Endogenous preference: The effect of financial education on preferences

(1)

Endogenous preference: The effect of financial education on preferences

Suthinee Supanantaroek (s2048337)

Supervisor: Prof. dr. B.W. Robert Lensink

Research Master Thesis

September 6, 2012

Faculty of Economics and Business

University of Groningen

Abstract: I design a questionnaire consisting of a survey on personal information and family background, and experiments to attain information on risk, time and social preferences of young adults. The data is used to elicit preferences in order to find the effect of financial education on preferences. I find that financial education at least partially affects preferences of subjects. In addition, I estimate the discount rates conditional on risk aversions using MLE technique. The estimation results are significantly lower than that assuming risk neutrality and hence more economically sensible.

Key words: Financial education, preferences, young adults, expected utility function, exponential

(2)

1. Introduction

Traditional Neoclassical economics assumes that preferences are exogenously determined and taken as stable (Stigler and Becker, 1977). However, not all economic phenomena can be explained by neoclassical economics. For example, no study has considered how economic institutions (both private and public), society, environment, and market affect preferences. Endogeneity of preferences is a critical underlying feature that determines behavior, economic and financial decisions – and hence economic outcomes. As Becker and Mulligan (1997) conclude in their work, “[b]y endogenizing discount rates, it appears possible to explain with a model of rational behavior many assertions in the literature that are claimed to imply irrational choices.”

In the last decades, many economists have come to accept the view that preferences are not exogenous and started to develop models that may explain changes in preference. Modern economics takes into account the role of psychology (see e.g. Meyers-Levy and Peracchio, 1995), biology (see e.g. Eisenberg et al., 2007; and Carpenter et al., 2009), neuroscience (see e.g. Camerer et al., 2004, 2005), and environment and culture differences (see e.g. Bowles, 1998; Malmendier and Nagel, 2007; and Palacios-Huerta and Santos, 2002) in determining preferences. Models have been developed to test for endogeneity of preferences and how preferences are formed. There have been several studies on the various potential determinants of preferences and how preferences have explanatory power for economic behavior and real-life decisions. It has also been shown that intertemporal choice decisions are affected by preferences. Becker and Mulligan (1997), for example, find that time preference is related to different levels of patience among individuals and leads to differences in individual wealth and wealth among nations. In a related study, a person with relatively less patience prefers a sooner but smaller payoff to a later but larger payoff and saves less (Bettinger and Slonim, 2007) and vice versa. A common feature of the majority of the existing research is that the data are collected from adults and therefore the understanding of young population’s preferences is rather limited.

Financial literacy has recently received greater attention among researchers as it is one of the factors contributing to economic growth and development (World Bank, 2009). Several studies show that financial literacy has positive effects on savings behavior of adults (see e.g. Fry et al., 2008; Meier and Sprenger, 2008, 2009; and Schreiner and Sherraden, 2007). Other works demonstrate that financial education is beneficial to younger people (e.g. Sherraden et al., 2009). A number of studies also suggest a positive relationship between education attainment and preferences, e.g. Shefrin and Thaler (1992) and Bauer and Chytilová (2007) for time preferences, Oreopoulos and Salvanes (2009) for risk preferences, and Jakiela et al. (2010) for social preferences. Even though there is an increasing amount of research on the effect of financial literacy on preferences (see e.g. Meier and Sprenger, 2009), little is known about the impact of financial education, if any, on young adults’ preferences.

(3)

subjects are asked to make decisions on three choice tasks1 and provide personal information. The choices they make contain information used to elicit risk, time and social preferences. Socio-economic characteristics, cognitive abilities, saving behavior, family background and financial education information provided are incorporated to explain differences in preferences. The insights and knowledge gained from this paper will contribute to the endogenous preferences literature as well as to the emerging body of research on young adults’ economic behavior and preferences. The findings will also be relevant and useful for economists and policymakers in the design and implementation of effective educational policies aimed at influencing preferences, economic behavior and educational outcomes of young people.

Time preference responses elicited by choice experiments may be influenced by risk attitudes, which are caused by the curvature of utility function under the expected utility model (Frederick et al., 2002; and Andersen et al., 2008a). Therefore, it is necessary to include risk aversion in the discounted utility model for an accurate value of the discount rate, i.e. a measure of time preference. This approach of discount rate elicitation is widely used in previous studies and generates sensible economic outcomes. Andersen et al. (2006) suggests that if risk neutrality is assumed instead of allowing for risk aversion, the estimated discount rate is significantly biased upward. Empirical evidence in Andersen et al. (2011, 2008) and Coller and Williams (1999) also points to the same direction. One example of omitting risk aversion is Benhabib et al. (2010) in which the discount rate of 472% is presented. Hence, it is dangerous to infer individual discount rates without knowing or assuming something about risk attitude. In this study, I consider a CRRA (constant relative risk aversion) parameter from an expected utility function in conjunction with an exponential discounting function in order to specify the relationship between risk and time preferences. The estimated CRRA parameter is very close to the empirical evidence of laboratory experiments in the United States conducted by Holt and Laury (2002, 2005) and Harrison et al. (2005). In order to prove that risk preference has an impact on time preference of individuals, I estimate the discount rate allowing for risk aversion and show that this results in significantly lower discount rate than assuming linear utility function (i.e. risk neutrality). The discount rate of 5.6% per year conditional on the estimated CRRA parameter is obtained, representing an impact of risk preference on time preference.

The results of this paper indicate that financial education has an impact on risk and time preferences of young adults, as the discount rate of first year students is higher than that of second year students (9.5% compare to 6.0%) and master students have the lowest rate (4.8%). First year students are also relatively less risk averse. Surprisingly, financial education does not have any effect of social preference as it does not lead to pro-social behavior. Lastly, financial education is not the only determinant of preferences as socio-economic characteristics, cognitive abilities, saving behavior and family background are also found to jointly have an impact on preferences.

1

For a survey of experimental studies, see Frederick et al. (2002).

2

I distinguish the (indirect) evolutionary selection mechanism from the genetic transmission.

3

See e.g. Lusardi and Mitchell (2007) and Bernake (2006) for surveys.

(4)

In section 2, I go through the existing literature on several determinants of preferences. In section 3, I provide support for the importance of education as well as financial education for preferences determination. In section 4, I develop hypotheses for this research paper. In sections 5 and 6, I cover the methodology and describe the data on the subject pool, experiments and summary statistics. In section 7, I cover statistical specifications. Results are presented in section 8. Section 9 deals with a robustness test. Finally, section 10 concludes and discusses possible extensions.

