The interaction between cognitive and non-cognitive factors for individual financial decision-making
Théodore Wanner a, *, supervised by Dr. V. Angelini a
a
Faculty of Economics and Business, University of Groningen, The Netherlands
ARTICLE INFO ABSTRACT
JEL classifications:
Using data drawn from the Longitudinal Internet Studies for
the Social Sciences (LISS panel), I analyze the interaction of cog- nitive and non-cognitive factors influencing decision-making with regards to financial assets and debts as well as the participation on the stock markets. I use financial literacy to measure cognitive skills and the Big Five personality traits to measure non-cognitive skills. I find that high financial knowledge impacts the decisions of individuals differently, depending on their personality. The re- search shows that different personality types make different fi- nancial decisions depending on their financial literacy. Among others, especially Neuroticism seems to have a substantial impact on the effectiveness of financial knowledge, diminishing the im- pact of financial literacy if Neuroticism is high. Furthermore, the results suggest that basic financial literacy more strongly interacts with the personality traits, while the impact of advanced financial literacy is more independent.
January 2018 A20
C24 D03 D14 D83 G11
Keywords:
Big Five personality traits Cognitive factors
Non-cognitive factors Portfolio choice Risk preference Behavioral Finance Financial Sophistication
The author is especially grateful to Viola Angelini, whose guidance and input in many fruitful discus- sions have been exceptionally helpful.
* Corresponding author at: Faculty of Economics and Business, student number s2927020, Nettelbosje 2,
9747 AE Groningen, NL. E-mail address: t.wanner@student.rug.nl (T. Wanner).
1. Introduction
Scholars have found various factors that may influence the financial decision-making of households. Two of those factors, cognitive and non-cognitive abilities of individuals, have received increasing attention.
Cognitive abilities are regarded as more intellectual abilities, such as reading skills and simple numerical calculations (Christelis, Jappelli, & Padula, 2010). The cognitive psychology literature identifies four do- mains of ability: orientation, memory, executive function and language (Richards, 2004). They depend on genetic and environmental factors, and may vary (e.g., taught or forgotten) over time. Those cognitive factors play a vital role in the outcome of the financial decision-making of individuals. Von Gaudecker (2015) for example has shown that households with low financial numeracy have lower investment returns.
Also, financial literacy strongly impacts the asset and debt accumulation of individuals, leading to higher or lower retirement incomes (Lusardi, Michaud, & Mitchell, 2017).
Another stream of literature argues that so-called non-cognitive factors, such as personality traits, play a vital role in the financial decision-making as well (Brown & Taylor, 2014; Bucciol & Zarri, 2015). Traditional finance theory predicts that the investor’s willingness to take financial risks mostly depends on risk aver- sion and investment opportunities (Christelis, Jappelli, & Padula, 2010). Furthermore, personality traits route in the personality structure of the individuals and can influence the decision-making process through, e.g., the risk aversion of the individual (Bucciol & Zarri 2015). Therefore, financial literacy might manifest itself differently for different individuals, depending on their non-cognitive factors. In the past decade, literature focused increasingly on the non-cognitive aspects, first introduced in combination with econom- ics in 2001, focusing on non-cognitive abilities on social and labor outcome (Heckman, 2001).
Since both cognitive and non-cognitive factors seem to impact the attitude of individuals to borrow and save, it is hard to determine the relationship between financial literacy and personality traits and their predicted investment behavior without analyzing them in a single study. In fact, if personality traits corre- late with financial literacy, it is not easy to determine whether the individual saves more and borrows less because he is drawn to it (and therefore has its roots in the personality traits), or because she/he is finan- cially literate. As financial literacy and personality factors both influence the financial decision-making, the degree of personality traits and financial literacy, ranging from “low to high”, might not be completely independent of each other. They might influence each other’s scale. As an example: some scholars found that personality traits are useful predictors for the decision-making under ambiguity and risky situations (Epstein & Breton, 1993). Situations naturally entail more risk and ambiguity if the decision-maker has less information (i.e., knowledge / financial literacy). Taken that into account, financial literacy might manifest itself differently depending on the personality of an individual. In different words: An individual, moder- ated by her/his personality, uses financial knowledge differently to make healthier financial decisions. This implies that there is an underlying dependence so that financial knowledge might lead to different decisions dependent on the personality.
However, little is known about this interaction between the cognitive and non-cognitive factors, and there
has not yet been a study combining the two approaches. The aim of this thesis is to shed light on the
relationship between cognitive and non-cognitive factors and therefore to disentangle the role of cognitive
and non-cognitive factors in the individuals’ financial decision-making. This reveals more about the key
drivers of debt and asset accumulation. In this thesis, I analyze potential interactions between financial literacy and the personality traits of individuals, which contributes to the literature in two ways:
(1) The effects of cognitive and non-cognitive effects are netted out, and potentially misguided conclusions can be detected.
(2) Those two arrays of factors might interact with each other. To detect any potential interaction, one must analyze both on the same data and in the same study.
For the reasons listed above, this research could be interesting for policymakers, especially with regards to the use of financial literacy. With the help of finance theory and the Five-Factor theory (also Five-Factor Model, i.e., FFM, Big Five Theory or OCEAN Big Five), a highly accepted personality measurement model, rating individuals’ personality based on five distinct traits, I have formulated the research question. The missing empirical evidence of an interaction effect between cognitive and non-cognitive abilities leads to the following:
RQ: How does the personality influence the impact of financial literacy on the financial decision-making of individuals?
I analyze the interaction using two indices about basic and advanced financial literacy, and the personal- ity traits as deduced from the Longitudinal Internet Studies for the Social sciences (LISS) data panel, which con- sists of a representative sample of the Dutch population.
The remainder of this paper is structured as follows: section two provides an overview of the current literature and shows where this thesis aims to add to the existing literature. Section three describes the methodology and the reason why this methodology has been chosen. Section four is about the data col- lection and the measurement methods. Section five provides the analysis and the results. Section six dis- cusses results and findings, and the last section serves as a conclusion. Robustness tests are reported in the appendix.
I find that Openness to Experience is mostly unrelated to the accumulation of assets and debts, except for investments, in which high financial literacy leads to more investment on the stock markets if the Open- ness is high. Conscientiousness interacts negatively with financial literacy on investments and debt making.
Highly agreeable individuals are more likely to make debt if their financial literacy is high. Though of the individuals who save, the ones with low Neuroticism save more if their financial literacy is high. The results suggest that there is potential to customize financial training and education to increase its efficiency.
2. Literature Review
2.1. Financial Literacy
Financial literacy has a substantial impact on individuals’ investment behavior and investment perfor-
mance. The investment behavior of individuals then has a strong impact on society, especially with regards
to asset and debt accumulation, which make a large share of individuals’ retirement preparations
(Binswanger & Carman, 2012). Retirement preparation is therefore among others dependent on financial
knowledge (van Rooij, Alessie, & Lusardi, 2011a). Financially more literate individuals tend to save more
and borrow less and have a higher probability to engage in activities that support investments, such as
pension funds information fairs. Financial literacy can to some degree be taught, which is essential for regulations (Lusardi & Mitchell, 2006).
