likert scale, with “does not apply at all” taking the value of 1, and “applies perfectly” taking a score of 5. Thus, the survey contains 50 questions related to personality. Table 2 shows the questions and relates them to each of the five traits. Although the annotations to the survey do not explicitly indicate to what personality trait each question is related, the asked questions are standardized and listed on the website of the International Personality Item Pool.
Table 2: Overview of the questions related to personality asked in the household survey, categorized with the personality trait from the “Big Five” index the question attempts to measure. Each category contains 10 questions in total. Some questions are reversed, i.e. asked in a negative connotation with respect to the personality trait. Personality questions are standardized internationally and derived from the International Personality Item Pool.
“Big Five” trait Survey question Reversed Mean (Standard Deviation) 1. Extraversion/introversion 1. I keep in the background Yes 3.15 (.99)
2. I am quiet around strangers Yes 2.95 (1.07) 3. I am the life of the party No 2.27 (1.03) 4. I do not talk a lot Yes 2.65 (1.04) 5. I feel comfortable around people No 3.69 (.83) 6. I start conversations No 3.34 (.91) 7. I have little to say Yes 2.38 (.96) 8. I talk to a lot of different people on
parties
No 3.19 (1.05) 9. I do not mind being the center of
attention
No 2.80 (1.06)
10. I do not like to draw attention to myself
Yes 3.36 (1.04)
2. Friendliness/hostility 1. I sympathize with others’ feelings No 3.89 (.79) 2. I take time out for others No 3.79 (.77) 3. I feel little concern for others Yes 2.06 (1.04) 4. I am interested in people No 3.89 (.80) 5. I insult people Yes 1.58 (.83) 6. I am not interested in other people’s
problems
Yes 2.21 (.97)
7. I have a soft heart No 4.08 (.70) 8. I am not really interested in others Yes 2.08 (.94) 9. I feel others emotions No 3.59 (.84) 10. I make people feel at ease No 3.59 (.78)
3. Conscientiousness 1. I like order No 3.87 (.898) 2. I leave my things lying around Yes 2.54 (1.18) 3. I do chores right away No 3.18 (.99) 4. I live my life according to schedules No 2.62 (1.08) 5. I neglect my obligations Yes 1.92 (.99) 6. I pay attention to detail No 3.76 (.89) 7. I am accurate in my work No 4.06 (.78) 8. I forget to put things back where
they belong
Yes 2.29 (1.11)
9. I am always well prepared No 3.51 (.83) 10. I often make a mess of things Yes 1.83 (.88) 4. Neuroticism 1. I have frequent mood swings No 2.13 (1.06)
2. I get stressed out easily No 2.38 (1.08) 3. I seldom feel blue Yes 3.34 (1.11) 4. I am relaxed most of the time Yes 3.56 (.84) 5. I worry about things No 3.17 (.99) 6. I am easily disturbed No 2.78 (1.00) 7. I get upset easily No 2.28 (.99) 8. I change my mood a lot No 2.19 (1.00) 9. I get irritated easily No 2.37 (.94) 10. I often feel blue No 1.98 (.98) 5. Openness to experience 1. I have excellent ideas No 3.39 (.81) 2. I have a vivid imagination No 3.34 (1.03) 3. I am full of ideas No 3.09 (.95) 4. I have a rich vocabulary No 3.63 (.91) 5. I have difficulty understanding
abstract ideas
Yes 2.53 (1.0)
6. I am not interested in abstract ideas Yes 2.66 (1.01) 7. I am quick to understand things No 3.83 (.78) 8. I do not have a good imagination Yes 2.39 (1.02) 9. I use difficult words No 2.68 (1.07) 10. I spend time reflecting on things No 3.64 (.82)
Table 3: Personality indices constructed from the individual questions asked in the survey. For each index, the mean score and standard deviation is shown. Additionally, Cronbach's alpha is calculated to verify the reliability of the indices. A Cronbach’s alpha above .70 is considered adequate.
After having constructed the indices, a correlation matrix can be used to look for obvious signs of relatedness. Table 4 gives this matrix including all five constructed indices and the variables for the natural logarithms of financial assets and unsecured debt. While none of the coefficients is high in absolute terms (<.10), conscientiousness, neuroticism and openness are consistently statistically significant with financial assets and unsecured debt. On the other hand, extraversion and friendliness are not statistically significant at all. It is important to keep in mind here that the correlation coefficient only measures the direct relation between any two variables, not accounting for more intricate relationships. Interestingly, there is a lot of correlation present between the personality indices.
Table 4: Correlation matrix for the personality indices and the natural logarithms of financial assets and unsecured debt (N=2,703). Correlations marked with *, ** or *** are statistically significant at the 10%, 5% or 1% level respectively.
Variable Extraversion Friendliness Conscientiousness Neuroticism Openness Extraversion - Friendliness .35*** - Conscientiousness .15*** .33*** - Neuroticism -.20*** -.22*** -.25*** - Openness .33*** .31*** .17*** -.16*** - Financial assets .00 -.00 -.04** .07*** .09*** Total debt -.03 -.02 .06*** -0.10*** .04**
Methodology Analyzing Financial Assets and Unsecured Debt
Firstly, the relationship between the Big Five personality traits and financial assets and unsecured debt will be modelled so that the first research question and associated hypotheses can be investigated. Following Brown and Taylor (2014) and using the work of Long (1997), financial assets (ahi) and unsecured debt (dhi) are treated as censored variables, since they
cannot have negative values. This makes it appropriate to use a censored regression model, Personality trait index Mean (Standard deviation) Cronbach’s alpha
Extraversion/introversion 2.95 (.60) .84
Friendliness/hostility 3.75 (.57) .84
Conscientiousness 3.50 (.57) .77
Neuroticism/emotional stability 2.41 (.66) .87
because the dependent variables are censored (from the left). Figure 2 and figure 3 clearly show this censoring point around zero. These censored regression models are utilized for data
where, hypothetically, standard regression models could predict values for the dependent variables that are not possible. In this case, a regression model could predict negative values for unsecured debt based on a combination of personality traits. However, since it is not possible to have negative values for unsecured debt or financial assets, it must be treated as a latent or censored variable.