2. Literature survey

Long since the discounted utility model proposed by Paul Samuelson (1937) have (time) preferences been assumed to be exogenously determined and taken as given (Becker and Mulligan, 1994). The majority of economists take the neoclassical view in which three fundamental assumptions are held: people have rational preferences; individuals maximize utility and firms maximize profits; and people act independently on the basis of full and relevant information. These assumptions provide ease and a solid foundation for economists to predict and explain individual economic responses (Palacios-Huerta and Santos, 2002). Friedman (1962) explains that preferences are not in fact stable but should be taken as given by psychologists, sociologists, anthropologists and biologists. However, in recent times, economists have come to focus on relationships between preferences and several factors along with other economic variables.

From the last decade onwards, there have been several studies on various potential determinants of preferences and how preferences have explanatory power for economic behavior and real-life decisions

.

There are six core determinants categorized based on diverse arguments among key literatures. The first is the intergenerational cultural transmission, which is claimed to define time, risk and social preferences. The second is the (indirect) evolutionary selection mechanism2. The third and fourth determinants are the environment and genes, respectively. It should also be noted here that some economists claim that the interaction between the two may change the genetic effect on preferences. The fifth determinant is the function of the brain as explained by neuroeconomics. The last determinant is the exposure to conflict or war of individuals.

Determinants of preferences

Intergenerational cultural transmission

The originators of the concept of intergenerational cultural transmission are Cavalli-Sforza and Feldman (1981), authors of the groundbreaking book: Cultural Transmission and Evolution. They study the cultural transmission and evolution and compare it to the genetic transmission and evolution. Cavalli-Sforza and Feldman classify a transmission into three categories: vertical, horizontal, and oblique transmission. A vertical transmission takes place between offspring and parents. A horizontal transmission, which can be thought of as a peer interaction, is where an individual learns from another individual in the same generation. The last type, an oblique transmission, is one where an individual

2

(5)

learns from some members of an earlier generation (Boyd and Richerson, 1985 and Cavalli-Sforza and Feldman, 1981). In terms of economic models, the intergenerational transmission can be classified into two categories: imperfect empathy (paternalistic altruism/paternalistic model) and perfect empathy (non-paternalistic model or utilitarian model). The former is the idea that altruistic parents are likely to prefer children with their own cultural traits and attempt to shape them to these traits. Perfect empathy illustrates the parents who evaluate their children’s welfares based on their children’s preferences. Parents are willing to invest in order to maximize their children’s utilities.

The intergenerational cultural transmission concept plays an important role in determining preference traits such as time, risk and social preferences even though there is only limited evidence in details to be found (Bisin and Verdier, 2000, 2008 and Cronqvist and Siegel, 2010). Knowles and Postlewaite (2005) provide evidence from their theoretical model that parental attitudes are related to children’s savings behavior. Dohmen et al. (2008) also observe a significant positive intergenerational correlation in terms of risk preferences and trust attitudes between parents and children. In other words, parents who are less risk averse and have more trust in other people raise their children to have similar traits compared to their own. Moreover, from their data, parents who are more similar to each other in terms of risk preferences have larger impact on risk preferences of their children. According to Bowles and Polanía-Reyes (2011), social preferences refer to motives such as altruism, reciprocity, intrinsic pleasure in helping others, inequity aversion, ethical commitments and other motives that induce people to behave more pro-socially (helping others) than would an own-material-payoff maximizing individual. Bisin et al. (2004) study the intergenerational cultural transmission of preferences for cooperation following a model of Bisin and Verdier (2001). They propose that parents spend time and invest resources on their children to socialize them to their preferred social norms, either cooperative norms or competitive individualistic norms.

(Indirect) Evolutionary selection mechanism

Dekel et al. (2007) propose a model of the indirect evolutionary selection approach to study general endogenous preferences. They suggest that behaviors caused by preferences determine success and preferences that yield success and maximize fitness tend to be popularized and become common in society. The formal model of the indirect evolutionary selection approach, however, is originally proposed by Güth and Yaari (1992) and Güth (1995). They assume that preferences are predetermined and change according to the evolutionary selection process.

(6)

there exists homogeneity of the preferences distribution in society whereas the long run distribution of preference traits in the intergenerational cultural transmission approach is heterogeneous. They further state that the (indirect) evolutionary selection mechanism approach omits socialization within a family and in society where an individual lives, which leads to the aforementioned heterogeneity in preferences.

Environment and genes

The view that preferences are endogenous to the environment is suggested and popularized by Bowles (1998). By conducting a survey on historical, anthropological, social psychological and other data, Bowles (1998) proposes that economic institutions may both directly (via, for example, the evolution of norms and incentives and constraints associated with institutional conditions) and indirectly (for example through institutional structures) affect preferences. Malmendier and Nagel (2007) take a macroeconomic shock as a specific environment that individuals go through and form risk preferences on financial investments accordingly. Palacios-Huerta and Santos (2002) study the role of markets in determining preferences and develop a model to explain risk preferences as a function of market risks, market incompleteness and non-market uncertainties. Carpenter (2003) also focuses on the extent of markets. He conducts experiments to test whether different aspects of markets affect social preferences and finds that, if markets are large and anonymous, individuals’ social preferences are likely to weaken as time passes.

The role of family environment (nurture) is also important in determining preferences. In this research paper, I distinguish the family environment from the intergenerational cultural transmission. The family environment here refers to a non-shared environment among individuals. Individuals who were brought up in different environments classified by different families and different treatments and rearing are different in terms of preferences and behaviors. However, there is an argument that the variance of preferences among individuals is due to (at least partially) nature (genes) rather than nurture. The argument is supported by confirming evidence, such as Eisenberg et al. (2007) and Carpenter et al. (2009) for time preferences and Zhong et al. (2009), Kuhnen and Chiao (2009), Barnea et al. (2010) and Cesarini et al. (2010) for risk preferences. Furthermore, a growing number of economic theorists’ interests are also devoted to this view. Netzer (2009), Robson and Samuelson (2009), and Brennan and Lo (2009) all propose that existing preferences are the result of natural selection. For this to be the case, preferences must be at least partially genetically determined (Cronqvist and Siegel, 2010).

(7)

genetic effects may determine preferences via environmental effects (nature via nurture). Their theory is that genotypes may either induce environmental responses or cause an individual to select a particular environment endogenously.