To give a better impression of what financial literacy is, I use the definition of Remund (Remund, 2010, p. 284): “Financial literacy is a measure of the degree to which one understands key financial concepts and possesses the ability and confidence to manage personal finances through appropriate short-term decision-making and sound, long-range financial planning, while mindful of life events and changing economic conditions.“
It has even been found that, in the US, 30-40 percent of the wealth inequality can be explained by financial knowledge (Lusardi, Michaud, & Mitchell, 2017). Individuals who have a low education are underrepre- sented in the financial markets, mostly not even having a transaction account (Hogarth, Anguelov, & Lee, 2005). Von Gaudecker (2015) studies the relationship between financial literacy, financial advice, and port- folio diversification, using the Dutch Household Panel Survey. He analyzes the impact of financial illiteracy on the loss that occurs to individuals due to under diversifying their portfolio. The results show that indi- viduals who have high financial literacy perform the best. Individuals with low financial literacy, but who seek advice from third parties, also have a decent return. The ones losing most are individuals with low financial literacy who do not seek for external help with their decision-making. Van Rooij, Lusardi and Alessie (2011b) estimate the relationship between financial literacy and wealth, using the Dutch Household Survey. They find that financial sophistication is linked to wealth and a higher probability to invest in stock markets. Jappelli & Padula (2013) present a theoretical intertemporal consumption model with which they also find that asset accumulation is linked to financial literacy. Thus, most the academic work has con- cluded that financial literacy is a key driver for a financially healthy society (Adams & Rau, 2011), while some scholars disagree with those statements (Willis, 2011). This research aims to answer the question which individuals respond most to financial knowledge, rather than does financial knowledge impact the financial decision- making.
2.2. The Five-Factor Model
The Five-Factor model came up in the late eighties and early nineties and found itself becoming more and more popular (Costa & McCrae, 1992; McCrae R. R., 1987). For an extensive review of the history and theory of the Big Five, consult De Raad (2000). Emergent from psychology, it has been used in a variety of different disciplines, among others to explain economic decision-making of individuals (e.g. (Almlund, Duckworth, Heckman, & Kautz, 2011; Borghans, Duckworth, Heckman, & ter Weel, 2008; Brown &
Taylor, 2014; Bucciol & Zarri, 2015; Donnelly & Howell, 2012; Nyhus & Webley, 2001), or to predict new venture success (Ciavarella, Buchholtz, Riordan, Gatewood, & Stokes, 2004). The model divides the per- sonality into five traits: Neuroticism, Openness to Experience, Conscientiousness, Extraversion and Agreeableness.
Neuroticism stands for emotional instability. It is one of the most widely accepted personality traits and therefore has before been described as one of the “Big-Two,” together with Extraversion (Wiggins, 1968).
It is striking that Neuroticism, unlike the other four personality traits, correlates with unfavorable charac-
teristics. Individuals who score high on Neuroticism tend to experience negative emotions, such as anxiety,
hostility, depression, impulsiveness, and vulnerability (Costa & McCrae, 1992). They can be described as
less confident and less relaxed (Zhao & Seibert, 2006). Neuroticism has also been found to be a relevant predictor of school attainment of children (Entwistle & Cunningham, 1968; Eysenck & Cookson, 1969), job satisfaction (Judge & Bono, 2001) and status in social groups (Anderson, John, Keltner, & Kring, 2001).
Openness to Experience (Openness, henceforth) has a long history with different names and has been popular for scholars to analyze. It has also been named Culture (Norman, 1963), Intelligence (Borgatta, 1964) and Imagination (Saucier, 1994). Throughout its history, Openness is associated with fantasy, intellectual curiosity, feelings, values, and ideas, that leads them to explore new ideas and try out new things (McCrae R. R., 1987). It has been linked to creativity (George & Zhou, 2001) and job performance (positive corre- lation) (Bing & Lounsbury, 2000), but negatively to salary (Seibert & Kraimer, 2001).
Conscientiousness is a personality trait where achievement (mostly work-related) is essential. Conscien- tious people tend to be more hardworking, persistent, competent and more motivated to pursue a goal in a consistent manner (Costa & McCrae, 1992). It has even been regarded as the ability to work hard (Barrick
& Mount, 1991).
Extraversion has, as already mentioned, Extraversion, been described as one of the Big-Two together with Neuroticism, due to its wide acceptance in the history of psychology (Wiggins, 1968). More extra- verted individuals tend to be more assertive, dominant, energetic, active, talkative and enthusiastic (Costa
& McCrae, 1992). They also tend to feel comfortable in larger groups, like spending time with other people and are cheerful and friendly. It has also been found to be positively related to the level of salary and promotions (Seibert & Kraimer, 2001).
Agreeableness is the personality trait with the briefest history. Agreeableness is a trait which is very con- cerned with interpersonal relationships. High Agreeableness can show itself in the desire to be a part of a larger commune (Wiggins, 1991). Agreeableness is focused more on other people, such as Extraversion, but then differently. Individuals scoring high on Agreeableness tend to be more trusting, altruistic, modest, forgiving, caring and unselfish (Almlund, Duckworth, Heckman, & Kautz, 2011; Costa & McCrae, 1992).
It is known that personality characteristics are important drivers for a high stock market participation (Luik
& Steinhardt, 2016). Prior research has mostly focused on the different settings in which personality traits
might influence decision-making. For example, Bucciol & Zarri (2015) analyzed the impact of different
personality traits on financial risk-taking. The authors use the US Health and Retirement Study (HRS)
from the years 2006-2012 and use the Big Five personality traits, and add other traits, such as cynical
hostility. They find that especially the “self-centered” personality traits, such as low Agreeableness and
high cynical hostility lead to higher financial risk taking. They use the share of risky assets of their total
financial wealth as a proxy for financial risk tolerance and found that Agreeableness has the largest influ-
ence on portfolio choice. Another study from Brown & Taylor (2014) also finds that the Big Five person-
ality traits have a significant impact on peoples’ financial decision-making. They analyze the influence of
personality traits on the amount of unsecured debt and financial assets held by households. They conclude
that Extraversion and Openness to Experience are strongly correlated with personal finances. Nyhus & Webley
(2001) assess the impact of personality traits on the savings and debts of individuals, using the Dutch
CentER Saving Survey (CSS) from 1996/1997, which provides them with detailed information on the
saving and borrowing behavior. They take the Big Five as well as another personality trait measurement, the 16PA (personality adjective), as developed by Brandstätter (1988). The 16PA personality trait model is comparable to the Big Five, but then with 16 different traits and therefore finer nuances between the traits.
