The effect of gender, age and education level on risk attitude and how the effect of age and gender are influenced by level of education

(1)

Bachelor thesis: Econometrics

Friso Blankenspoor (10149554)

The effect of gender, age and education level on risk

attitude and how the effect of age and gender are

influenced by level of education

(2)

1. Introduction

In what manner people choose between options in certain circumstances is a difficult question to answer because every person unique and acts differently. Decision-‐making is a subject that has been studied several times. This paper will focus on choosing between a safe certain payment or a more risky higher uncertain payment. These kinds of decisions are influenced by someone’s attitude towards risk. Risk is defined as a certain chance of winning something of value weighted to losing something of value. Risk attitude is influenced by several personal characteristics. What characteristics this are and what these characteristics are influenced by has been studied by multiple researchers (Bakshi and Chen, 1994; Barber and Odean, 2001; Hartog, Ferrer-‐i-‐Carbonell & Jonker, 2002; Jianakoplos and Bernasek, 1998; Morin and Suarez, 1983; Shaw, 1996; Wang and Hanna, 1997). The effect of age, gender and education on risk aversion has been investigated several times. Highly educated individuals are more likely to be risk takers than low educated individuals. Risk aversion seems to be slightly increasing in age and there is limited evidence that the effect of age on risk aversive behavior is not linear. In general, men are less risk avers than women.

In this article, the general effect of age, gender and education on risk aversion is studied. Secondly, the effect of education on the association between age and gender and risk aversion is observed. My hypothesis consists of two parts. Part one is that male are more likely to seek risk than women and a higher education level and increasing age cause both a decrease in risk aversion. Part two is that the influence of education on the effect of gender will cause a decrease in difference between men and women, the influence of education on age will moderate the effect of age. This theory has not been investigated up until now.

I will start with studying the general effect of gender, age and education on risk aversion and if the effect of age is linear followed by if the effect of age and gender are influenced by education.

In other studies (Bakshi and Chen, 1994; Barber and Odean, 2001; Hartog, Ferrer-‐i-‐ Carbonell & Jonker, 2002; Jianakoplos and Bernasek, 1998; Morin and Suarez, 1983; Shaw, 1996; Wang and Hanna, 1997) different methods to measure risk aversion have been used. It is not clear what the best way to measure risk aversion is and what kind of implications should be accounted for. I will measure risk attitude by absolute risk aversion, relative risk aversion and normal risk aversion deducted from the participants’ lottery questionnaire answers. This questionnaire is part of the data that I will use to study my hypothesis. The data is from a LISS panel is managed by CentERdata. This organization is affiliated with the University of Tilburg. There are a lot of background variables available and a panel risk preferences. When the

(3)

measure of risk aversion is deduced from the data I will regress the background variables gender, age, education, gender combined with education, age combined with education and a constant term on the risk aversion measure.

In the next chapter of this article I will discuss the most important literature and clarify the types of risk measurement I used. In the section method and data I will clarify the method I used to get to my results. The final part of this article consists of conclusion and discussion followed by a list of references.

(4)

2. Literature review

A lot of research has been performed on the characteristics that influence risk aversion. In this section I will discuss the most important articles that are of interest to my study. Secondly, I will discuss different kind of methods that are currently available to measure risk aversion and show how absolute and relative risk aversion measures are derived. I will also discuss how to

interpret someone’s risk attitude.

2.1 Risk aversion measurement

Risk aversion is usually measured by two different measurements; (1) the proportion of risky assets in someone’s portfolio or (2) through a risk aversion measure computed from a lottery panel with a risk preference questionnaire. The proportion of risky assets compared to total assets is a measure that can be influenced by many factors that are difficult to control for, since it would require a lot of personal information. For example, does a participant receive a big

financial legacy or does someone have a shorter life expectancy because of a medical condition. These kinds of factors could have a big influence on the portfolio composition. These factors taken into account, I have chosen to measure risk by the data from a lottery questionnaire for risk preference.