Truncated regression models are not suitable in this context, since they assume that observations are only included for situations where the dependent variable is not censored, which is not the case. Thus, the observable dependent variables 𝑙𝑛(𝑎ℎ𝑖) and 𝑙𝑛(𝑑ℎ𝑖) are modeled using a random-effects Tobit regression model, where respectively 𝑙𝑛(𝑎ℎ𝑖∗ ) and 𝑙𝑛(𝑑ℎ𝑖∗ ) stand for the latent variables. Equations (2) and (3) show this model for financial assets, while equations (4) and (5) show it for unsecured debt.
𝑙𝑛(𝑎ℎ𝑖∗ ) = 𝛼0+ 𝛽1′𝑋ℎ𝑖+ ∑5𝑗=1𝐶𝑗𝐼𝑗ℎ𝑖+𝜀ℎ𝑖1 (2) 𝑙𝑛(𝑎ℎ𝑖) = { 𝑙𝑛(𝑎ℎ𝑖∗ ) 𝑖𝑓 𝑙𝑛(𝑎ℎ𝑖∗ ) > 0 0 𝑖𝑓 𝑙𝑛(𝑎ℎ𝑖∗ ) ≤ 0 (3) 𝑙𝑛(𝑑ℎ𝑖∗ ) = 𝛼1+ 𝛽2′𝑋ℎ𝑖+ ∑5𝑗=1𝐷𝑗𝐼𝑗ℎ𝑖+𝜀ℎ𝑖2 (4) 𝑙𝑛(𝑑ℎ𝑖) = { 𝑙𝑛(𝑑ℎ𝑖∗ ) 𝑖𝑓 𝑙𝑛(𝑑ℎ𝑖∗ ) > 0 0 𝑖𝑓 𝑙𝑛(𝑑ℎ𝑖∗ ) ≤ 0 (5)
In equations (2) through (5), the amount of financial assets and unsecured debt is given by 𝑎ℎ𝑖 and 𝑑ℎ𝑖 for person i (i = 1, …, 10) in household h (h = 1, …, 2116). The first term, 𝛼 is a
Figure 3: Histogram showing the censoring of the distribution of the natural logarithm of unsecured debt. As visible, there is a clear censoring point around zero. Almost 80 percent of the individuals in the dataset carry no unsecured debt.
constant. Furthermore, 𝑋ℎ𝑖 represents a vector of control variables, which are drawn from existing literature, consisting of gender, age, number of adults in the household, number of children in the household, employment status, house ownership and education level. The variable 𝐼𝑗ℎ𝑖 (j = 1, …, 5) represents each of the Big Five personality trait index scores calculated earlier. Finally, Ɛℎ𝑖1 and Ɛℎ𝑖2 represent the error terms consisting of two parts: 𝜀ℎ𝑖 = 𝛼ℎ+ 𝜑ℎ𝑖, where 𝛼ℎ is a household-specific effect and 𝜑ℎ𝑖 a normally-distributed
random error term. The vector 𝛽′ and each 𝐶
𝑗 and 𝐷𝑗 stand for the parameters that must be
estimated.
Using Cameron and Trivedi (2010)’s tests for heteroskedasticity and nonnormality, we find that the initial models estimated in Stata suffer significantly from these statistical flaws (Appendix B provides the Stata codes to run these tests). Thus, it cannot be used since the estimated coefficients will be biased and unreliable. Fortunately, the authors present an alternative in the form of combining a binary regression to model Pr(total financial assets > 0) with a linear regression to model E(ln total financial assets|total financial assets > 0). The same is done for unsecured debt. Specifications for these models are given in equations (6) through (9). The independent variables are the same as for the original Tobit model.
ℎ𝑎𝑠 𝑓𝑖𝑛𝑎𝑛𝑐𝑖𝑎𝑙 𝑎𝑠𝑠𝑒𝑡𝑠 = 𝛼0+ 𝛽3′𝑋ℎ𝑖+ ∑5𝑗=1𝐸𝑗𝐼𝑗ℎ𝑖+ 𝜀ℎ𝑖3 (6)
𝑙𝑛(𝑎ℎ𝑖) = 𝛼0+ 𝛽4′𝑋ℎ𝑖+ ∑5𝑗=1𝐹𝑗𝐼𝑗ℎ𝑖+ 𝜀ℎ𝑖4 (7)
ℎ𝑎𝑠 𝑢𝑛𝑠𝑒𝑐𝑢𝑟𝑒𝑑 𝑑𝑒𝑏𝑡 = 𝛼0+ 𝛽5′𝑋ℎ𝑖+ ∑5𝑗=1𝐺𝑗𝐼𝑗ℎ𝑖+ 𝜀ℎ𝑖5 (8)
𝑙𝑛(𝑑ℎ𝑖) = 𝛼0+ 𝛽6′𝑋ℎ𝑖+ ∑5𝑗=1𝐻𝑗𝐼𝑗ℎ𝑖+ 𝜀ℎ𝑖6 (9)
Using this procedure, a probit model is combined with a standard linear regression model. Overall fit as measured by summing the likelihood for the two models using this procedure is significantly better than the initial model for both financial assets and unsecured debt. Next, a Hausman test is used to compare a fixed-effects model with a random-effects model. Since the null hypothesis of the test cannot be rejected for either financial assets and unsecured debt, a model using random-effects is deemed appropriate. Finally, clustered robust standard errors are used at the household level.
knowledge. Some examples exist of studies that found interaction effects while looking at other socio-economic variables. One example comes from Witt, Burke-Smalley, Barrick, and Mount (2002) who found such effects in a study relating conscientiousness and friendliness to job performance. The authors find an interaction effect between these two traits strenghteing the relationship. Nyhuis and Pons (2005) examine the effect of personality on wage earnings, and also include interaction terms between personality and tenure and education. They find interaction effects between conscientiousness and tenure, and agreeableness and tenure. Based on findings in earlier literature about which traits exert statistically significant influence (see literature review, household finance section) and my own hypotheses, the scope is restricted to three of the traits, namely: conscientiousness, friendliness and openness.