They find that, from the Big Five, mostly emotional stability (low Neuroticism) and high Extraversion are stable predictors of the saving and borrowing behaviors.
2.3. The interaction of Financial Literacy and the Five-Factor Model
As shown in the previous sections, cognitive and non-cognitive factors have long played an essential role in research. However, their interactions have not been explored as thoroughly. There have been studies on the interaction of intelligence and personality, (e.g., Ackerman & Heggestad, 1997). Other authors studied the influence of personality on intelligence or vice versa (Furnham, Forde, & Cotter, 1998;
Saklofske & Zeidner, 1995).
Personality traits impact not only the learning ability of individuals but also the decision-making process.
This implies that individuals, based on their personality, react differently to uncertainty and risk- and return estimations. Moreover, financial literacy seems to provide individuals with the tools to know how they should make financial decisions to reach a healthy financial situation, while personality provides them (or not) the discipline and control to do so (Costa & McCrae, 1992). Since personality influences the way an individual uses available information, the information might be of lesser or higher importance, depending on personality. To date, we do not know whether the decision to accumulate assets and debts originates from knowledge, from the personal predisposition or a combination. This can have serious policy impli- cations (Bucciol & Zarri, 2015). Since, as discussed above, both personality traits, as well as financial liter- acy, have seem to have an enormous impact on the economic decision-making process of households, it is not yet clear which dominates and how they interact. This information can be of great value for the government, its schools as well as the individuals themselves. Both cognitive and non-cognitive factors seem to have a substantial impact, and finally disentangling and isolating their effect, as well as assessing their interaction might bring two different literature streams together.
2.4. Interaction Effects & Hypotheses
Neuroticism: Individuals who score high on Neuroticism tend to save less and borrow more, among others due to their impulsiveness in their buying behavior (Bucciol & Zarri, 2015; Brougham, Jacobs-Lawson, Hershey, & Trujillo, 2011; Mowen & Spears, 1999; Nyhus & Webley, 2001). They also have self-control problems, which is related to the accumulation of debt and low financial literacy (Gathergood, 2012). Financial literacy though helps people to understand the mechanisms of financial markets, and individuals who are more literate, tend to save more, borrow less and are also more likely to keep a part of their wealth in stocks (Brown & Taylor, 2014). However, high Neuroticism brings a lack of control and execution, which cannot be mitigated with financial knowledge. Therefore, I assume that knowledge, such as financial literacy, does not mitigate the self-control problems of Neuroticism.
Hypothesis 1: Individuals scoring high on Neuroticism save less and borrow more independently of their financial literacy.
Openness: Openness is connected to optimism. Optimistic individuals tend to overestimate re- turns from financial markets and therefore tend to have more unsecured debt than other individuals. Ad- ditionally, it has been found that they have a large tendency to accumulate debt, especially credit card debt (Brown & Taylor, 2014; Nyhus & Webley, 2001), but tend to be more risk-averse (Almlund, Duckworth, Heckman, & Kautz, 2011). Prior research has shown that financial literacy and education not only im- proves the accuracy of financial decisions made but increases the confidence to an even larger degree (Puri
& Robinson, 2007). Therefore, since individuals with an already optimistic personality, have high financial literacy, it might turn their optimism into over-optimism to a degree in which they find it less important to accumulate savings and limit debt (Puri & Robinson, 2007). Concluding, individuals who score high on Openness to Experience react to higher financial literacy, but in a non-optimal way. I therefore formulate my second hypotheses:
Hypothesis 2: Individuals scoring high on Openness to Experience, save less and borrow more if their financial literacy is high than if it is low.
Conscientiousness: Individuals scoring high on Conscientiousness, other than individuals scor- ing high on Neuroticism, tend to have less debt, irrespective of the kind of debt. They tend to buy less impulsively and plan their finance more through (Brown & Taylor, 2014; Brougham, Jacobs-Lawson, Her- shey, & Trujillo, 2011; Mowen & Spears, 1999). Introverts in general, tend to save more and borrow less (Nyhus & Webley, 2001). Individuals who score high on Conscientiousness, therefore, are more thorough with the investment decisions and their financial literacy has an only little impact on their financial deci- sions. Furthermore, financially more literate people are more likely to participate in the stock markets and therefore tend to take more risk. This implies that financial literacy and Conscientiousness could act as substitutes. Both resulting in individuals tending to have a proper financial plan, with less borrowing and more asset accumulation, and especially sound financial investment. Individuals scoring low on Conscien- tiousness though might benefit more from financial literacy since they do not have the natural propensity to plan their finances soundly. I deduce the following hypothesis:
Hypothesis 3: Individuals scoring low on Conscientiousness invest more in stocks if their financial literacy is high than if it is low.
Extraversion: Individuals scoring high on Extraversion tend to make more debt, especially over- draft debt, compared to less extraverted individuals. They also tend to have substantial investments such as stocks, which is one of the riskiest asset classes and can be connected to their impulsive lively style (Brown & Taylor, 2014). Almlund, Duckworth, Heckman, & Kautz (2011, p. 17) describe Extraversion as follows: “An orientation of one’s interests and energies toward the outer world of people and things rather than the inner world of subjective experience; characterized by positive affect and sociability.” Therefore, people who score high on Extraversion, and are financially literate, have an exceptionally high interest in financial topics. Extraver- sion opens them a way to be more sophisticated, experimental and knowledgeable about their investments.
Barrick & Mount (1991) found that extraverted people respond more effectively to training. This implies
that financial literacy improves the financial decision-making. I therefore assume that individuals who
score high on Extraversion, save more and borrow less, and invest more in risky assets.
Hypothesis 4a: Individuals scoring high on Extraversion, save more and borrow less if their financial literacy is high, than if it is low.
Hypothesis 4b: Individuals scoring high on Extraversion invest more in stocks if their financial literacy is high, than if it is low.
Agreeableness: All in all, more agreeable people tend to be less money focused than less agreeable individuals (S.E. & Kraimer, 2001). As found in multiple studies, they tend to save less and borrow more (Brown & Taylor, 2014; Nyhus & Webley, 2001). Additionally, individuals scoring high on Agreeableness tend to take less financial risks when investing their savings (Almlund, Duckworth, Heckman, & Kautz, 2011; Bucciol & Zarri, 2015). Agreeableness does not seem to be largely influenced by financial interests, and I therefore predict that financial literacy has no impact on the financial decision-making of individuals that score high on Agreeableness.
Hypothesis 5: Individuals scoring high on Agreeableness save less and borrow more independently of their financial literacy.
3. Methodology
A substantial fraction of the dataset I use contains 0-values for assets and debts. This implies that not all individuals in the study decided to make debt, assets or participate in the stock markets. To deal with those zero-values, a two-part model is applied. The first part consists of a Probit regression, and the second part of an ordinary least squares (OLS, henceforth) regression, conditional on asset/debt ownership.