Role of neuroeconomics /neuroscience

From the last decade, neuroscience has increasingly played a role in economics with an aim to explain puzzles in economic decisions, giving risk to the so called neuroeconomics. As strong advocates of the field, Camerer et al. (2004, 2005) emphasize that neuroeconomics, by incorporating technique used in neuroscience, clarifies inconsistency and irrationality of individual behavior, which are two of the key issues in traditional economics. Neuroeconomics makes use of the possibility of the direct measurement of thoughts and feelings as well as the knowledge of how the brain works to form a better understanding of how economic decisions are made. Experiments in neuroeconomics apply a combination of brain imaging or stimulation experiments developed in the cognitive neurosciences and microeconomics methodology or game theory experiments developed in the economic sciences (McCabe, 2008). In addition, neuroeconomics, with the aim to improve an understanding on individuals’ objectives, emphasizes the physiological and psychological processes of decision making. The objective is to relate the decision making process to the physiological process in the brain or to descriptions of emotional experiences (Gul and Pesendorfer, 2005) Several researchers therefore claim that neuroecnomics enables economists to clarify certain puzzles in economic research, namely intertemporal choices, decisions under risk and uncertainty, and social decision making.

Parochial altruism

Charles Darwin (1873) saw wars and between-group conflicts as dominant evolutionary forces for social solidarity and altruism among within-group members to exist. Bowles and Choi (2004), Hammond and Axelrod (2006), Choi and Bowles (2007), Bowles (2008, 2009) and Lehmann and Feldman (2008), to name a few, attempt to clarify traits evolution based on parochial altruism by taking the view of Charles Darwin. Samuel Bowles is the most prominent proponent of the role of war and the exposure to conflict as influential features for the evolution of traits. Bowles (2006) investigates whether between-group competition can lead to the evolution of altruism. His empirical analysis indicates that altruism may be the result of the gene-culture co-evolutionary process that arises from a group conflict. Bernhard et al. (2006b) also support co-evolution of the two. They take the social norms approach and suggest that, as altruistic norm compliance and norm enforcement often arise in the environment of between-group conflict, they are likely to be formed by parochialism. Bowles and Choi (2004) suggest that altruism and parochialism co-evolve and each provides an environment favoring the evolutionary success of the other. Furthermore, they emphasize that each, if operates separately, cannot explain propagation of humankind under the exposure to warfare such as that among ancestral humans.

(8)

behavior. Their main finding is that shocks, even temporary, have long-term consequences, which can be permanent if there are shifts in preferences. They find that the exposure to violence affects preferences: Individuals who have experienced violence are more altruistic towards members in their society, more risk-taking, and more impatient.

3. Role of education

Role of education on preferences

In the last decades, the role of education as a determining factor of economic behavior, pecuniary and nonpecuniary returns and preferences has been the focus of many researchers. Friedman (1962) suggests that education may have both economic and non-economic benefits in addition to its effect on labor market payoffs. Oreopoulos and Salvanes (2011) show empirical evidence that education generates not only pecuniary return but also other economic and non-pecuniary benefits. Education may also affect many aspects of an individual’s life both in and outside the labor market. For instance, education can affect the degree to which an individual enjoys working as well as lead to better decision making in the long-term in terms of health, marriage and parenting style. Yet, there are many studies focusing solely on the return on education based on income generated. Henderson et al. (2011) argue that the rate of return on education is heterogeneous. Their empirical results support this argument and indicate that on average blacks have higher returns on education than whites, natives have higher returns than immigrants and younger workers have higher returns than older workers. In addition, they find a significant heterogeneity in the rate of return within groups. However, when consider non-pecuniary effects of education, one must separate the effect of education taken alone from the effect of higher income generated from education. This is because as higher education leads to higher income, higher income will affect an individual’s life. In addition to private returns, education also generates social returns. Such educational spillovers can be seen differently. Lucas (1988) and Jovanovic and Rob (1989) explain that positive spillovers are derived from the sharing of knowledge and skills through formal and informal interactions across workers. Acemoglu (1996, 1998) argue that spillovers from education may arise through search externalities or endogenous skill-biased technical change. All in all, the evidence found in Moretti (2002) confirms the spillover effect of education.

(9)

Along the same line with Shefrin and Thaler (1992), Oreopoulos and Salvanes (2009) mention in their review that “Some suggest schooling improves patience, making individuals more goal-oriented and less likely to engage in risky behavior.” Furthermore, Becker and Mulligan (1997) explain how education may affect individuals’ time preferences, “Schooling focuses students’ attention on the future. Schooling can communicate images of the situations and difficulties of adult life, which are the future of childhood and adolescence. In addition, through repeated practice at problem solving, schooling helps children learn the art of scenario simulation. Thus, educated people should be more productive at reducing the remoteness of future pleasures.” (p. 735-736). In terms of empirical support, Bauer and Chytilová (2007), after controlling for age, sex, income, marital status, and clan linkage and addressing possible endogeneity problem, present a significant causal effect of education on time preference. Perez-Arce (2011) also draws a similar conclusion from his study, indicating that education has a causal effect on children’s time preferences. On the other hand, Bettinger and Slonim (2006) do not find a significant effect on children’s patience.

Sutter et al. (2010) show that mathematics grade used as a measure for cognitive abilities gained from education is positively correlated with the level of patience of children and adolescents. This finding is consistent with Benjamin et al. (2006), Steinberg et al. (2009) and Castillo et al. (2011). In addition, they find a significant correlation between risk and time preferences, i.e. individuals with more risk aversion are more patient. The same finding is also reported in Keren and Roelofsma (1995). They further investigate the relation between time and risk preferences and field behavior; their results are similar to those for adults (see e.g. Chabris et al., 2008). The less patient an individual is, the more likely the individual smokes, drinks alcohol and has a higher BMI.

Role of financial education on preferences

The Organization for Economic Cooperation and Development (OECD, 2005) terms financial education as “[t]he process by which financial consumers/investors improve their understanding of financial products and concepts and, through information, instruction, and/or objective advice, develop the skills and confidence to become more aware of financial risks and opportunities to make informed choices, to know where to go for help, and to take other effective actions to improve their financial well-being.” Financial education is aimed at increasing financial literacy which is defined as “[t]he ability to use knowledge and skills to manage financial resources effectively for a lifetime of financial well-being” (PACFL, 2008). Financial education is considered highly important, especially at an early stage in life. The OECD addressed in the 2006 policy brief that “[f]inancial education should start at school and should be clearly distinguished from commercial advice.” Financially literate people make fewer mistakes and are in better financial condition than financial illiterates3 (Meier and Sprenger, 2009). In addition, previous studies have shown that financial literacy has important implications on financial decision making and financial behavior. Positive relationship between financial knowledge and financial behavior is documented (see e.g. Hilgerth et al., 2003). Calvert et al. (2005) find that more financially sophisticated individuals are more likely to buy risky assets and invest more efficiently. Along the same line, Kimball and Shumway (2006) provide evidence for a positive association between financial sophistication and

3

(10)

portfolio choice. People with low financial literacy are more likely to be indebted (Lusardi and Tufano, 2009), less likely to invest in stock markets (van Rooij et al., 2007), less likely to choose mutual funds with lower fees (Hastings and Tejeda-Ashton, 2008), less likely to accumulate wealth and manage wealth effectively (Stango and Zinman, 2009; and Hilgert et al., 2003) and less likely to plan for retirement (Lusardi and Mitchell, 2006, 2007a, 2009)4.