To ease interpretation of the results, the three included personality traits are centered in this regression. Centering means to substract the mean from every observation, so that the interpretation of the other regression coefficient should be with respect to this average. As an illustration, consider the following model that has as independent variables friendliness and openness and an interaction term between the two. Both of the independent variables are centered around the average (example adjusted from Williams, 2015).
𝐸(𝑌) = 𝛼 + 𝛽1𝐹𝑅𝐼𝐸𝑁𝐷𝐿𝐼𝑁𝐸𝑆𝑆 + 𝛽2𝑂𝑃𝐸𝑁𝑁𝐸𝑆𝑆 + 𝛽3𝐹𝑅𝐼𝐸𝑁𝐷𝐿𝐼𝑁𝐸𝑆𝑆 ∗ 𝑂𝑃𝐸𝑁𝑁𝐸𝑆𝑆 (10)
= 𝛼 + 𝛽1𝐹𝑅𝐼𝐸𝑁𝐷𝐿𝐼𝑁𝐸𝑆𝑆 + (𝛽2+ 𝛽3𝐹𝑅𝐼𝐸𝑁𝐷𝐿𝐼𝑁𝐸𝑆𝑆) ∗ 𝑂𝑃𝐸𝑁𝑁𝐸𝑆𝑆 (11)
= 𝛼 + 𝛽2𝑂𝑃𝐸𝑁𝑁𝐸𝑆𝑆 + (𝛽1+ 𝛽3𝑂𝑃𝐸𝑁𝑁𝐸𝑆𝑆) ∗ 𝐹𝑅𝐼𝐸𝑁𝐷𝐿𝐼𝑁𝐸𝑆𝑆 (12)
This example of these three mathematically identical models shows that the effect of friendliness on Y depends on openness, and the effect of openness on Y depends on friendliness. Now consider what happens for an individual with “average” openness (i.e. openness = 0).
𝐸(𝑌) = 𝛼 + 𝛽1𝐹𝑅𝐼𝐸𝑁𝐷𝐿𝐼𝑁𝐸𝑆𝑆 (13)
The model simplifies to an equation with only the intercept and one independent variable. Based on this fact, we can now say that the coefficient of friendliness measures the effect of friendliness for a person with average openness. The same reasoning can be used for openness by setting friendliness = 0.
ln(𝑎ℎ𝑖) = 𝛼0+ 𝛽5′𝑋ℎ𝑖+ ∑𝑗=13 𝐾𝑗𝐽𝑗ℎ𝑖+ ∑3𝑗≠𝑘𝐿𝑗𝐽𝑗ℎ𝑖𝐽𝑘ℎ𝑖 + 𝜀ℎ𝑖7 (14)
ln(𝑑ℎ𝑖) = 𝛼0+ 𝛽6′𝑋ℎ𝑖+ ∑𝑗=13 𝑀𝑗𝐽𝑗ℎ𝑖+ ∑3𝑗≠𝑘𝑁𝑗𝐽𝑗ℎ𝑖𝐽𝑘ℎ𝑖+ 𝜀ℎ𝑖8 (15)
In these equations, the first term, α is a constant. Xhi represents the vector of control variables, and the variable Jjhi (j = 1, ..., 3) represents the centered personality indexes for
conscientiousness, friendliness and openness. 𝐾 and 𝑀 are the estimated regression coefficients for the individual demeaned personality traits included. Then, 𝐿 and 𝑁 are the coefficients for each of the included interaction terms. Finally, Ɛhi stands for the error term, which again consists of a household-specific error and a normally-distributed error term. In the estimation of these models, only observations with positive financial assets or unsecured debt are included.
Table 5 gives an overview of the number of observations categorized along whether they score higher or lower than the average for a particular trait. Using the three included traits, this results in eight possible categories. This overview clearly shows that some profiles are more common than others, hinting at the possibility that some traits move in tandem. The correlation results as were shown in Table 4 also provide evidence for this suggestion.
Table 5: Number of observations categorized using the dummy variables indicating for each individual in the dataset whether he/she scores higher (or lower) than the average on a particular trait. It is immediately visible that some combinations are overrepresented versus other combinations.
Openness Conscientiousness Friendliness Number of observations
High High High 437
High High Low 225
High Low High 215
Low High High 341
Low High Low 346
High Low Low 268
Low Low High 215
Low Low Low 656
Methodology Analyzing Proportion in Risky Assets
Furthermore, no adaptations are required if a large portion of the observations is at either end of the bounds. Equation (16) describes the model.
𝑝𝑒𝑟𝑐𝑒𝑛𝑡𝑎𝑔𝑒 𝑟𝑖𝑠𝑘𝑦 𝑎𝑠𝑠𝑒𝑡𝑠 = 𝛼5+ 𝛽5′𝑋ℎ𝑖+ ∑5𝑗=1𝜑𝑗𝐼𝑗ℎ𝑖+𝜀ℎ𝑖5 (16) For the most part, this model is nearly identical to the censored regression model described earlier. Thus, α is the regression constant. 𝑋ℎ𝑖 represents the vector of control variables. In
addition to the control variables mentioned earlier, also included is the natural logarithm of total financial assets as control variable in this model, since logically only individuals having sufficient financial wealth are willing to invest part of their wealth in risky assets. The variable Ijhi (j = 1, ..., 5) represents each of the Big Five personality trait index scores
calculated earlier. Finally, 𝜀ℎ𝑖5 is the error term. As in the earlier models, clustered robust
standard errors are used at the household level. Robustness Checks
To check the robustness of the results from the models, analysis is also performed using a different dataset, the DHS Household Survey of 2016 which was described above. Since it contains less observations and less observations related to personality, it was not included in the main dataset. It suffices however for robustness checks. Additionally, the analysis is performed using a subsample of the original dataset, only including respondents between the ages 30 and 65. Empirical evidence shows that, while personality changes throughout life, it is most stable during adulthood (see Roberts and DelVecchio, 2000; Roberts, Walton, and Viechtbauer, 2006; Terracciano, McCrae, and Costa, 2010), as also explained in the literature review.