I used three types of risk aversion measurements for each participant: relative risk aversion, absolute risk aversion and normal risk aversion. Each individual could be categorized in risk loving, risk neutral or risk averse. In the following part I will explain a few important characteristics of each preference towards risk and how to compute this absolute and relative risk aversion measure. In the eyes of a person who has a risk neutral utility, the expected value of the uncertain game is equal to the expected value of the utility of the uncertain game. Risk lovers prefer the expectation of the utility of the uncertain game and risk avoiders vice versa. This can be explained by a person’s utility function. This function represents a person’s

preferences. Risk loving refers to a convex utility function, risk aversion refers to concave utility function and risk neutrality refers to linear utility function. From Wakker (2010) it follows that when a utility function U is twice continuously differentiable, a common used measure for this concavity or risk aversion is the Pratt-‐Arrow measure, defined as –U’’/U’. On the domain of positive outcomes α>0, an alternative popular measure of risk aversion, or concavity of utility, is α ∗ (-‐U’’(α)/U’(α)). This is sometimes addressed as the relative (or proportional) measure of risk aversion. The Pratt-‐Arrow measure is called the absolute measure of risk aversion (Wakker, 2010). Because it is impossible to measure a person’s exact utility function I will use two general

(5)

forms that are mathematically convenient and are proven to be good estimates of the utility function: power utility function and exponential utility function.

Power utility with parameter θ for α>0:

𝑓𝑜𝑟 𝜃 > 0, 𝑈 α = 𝛼!

𝑓𝑜𝑟 𝜃 < 0, 𝑈 α = −𝛼!

𝑓𝑜𝑟 𝜃 = 0, 𝑈 α = ln (𝛼)

When θ gets smaller the more concave U and so the more risk averse. With the power utility function the Pratt-‐Arrow measure for relative risk aversion can be derived. The relative risk aversion measure will be 1-‐θ.

Exponential utility function with parameter θ for every α:

𝑓𝑜𝑟 𝜃 > 0, 𝑈 α = 1 − 𝑒!!"

𝑓𝑜𝑟 𝜃 = 0, 𝑈 α = 𝛼 𝑓𝑜𝑟 𝜃 < 0, 𝑈 α = 𝑒!!"_{− 1}

The parameter 𝜃 is again an index of concavity, with linear utility for 𝜃=0, concave utility for 𝜃>0 and convex for 𝜃<0 (Wakker, 2010). When I compute the absolute risk aversion with an

exponential utility function I find 𝜃 that is independent of α.

1.2 Research review

Several researchers (Bakshi and Chen, 1994; Wang and Hanna, 1997) have studied the influence of age on risk aversion. Wang and Hanna (1997) assessed if risk tolerance was influenced by age. The authors used a panel of the Survey of Consumer Finances 1983-‐89 to test their life-‐cycle investment hypothesis, which stated that risk tolerance decrease with age. The proportion risky assets compared with the total wealth were the measure for risk tolerance. The participant’s wealth was defined as the sum of net worth and human capital. Human capital had been defined as a person’s competencies that have economic value.

Wang and Hanna (1997) conclude that the risk tolerance increases with age, what indirectly means that risk aversion decreases with age. The dataset Wang and Hanna (1997) used lacked information about household’s present value of future pension and Social Security wealth and the measurements of future earnings. They therefore imputed data from the 1983 cross-‐sectional data of Surveys of Consumer Finances (Avery & Elliehausen, 1990). However, this could have affected the reliability of their results, as there could have been a generational effect. They mention that human capital for young people a relative bigger part of their net wealth compared to older people who have bigger part financial wealth accounting for their net wealth (Wang and Hanna, 1997). This means that the proportion of investments in risky assets

(6)

increases when people grow older. Younger people may seem a bit more risk averse since they

cannot bear short-‐term losses with limited financial resources (Wang and Hanna, 1997). Bakshi and Chen (1994) concluded the exact opposite. Their hypothesis was that the

relative risk aversion of investors increases with age. They tested this hypothesis by computing an Euler equation. This equation is a discrete-‐time model in which the investor’s makes

consumption-‐portfolio decisions form time 0 to T, with time intervals of Δt. In order to test life-‐ cycle risk aversion hypothesis and the Euler function they needed to specify a functional form for utility function u(•,•). They specified the utility function as a power utility function since it offers mathematical convenience and makes interpretations intuitive. They computed the Arrow-‐Pratt relative risk aversion measure, which seems to be linear in average age.

Substituting the utility function in the Euler equation gives the equation they tested with several methods to verify the hypothesis. Intuitively they say that if you get older, the number of

paychecks you will receive declines so they suggest that we can see that human capital as a decreasing function of age (with human capital defined as mentioned previously). When relative risk aversion is decreasing in human capital, it follows that risk aversion is an increasing

function of age. While aging, you have fewer opportunities to bear a financial loss with labor income.