Studies are conducted to examine the impact of financial education on young people (see e.g. Danes and Haberman, 2007; Mandell, 2006a,b, , 2008a,b, 2009; Mandell and Klein, 2007; Peng et al., 2007; Valentine and Khayum, 2005; and Varcoe et al., 2005). For the U.S. sample, Danes et al. (1999) evaluate the National Endowment for Financial Education’s High School Financial Planning Program (HSFPP) in 1997-1998 and report increase in knowledge, self-efficacy and savings rates. Lusardi and Mitchell (2007) empirically investigate and show that individuals who were financially educated when young are more likely to plan for retirement. Carlin and Robinson (2012) conclude from their study that students who participate in financial education program have higher savings rates, pay off debt faster and spend considerably less on entertainment and dining out. Agarwal et al. (2007) show that it is very common for young people – a population group that rarely has financial knowledge – to make financial mistakes. Lusardi et al. (2010) found that financial literacy is severely lacking among young adults, especially in women. Differences between genders exist even though differences in demographic, family background and peer characteristics are taken into account. However, the literature on financial education to young people is relatively limited, and the effect of financial educational programs on young adults is understudied.

4. Hypotheses development

Recently, financial literacy has been a focus of several researchers as it is one of the factors contributing to economic growth and development (World Bank, 2009). A number of studies on adults show that financial knowledge positively affects saving behavior (Fry et al., 2008; Meier and Sprenger, 2008, 2009; and Schreiner and Sherraden, 2007). In this study, I empirically examine whether financial education has an impact on preferences of adolescents. The main presumption is that being financially educated at an earlier stage in life is more beneficial. The results of this study are especially relevant given the fact that, increasingly, individuals have to make important and complicated economic and financial decisions at an earlier stage in life. This argument is supported by recent empirical evidence. First, financially literate individuals are found to be able to make better decisions for present and future life events (Lusardi and Mitchell, 2006, 2007a, 2008, 2009; and Meier and Sprenger, 2008). Furthermore, there is a relationship between financial education and economic behavior, as Lusardi and Mitchell (2006b) find that financial literacy independently determines the ability of individuals to plan for retirement. In their related works, Lusardi (1999) and Lusardi and Mitchell (2006a) indicate that an absence of retirement planning is identical to an absence of saving. Lusardi and Mitchell (2007c) prove that financial knowledge is strongly related to economic knowledge. Other researchers propose that financial knowledge and financial behavior are positively correlated (see e.g. Hilgerth et al., 2003;

4

(11)

Calvert et al., 2005; and Kimball and Shumway, 2006). Their empirical evidence indicates that financial literacy enables individuals to make more efficient financial decisions.

Previous studies have shown the benefit of education on young people (e.g. Sherraden et al., 2009). A number of works also suggest a positive relationship between education attainment and preferences, e.g. Shefrin and Thaler (1992) and Bauer and Chytilová (2007) for time preferences, Oreopoulos and Salvanes (2009) for risk preferences, and Jakiela et al. (2010) for social preferences. In recent years, there has been greater interest in the relationship between financial education and preferences. It can be logically expected that financially literate individuals are more likely to be patient and more willing to postpone consumption to the future and therefore have lower discount rate. They tend to proportionally save more than people with comparable socio-economic characteristics and family background, as they gain more knowledge about finance and incorporate financial concepts learned in their economic and financial decision makings. Also, it is expected that patient individuals have better financial outcomes, make an earlier plan for retirement (Fang and Silverman, 2009), have higher credit scores and are less likely to default on loans (Meier and Sprenger, 2008). In terms of risk preferences, individuals with financial knowledge are more likely to be risk-taking and thus more likely to invest in risky assets, such as stocks and more sophisticated financial instruments. This is so because they are equipped with understanding about inflation, concepts of interest rate (time value of money) and risk diversification. Benjamin et al. (2006) and Dohmen et al. (2007) indicate that knowledge and cognitive ability may have an effect on risk and time preferences and hence on financial decision making. Financial education may have a positive effect on trust and trustworthiness, which represent social preferences. People with financial knowledge are expected to act more pro-socially to others than would be an own-material-payoff maximizing. The underlying reason might be that financial literate individuals are more able to understand the economic well-being inequality among people and the concept of social welfare. They, therefore, take these into consideration and act accordingly. Even though there is an increasing amount of research on the effect of financial literacy on preferences (see e.g. Meier and Sprenger, 2009), little is known about the impact of financial education, if any, on young adults’ preferences.

In summary, I aim to investigate the following:

Hypothesis 1: Financial education leads to higher preferences towards risks among young adults. Hypothesis 2: Financial education leads to lower discount rates among young adults.

Hypothesis 3: Financial education leads to more pro-social behavior among young adults.

(12)

policies to alter them. Their finding indicates that risk preferences vary with the age range. Supporting empirical evidence is shown in Dohmen et al. (2005) in the form of a negative relationship between age (as well as gender) and willingness to take risks. However, Sutter et al. (2010) do not find any correlation between risk preference and age.

In terms of the consequences of risk preferences on economic behavior, some scholars (see e.g. Brown and Taylor, 2005, 2007; Shaw, 1996; and Brunello, 2002) provide evidence that risk preference has a significant influence on human capital accumulation. Their empirical findings show that the level of educational attainment is negatively correlated with the degree of risk aversion. As the level of education determines earnings, leaving risk preference out of the picture would distort an estimate of the return on education. Apart from this, time preference is proved to be influential on economic outcomes as well. Castillo et al. (2011) find that educational outcomes partially depend on how much the future is valued. On the other hand, cognitive ability, which determines educational success, socio-economic background and risk aversion are not able to explain varieties in time preference across gender and nationality. Their conclusion contradicts that of Dohmen (2010), which states that the low level of cognitive ability is significantly correlated with risk aversion and impatience.