Results
Results Analyzing Financial Assets and Unsecured Debt
In this section, the results from the Tobit regression relating personality traits to financial assets and unsecured debt are presented. As mentioned, the Tobit models are estimated using a two-step procedure. Thus, results are given for both the linear models and the probit models. Additionally, the results for the models including interaction effects are presented.
Table 6: Results of the linear regression model relating the "Big Five" personality traits to financial assets and unsecured debt. Regression coefficients and standard errors (between) parentheses are included in the table. *, ** and *** mark statistically significant coefficients at the 10%, 5% and 1% level respectively. In addition, the table provides the Wald Chi2 and pseudo-R2 for the models. Only observations are included that have positive financial
assets and unsecured debt.
Financial assets Unsecured debt Coefficient (standard error) Coefficient (standard error)
Constant 5.61 (.64)*** 9.27 (1.25)*** Explanatory variables Extraversion -.08 (.07) -.04 (.15) Friendliness -.28 (.08)*** -.04 (.18) Conscientiousness .22 (.07)*** -0.05 (.18) Neuroticism -.07 (0.67) .01 (.15) Openness .29 (.08)*** -.07 (.20) Control variables Male .46 (.08)*** .36 (.20)* Age .03 (.00)*** -.03 (.01)*** Adults in household -.50 (.16)*** -.51 (.29)* Children in household -.15 (.04)*** -.14 (.09) Employed .11 (.11) .30 (.22) Self-Employed .43 (.13)*** .40 (.33) Partner .29 (.19) .46 (.37) Home owner 1.04 (.11)*** .25 (.23)
Household income: poor (10k – 14k)
-0.07 (.27) -.48 (.57)
Household income: lower-class (14-22k) .19 (.25) -.15 (.41) Household income: middle-class (22k – 40k) .37 (.23) -.48 (.36) Household income: upper-class (>40k) .55 (.24)** -.18 (.38) Education: high school .36 (.21)* 1.01 (.41)**
Education: college .49 (.09)*** .55 (.21)** Health: fair .25 (.22) .41 (.43) Health: good .53 (.21)** .07 (.42) Health: excellent .64 (.22)*** .59 (.46) Model statistics Included observations 2,337 572 Wald Chi^2 517.08*** 81.26*** R^2 19.4% 12.42%
natural logarithm of unsecured debt was used as independent variable should be interpreted with a grain of salt.
Looking briefly at the controls before turning to the personality traits, it is clear that there are several socio-economic indicators that are highly statistically significant with respect to both financial assets and unsecured debt. Age has a positive effect on financial assets, while having a negative effect on unsecured debt. The number of individuals in the household (both adults and children) has a negative effect on financial assets, indicating that financial resources are used for supporting the individuals that would otherwise be available for other purposes. Self-employed individuals tend to hold significantly more financial assets than unSelf-employed individuals or regular employees, which makes sense keeping in mind that these individuals generally have to have larger buffers because their fate is more dependent on the economic cycle. Interestingly, only for upper-class households does there appear to be a statistically significant positive effect on financial assets. For households with an income <40k, no significant relationship can be found. Lastly, both education and health status appear to have a positive effect on financial assets. For unsecured debt, only a positive effect can be found for education.
any relationship. Our results align on openness to experience and friendliness with regard to financial assets. Moreover, while we find a positive relationship for conscientiousness, they fail to find any consistent relationship. Related to unsecured debt, they find a positive relationship for extraversion and openness to experience. This highly suggests that my lack of finding any results with regard to unsecured debt is likely related to my limited sample size. Moving on, Table 7 shows the results of the probit models used to estimate the probability of holding financial assets or unsecured debt. The equations for the models can are shown in equation (6) and (8). For brevity, the results with respect to the control variables are omitted and the discussion limits itself to the explanatory variables. A full overview of the results can be found in appendix C. Since this regression model has a binary dependent variable, it provides result about what personality traits influence the probability of having financial assets and unsecured debt per se, rather than the exact level of these holdings.
Table 7: Results from the probit model using a binary variable indicating whether the respondent has financial assets or unsecured debt. Only results for the explanatory variables are shown. Regression coefficients and standard errors (between parentheses) are given. *, ** and *** mark statistically significant results at the 10%, 5% and 1% level respectively. Additionally, the Wald Chi2 statistic is given for both models.
Has financial assets Has unsecured debt Coefficient (standard error) Coefficient (standard error) Explanatory variables Extraversion -.12 (.06)* -.05 (.07) Friendliness .01 (.07) .11 (.08) Conscientiousness .10 (.06) -.05 (.07) Neuroticism -.06 (.05) .15 (.07)** Openness -.03 (.07) .22 (.09)** Model statistics Number of observations 2,703 2,703 Wald Chi^2 34.35** 105.73***
Looking at unsecured debt, the results show that neuroticism and openness to experience do show some statistically significant impact on the probability of holding unsecured debt. Since just 26 percent of the observations hold any unsecured debt, and that in general it is less likely for any one individual to have (considerable) unsecured debt, it is likely that personality does exert some influence on this probability. Regarding the found relationships, it is interesting that neuroticism has a positive relationship with the probability of holding unsecured debt. Individuals that score high on neuroticism are considered to be emotionally unstable and thus experiencing anxiety and vulnerability more easily. As such, one would expect that neuroticism should have a negative relationship with unsecured debt, because these individuals would be wise to avoid the risk associated with debt. Finally, openness to experience also increases the probability of holding unsecured debt, which is in line with the hypothesized effect and the findings of Brown and Taylor (2014).