Different implication in human capital and housing could lead to different results for the effect of age. The difference between Wang and Hanna (1997) and Bakshi and Chen (1994) could be explained by the fact that Wang and Hanna (1997) state that human capital is an increasing function with age and Bakshi and Chen states (1994) implicates that human capital is decreasing with age.

Morin and Suarez (1983) investigated the effect of wealth on risk aversion. Wealth was defined as a person’s net worth, total asset holdings minus their total debt. It points out that relative risk aversion will decrease with wealth. The study also showed that age increases the level of risk aversion. This supports Bakshi and Chen (1994). However, there was an exception for the group with the highest wealth. They observed no significant effect and the degree of risk aversion was constant over age (Morin and Suarez, 1983). It seems that wealth is influencing risk attitude. However, income, wealth and education are highly correlated so this could also be caused by education.

Riley and Chow (1992) conclude in their article that risk aversion is decreasing with education. This means that higher educated individuals are more likely to be risk tolerant because they are better in estimating the amount of risk that they are exposed to. The study of Donkers, Melenberg & van Soest (2001) confirms that education has a positive effect on risk taking.

(7)

Hartog, Ferrer-‐i-‐Carbonell & Jonker (2002) studied the influence of individual

characteristics on risk aversion. The data used consisted of multiple lottery questions. They used the expected utility theory to derive the Arrow-‐Pratt measure of risk aversion based on answers of the lottery questions. They found interesting results; no significant effect of parental

background or marital status on risk aversion was observed, except for highly educated mothers. Mothers that are highly educated reduce risk aversion. They investigated if this effect could be possibly caused by the fact that a relatively large part of the women did not have a job and or steady income. Therefore, they would be more likely to be risk averse since it was not their income that they are risking. However, this did not seem to be the case (Hartog et al., 2002). Similar with other studies (Jianakoplos and Bernasek, 1998; Barber and Odean, 2001), it appears that in general women are more risk averse than men.

In this study I have done research to the effect of gender, age and education on risk aversion and will test the conclusions of several mentioned articles (Bakshi and Chen, 1994; Barber and Odean, 2001; Hartog, Ferrer-‐i-‐Carbonell & Jonker, 2002; Jianakoplos and Bernasek, 1998; Morin and Suarez, 1983; Shaw, 1996; Wang and Hanna, 1997). Secondly, does education influence the effect of gender? Hartog, Ferrer-‐i-‐Carbonell & Jonker (2002) give reason to believe so. Could it be that the effect of age on risk aversion will be smaller for highly educated

individuals? Higher educated individuals are better in estimating the amount risk they are exposed to and will use this in their decision. Because men are in general already more risk seekers the effect of education is expected to be smaller. This could explain why the difference between highly educated men and women in risk aversion is smaller. Likewise education could influence the effect of age on risk aversion and could cause a moderating effect. So I would expect that the effect of age is smaller for higher educated individuals, assumed that the effect of age is negatively correlated with risk aversion based on Wang and Hanna (1997).

(8)

3. Data & Methods

For my research I have used the LISS panel that is managed by CentERdata. This is an organization affiliated with the University of Tilburg. The sample of the LISS panel is representative for the general Dutch population. The panel consists of approximately 9000 participants. Every month the participants completed an online questionnaire. They received a financial compensate for their participation four times a year. There is a risk panel available and a larger number of background variables. The panel measuring higher order risk attitudes of the general population consisted of 3271 participants that completed the whole questionnaire. Within the questionnaire, there were 4 groups with two groups with real payoffs and two with hypothetical payoffs. These groups where randomly distributed. Group 1 consisted of 30% of the participants and had normal real payoffs, group 2 consisted of 10% of the participants and had low but real payoffs. Group 3 and 4 consisted both of 30% of the participants corresponding with high hypothetical payoffs and normal hypothetical payoffs. The participants of group 1 and 2 knew that they could earn real money. The questionnaire consisted of 4 parts that all

contained lottery questions. For this research I only used the results of part 1. Part 1 of the questionnaire consists of 5 lottery questions. The participant could choose between a sure payment and an uncertain variable payment, the sure payment will differ for all 5 questions. The sure payment was 25 euro’s going up to 40 euro’s with steps of 5 euro’s or starting at 40 euro’s and going down to 25 euro’s. The uncertain payment will be 65 euro’s with chance of 50% or 5 euro’s with equal chance 50%.