With regards to previous studies on endogenous preferences, there are contradicting empirical results on the relationship between risk and time preferences. The first stream of research supports no relationship between risk and time preferences, as mentioned above. The second stream, however, argues that risk and time preferences correlate. Sutter et al. (2010) present evidence on a significant relation between the two, with relatively more risk averse subjects being more patient. The same result is shown by van Praag and Booij (2003) with a correlation of about -0.35 between a CRRA parameter and discount rate. As both risk and time preferences affect individual behavior, it can be rationally expected that risk and time preferences are correlated and must be jointly considered when trying to analyze how economic and financial decisions are made. Frequently, individuals must make decisions in circumstances where outcomes are not foreseeable. For instance, investment is always accompanied with uncertainty as the return on investment cannot be a priori realized. Omitting risk attitudes may result in biased discount rates measuring time preferences of individuals (see e.g. Andersen et al., 2008 for empirical evidence).

My final hypothesis is thus that:

Hypothesis 4: Risk preferences affect time preferences of individuals.

(13)

5. Methodology

Risk preference

To test hypothesis 1 – that financial education leads to higher preferences towards risk of young adults – risk preference is estimated at group level (by the 3 groups of university students) classified according to the level of financial education. The first group consists of students who have no knowledge in finance; the second group those with basic financial knowledge; and the third group students with more advanced financial knowledge. Details on the sample set are discussed in the next section. The risk preference of each group is estimated and compared. It should be noted that I solely compare different estimated risk preferences among the 3 groups of students who have different levels of financial education without doing additional statistical test. The underlying reason for this is that the estimated risk preference is just a constant term. It is basically based on choices made in a binary choice list task for risk preference in a questionnaire by all subjects in each group. The estimated risk preference, therefore, reflects risk preference of each group. The methodology employed here allows for an investigation of the effect of financial education on risk preference. If the estimated risk preferences among groups differ, it purely reflects the impact of financial education on risk preferences of young adults. Hypothesis 1 is proved to be true if a group of master students has relatively higher estimated risk preference than a group of second year students, and a group of first year students has the lowest among all.

Risk preference: Measuring CRRA parameter (risk aversion)

The estimation is based on data obtained from an experiment for risk preference in a questionnaire (see figure 1 below). Subjects are asked to make choices between risky options and sure options as of today. All these choices contain information about utility, which could then be used to identify his/her utility function. However, an individual utility function is unobserved and hence is a latent function. Each of the chosen options from a binary choice list of all subjects in each group is taken as the data and analyzed using a CRRA utility function5. These data then transformed into the expected utility functions of risky choices (if subjects choose option A) and of sure choices (if subjects choose option B) based on the expected utility theory. These two expected utility functions are then linked to each other to construct the latent index, which is in the form of a cumulative probability distribution function allowing for some errors from the perspective of the deterministic EU theory model by including a structural noise parameter and estimated using maximum likelihood to infer the probability of the observed choices. The estimated risk preference is a CRRA parameter at group level. Details on its statistical specification are discussed below.

5

(14)

Time preference

Hypothesis 2 – financial education leads to lower discount rates among young adults – is tested by the same methodology. Time preference, conditional on the CRRA parameters6 estimated above, is also estimated and compared at group level. This methodology allows the impact of financial education on time preference to be assessed. If the estimated time preferences of different groups differ, it means that financial education has an impact on time preferences of students. Hypothesis 2 is not rejected if the estimated discount rate of a group of first year students is the highest (representing the least patience) among the 3 groups, and that of a group of master students is the lowest (representing the most patience).

Time preference: Measuring discount rate

The data obtained from a time preference experiment is used in the estimation of time preference (see the first half of figure 2 below). Subjects are asked to make choices between different amounts of money receiving today and a certain amount of money receiving in a year. The same procedure taken in risk preference estimation is also applied for time preferences except that it is an intertemporal utility function instead of an expected utility function. Expected utility functions of risky choices and sure choices, therefore, are replaced by utility functions of present payoffs and utility functions of future payoffs. These two utility functions are then linked to each other to construct the latent index, which is in the form of a cumulative probability distribution function allowing for some errors by including a structural noise parameter. To determine a time preference (discount rate) allowing for risk aversion, time and risk preferences are linked by conditioning the estimated CRRA parameter on the estimated discount rate and estimated using maximum likelihood to infer the probability of the observed choices. The estimated time preference is a discount rate at a group level. Details on the statistical specification are discussed below.

Social preference

Hypothesis 3 states that financial education leads to more pro-social behavior among young adults. This is tested by performing a probit regression using information at an individual level. The dependent variable is binary, indicating whether a subject is pro-social. By including 2 indicator variables for 2 of the 3 student groups as independent variables (hence the constant denotes the third group), the likeliness of being pro-social per group of financial education can be evaluated. I assign the first indicator variable to refer to the first year student (1 is assigned if she/he is a first year student, 0 otherwise) and the second indicator variable to refer to the second year student (1 is assigned if she/he is a second year student, 0 otherwise). The master student is denoted by a constant term in a regression equation

6

(15)

where is a coefficient (a constant term) for a group of master students; and are indicators whether a subject is in a first year and a second year, respectively; and denotes the error term. The influence of financial education is then assessed by checking the p-value for statistical significance of each indicator variable. If at least one of the indicators for student groups is statistically significant, it would mean that a difference in financial education affects the likeliness of being pro-social. Hypothesis 3, therefore, is not rejected if the group of master students has the highest likelihood to be pro-social and the group of first year students has the lowest.

Social preference: Measuring SVO angle in degrees

The method for social preference (or SVO – Social Value Orientation) measurement called SVO Slider Measure based on Murphy et al. (2011) is applied to classify a student as altruistic, pro-social, individualistic or competitive. The SVO Slider Measure has six primary items and nine secondary items. Secondary SVO slider items are optional. They are used to separate joint maximization from inequality aversion as both of them are listed as pro-social SVO. In this study, however, only six primary items are considered because being classified as pro-social or not is enough for hypothesis testing. Each item is a resource allocation choice between a decision maker (a student) and the other person over a range of joint payoffs. A student basically has to sequentially make 6 decisions on a list of multiple choices (see figure 3 below for an example). The obtained data is then processed by a simple calculation as shown in the next section. The last step is to convert a measure into degrees and assign the student to a particular specified interval according to SVO angle in degrees derived from choices she/he made. A student can be classified as altruistic if ; pro-social if ; individualistic if and competitive if . Details and example are discussed in the next section. The distribution of elicited SVO angle in degrees is displayed in figure 2 in the Appendix.

The SVO Slider Measure enables the use of social preferences, in this case pro-social, as a dependent variable in a probit regression. In addition, it yields a high resolution output, which makes it sensitive to inter- and intra-individual differences, facilitates comparison, and allows for greater explanatory potential of SVO through increased statistical power (Murphy et al., 2011).

The effect of risk attitudes on time preferences

(16)

neutrality is higher than the estimated discount rate allowing for risk aversion, risk preferences of individuals would be proved to have an influence on their time preferences and hypothesis 4 would be validated.