After having looked at the personality traits individually, it is now time for a more in-depth investigation and allow for more complex relationships to exist. Specifically, equations (14) and (15) allow for the existence of interaction effects between the independent variables. Based on the findings in the previous section, and findings by earlier authors, the scope for interaction effects is limited to the personality traits conscientiousness, friendliness and openness. As explained in the methodology, the explanatory variables are centered in this model (the control variables are not). This allows for easier interpretation of the results, which are reported in Table 8.
Table 8: Results of a linear regression model including interaction terms between openness, conscientiousness and friendliness. The main explanatory variables are centered around the mean. The table first shows the results for the individual traits and then for the interaction terms. *, ** and *** indicate statistical significance at the 10%, 5% and 1% level respectively. Only observations with positive financial assets or unsecured debt are included.
Financial assets Unsecured debt Coefficient (standard error) Coefficient (standard error) Personality traits (demeaned)
Extraversion -.07 (.07) -.05 (.15) Friendliness -.27 (.08)*** -.05 (.18) Conscientiousness .23 (.07)*** -.04 (.19) Neuroticism -.08 (.07) .01 (.16) Openness .30 (.08)*** -.08 (.21) Openness * Friendliness -.06 (.12) .12 (.27) Openness * Conscientiousness .07 (.13) -.17 (.33) Friendliness * Conscientiousness -.27 (.12)** .14 (.27)
Model statistics
Number of observations 2,337 572
Wald Chi^2 541.99 82.46
R^2 19.6% 12.51%
Firstly, results for the individual personality traits are identical in terms of sign and statistical significance to those shown in Table 6. Friendliness, conscientiousness and openness all show a strong relationship with financial asset holding. None of the traits show any obvious relationship with respect to unsecured debt holding. Continuing, the results show the presence of an interaction effect between friendliness and conscientiousness. To investigate whether the interaction term enhances the model, a Wald test can be used, which uses the following hypotheses:
𝐻0: 𝐹𝑅𝐼𝐸𝑁𝐷𝐿𝐼𝑁𝐸𝑆𝑆 ∗ 𝐶𝑂𝑁𝑆𝐶𝐼𝐸𝑁𝑇𝐼𝑂𝑈𝑆𝑁𝐸𝑆𝑆 = 0 𝐻1: 𝐹𝑅𝐼𝐸𝑁𝐷𝐿𝐼𝑁𝐸𝑆𝑆 ∗ 𝐶𝑂𝑁𝑆𝐶𝐼𝐸𝑁𝑇𝐼𝑂𝑈𝑆𝑁𝐸𝑆𝑆 ≠ 0
The test statistic of 5.37 is statistically significant at the 5% level (p = .02) thus rejecting the null-hypothesis that the coefficient is not different from zero. Figure 4 displays the relationship graphically. For a person with “average” friendliness (friendliness = 0), the relationship between conscientiousness and financial assets is upward sloping, in line with the positive coefficient for
conscientiousness as depicted in Table 8. If the person is “less-than-average” friendly, the relationship becomes even more positive. However, for individuals that are “more-than-average” friendly, the relationship becomes downward sloping. No such relationship is observed for unsecured debt, which is not strange since none of the personality traits themselves are significant.
In addition to the results shown in Table 8, further models were estimated including interaction effect with and between the other two personality traits: extraversion and neuroticism. They did however not yield any added insights. Moreover, in line with the models depicted in equations (6) and (8), we also estimated two probit models on the probability of holding financial assets or unsecured debt, with the interaction terms included in the estimation, but none of the terms showed any statistical significance. Since none of these models provided supplementary insights, their results are omitted.
Robustness Checks
The same two datasets were also used to check the robustness of the interaction effect found between friendliness and conscientiousness. In line with the other robustness checks, the effect was reproducible in the subsample of individuals between 30 and 65, but not in the Household Survey 2016.
Results Analyzing Proportion in Risky Assets
After looking at financial assets and unsecured debt in general, focus now shifts to a specific type of financial assets, specifically risky assets. This asset class is defined to include: growth funds, investment funds, bonds, stocks, purchased options and substantial interests. In comparison to other financial asset classes, this set carries considerably more risk. To analyze the effect of personality on risky asset holding, a fractional probit model is used as formalized by Papke and Wooldridge (1996). Table 9 presents the results of the model. To give additional perspective, the model is also ran on a set of observations with positive risky asset holding. Only results for the explanatory variables are shown. The full results are available in appendix E.
Table 9: Results from the fractional probit model using risky asset holding (as percentage of total assets) as dependent variable. The explanatory variables are the Big Five personality traits. Also included are the results of a model only including observations with positive risky asset holding. *, ** and *** denote statistically significant results at the 10%, 5% and 1% level respectively. Also included are the Wald Chi2 and pseudo-R2 for both
models.
Entire sample Only having risky assets Coefficient (standard error) Coefficient (standard error) Explanatory variables Extraversion -.14 (.06)** -.26 (.09)*** Friendliness .10 (.07) .16 (.08)** Conscientiousness -.03 (.06) .03 (.08) Neuroticism .01 (.06) -.06 (.08) Openness .03 (.07) -.01 (.09) Model statistics Included observations 2,703 413 Wald Chi^2 222.03*** 63*** R^2 17.2% 3.4%
significant, with a positive sign. This is notable, since the results in Table 6 show a negative relationship between friendliness and financial assets. Apparently, individuals that show more empathy to others, hold less financial assets in general, but tend to invest a greater portion of their financial assets in risky assets. The other three traits: conscientiousness, neuroticism and openness to experience fail to exert any influence. As such, the hypothesized effect cannot be found. Comparing these findings to Bucciol and Zarri (2015), they find a small negative relationship between friendliness and risky asset holding, which contradicts the results in Table 9. Brown and Taylor (2014) find also find a negative relationship with friendliness and additionally a positive relationship with openness to experience.