I have derived three different risk aversion measures the normal, the Arrow-‐Pratt-‐De Finetti measure of relative risk aversion1_{and
the
Arrow-‐Pratt
measure
of
absolute
risk}

aversion2_{.
The
normal
measure
is
defined,
as
the
estimate
of
the
sure
value
at
what
that}

participants is indifferent between that sure payoff and the variable payoff. To measure this I took the middle value of the last sure payment value where the participant choses the sure payoff and the first sure payment value where the participant choses the variable payoff. To compute the absolute risk aversion measure θ, I have used the exponential utility function.

1

The Arrow-‐Pratt-‐De Finetti measure of relative risk-‐aversion (RRA) or coefficient of relative risk aversion is defined as: 𝑅 𝑐 = 𝑐𝐴 𝑐 =!!"!!(!)_!!(!) , where c is constant and 𝑢(∙) is a utility function.

2_{Arrow-‐Pratt
measure
of
absolute
risk-‐aversion
also
known
as
the
coefficient
of
absolute
risk
aversion,
defined
as
𝐴 𝑐 = −}!!!(!)

!!(!),

where c is constant and 𝑢(∙) is a utility function.

(9)

I have solved the following equation towards θ, where 𝐶 is the normal risk aversion measure: 1 2∗ 1 − 𝑒!!!" /𝜃 + 1 2∗ 1 − 𝑒!!! /𝜃 = (1 − 𝑒!!")/𝜃

To find the relative risk aversion measure defined by ϕ, I have solved the following equation for every participant towards ϕ, where 𝐶 is normal risk aversion measure:

1 2∗ 65!/𝜑 + 1 2∗ 5!/𝜑 = 𝐶!/𝜑

The relative risk aversion measure is: 1−𝜑.

The next step was, to find a proper model that explains these risk aversion measures. The characteristics that I used are listed in the background variables namely gender, age and education. To find good results I computed 3 different groups. Group 1 consisted of all 3271 participants, this where all participants with all types of payoffs and inconsistent individuals included. Inconsistent individuals did switch more than ones between the safe and risk option.

However to make a good risk aversion estimate of this individuals I took the first switching point to compute the normal risk aversion measure. In group 2 I included all types of payment but excluded the inconsistent participants, this resulted in a group of 2248 participants. For group 3 took in account only the normal real payments and the consistent individuals that resulted in a group size of 711.

In the background variables gender was give by a 2 for male individuals and 1 for female. To make a good estimate of the gender differences I created a dummy variable for gender where value 1 corresponds to male participants and 0 for female. In the background variables

education was categorist as 1 for primary education, 2 for lower vocational education, 3 for higher general secondary education or pre-‐university education, 5 for bachelors degree and 6 for masters degree. In the questionnaire each participant filled in his or her age.

To investigate if the differences in risk attitude between high-‐educated women and men declines compared to differences between lower educated men and women I have constructed a cross product of gender and education to the model. For the influence of education on the effect of age on risk aversion I have constructed a cross product of age and education. I added a square age product to investigate if there was a quadratic effect of age.

(10)

4. Results

In table 1 the average risk measure is presented for each individual group. As stated in this table, all groups were risk averse on average. Also I found that if you include inconsistent individuals and classify them by their first switch, this influences the group average by getting more risk averse.

Table 1. Average risk measures for each group

Group 1:

All respondents Group 2: Only

consistent

Group 3: Normal real payments and consistent

# Observations 3271 2248 771

Avg normal risk-‐

aversion 26.8182513 30.26023132 30.25668073

Avg relative risk-‐

aversion 0.024677155 0.015167901 0.015626365

Avg absolute risk-‐

aversion 0.445968238 0.225684462 0.219063513

The following descriptive statistics for the three different groups are presented in Table 2. Computed the three average risk aversion measures for each group with education level, gender and age taken into account. Education divided into highly educated or low educated with highly educated defined as individuals with Master’s or a Bachelor’s degree. The other variables I accounted for are gender and age, with age categorized as up and below the age of 55. The effect of age and education is different for women.