Risk and time preferences estimation procedure for hypothesis 4 is the same as for hypothesis 1 and 2 testing with two differences. First, to test hypothesis 4, preferences are estimated based on data at an aggregate level. Second, for hypothesis 4, time preferences (discount rates) are estimated both with and without conditioning on the estimated CRRA parameter, which correspond to assuming risk aversion and assuming risk neutrality, respectively. The estimation is also done with maximum likelihood to infer the probability of the observed choices. The estimated time preferences are discount rates at an aggregate level.

Individual economic and financial decision making is correlated with preferences. Risk and time preferences are considered main factors determining behavior and hence economic and financial outcomes. In this study, I conduct an experiment to obtain information to estimate preferences of 3 groups of economics and business economics students at the University of Groningen who possess different levels of financial knowledge7. I use statistical methods to estimate risk and time preferences and a measurement method for social preferences based on experimentally generated data and link them to the level of financial literacy. In addition, risk, time and social preferences are estimated at an individual level because I would like to know whether differences in risk, time and social preferences among the 3 groups of students are solely consequences of the differences in the level of financial education. Therefore, socio-economic characteristics, cognitive abilities, saving behavior and family background will be taken into consideration in a robustness check. The empirical study consists of (a) designing a questionnaire to obtain information on personal information and family background (the first part of a questionnaire); (b) designing experiments used in the questionnaire to elicit risk, time and social preferences of subjects (the second part of a questionnaire); (c) applying the information on risk and time preferences (social preference) obtained from the questionnaire to economic models (measurement methodology) and estimating preferences using a maximum likelihood estimation (the SVO Slider Measure introduced by Murphy et al. (2011)); (d) comparing risk, time and social preferences among the 3 groups of students and, for a robustness test, allowing them to be determined by individual characteristics – i.e. socio-economic characteristics, cognitive abilities, saving behavior and family background; and (e) comparing time preference assuming risk neutrality to time preference assuming risk aversion.

In this study, 3 experiments for risk, time and social preferences along with a survey on personal information and family background are conducted in a classroom setting (i.e. lab experiment), where subjects are asked to provide their answers in a paper questionnaire. This approach is proved to be the best for this study given time and resources available. As mentioned above, information regarding

7_{For a survey on time preferences measurement, see Frederick et al. (2002). Several methodologies on risk}

(17)

individual risk and time preferences is obtained from choices made in simple risk and time tasks. All these choices contain information about utility, which could then be used to identify his/her utility function. However, an individual utility function is unobserved and hence is a latent function.

6. Data

Subject pool and summary statistics

The sample set contains 3 groups of economics and business economics students, in both Dutch and English programs, at the University of Groningen. The first group consists of Dutch program students who are about to begin with the financial accounting I course for bachelor’s students (first year students). This group represents students who have not yet followed any finance course. The second group represents adolescents with basic financial knowledge and includes bachelor’s students from both Dutch and English programs who are about to follow the finance II course (second year students). Finally, the third group is made up of master in finance students following the program’s corporate finance course. Students in this group can be seen as young adults with more advanced financial knowledge. All courses are in the first half of second semester of academic year 2011/2012.

Data collection took place between 13and 17February 2012, which was the first week of the second semester of the academic year 2011-2012. The survey on the subjects from the financial accounting I and finance II courses was conducted in the first tutorial sessions, right after the first lecture classes of both courses had started. For master in finance students following the corporate finance course, a survey was done in the first lecture class. The questionnaires were distributed at the beginning of class and collected either during the break or at the end of class depending on how the instructors/lecturers had agreed with the students. Therefore, students were given timing flexibility in terms of filling in a questionnaire8. However, they were asked to independently complete all parts of the questionnaire by themselves. Even though the instructors/lecturers were requested to ask students to cooperate by completing and returning a questionnaire, on average approximately 80% of students in each of 3 courses collaborated. In total, 448 questionnaires were collected, 300 of which contained complete information in all parts.

A potential drawback of using this sample set is that all students are receiving education in the Netherlands, which is a developed country. Subjects who supposedly represent young adults with no knowledge in finance may have already acquired some financial knowledge or concepts from previous education or courses prior to entering the university.

Questionnaire

A questionnaire was designed for conducting survey on students in the sample set. Students were asked to provide information on background and certain behavioral characteristics (e.g., personal

8_{Before the questionnaire was used in this study, I pre-tested it by asking students both at the faculty of}

(18)

information, health, family background, cognitive abilities and saving behavior) in the first part of the questionnaire. Students were also asked to respond to binary and multiple choice lists used in standard experimental games for eliciting risk, time and social preferences in the second part of a questionnaire. Questionnaires for the 3 groups of students are identical except questions on study program and grades on core courses that they have completed. Instructions and decision tasks on questions for preferences are the same across questionnaires (see the Appendix for a sample of the questionnaire used in the study). All information provided by students is kept anonymous throughout the study.

Risk preference

I apply the framework of the Ellsberg two-color choice task9 to obtain information on risk preferences of subjects. Students were presented with a binary choice list. They were asked to make a decision between drawing a ball from a bag containing 5 red and 5 white balls (option A) representing a risky choice and receiving a certain amount of money (option B) representing a sure choice. The same probability of getting a red and a white ball from each draw (0.5) is given clearly in the instruction. If a student chooses to draw a ball from a bag, she/he has to choose a color first. If she/he chooses red (white), she/he will receive 5 euro if the ball drawn is red (white) and nothing otherwise. Hence, a probability of getting money for this risky choice (option A) is 0.5 whereas a probability of getting money for a sure choice is 1 throughout the experiment. A guaranteed payoff varies in an ascending order of 0.50 euro from 0.50 euro to 5.50 euro10. Figure 1 presents an example of a decision table used in the questionnaire.

Please give your preference in column C by writing down A or B.

A B C

draw from a bag guaranteed 0.50 euro A

draw from a bag guaranteed 1 euro A

draw from a bag guaranteed 2 euro A

draw from a bag guaranteed 3 euro B

draw from a bag guaranteed 3.50 euro B

9_{See e.g. Ellsberg (1961) and Sutter et al. (2010).} 10

(19)

Figure 1: Choice list for risk preference

All ten choices chosen by students contain information about their risk attitudes. Information acquired from this experiment will be used in a maximum likelihood estimation to elicit a CRRA parameter representing the level of risk aversion of subjects. The elicitation procedure is based on the expected utility theory assuming that it holds for the choices over risky alternatives; that the subjects employ a CRRA utility function defined over the prizes over which they made choices; and that discounting is exponential (Coller et al., 2005, 2011; and Andersen et al., 2006). Further details can be found in section 7 below.