Lastly, the results of the fractional probit model expressed in equation (16) are checked on the DHS Household Survey of 2016. Results are given in Appendix F. Regrettably, the results as in Table 9 are not reproducible in this dataset. This absence of consistency is unfortunate. Coupled with the results from earlier research, the evidence from this particular research is inconclusive and should be researched more extensively, preferably with a larger set of data.
Discussion
The presented results cleary show that personality has an effect on household finances in multiple ways. After controlling for several socio-economic factors and verification against different samples, the results are largely persistent and in accordance with expectations derived from prior literature. Personality, as measured by the “Big Five” psychological framework influences the absolute amount of financial assets held. Friendliness, conscientiousness and openness all have an effect on this aspect of household finance. Additionally, evidence is found for an interaction effect between conscientiousness and friendliness on financial asset holding. Interestingly, a combination of high conscientiousness and low friendliness leads to a much higher level of financial assets, all else equal. This result is found on top of the individual importance of the personality traits. Regarding unsecured debt, the effect of personality is less clear. No obvious relationships are found in the main dataset, but conscientiousness does have a strong negative effect in another set. This effect also is originally hypothesized.
This research has also looked at the probability of holding financial assets or unsecured debt
per se, rather than at the exact level of these values. In this context, personality does not seem
making it unlikely that personality exerts any influence. Regarding debt, again mixed results are found. The original sample shows positive effects for neuroticism and openness on the probability of holding unsecured debt, but the effect of neuroticism vanishes in another set of
data. Additionally,
conscientiousness then
becomes negatively
statistically significant. Interestingly, this finding is again consistent with the originally hypothesized effect.
Finally, we also investigated any link between personality and investing behavior. We find that extraversion decreases risky asset holding, while friendliness increases risky asset holding. These findings seem to contradict earlier research and deserve more attention.
None of the estimated linear models display high values for 𝑅2, topping at just over 19 percent for the models displayed in Tables 6 and 8. It is important to put this number in the appropriate context before coming to any conclusions. This research has looked at the effect of personality on household finances. It is obvious that there are many more variables not included in the current analysis that can be expected to have an effect on household finances. Comparing our results to prior literature, we find mostly similarities, with a few exceptions (particularly risky asset holding). Most notably, while we find a positive relationship for conscientiousness, Brown and Taylor (2014) fail to find any consistent relationship with respect to financial assets. Other differences can mostly be explained by our explicit two-step estimation procedure, whereas they estimate the Tobit model directly. This finding of consistent results is significant, since the populations that were studied differ on cultural aspects. This is depicted in the subsequent graph, which shows the five Hofstede cultural dimensions (Hofstede, 1983), together forming a framework often used to compare countries
0 25 50 75 100 Power Distance
Individualism Masculinity Uncertainty Avoidance
Long-Term Orientation
Hofstede's cultural dimensions
Netherlands United Kingdom
on cultural aspects. Other research has hypothesized the relationship between culture and personality and found significant correlations between the different cultural dimensions and personality traits (Hofstede and McRae, 2004). All in all, this gives rise to the conclusion that the demonstrated effects of personality on household finance are consistent across different cultures.
The results of this study are highly relevant, since household finance plays an important topic in everyone’s live. My findings shine additional light on the economic decision-making process within households and has the potential to improve our understanding of it. Financial ineptitude is a socially very relevant issue, and every opportunity to understand it better should be grabbed with both hands. Financial service providers can incorporate these results to improve the advice and products offered to consumers. Governments and other educational providers can use these results to provide better financial education and specifically target individuals susceptible to unnecessary risk-taking (through the form of unsecured debt holding or excessive investments in risky assets) or individuals showing an aversion toward saving money.
While this research gives additional insights into the effects of personality on household finance, it does suffer from limitations that open the door for future research. Some of the results are not reproducible in a different dataset or become less statistically significant. Only 572 individuals in the main dataset have positive unsecured debt, making it difficult to find reliable results. Most notably, the interaction effect between friendliness and conscientiousness which is found in the main dataset and also present in the subsample of individuals between 30 and 65 is not reproducible in the Household Survey of 2016.
Conclusion
References
Borghans, L., Duckworth, A. L., Heckman, J. J., Weel, B., 2008. The Economics and Psychology of Personality Traits. Journal of Human Resources 4, 972-1059.
Boyce, C. J., & Wood, A. M., 2011. Personality and the marginal utility of income: Personality interacts with increases in household income to determine life satisfaction. Journal of Economic Behavior & Organization 78, 183-191.
Brown, S., & Taylor, K., 2014. Household Finances and the ‘Big Five’ Personality Traits. Journal of Economic Psychology, 45, 197–212.
Bucciol, A., & Zarri, L., 2015. Does Investors' Personality Influence Their Portfolios? Netspar Discussion Paper.
Cameron, C. A., & Trivedi, P. K., 2010. Tobit and selection models. In C. A. Cameron, & P. K. Trivedi, Microeconometrics for Stata. Texas: Stata Press, pp 540-543
CentERdata, 2016. DNB Household Survey. Retrieved from https://www.dhsdata.nl/site/ Cobb-Clark, D. A., Kassenböhmer, S. C., Sinning, M., 2016. Locus of Control and Savings.
Journal of Banking & Finance 73, 113-130.