Therefore, in a first step, I analyze whether financial literacy, personality, and their interactions impact the propensity of individuals to make an initial decision to accumulate assets (debts). In a second step, I analyze, only for individuals who have assets (debts) how it affects the accumulation of it.
For the first part of this two-part model, I have a binary dependent variable with the values 1 for owning assets (debts) and 0 for holding no assets (debts). I use the Probit model to analyze this relation- ship. Another possibility could be to apply a linear probability model (Hill, Griffiths, & Lim, 2011). In this type, a simple OLS regression is performed on the binary dependent variable. However, the linear proba- bility model has several shortcomings. For example, it can estimate values outside the [0,1] range, as it implies marginal effects of changes in continuous explanatory variables are constant. The Probit or Logit models overcome the shortcomings of the linear OLS model. Both are non-linear models. They transform the linear index that can range from [−∞,∞] into a binary variable that ranges over [0, 1]. The Probit model uses a cumulative distribution function (CDF) instead of a linear one. The logit model uses a logistic function.
3.1.1. The Two-part model
A two-step model is being applied, in which I first conduct a Probit analysis, which estimates the proba-
bility of an individual holding assets or debt, which is
P "
#= 1 & = P "
#∗> 0 & = F x
+β = - . /.
012
34
(1)
while F is the standard normal cumulative density function, and - is the standard normal density (Cameron & Trivedi). "
#stands for either assets 6
#of the individual or debts 7
#of the individual. The binary variable "
#is either 0 or 1, depending on the latent variable "
#∗.
"
#= 1 89 "
#∗> 0
0 89 "
#∗≤ 0 (2)
The estimation estimates if 7
#respectively 6
#is 1 or 0 based on the financial literacy. In the cases of this thesis, this means that:
6#= 1 89 6#∗= ; + =>?@ABCD8E" E?C8E@ #F + 98B. D8E>?CH" #I + HABE?AD@ #J + K#> 0
0 >D@>
(3)
7#= 1 89 7#∗= ; + =>?@ABCD8E" E?C8E@ #F + 98B. D8E>?CH" #I + HABE?AD@ #J + K# > 0
0 >D@>
(4)
Here, F, I CB/ J are the vectors of the associated parameters, and K
#is an error term.
After estimating F, I CB/ J using the Probit estimations above, I now estimate the parameters M in the second part. To do so, I conduct a linear regression (OLS) conditional on asset or debt ownership. This means that a normal regression is applied to all the values in assets (debts) which are > 0. Then, the estimation of the assets and debts held consists of both parts, the Probit part and the OLS part. The OLS regression looks as follows:
N#= ; + M#&#+ K# ∀ 6∗# > 0
(5)
P#= ; + M#&#+ K# ∀ 7#∗ > 0
(6)
While S stands for assets and B for debts. Because of the limited interpretability of the coefficients of the
Probit model, I display the average marginal effects (A.M.E.) for the Probit results without interaction. It
is incorrect to take the marginal effects of the interaction terms that are calculated by Stata since one
cannot alter the one variable without altering another variable as well (Frondel & Vance, 2011). In a second
step, I add the interaction effects and report them in graphs rather than the Stata regression output, to
increase the interpretability. For the estimations with an interaction term, I show the coefficients at differ-
ent values of the personality traits and different values for the financial literacy. The interaction effects are
between two continuous variables, namely the personality traits and the financial literacy. There are in total
ten interaction effects: The five personality traits with both basic and advanced financial literacy. In the
second part, the OLS regression, I also conduct the regression without the interaction term in a first step.
In a second step, I add the interaction terms. I only report the average marginal effects for non-interaction terms. The log of the monetary value of assets and debts are used.
3.1.2. Tobit model
Once I conducted the two-part model, I check it for robustness using the Tobit analysis with left-censored data on the value 0. The first part of the two-part model is supposed to return the propensity of an indi- vidual to hold assets or debts under certain circumstances, while the second part finds out that once an individual decides to hold assets or take on debt how those same circumstances influence the asset or debt holding.
The Tobit analysis is supposed to analyze both (McDonald & Moffitt, 1980). The Tobit model is a cen- sored normal regression and consists of two parts. The first part is the maximum-likelihood estimation.
The second part is the linear model for the part that is uncensored.
"
#= "
#∗89 "
#∗> 0
0 89 "
#∗≤ 0 (7)
where "
#∗is the latent variable for debts, assets and stock market participation. Here, in contrast to equa- tion (4), the continuous dependent variable only takes on 0 for the values of the latent variable of 0 or less (left-censoring at 0) and takes on the value of the latent dependent variable otherwise (of the €-amount of assets or debts).
4. Data collection & measurement methods
4.1. Data Requirement & Data Availability
This thesis requires data about people, specifying their personality traits (Big Five, i.e., Extraversion, Agree- ableness, Openness, Conscientiousness, and Neuroticism). Additionally, economic data about the debt accumulation, asset accumulation, risk preferences and stock market participation are needed. The Longi- tudinal Internet Studies for the Social sciences (LISS) data panel is a suitable candidate for this purpose. It is a Dutch online household survey administered by Statistics Netherlands. The survey has been conducted by the CentER data research institute. The panel is based on a true probability sample of households.
Households without a computer or internet connection are provided with it (CentERdata, 2017).
4.2. Measurement of Variables & Dataset
The LISS data panel consists of ten questionnaires, all covering a certain aspect of peoples’ lives. There are eleven core studies, namely panel 1) Background variables, 2) Health, 3) Religion and Ethnicity, 4) Social Integration and Leisure, 5) Family and Household, 6) Work and Schooling, 7) Personality, 8) Politics and Values, 9) Economic Situation: Assets, 10) Economic Situation: Income, 11) Economic Situation:
Housing. The LISS panel consists of more than 13000 individuals in 11500 households (CentERdata,
2017). No other data is necessary. The eleven questionnaires are provided in different datasets, which need
to be merged based on a unique indicator variable, which is the identification number of the person filling
in the questionnaire (nomem-encr). Next to the eleven core studies, there are different side questionnaires about more narrow topics. I need one of them, namely number 68 about financial literacy. Not every panel is conducted every year, hence I have some panels from different years. In this research, I use the ques- tionnaire 1 (version 6.0, from January 2012), the questionnaire 6 (Wave 5, version 1.0, from January 2013), the questionnaire 7 (Wave 6, version 1.0 from July 2013), the questionnaire 9 (LISS core study, Wave 3, version 1.0 from January 2013) and the questionnaire about financial literacy as mentioned above (version 1.0, conducted in March 2012).