Table 2. Average risk measures education, age and gender accounted for each group

Group 1: All respondents Group 2: Only consistent Group 3: Normal real

payments and consistent

Msc/

Bsc Female Over 55 # obs Avg Nor RM Avg Abs RM Avg Rel RM # obs RM Avg Nor Avg Abs RM Avg Rel RM # obs Avg Nor RM Avg Abs RM Avg Rel RM

Yes Yes No 346 26.1561 0.4865 0.0267 248 29.1936 0.293 0.01827 84 29.5238 0.2623 0.01802

Yes Yes Yes 134 27.3881 0.4121 0.023 92 30.8152 0.192247 0.01367 32 33.2813 0.029 0.007

Yes No No 302 27.351 0.4175 0.0229 219 30.8333 0.1978 0.01277 73 31.9521 0.1096 0.0108

Yes No Yes 233 27.8219 0.38303 0.02202 174 30.7758 0.195 0.01366 59 30.4661 0.2075 0.0149

No Yes No 697 27.0983 0.4266 0.02388 486 30.7369 0.21653 0.01496 148 30.4054 0.2121 0.015

No Yes Yes 536 26.4366 0.4671 0.02582 371 29.4677 0.27113 0.01768 113 29.3584 0.27197 0.01824

No No No 581 26.4157 0.4757 0.02557 363 30.7369 0.2018 0.01359 119 29.3487 0.28518 0.01556

No No Yes 442 26.8213 0.44297 0.02484 295 30.5 0.2057 0.01483 83 30.4518 0.19904 0.01556

(11)

I did linear regressions with different control groups, group A consist of a constant, age, dummy variable for gender with 1 corresponding for males, education category. Education is categorist in 1 for primary education or lower vocational education, 2 for higher general

secondary education or pre-‐university education and 3 for bachelor’s degree or masters degree. Group B consists of same variables with the cross terms of gender, age with education as added categories. Group C consists of the same variables as A and the square product of age. In table 3, table 4 and table 5 the results of the regressions are presented.

Tabel 3. risk aversion group 1

Normal RM

Relative RM

Absolute RM

Control: A B C A B C A B C

Age .016574 (.1279) -‐.009257 (.7381) .101368 (.0857*) -‐.000039 (.1896) (.5975) .000049 -‐.000289 (.0796*) -‐.001172 (.0948*) .000442 (0.8042) -‐.006535 (.0856*) Dumgender .129705 (.7279) -‐.368498 (.7047) .14039 (.7065) -‐.000504 (.6287) (.6818) .001115 -‐.000535 (.6074) -‐.004456 (.8529) .024172 (0.6998) -‐.005132 (.8309) EduCat .292894 (.2039) -‐.521078 (.4953) .0237124 (.3101) -‐.000739 (.2514) (.3944) .001819 -‐.000575 (.3784) -‐.018157 (.2217) .032192 (0.5133) -‐.01463 (.3313) SqAge -‐ -‐ -‐.00088 (.1435) -‐ -‐ .00000286 (.1242) -‐ -‐ .0000556 (.1512) CrossAgeEdu -‐ .013757 (.3220) -‐ -‐ -‐.000043 (.2683) -‐ -‐ -‐.000862 (.3358) -‐ CrossGenderEdu -‐ .234751 (.6098) -‐ -‐ -‐.000765 (.5518) -‐ -‐ -‐.013386 (0.6516) -‐

Dependent variable risk aversion measure; Linear regression coefficients reported and p-‐values. */** indicate significance at 10% and 5%.

Normal

Relative RM

Absolute RM

Control: A B C A B C A B C

Age .005128 (.7102) -‐.035921 (.3214) .081419 (.2768) -‐.00000187 (.9610) (.2484) .000116 -‐.000230 (.2676) -‐.000512 (0.5670) . 002110 (.3688) -‐.005248 (.2795) Dumgender .793222 (.0893*) .871372 (.4785) .810484 (.0827) -‐.0002549 (.0486**) (.5448) -‐.002061 -‐.002601 (.0444) -‐.044592 (.1406) -‐.054060 (.4975) -‐.004566 (.1315) EduCat .084029 (.7703) -‐.998482 (.3033) .046936 (.8714) -‐.000337 (.6725) (.2779) .002914 -‐.000226 (.7783) -‐.003676 (.8438) .064546 (.3046) -‐.001373 (.9418) SqAge -‐ -‐ -‐.000787 (.2998) -‐ -‐ .00000235 (.2633) -‐ -‐ .000049 (.3207) CrossAgeEdu -‐ .021636 (0.2198) -‐ -‐ -‐.0000615 (.2075) -‐ -‐ -‐.001385 (.2255) -‐ CrossGenderEdu -‐ -‐.068491 (.9050) -‐ -‐ -‐.000163 (.9184) -‐ -‐ .006632 (.2255) -‐