Conlin, A., Kyröläinen, P., Kaakinen, M., Järvelin, M.-R., Perttunen, J., Svento, R., 2015. Personality traits and stock market participation. Journal of Empirical FInance 33, 34-50.
Cortina, J. M, 1993. What is Coefficient Alpha? An Examination of Theory and Applications. Journal of Applied Psychology 78, 98-104.
Costa, P. T., McCrae, R. R., 1992. Normal personality assessment in clinical practice: The NEO Personality Inventory. Psychological assessment.
Diener, E., Oishi, S., Lucas, R. E., 2003. Personality, Culture, and Subjective Well-Being: Emotional and Cognitive Evaluations of Life. Annual Review of Psychology 54, 403-425.
Digman, J. M., 1990. Personality Structure: Emergence of The Five-Factor Model. Annual Review of Psychology 41, 417-440.
Dohmen, T., Falk, A., Huffman, D., Sunde, U., Schupp, J., Wagner, G. G., 2011. Individual risk attitudes: measurement, determinants and behavioral consequences. Journal of the European Economic Association 9, 522-550.
Evans, J. S., 1984. Heuristic and analytic processes in reasoning. British Journal of Psychology 75, 451–468.
Fenton-O'Creevy, M., Soane, E., Nicholson, N., Willman, P., 2011. Thinking, Feeling and Deciding: The Influence of Emotions on the Decision-Making and Performance of Traders. Journal of Organizational Behavior 32, 1044-1061.
Gigerenzer, G., Selten, R., 2002. Bounded Rationality: The Adaptive Toolbox. Massachusetts: MIT Press.
Goldberg, L. R., 1993. The Structure of Phenotypic Personality Traits. American Psychologist 48, 26-34.
Guiso, L., Sodini, P., 2013. Household Finance: An Emerging Field. Handbook of the Economics of Finance. Elsevier B.V, pp 1397-1519.
Heckman, J. 2011. Integrating Personality Psychology into Economics. National Bureau of Economic Research.
Heineck, G., Anger, S. 2010. The returns to cognitive abilities and personality traits in Germany. Labour Economics 17, 535-546.
Hofstede, G. 1983. National Cultures in Four Dimensions: A Research-Based Theory of Cultural Differences among Nations. International Studies of Management & Organization 13, 46-74.
Hofstede, G., McRae, R. R., 2004. Personality and Culture Revisited: Linking Traits and Dimensions of Culture. Cross-Cultural Research 38(1).
International Personality Item Pool, 2017. Retrieved from http://ipip.ori.org/newBigFive5broadKey.htm
Jeronimus, B. F., Riese, H., Sanderman, R., Ormel, J., 2014. Mutual reinforcement between neuroticism and life experiences: A five-wave, 16-year study to test reciprocal
causation. Journal of Personality and Social Psychology 107, 751-764.
Kahneman, D., 2003. A perspective on judgment and choice: Mapping bounded rationality. American Psychologist 58, 697-720.
Kahneman, D., Tversky, A., 1979. Prospect Theory: An Analysis of Decisions under Risk. Econometrica 47, 263-291.
Lo, A. W., Repin, D. V., Steenbarger, B. N., 2005. Fear and Greed in Financial Markets: a Clinical Study of Day-Traders. National Burea of Economic Research(No. w11243). Long, S. J., 1997. Regression Models for Categorical and Limited Dependent Variables.
SAGE Publications.
Markowitz, H., 1952. Portfolio Selection. Journal of Finance 7, 77–91.
Mayfield, C., Perdue, G., Wooten, K., 2008. Investment management and personality type. Financial Services Review 17, 219-236.
Melville Wealth Management. (n.d.). Retrieved from
https://www.raymondjames.com/melvillewealthmanagement/pdfs/long_term_investm ent_performance.pdf
Neumann, J. v., Morgenstern, O., 1947. Theory of Games and Economic Behaviour. Princeton: Princeton University Press.
Nyhuis, E. K., Pons, E., 2005. The Effects of Personality on Earnings. Journal of Economic Psychology, 363-384.
Papke, L. E., Wooldridge, J. M., 1996. Econometric methods for fractional response variables with an application to 401(k) plan participation rates. Journal of Applied Econometrics 11, 619-632.
Robert, B. W., 2009. Back to the Future: Personality and Assessment and Personality Development. Journal of Research in Personality 43, 137–145.
Roberts, B. W., & DelVecchio, W. F., 2000. The Rank-Order Consistency of Personality Traits from Childhood to Old Age: A Quantitative Review of Longitudinal Studies. Psychological Bulletin 126, 3-25.
Roberts, B. W., Walton, K. E., Viechtbauer, W., 2006. Patterns of Mean-Level Change in Personality Traits. Psychological Bulletin 132, 1-25.
Sadi, R., Asl, H. G., Rostami, M. R., Gholipour, A., Gholipour, F., 2011. Behavioural Finance: The Explanation of Investors' Personality and Perceptual Biases Effects on Financial Decisions. International Journal of Economics and Finance 3.
Smith, A., 1776. An inquiry into the Nature and Causes of the Wealth of Nations. London. Terracciano, A., McCrae, R. R., Costa, P. J., 2010. Intra-individual change in personality and
age. Journal of Research in Personality 44, 31-37.
Uysal, S. D., & Pohlmeier, W., 2011. Unemployment duration and personality. Journal of Economic Psychology 32, 980-992.
Williams, R. (2015, February 20). Interaction effects between continuous variables. Retrieved from University of Notre Dame: https://www3.nd.edu/~rwilliam/stats2/l55.pdf
Appendix B: testing the assumptions of a Tobit regression model
The following snippet tests the assumptions of non-normality and heteroscedasticity for a Tobit model. First estimate the model in Stata and then run the commands. It assumes that all regressors are stored in a variable $xlist.