All the datasets merged, the file consists of 16583 observations. Consequently, I drop the individuals who did not participate in all of the surveys. 5156 individuals filled in all the personality trait questions, and 4859 individuals who have a score for financial literacy. Combining the conditions, I end up with 3721 individuals who have a value for both. While doing the analyses, I am additionally dependent on data about assets, debts and their sub-categories. There are 5573 individuals in total answering the questions on assets and debts. 3142 do have assets and a personality score and a financial literacy score as well as values for all the background variables, providing the sample size for the final analysis.
4.2.1. Dependent variables
Six major dependent variables are used to answer the hypotheses: for each, assets, debts and investments a binary variable for the Probit part and a continuous variable on the amount for the OLS part. Other asset- and debt-classes (sub-categories) are also analyzed. The LISS data panel entails several questions to assess the total amount of assets and the total amount of debts. The LISS panel distinguishes between five different assets classes: The first assets class are the savings, which consist of banking accounts, savings accounts, term deposit accounts, savings and bonds or savings cert and bank saving schemes. The second one of single-premium insurance policy, life annuity insurance and endowment insurance (not linked to mortgages). The third category are investments (e.g. different types of funds, bonds, stocks, warrants etc.).
The fourth one is real estate other than the first home. The fifth category asks for other assets, such as motorcycles, boats and similar assets. A last question (ca12c009) asks whether the individual owns any of those (0=yes, 1=no), which is used as the dummy variable for assets, but in reversed order (1=yes, 0=no).
In follow-up questions, there are for each of the five categories two questions about the €-amount. The first is a direct question in which the individual can insert the amount. If the individual chooses the option
“I prefer not to say”, the observation is dropped. If the individual chooses the option “I don’t know”, another question asks with categorical values. In that case, I take the middle of the range (e.g. the category
€500-€1000 then equals €750 ((500+1000)/2). Again, in case the individual chooses the option “I prefer
not to say”, the observation is dropped. The same procedure counts for all the categories of assets and
debts. The total savings of an individual consists of the first category with the LISS code ca12c078. Debt
holding consists of the sum of its sub-categories: mortgages, student grants and loans. For the monetary
values discussed above, the natural log has been used to normalize the positive skewness. I add 1 to the
variable I take the log of, to retain the 0-values. Assets should be a strictly non-negative value and to deal
with the 38 negative values for assets, I take the inverse hyperbolic sine (Ihs, henceforth) as an alternative
transformation to the natural log. This transformation is appropriate for wealth because it deals with skew- ness, but also retains zero and negative values, and allows to explore sensitive changes in the distribution, and can avoid stacking and misinterpretation (Friedline, Masa, & Chowa, 2015). The IHS can be written as follows:
8ℎ@ 6 = log 6
#∗+ 6
#∗∗ 6
∗#+ 1 (8)
With 6
#∗being the €-value for assets for individual 8 . The same is being done for the deposit account.
It is important to note that this is done because it is possible to have negative assets, for example with a negative balance on a banking account. But the observations for the other asset categories as well as debt for the individuals who state that they do have assets in the binary question in the LISS, but then give 0 as the amount of assets they have, are dropped in the linear model, since they do neither belong to the category of having no assets nor to the category of having assets. Additionally to the continuous amount of assets and debts, I also create a dummy variable 6
#for asset holding, as well as 7
#for debt holding. Both variables take the value 1 if the individual owns assets (debts), and the value 0 if the individual does not own assets (debts). There are three categories for debt: the first one is loans, the second one study grant and the third one mortgages. While the continuous €-amount for the three categories is calculated in the same way as for assets, the dummy for mortgages is constructed differently. The reason is that the dummy for mort- gages in the LISS panel is conditional on real estate ownership, meaning that only individuals who indi- cated to have a real estate property answered the question on the mortgage. This makes sense, since one cannot have a mortgage without real estate. For the analyses though, I create a more complete picture by including all individuals of which I know that they have no mortgage, independently of them owning real estate. I therefore recode the mortgage dummy variable and add 0’s (no mortgage) for the individuals which have no real estate, increasing the number of observations for real estate from 356 to 5570, with 3.30% of the sample having a mortgage.
4.2.2. Sample size issues
As can be seen from the results on the sub-categories in the appendix, some of the regressions have unacceptably low sample sizes. For example, the Probit analysis on mortgages has only 93 observations (despite the adjustments as mentioned in the previous section). The results are not interpretable, yet I still report the results for completeness. The sample size comes from the fact that there were only 355 indi- viduals even have real estate, and potentially could have a mortgage. 184 of them have a mortgage, using the wave 3 of the LISS core study on assets (Streefkerk, 2017). 104 of them have a score for financial literacy and personality. Adding the control variables, I end up with 93 observations. The case is similar for other asset a debt sub-categories.
4.2.3. Independent variables
The independent variables are about financial literacy and personality traits. In the LISS panel, there is an
extra module called financial literacy, which contains four questions about financial literacy, concerning 1)
compound interest effect, 2) money illusion, 3) diversification effect and 4) bond prices. The four variables range from ew11a002 – ew11a005. To reduce the four variables to a more interpretable measure, I create an index from the questions. To derive an index measuring financial literacy, I follow the methodology of Lusardi and Mitchell (2010). I first create a dummy from the answers of the individual, equaling 1 if the answer is correct and 0 if it is not correct. Then, I create another dummy, equaling 1 if the individual states
“I don’t know”, rather than answering wrong. This is important because “I don’t know” is an answer that shows that the respondent does not understand the matter (Lusardi & Mitchell, 2010). The questions and possible answers of the questionnaire are shown in Appendix C – Financial Literacy.
As Lusardi & Mitchell (2010), I first perform a factor analysis on the four questions about financial literacy, to reduce the amount of data. I run a principal component analysis (PCA) on the four questions for 4859 individuals. The Kaiser-Meyer-Oklin (KMO) measure is 0.68 with all individual KMO measures of >0.6 (see Appendix C – Financial Literacy). The KMO measure is a measure of sampling adequacy, which can be regarded as “mediocre” if it is above 0.6. (Kaiser & Rice, 1974). The Barlett’s Test of Sphericity (p <
0.005) indicates that the data is factorizable (Kaiser & Rice, Little Jiffy, Mark Iv, 1974). Those tests provide the minimum standards to proceed to a Factor Analysis. I end up with two factors with eigenvalues above 1, explaining 38.09% and 19.92% of the total variance respectively, as shown in Table 1. An eigenvalue of above 1 can be regarded as high enough to keep the component (Kaiser, 1960). I therefore only take two factors. This two-factor solution explains 58.00% of the total variance. The first one loads largely on the first two questions about interest compounding and the money illusion as shown in Table 2.