(12)

Normal RM

Relative RM

Absolute RM

Variable group A B C A B C A B C

Age .027347 (.2823) -‐.032930 (.6416) .162725 (.2453) -‐.00000668 (.3436) (.4677) .000142 -‐.000428 (.2708) -‐.001977 (.2315) .001371 (.7655) -‐.010695 (.2396) Dumgender .043977 (.9596) -‐.147081 (.9498) .085647 (.9215) -‐.000168 (.9445) (.8881) .000912 -‐.000279 (.9080) .0000018(.9997) . 010429 (.9452) -‐.002666 (.9624) EduCat .994233 (.0634*) -‐.602680 (.7473) 0.948910 (.0775*) -‐.002683 (.0708*) (.5642) .002946 -‐.002563 (.0856*) -‐.065451 (0.0599*) . 023186 (.8464) -‐0.06253 (.0733*) SqAge -‐ -‐ -‐.001396 (.3255) -‐ -‐ .000004 (.3446) -‐ -‐ .0000899 (.3296) CrossAgeEdu -‐ .030364 (0.3638) -‐ -‐ -‐.000105 (.2569) -‐ -‐ -‐.001686 (.4374) -‐ CrossGenderEdu -‐ .050411 (.9625) -‐ -‐ -‐.000381 (.8981) -‐ -‐ -‐.002701 (0.9691) -‐

Table’s 3,4 and 5 give different results. Table 3 shows that the effect of age is negative on the absolute risk aversion measure at 10% significance level. In for group 2 I found at 10% significance level that education has positive effect on normal risk aversion measure and at 5% significance has negative effect on relative risk aversion measure. Table 5 shows that there is at 10% significance level effect on all three risk aversion measures, that education cause a decrease in risk aversion. There are no significant effects of the cross products in all groups.

In table 6 results for group 1 are presented from a regression with dependent variable the normal risk aversion measure and with controls a constant, age and age squared. There is at 10% significance level positive effect of age and small negative effect of square of age on the normal risk aversion measure.

Table 6. group 1 regression

Dependent Variable Normal Risk aversion measure

Variable Coëfficiënt Std. Error Prob.

Constant 24.08641 1.337476 0.0000 Age 0.109511 0.058425 0.0610 Square Age -‐0.000978 0.000593 0.0992

(13)

5. Conclusion

In the used panel there where a lot of inconsistent individuals, table 1 shows that on average the group with inconsistent individuals is more risk avers. The individuals that where inconsistent did most likely not completely understand the questionnaire and then they would tend to choose the safe option. Table 1 also shows that on average all groups where risk avers.

Taken into account education, gender and age table 2 shows that high-‐educated females become more risk seekers in age. The risk aversion level for men stays almost the same in age. For lower educated females I found mixed results. Low educated men get slightly more risk seeking in age. This suggests that education works harder for women than for men. This could be caused by the fact that high educated men and women make their decision based on an estimate of the actual risk so are more likely risk seekers, because men are in general already more risk seekers this effect will be smaller for men. There is in table 2 no good sign of that age is stimulated by education.

This study supports the first part of my hypothesis. In general, males are more likely to be risk takers this founding support several mentioned articles (Barber and Odean, 2001; Hartog, Ferrer-‐i-‐Carbonell & Jonker, 2002; Jianakoplos and Bernasek, 1998). Furthermore, highly educated and older people tend to be less risk averse than low educated and younger individuals. These results are consistent with several previously performed studies (Wang and Hanna, 1997; Jianakoplos and Bernasek, 1998; Barber and Odean, 2001). The effect of age is the opposite of what is found by Bakshi and Chen (1994). In group 1 I have found results of 10% significance that indicate that people get less risk avers in age. In group 2 I found that men are more likely to be risk seekers than women at a 5% significance level and in group 3 I found that highly educated are more likely to be risk seekers at 10% significance level. It is not really strong evidence because it is at 10% significance level. The effect of gender could be explained by the fact that men are more competitive than women. Higher educated individuals are better in making an estimation of the actual risk they are facing so are more likely to be risk takers. When you get older your life expectancy decreases and from a certain age you will have less financial obligations so that could cause that individuals be come more risk seekers.