Student number: s2150018 Name: Remco Pander Study Program: M.Sc. Finance
Field keyword: personality traits, financial investments, the big “five” Supervisor: dr. Artem Tsvetkov
predict xb, xb matrix btobit = e(b)
scalar sigma = btobit[1, e(df_m)+2] generate threshold = (0-xb)/sigma
generate lambda = normalden(threshold)/normal(threshold) generate dy = total_financial_assets > 0
quietly generate uifdyeq1 = (log_total_financial_assets - xb)/sigma if dy == 1 quietly generate double gres1 = uifdyeq1
quietly replace gres1 = -lambda if dy == 0 summarize gres1
quietly generate double gres2 = uifdyeq1^2 - 1 quietly replace gres2 = -threshold*lambda if dy == 0 quietly generate double gres3 = uifdyeq1^3
quietly replace gres3 = -(2 + threshold^2)*lambda if dy == 0 quietly generate double gres4 = uifdyeq1^4 - 3
quietly replace gres4 = -(3*threshold + threshold^3)*lambda if dy == 0 foreach var in $xlist {
generate score`var' = gres1*`var' }
global scores score* gres1 gres2 generate one = 1
quietly regress one gres3 gres4 $scores, noconstant * Testing model non-normality
display "N R2 = " e(N)*e(r2) " with p-value = " chi2tail(2, e(N)*e(r2)) foreach var in $xlist {
generate score2`var' = gres2*`var
}
global scores2 score* score2* gres1 gres2 summarize $scores2
quietly regress one gres3 gres4 $scores2, noconstant * Testing model heteroscedasticity
Appendix C: regression results probit model of personality traits on
financial assets and unsecured debt
Has financial assets Has unsecured debt Coefficient (standard error) Coefficient (standard error)
Constant 1.35 (.52)*** -.58 (.64) Explanatory variables Extraversion -.11 (.060)* -.05 (.07) Friendliness .011 (.07) .11 (.08) Conscientiousness .10 (.06) -.05 (.07) Neuroticism -.06 (.05) .15 (.07)** Openness -.03 (.07) .22 (.09)** Control variables Male .11 (.07) .33 (.08)*** Age -.00 (.003)* -.03 (.00)*** Adults in household -.13 (.14) .15 (.17) Children in household -.04 (.03) -.07 (.04) Employed -.08 (.09) .26 (.11)** Self-Employed -.23 (.102)** .00 (.14) Partner -.014 (.16) -.25 (.20) Home owner .08 (.08) -.21 (.10)**
Household income: poor (10k – 14k)
-.00 (.22) .09 (.27)
Household income: lower-class (14-22k) -.04 (.20) .21 (.24) Household income: middle-class (22k – 40k) -.14 (.19) .19 (.22) Household income: upper-class (>40k) -.12 (.19) .10 (.23)
Education: high school .14 (.15) -.20 (.20)
Appendix D: robustness checks two-step Tobit model
Sample of individuals between 30 and 65Financial assets Unsecured debt Has financial assets Has unsecured debt Coefficient (standard error) Coefficient (standard error) Coefficient (standard error) Coefficient (standard error) Explanatory variables Extraversion -.06 (.08) .07 (.17) -.11 (.07) .00 (.08) Friendliness -.17 (.10)* -.03 (.19) -.02 (.08) .04 (.09) Conscientiousness .27 (.09)*** -.11 (.21) .16 (.09)* .02 (.09) Neuroticism .04 (.08) .04 (.17) -.02 (.07) .13 (.08)* Openness .21 (.10)** -.21 (.21) -.08 (.09) .21 (.11)** Model statistics Number of observations 1,421 436 1,648 1,648 Wald Chi^2 298.67 44.91*** 21.59 49.79*** R^2 (if applicable) 19,1% 0,9% Household Survey 2016 Financial assets
Appendix E: robustness check interaction effects
Sample of individuals between 30 and 65Financial assets Unsecured debt Coefficient (standard error) Coefficient (standard error) Personality traits (demeaned)
Extraversion -.05 (.08) .08 (.17) Friendliness -.17 (.10) -.01 (.20) Conscientiousness .28 (.09)*** -.05 (.24) Neuroticism .05 (.08) .04 (.17) Openness .21 (.10)** -.23 (.22) Interaction terms Openness * Friendliness -.11 (.16) -.00 (.30) Openness * Conscientiousness .24 (.17) -.27 (.39) Friendliness * Conscientiousness -.36 (.15)** .13 (.33) Model statistics Number of observations 1,421 436 Wald Chi^2 313.9 45.77 R^2 19.45% 9.45% Household Survey 2016
Financial assets Unsecured debt Coefficient (standard error) Coefficient (standard error) Personality traits (demeaned)
Appendix E: full model fractional response probit
Entire sample Only having risky assets Coefficient (standard error) Coefficient (standard error)
Constant -5.209 (.657)*** -.205 (.893) Explanatory variables Extraversion -.141 (.064)** -.264 (.078)*** Friendliness .101 (.070) .164 (.078)** Conscientiousness -.027 (.063) .034 (.081) Neuroticism .008 (.060) -.059 (.078) Openness .031 (.073) -.007 (.092) Control variables Male .146 (.077)* -.029 (.094) Age .006 (.003) .005 (.004) Adults in household .035 (.170) .209 (.276) Children in household .030 (.039) .163 (.058)*** Employed -.107 (.098) -.312 (.106)*** Self-Employed -.092 (.120) .204 (.150) Partner -.089 (.179) -.248 (.293) Home owner .223 (.093)** .235 (.107)**
Household income: poor (10k – 14k)
.071 (.262) .074 (.353)
Household income: lower-class (14-22k) -.179 (.249) -.112 (.338) Household income: middle-class (22k – 40k) -.169 (.237) -.158 (.322) Household income: upper-class (>40k) -.099 (.238) -.124 (.322) ln(Total financial assets) .259 (.022)*** -.049 (.032) Education: high school .321 (.234) .121 (.360)