The second factor loads largely on the second two questions, which are about diversification and bond prices. The first two questions can be regarded as more basic questions, whereas the second two are more sophisticated. Other than Lusardi and Mitchell (2010), I keep my two variables, and use them both as independent variables for my analyses. Lusardi and Mitchell (2010) now run factor analyses on both the basic and the advanced set, since they have more questions about financial literacy. I don’t follow their methodology in this part, and re-run one factor analysis with two factors, as discussed above. This is because Lusardi and Mitchell (2010) have more questions on both, basic (6 questions) and advanced fi- nancial literacy (8 questions), while the LISS panel only entails two questions each. In Table 2, the correct answers have a negative value, and the “I don’t know” answers have a positive value. For the regression analysis, I prefer positive values for correct answers and negative values for wrong or “I don’t know”
answers. Therefore, the variables can be regarded as financial illiteracy. I take the negative of the values,
and have “financial literacy”. I now reshape the index so that it ranges from 1 (lowest value) to 10 (highest
value).
Table 1: Components of the PCA. 8 Initial dummy variables are taken. Two components have an Eigenvalue above 1, which are then taken as the financial literacy indices. The number of observations are 4859. The data is from the LISS panel of 2012.
Component number Eigenvalue Variance explained Cumulative Variance Explained
1 3.0470 38,086 38,086
2 1.5930 19,917 58,004
3 0.9460 11,823 69,826
4 0.8370 10,469 80,295
5 0.5810 7,262 87,557
6 0.4190 5,239 92,796
7 0.3080 3,851 96,648
8 0.2680 3,352 100,000
Table 2: This table presents factor loadings on the financial literacy module of the LISS panel of 2012. The wording of the ques- tions are presented in Appendix C – Financial Literacy. All questions also have and ”I don’t know” option. Choosing this option is also included separately in the factor analysis. The Principal Component method is used to obtain these factor loadings. From the loadings of the factor analysis I conclude that the first two questions measure basic financial knowledge and the third and fourth question advanced financial literacy. The number of observations are 4859.
Factor 1 Factor 2
Question Factor Correct Don't know Correct Don't know
Compounding Basic -0.772 0.797 -0.017 0.041
Money illusion Basic -0.713 0.760 -0.228 0.219
Diversification Advanced -0.170 0.156 -0.773 0.806
Interest rates and bonds Advanced -0.046 0.091 -0.624 0.723
Personality traits: For the non-cognitive factors, I use the Big Five personality traits, one of the most common personality models in the past 20 years (McCrae & Costa, 2006). In the LISS data panel, the five personality traits are assessed in 50 personality items, 10 questions for each of the five traits. The high number of questions is done to increase the reliability of the outcome. The variable codes range from cp13f020 to cp13f069, while the first question measures Extraversion, the second Agreeableness, the third Conscientiousness, the fourth Neuroticism and the fifth Openness to Experience, and then repeating it- self. For more information, see Appendix A – Personality. All items are measured on a 5-Point Likert Scale, ranging from 1 “very inaccurate” to 5 “very accurate”. 18 out of the 50 questions are asked reversely.
It is important to note that the original questions were asked in Dutch, while the questions in Appendix A – Personality are only a translation of the questions. A precise translation of the traits is crucial for the respondent to give an accurate answer. In 2000, a study proved the generalizability of the questionnaire (Caprara, Barbaranelli, Bermúdez, Maslach, & Ruch, 2000). To see the questions in their original language, please consult the LISS panel (LISS Panel: CentER data, 2017). Despite that there is still an ongoing debate about whether personality traits stabilize over time or may change, I do not take a change into account, and assume for this study that they do not, or only marginally, change, following (Brown & Taylor, 2014;
Caspi, Roberts, & Shiner, 2005; Borghans, Duckworth, Heckman, & ter Weel, 2008). To get one variable
for each personality trait, I first invert the reversely formulated personality scores. The Cronbach’s alpha is used to control for internal consistency for each of the five personality traits (Cronbach alpha scores are for Openness to Experience: 0.767; Conscientiousness: 0.777; Extraversion: 0.875; Agreeableness: 0.819;
Neuroticism: 0.886;). The variables are consistent if the Cronbach alpha is at least 0.7 (Song, Podoynitsyna, van der Bij, & Halman, 2008). I then take the mean of the values for each personality type, leaving me with five variables. This yields comparable results to a factor analysis (Brown & Taylor, 2014).
4.2.4. Control variables
I take common demographic control variables, such as gender, age, household position, net income and education. Household position is a variable ranging from 1 “household head” to 7 “family member or broader”. I take the natural log of the €-value of net income +1, and drop the few initially negative values.
Education has been recoded into three dummies, representing low, mediate and high education levels,
which I derived from the variable oplmet, a categorical variable about the highest level of education with
diploma. The observations for “other”, “not yet completed” and “not yet started” have been dropped,
leaving the variable with 6 possible answers and 9623 observations. The first two (primary school & vmbo)
are recoded to the binary variable “low education”, the second two (vwo & mbo) “medium education”,
and the last two (hbo & wo) “high education”. As another control variable, I take the size of the household,
due to the additional income and cost factors that can potentially arise. I also take a dummy variable for
partner, with 1 meaning that the individual lives with a partner and 0 for not living with a partner (inde-
pendent of children). I have a categorical variable for the urbanity of the place of living, since that might
influence the holding of mortgages. The urbanity variables range from 1 “extremely urban” to 5 “not
urban”. Another control variables is schooling satisfaction, which might be affected by the motivation to
learn. Schooling satisfaction is coded in variable cw16i004 in the “work and schooling LISS panel”. The
score ranges from 0 “not at all satisfied” to 10 “fully satisfied”. Also, I create dummy variables for eco-
nomic degrees (variable cw12e016, economics, econometrics, finance, business administration or similar)
and mathematics degrees (cw12e018). This can have an impact on the ability to deal with the financial
literacy questions, especially the ones about compound interest and diversification. Lastly, I also use dum-
mies for being paid-employed, self-employed, student-status and disability-employed (dummy with 1 =
yes, 0 = no, created from the LISS variable cw16i525).
5. Analysis
In the following section, I report the results. I start with the descriptive statistics of all variables included.
Then I report on the two-part model. The Tobit analyses are reported in the Appendix E – Robustness Tests.
5.1. Descriptive Statistics
Table 3: Summary Statistics dependent variables. The data is from the LISS panel of January 2013. The variables are in € values respectively their natural logarithm including 0-values. For total debts and its logarithm as well as total loans and its logarithm, one outlier has been re- moved. The number of Observations is 3142.