This article shows that the predicted influence of education on the effect of gender could be true because the coefficients and descriptive statistics point out the predicted effect, however evidence is limited. For group 2 I found that the sign of the coefficients of the measured cross term between gender and education is negative and this is what I expected. This means that the difference between men and women gets smaller when focusing on highly educated individuals. However, this is different in group 1 and 3. In these groups there are no significant results. In all groups the coefficient of the cross term between age and education is positive and the effect of

(14)

age was positive so there was an moderating effect only not the way I expected. Because of the lack of significance it is not a reliable statistic and there cannot be conclude any effect, further research is necessary.

This study suggested that the difference gets smaller for the effect of age. With the results of variable group C and the test results of the redundant variable Test and Omitted Variable Test I found signs that the square of age could be included into the model. Table 6 shows significant results for the square term of age. This supports what I expected and means that the change in risk aversion is expected to decrease in age, however the effect is limited.

Parts of my hypothesis are support by the results. However, because of the lack of significance in this study my hypothesis should be investigated further.

(15)

6. Discussion

Since the large amount of inconsequent participants I have my questions about the description of the questionnaire. It seems a lot of people did not understand the tasks given. I compared the Measuring Higher Order Risk Attitudes of the general Population panel with True Risk

Preferences panel that is also managed by CentERdata. This panel gave on average different results. This is remarkable as the exact same participants filled in both questionnaires. The Liss panel listed that the data was representative for the Dutch population. I think this is not true because all participants did get paid, so probably a lot did not participate if they did not get paid and so is this group is not representative for the whole population. In the panel there was one questions where participants could scale their risk aversion level this did not match the scores deducted by the answers of the lottery questions. These arguments give reason to doubt the reliability of the panel that I used.

The amount of payments plays an important role. With really high payments participants would tend to chose the safe option since there is a lot on stake, when the payments are low the participants will take more risk because they care less if they lose the small amount. Also the differences between the payments play an important role to determine clear differences. I think the steps of 5 euros that where used are too small, what gives small difference between

participant’s and that is one of the reasons that causes the lack of significant results.

The effect of age should be investigated by following individuals for a period of time. In the data that I used there could be a generation effect.

I also question if the lottery question method that is used to measure risk aversion is a proper method to identify someone risk aversion measure. It still is not clear what is the best way to measure someone’s risk aversion level, this is one of the main reasons why there is still a lot knowledge to gain from research in this field. Further research using different panels and different methods is therefore necessary.

(16)

7. References:

Bakshi, G.S. & Chen, Z. (1994). Baby boom, population aging, and capital markets. Journal of

Business, 67(2), 163-‐202.

Barber, B.M. & Odean, T. (2002). Boys Will be Boys. Oxford Journals, Quaterly Journal of

Economics. 116(1), 261-‐292.

Croson, R. & Gneezy, U. (2009). Gender Differences in Preferences. Journal of Economic

Literature, 47(2), 448-‐474.

Donkers, B., Melenberg, B. & van Soest, A. (2001). Estimating Risk Attitudes usaing lotteries: A Large Sample Approach. Journal of Risk and Uncertainty, 20(2), 165-‐195

Hartog, J., Ferrer-‐i-‐Carbonell, A. & Jonker, N. (2002). Linking Measured Risk Aversion to Individual Charaxteristics. Kyklos, 55(1), 3-‐26.

Jianakoplos, N.A. & Bernasek, A. (1998). Are Women More Risk Averse.

Economic Inquiry, XXXVI (1998), 620-‐630.

Morin, R.A. & Suarez, A.F. (1983). Risk Aversion Revisited. The Journal of Finance, 38(4), 1201-‐ 1216.

Peter P. Wakker (2010). Prospect theory: for risk and ambiguity. Cambridge University Press. UK (Chapter 3)

Riley W.B. & Chow K.V. (1992). Asset Allocation and Individual Risk Aversion. Financial Analysts

Journal, 48(6), 32-‐37.

Shaw, K.L. (1996). An Emperical Analysis of Risk Aversion and Income Growth. Journal of Labor

Economics, 14(4), 626-‐653.

Wang, H. & Hanna, S. (1997). Does risk tolerance decrease with age. Financial Counseling and

Planning. 8. 27-‐32.