The method to measure risk preference employed in this research paper is appropriate for the sample set according to its simplicity and suitability to operate in the classroom setting compared to other methodologies. Even though Holt-Laury risk price lists (see Holt and Laury, 2002) using two sets of lottery and survey questions used to elicit risk preferences (see e.g. Dohmen et al., 2005 and Brown and Taylor, 2007) are widely used by several researchers, they are an inferior risk attitudes measurement methodology for this study. This is because, given the straightforwardness of this experiment for getting information about risk attitudes as well as the nature of sample set (university students), the Ellsberg two-color choice is suitable for a study conducted via a questionnaire depending solely on written instructions. In addition, the Ellsberg two-color choice task differs from the Holt-Laury risk price lists in that the probability distribution is the same throughout the choice lists and the payoff from taking a gamble is fixed while a sure payoff increases monotonically. This is relatively easier for subjects to follow and potentially makes it possible to minimize a stochastic error (a structural noise parameter), which used to allow some errors from the perspective of the deterministic expected utility theory model11 (see e.g. Andersen et al., 2006 and Coller et al., 2005, 2011).

Time preference

The multiple price lists (MPLs) method12 is employed to obtain information on individuals’ time preferences. MPLs impose a linear utility function over experimental outcomes13. By using money as a payoff in MPLs, patience (represented by discount rate) of each individual can be measured (Meier and Sprenger, 2010). Students faced two sets of choice lists classified based on different timing combinations of payoffs. In the first set, they were asked to choose between receiving a smaller payoff for sure today and receiving a larger payoff for sure in one year (no front-end delay). In the second set,

11_{See Harless and Camerer (1994), Hey and Orme (1994) and Loomes and Sugden (1995) for earlier empirical}

studies which include some formal stochastic specification in the version of the EUT tested.

12

See e.g. Coller and Williams (1999) and Harrison et al. (2002).

13

(20)

they were asked to choose between receiving a smaller payoff for sure in one year and receiving a larger payoff for sure in two years (with front-end delay)14. A later payoff is fixed at 200 euro in both lists. An earlier payoff varies in a descending order (195, 190, 185, 180, 175, 170 and 160). However, I limit the scope of this study to no front-end delay. In future research, present bias (β < 1) or the passion for the present will be included along with discount rates to explain time preference. Figure 2 presents an example of a decision table used in a questionnaire.

Please give your preference between the earlier amount (option A) and the later but larger amount of money (option B) by writing down A or B in column C.

A B C

195 euro today 200 euro in 1 year A

180 euro today 200 euro in 1 year B

195 euro in 1 year 200 euro in 2 year A

190 euro in 1 year 200 euro in 2 year A

185 euro in 1 year 200 euro in 2 year B

Figure 2: Choice list for time preference

With MPLs, the point where a student switches from choosing a sooner payoff to a later payoff (from A to B) provides information about her/his intertemporal preference. This implicitly assumes a

14

(21)

linear utility over binary choices and is consistent with the expected utility theory (Rabin, 2000). Data about time preferences of individuals obtained from this experiment will be treated using the maximum likelihood estimation technique in order to estimate the discount rate. The elicitation procedure is based on the expected utility model with exponential discounting and a CRRA utility function defined over the prizes over which students made choices. Details on the estimation can be found in section 7.

Even though the Convex Time Budget (CTB) approach has been proposed as a superior methodology to the MPLs (see e.g. Andreoni and Sprenger, 2011a), I find that the MPLs for time preferences represent a more suitable methodology for this paper because the developed questionnaire is also designed to generate information on risk preferences. As a result, the curvature of the utility function (Frederick et al., 2002 and Andersen et al., 2008a), which generates risk aversion, and an experimental methodology controlling for the curvature of the utility function can be employed in this study by conditioning the estimated discount rate on risk aversion (a CRRA parameter)15. While this approach is modified from those in Harrison et al. (2002, 2005, 2008, 2009, 2010), Harrison and Rutström (2008a), Harrison (2008), Andersen et al. (2006, 2008) and Coller et al. (2005, 2011) according to a time constraint and the quality of data used in the study, it is proved to yield the same results. Details are provided below. By incorporating risk and time preferences following this methodology (see e.g. Coller et al., 2005, 2011; Andersen et al., 2006; and Harrison and Rutström, 2008 for joint likelihood of the risk aversion and discount rate maximization), the resulting estimated discount rates are statistically significantly lower and economically more sensible than Andreoni and Sprenger (2009a). Social preference

The measure of social preference or Social Value Orientation (SVO) called the SVO Slider Measure, as introduced by Murphy et al. (2011), is applied16. However, as explained above, only 6 primary SVO slider items are used in this study to elicit social preferences17. Students were asked to make a series of decisions on resources allocation between, and hence affect payoffs for, themselves and the other person. The amount of resources in each item either remains the same or varies, in an ascending and a descending order, from 15 to 100 and vice versa. Figure 3 presents an example of a slider item used in the questionnaire.

Below you will find an example. A person chose to distribute 50 euro for him/herself and 40 euro for the other person. There are no right or wrong answers. All questions aim to measure personal preferences. After you have made your decision, please write the resulting distribution on the space below. As you can see, your decisions will influence both the amount of money you receive and the amount of money the other receives.

15

In the future research, a joint estimate of a discount rate and a CRRA parameter by maximum likelihood estimation (MLE) will be employed.

16_{There are two versions (A and B) of the paper based version of the SVO Slider Measure. Only version A is}

employed in the study. More information can be found at http://vlab.ethz.ch/svo/SVO_Slider/.

17

(22)

You: 50 Other: 40

Figure 3: SVO Slider Measure for social preference

With 6 primary SVO slider items, the social preference type – altruistic, pro-social, individualistic, or competitive – of each student is specified. First, the mean of payoffs allocated to her/himself across all 6 primary items ( ̅̅̅) is calculated:

̅̅̅

where is the resource allocated to self in each of six primary items. Second, the mean of payoffs allocated to the other person across all 6 primary items ( ̅̅̅̅) is calculated:

̅̅̅̅

where is the resource allocated to the other in each of six primary items. In the third step, both means are subtracted by 50 in order to shift the base of the resulting angle to the center of the circle (50, 50) rather than having its base start at the Cartesian origin (Murphy et al., 2011)18: ̅̅̅ and ̅̅̅̅ . Fourth, the ratio of the mean of payoffs to the other minus 50 to the mean of payoffs to the self minus 50 is determined: ̅̅̅̅̅̅̅̅

.