Variable Mean Std. Dev. Min Max
Log Debts 0.9083 2.6922 0.0000 15.5084
Dummy Debts 0.1840 0.3875 0.0000 1.0000
Ihs Assets 8.0918 4.4290 0.0000 17.2191
Dummy Assets 0.9214 0.2692 0.0000 1.0000
Log Loans 0.6520 2.1398 0.0000 13.1224
Dummy Loans 0.1346 0.3414 0.0000 1.0000
Log Study Grant 0.2281 1.3949 0.0000 10.6690
Dummy Study Grant 0.0315 0.1747 0.0000 1.0000
Log Mortgages 0.2679 1.7306 0.0000 14.8088
Dummy Mortgages 0.0296 0.1695 0.0000 1.0000
Ihs Banking Account 6.2149 4.6612 0.0000 16.6048
Dummy Banking Account 0.8931 0.3091 0.0000 1.0000
Log Insurance 0.9971 2.9494 0.0000 16.5259
Dummy Insurance 0.1388 0.3458 0.0000 1.0000
Log Investment 1.1776 3.1749 0.0000 14.1520
Dummy Investment 0.1537 0.3607 0.0000 1.0000
Log Real Estate 0.5590 2.5221 0.0000 14.7318
Dummy Real Estate 0.0640 0.2447 0.0000 1.0000
Log Other Assets 5.2448 4.3200 0.0000 13.9978
Dummy Other Assets 0.6935 0.4611 0.0000 1.0000
Table 3 shows the summary statistics of the dependent variables. The mean probability of having debt is 18.40%, and the probability of having assets is 92.14%. Only 15.37% of the individuals have investments.
Almost everyone who has assets also has a banking account, and the amount on it varies largely, as can be
seen from the standard deviation which is the highest by far with 4.6612. There are 3142 observations for
both assets and debts.
Table 4: Summary Statistics independent variables. The data is from the LISS panel of 2012/2013. The column “Variables” indicates the names of the independent variables, including the control variables. The number of Observations is 3142.
Variable Mean Std. Dev. Min Max
Openness 3.4378 0.5022 1.5 5
Conscientiousness 3.7517 0.5162 1.7 5
Extraversion 3.2222 0.6587 1.1 5
Agreeableness 3.8550 0.5082 1.5 5
Neuroticism 2.4789 0.6966 1 5
Basic financial literacy 8.9234 1.6271 1 10
Advanced financial literacy 4.2649 2.4169 1 10
Gender 0.4771 0.4996 0 1
Age 52.9523 16.6157 16 92
Household position 1.7212 1.1700 1 7
Log net income 6.4497 2.2962 0 12.1
Low education dummy 0.2944 0.4558 0 1
Medium education dummy 0.3215 0.4671 0 1
High education dummy 0.3686 0.4825 0 1
Number of household members 2.4971 1.2744 1 8
Partner dummy 0.7406 0.4384 0 1
Urbanity of living 2.9987 1.2642 1 5
Schooling satisfaction 7.2546 1.6334 0 10
Math degree dummy 0.0433 0.2035 0 1
Economics degree dummy 0.1747 0.3798 0 1
Paid employment dummy 0.4246 0.4944 0 1
Self employment dummy 0.0356 0.1854 0 1
Student dummy 0.0627 0.2425 0 1
Disability employment dummy 0.0481 0.2139 0 1
Table 4 shows the summary statistics of the independent variables. The highest average personality trait is
Agreeableness with an average value of 3.86 out of 5. Neuroticism is the lowest with a score of 2.48. Basic
financial literacy is obviously much more spread with an average value of 8.92 out of 10. Advanced finan-
cial literacy is less spread with a mean score of 4.26 out of 10. The average respondent of the survey in the
sample is almost 53 years old. The schooling satisfaction among the respondents is rather high with a
mean score of 7.25 out of 10. 4.33% of the individuals have a degree in math, and more than 17.47% have
a degree in economics related studies. 42.46% of the individuals are in labor work, while only 3.56% are
self-employed, and 6.27% are still in school/university.
Table 5: Summary Statistics independent variables per debt & asset holder. The data is from the LISS panel of July 2013.
The sample size is 3142. A t-test has been applied for the difference in mean. * = p-value < 0.1, ** = p-value < 0.05, *** = p-value < 0.01.
Assets Debt
Yes No Difference Yes No Difference
Openness 3.5570 3.4109 -0.1461*** 3.4511 3.2811 -0.1699***
Conscientiousness 3.7174 3.7593 0.0419* 3.7586 3.6699 -0.0887***
Extraversion 3.2396 3.2182 -0.0214 3.2245 3.1946 -0.0299
Agreeableness 3.8397 3.8584 0.0187 3.8589 3.8095 -0.0494
Neuroticism 2.5138 2.4709 0.1954*** 2.4656 2.6334 .1677***
Literacy Basic 9.0158 8.2349 8.9650*** 8.9862 8.1867 -0.7995***
Literacy Advanced 4.3676 3.7880 -0.5795*** 4.3191 3.6288 -0.6903***
Table 5 shows the descriptive statistics of the personality scores for individuals holding assets and debt.
One can see here that individuals holding assets have a higher mean score for both categories of financial literacy than the average respondent. Debt holders though have a higher value in advanced financial liter- acy.
Table 6: Summary Statistics independent variables per debt & asset holder. The data of the personalities is from the LISS panel of July 2013. The Data of the financial literacy from May 2012. The sample size is 3142. A t-test has been conducted to test the significance of the means. * = p-value < 0.1, ** = p-value < 0.05, *** = p-value < 0.01.
Basic Literacy Advanced Literacy
Top 25% Bottom 25% Difference Top 25% Bottom 25% Difference
Openness 3.3597 3.3164 -0.0432* 3.5526 3.3379 -.2234***
Agreeableness 3.8838 3.8465 -0.0373 3.7729 3.7134 -0.0594**
Conscientiousness 3.1534 3.2480 0.0947** 3.2806 3.1783 0-.1022***
Extraversion 3.1431 3.2497 0.0166*** 3.8128 3.8698 .05686**
Neuroticism 2.5283 2.6514 0.1231** 3.8150 3.8819 0.0669***
Table 6 shows the difference in the mean personality trait per basic and advanced financial literacy, each
category split in the top and bottom 25% of the total respondents. The differences of the personality
scores between the financial more literate and the less literate seem remarkably small, with the largest
significant difference for both literacies in Openness, in which in both cases the top 25% of the respond-
ents with higher literacy also have higher Openness. The differences are significant for all differences but
Agreeableness in basic financial literacy.
Table 7: Correlation table between the personality traits and financial literacy. The Data is from the LISS panel for the year 2012/2013.
The personality traits have been determined with 10 questions per personality trait, each to be answered with on a Likert scale from 1
"very inaccurate" to 5 "very accurate". The individual personality scores have been transformed to one score, with all valid Cronbach alpha scores. The financial literacy scores have been deduced with a PCA from a 4-question questionnaire. A t-test has been conducted to test the significance of the means. * = p-value < 0.1, ** = p-value < 0.05, *** = p-value < 0.01.
Basic Literacy Advanced Literacy
Openness 0.1092*** 0.1776***
Conscientiousness 0.0827*** 0.0359
Extraversion -0.0200 0.0676***
Agreeableness 0.0371 -0.0640***
Neuroticism -0.1191*** -0.1074***
Table 7 shows the correlation between the five personality types and the two literacy indices. Basic finan- cial literacy correlates positively with Openness and Conscientiousness, and negatively with Neuroticism, all on a 1%-significance level. Advanced financial literacy correlates positively with Openness and Extra- version, but significantly negative with Agreeableness and Neuroticism.