The next step is to calculate the SVO angle (in radians) by computing the arc tangent of the ratio:

₍( ̅̅̅̅̅̅ )

(

̅̅̅̅̅̅₎) SVO angle in radians is then converted into SVO angle in degrees by:

( )

18

See Figure 1A in Appendix for the self/other allocation plane imported from Murphy et al. (2011).

You receive

Other receives

35

30 40 45 50 55 60 65 70

(23)

where pi (π) is approximately equal to 3.14. Finally, a student can be classified as altruistic if ; pro-social if ; individualistic if and competitive if . The summary statistic of participants is shown in Table A1 in the Appendix. The obtained information on social preferences is at an individual level with the majority of the subjects being characterized as pro-social, followed by individualistic..

This new SVO measure is employed in the research study because of its several advantages over other existing methods. The SVO Slider Measure has superior psychometric properties compared to other popular methods in use (e.g. the 9-Item Triple-Dominance scale, the Ring Measure, the Regression and clustering approach, the Utility measurement, the Altruism scale, the Social Behavior scale and Schulz and May’s Sphere Measure)19. Moreover, it overcomes many limitations as well as combines strengths of others. Murphy et al. (2011) report that a test-retest reliability of the SVO Slider Measure compares favorably against that of the 9-Item Triple-Dominance scale and the Ring Measure. In addition, they test for a convergent reliability of the SVO Slider Measure with those two methods. The SVO Slider Measure classifies 74% and 75% of the same subjects to the same SVO category as the 9-Item Triple-Dominance scale and Ring Measure, respectively. Furthermore, the most powerful property of the SVO Slider Measure is that it yields a high resolution output (a ratio). This property fits perfectly with the nature of Social Value Orientation, which is a continuous scale, and therefore generates a high statistical power of the measure.

Summary statistics

The summary statistics on the sample set are presented in Table 1. Panel A shows the socio-demographic summary statistics. From the total of 448 students surveyed, 389 provide complete basic demographic information (i.e. gender, age, nationality, weight, height and sibling related information). It can be observe from the table that the average student in the study is male, around 21 years old, Dutch, of normal weight, a second child and grew up with 2 siblings. Of these, 301 students also gave all information on financial education, cognitive ability (average grade and average mathematics grade) and saving behavior. The average student has around 1 year of financial education as the majority of students in a sample set are second year students; average grade and average mathematics grade rounded up to 7; and more or less savings of approximately 12 percent of his monthly income. Panel B shows the summary statistics of social preferences of the entire sample set. The average student is unlikely to be competitive. This aggregate sample set is used for the main analysis in the study.

19_{For an overview of measurement methods for social preferences see Murphy and Ackermann (2011). For more}

(24)

Table 1: Summary statistics for an aggregate sample set

Variable Observation Mean Std. Dev. Min Max

Panel A: Socio-demographics Gender 448 0.671875 0.4700555 0 1 Age 441 21.18821 1.837352 18 30 Nationality 448 0.84375 0.3634981 0 1 BMI 432 21.99449 2.392096 16.47359 31.25 Sibling 445 0.8808989 0.324272 0 1 Number of sibling 400 1.815 1.059992 0 9 Birth order 393 1.842239 0.9452065 1 7

Grew up with sibling 389 0.9794344 0.1421074 0 1

Year of financial education 448 0.9620536 0.7661095 0 3

Average grade 322 6.872671 0.8497379 4 10

Average mathematics grade 301 6.72093 1.359606 1 10

Saving 368 0.5461957 0.4985392 0 1

Percentage of monthly income saved 368 11.90217 17.91004 -20 100

Panel B: Preferences

Pro-social 432 0.5486111 0.4982083 0 1

Individualistic 432 0.4467593 0.4977338 0 1

Competitive 432 0.0046296 0.0679624 0 1

(25)

varies from 4 to 10 in the aggregate sample set. An average mathematics grade measures numerical skills and varies from 1 to 10. There is a large difference in terms of the percentage of monthly income saved, ranging from not being able to save and having to borrow up to 20 percent of monthly income to being able to save the entire amount. A number of aggregate observations contain missing values, a majority of which relate to information on siblings, saving behavior and especially cognitive abilities. Table 2: Summary statistics on family background for an aggregate sample set

Age of father 427 52.85948 4.915619 38 78 Age of mother 431 50.79118 5.021629 36 68 Education of father 401 4.009975 1.24294 1 7 Education of mother 381 3.606299 1.164095 1 7 Occupation of father 350 2.211429 0.8017786 1 4 Occupation of mother 300 2.02 0.5950745 1 4

Table 2 presents the summary statistics of family background for the aggregate sample set. 300 students provide complete information on age, education and occupation of their parents. Parental education attainment level is set to range from 1 (primary school) to 7 (PhD and higher)20 and parental occupation from 1 (worker and logistics worker) to 4 (managing director, doctor and professor)21. The age of parents in the sample set varies from 36 to 78 years old with an average of 53 years old for fathers and 51 for mothers. A majority of fathers in the sample set have HBO or HTS education and most likely work as a(n) teacher, accountant, nurse, entrepreneur or business owner. Most mothers have higher education than MBO or MTS and also have similar occupations as fathers.

20_{The level of 1 represents elementary school education; the level of 2 represents secondary school, ITO, MULO,}

ULO, HBS, MMS, gymnasium, lyceum, ITS, MAVO and atheneum education; the level of 3 represents MBO and MTS education; the level of 4 represents HBO and HTS education; the level of 5 represents BSc education; the level of 6 represents MSc education; and the level of 7 represents PhD and higher education.

21

(26)

Table 3: Summary statistics for Dutch program students following financial accounting I course

Panel A: Socio-demographics Gender 99 0.6767677 0.4700908 0 1 Age 98 19.87755 1.890101 18 29 Nationality 99 1 0 1 1 BMI 94 21.31026 2.428879 17.357 30.18959 Sibling 99 0.959596 0.197907 0 1 Number of sibling 99 1.757576 1.246144 0 9 Birth order 95 1.852632 1.100951 1 7

Grew up with sibling 94 0.9787234 0.1450787 0 1

Average grade 97 6.525773 0.9475053 4 10

Average mathematics grade 77 5.649351 1.782711 1 10

Saving 84 0.6547619 0.4783014 0 1

Percentage of monthly income save 84 19.38095 25.23332 0 100

Panel B: Preferences

Individualistic 88 0.3977273 0.4922333 0 1

Pro-social 88 0.6022727 0.4922333 0 1

Table 4: Summary statistics for students following finance II course

Panel A: Socio-demographics

Gender 308 0.6525974 0.4769198 0 1

Age 302 21.2351 1.352134 18 28