5.2. Regressions
In the following section, I show the results of the regressions. The first part shows the analyses without interaction effect, for an impression of the single impact of the personality traits and the financial literacy on debt and asset holding and stock market participation. The results of assets, debts and the asset category investments are shown in the main analyses. Other sub-categories of assets and debt are found in Appen- dix D – Additional Analyses.
5.2.1. Results without interaction effects
Table 8 shows the average marginal effects with its associated standard errors of the Probit analysis on assets, its sub-category investments and on debts.
Openness, in line with prior literature, increases the probability to accumulate debt with 25.70% per unit
increase in Openness, significant on the 1%-level. Furthermore, it does not affect the propensity to hold
assets or invest in the stock markets. Conscientious individuals, in line with prior literature, are less likely
to accumulate debt, while not impacting the propensity to accumulate assets. Extraversion does not impact
the general asset nor debt holing, but negatively impacts the probability of individuals to invest. More
agreeable individuals are more likely to accumulate assets. The impact on general assets is with approxi-
mately 18.22% per unit increase in Agreeableness very large and is significant on a 5%-level. This means
that highly agreeable individuals tend to save more, which contradicts the findings of Brown and Taylor
(2014). Neuroticism negatively impacts the propensity of individuals to invest by around 14.69% per unit
increase in Neuroticism. The marginal effect is significant on the 1%-level. Basic financial literacy posi-
tively impacts both the holding of assets and investments, but not the holding of debt. The impact on
investments is with 15.20% more than twice as high as the impact on general assets, with 6.93%. Advanced
financial literacy has a positive but small impact on holding debt, and a large impact of 16.76% per unit
increase in advanced financial literacy, on investments.
The second part, the OLS model describes the impact of the personality traits on the amount of assets collected, once the individuals did so. The results are presented in Table 9. Openness has no impact on any of the three categories, assets, debts and investments. This means that the accumulation of assets, debts and invest- ments is independent on the Openness of an individual. Therefore, while Openness increases the proba- bility of an individual to make debts, it does not influence the amount. Conscientiousness has a significantly positive impact on assets, with a 26.45% change in assets per unit increase in Conscientiousness. Other than that, there is no significant effect. Extraverted individuals tend to accumulate much more debt. The impact is positive with a 60.35% increase per unit change of Extraversion, significant on the 10%-level.
As shown in the Probit model, Extraversion does not impact the tendency of individuals to take on general debt, but once they decided to do so, higher Extraversion leads to higher amounts of debts accumulated.
Agreeableness again has no significant impact on any of the categories. Neuroticism, as expected from the existing literature, negatively impacts the amount of assets held by 23.91% per unit increase in Neuroti- cism. Also shown in table 9, both basic and advanced financial literacy tend to have a positive impact on the amount of assets accumulated as well as on the investment amount. Especially basic financial literacy is striking with a 92.16% increase in the amount of investments held per unit increase in basic financial literacy, thereby exceeding the effect of advanced financial literacy with a value of 21.96%.
As of the sub-categories, it is striking that both Conscientiousness and Agreeableness have a huge impact
of over 80% on the insurance amount, as shown in Table 18 in the Appendix E. Also, Neuroticism has
an evenly large negative impact. Extraversion has a negative impact on the €-amount held on the banking
account. Both financial literacy measures have positive impacts on most of the asset categories. There are
no significant impacts on the debt values, as shown in table 20.
Table 8: First part of the two-part model. Probit analysis without interaction. The Data are from the LISS panel for the year 2012/2013. Average marginal effects reported throughout with its associated standard errors in parentheses. * = p-value < 0.1, ** = p-value < 0.05, *** = p-value < 0.01
Holding Assets Holding Debt Holding Investments
A.M.E. Std. Err. A.M.E. Std. Err. A.M.E. Std. Err.
Openness 0.1442 (0.0879) 0.2570*** (0.0643) 0.0212 (0.0728)
Conscientiousness 0.0455 (0.0777) -0.1053* (0.0572) 0.0897 (0.0652)
Extraversion -0.0809 (0.0632) -0.0508 (0.0460) -0.0993* (0.0525)
Agreeableness 0.1822** (0.0850) 0.0642 (0.0614) -0.0057 (0.0709)
Neuroticism -0.0610 (0.0564) 0.0658 (0.0421) -0.1469*** (0.0498)
Basic Financial Literacy 0.0693*** (0.0181) -0.0029 (0.0177) 0.1520*** (0.0368)
Advanced Financial Literacy 0.0227 (0.0161) 0.0304** (0.0122) 0.1676*** (0.0148)
Gender 0.1852** (0.0879) 0.1523** (0.0640) -0.0190 (0.0743)
Household Position -0.1284*** (0.0419) -0.0701** (0.0338) -0.0418 (0.0450)
Age -0.0085** (0.0039) -0.0129*** (0.0028) 0.0159*** (0.0033)
Log income 0.0860*** (0.0173) 0.0332** (0.0167) 0.0473** (0.0215)
Low education -0.2735 (0.2751) -0.0180 (0.2291) 0.1414 (0.3059)
Medium education 0.0034 (0.2785) -0.0435 (0.2292) 0.1732 (0.3044)
High education 0.0780 (0.2831) 0.0946 (0.2307) 0.5768* (0.3042)
Household Members -0.0756** (0.0368) -0.0630** (0.0286) 0.0529 (0.0327)
Urbanity 0.0445 (0.0287) -0.0138 (0.0216) -0.0154 (0.0245)
Paid employment -0.0237 (0.1096) 0.0072 (0.0810) 0.1578* (0.0928)
Self employment -0.2674 (0.2042) 0.1143 (0.1446) 0.4927*** (0.1499)
School employment 0.3750* (0.2053) 0.2136 (0.1598) 0.1911 (0.2424)
Disability Employment -0.2710* (0.1566) 0.0555 (0.1318) -0.0569 (0.1771)
Schooling satisfaction -0.0068 (0.0216) -0.0228 (0.0183) -0.0019 (0.0221)
Partner -0.0559 (0.0992) 0.0821 (0.0743) -0.0122 (0.0847)
Math degree 0.0399 (0.2260) -0.0224 (0.1269) 0.1075 (0.1332)
Economics degree 0.0950 (0.1042) -0.0123 (0.0708) 0.0545 (0.0760)
Observations 3142 3142 3142
Table 9: Second part of the two-part model without interaction: Data are from the LISS panel for the year 2012/2013. Estimates based on OLS regression conditional on ownership. Coefficients reported throughout with its associated standard errors in parentheses. * = p-value < 0.1, ** = p-value < 0.05, *** = p-value < 0.01.
