Do individuals become more risk averse during a recession?

(1)

Do individuals become more risk

averse during a recession?

Ilse Faber

S1983466

University of Groningen

MSc Finance

Supervisor: prof. dr. T.K. Dijkstra

Word count: 10.777

Abstract

This thesis investigates the influence of a recession on people’s financial risk tolerance. The database of the Dutch DNB household survey was used for this study. It distinguishes itself from other studies because it has a combination of long term observations and a sample which forms a reflection of the Dutch population. The main finding is that risk aversion is a rather stable trait. Although the state of the economy has a statistically significant influence on people’s individual risk aversion, this influence is considerably minor.

(2)

1. Introduction

During the period of 2007-2008, one of the most abhorrent economic recessions in history took place. Market indices broke several loss records during that period, the Dutch AEX for example, had a decrease of 52% in the year 2008. (beleggen.com) This crisis made an abundance of investors lose their trust in the financial markets. Several studies have been conducted regarding the effects of a recession on investors, such as that of Weber, Weber & Nosic (2012). They concluded that the risk preferences of British investors were rather stable. In addition to their result, a study of Gerrans, Faff & Harnett (2015) showed similar results. Loss aversion describes the effect that losses loom larger than gains, it is however something different than risk aversion (Ackert and Daves, 2010). The studies of Weber, Weber & Nosic (2012) and Gerrans, Faff & Harnett (2015) make it seem that loss aversion is out of the question. Loss aversion mainly describes the difference between the strength of the feelings that a financial gain generates and the strength of the feelings that a financial loss (of the same amount) generates. However this thesis is about risk aversion. Ackert and Daves (2010) state that “Risk aversion implies that the certainty equivalent, namely a certain wealth level such that you

are indifferent between this wealth level and a particular prospect, is less than the expected value of wealth of the prospect”. For example, if you had the choice between €20 or a 50% chance on €50,

(3)

2. Literature review and hypotheses formulation

2.1 What is a recession?

For this thesis we are not interested in how to predict a recession. We are interested in clearly indicating if and when an economy was in a state of recession during a period in the past. There are many different ways to measure if an economy is in a state of recession, however not all of the methods are equally as good in indicating a recession. A very common used method is indicating the start of a recession as two consecutive quarters of negative growth in GDP and indicating the end of a recession as two consecutive quarters of positive GDP growth. However, this method is in fact a highly inaccurate method. According to Layton and Banerji (2003), this method was not able to indicate three of the nine big recessions since 1948. Also the start and ending points of the

recessions are indicated at the wrong moments at least 40% of the time. Layton and Banerji tested if using the monthly GDP instead of the quarterly GDP helped to improve the method, however it barely helped. The only improvement was that this new method had a smaller deviation from the actual start and ending point of the recession. Then they tested the ability of two coincident indices to indicate the recession. The first one used was the ECRI index, which contains the chain-weighted real GDP, industrial production, real manufacturing and trade sales, real personal income, non-farm employment and the (inverted) unemployment rate. However, a note should be made about the unemployment rate. Sometimes the unemployment rate decreases because the labor force

decreases, this could happen because people give up in their search for a job. (cbs.nl) This can make the unemployment rate inaccurate at times. Since this study is about people’s individual experience of the recession, a number such as the unemployment rate might give a wrong impression of what people actually experience from the unemployment. Layton and Banerji also constructed a similar index which contained the fixed-weight GDP instead of the chain-weighted real GDP. Both of these indices showed convincing improvements compared to the previously used methods. Both of the indices did not miss any of the nine big recessions. Also the deviations from the start and ending points of the recessions were much smaller. Still the start and ending points were placed in the wrong quarters 33% of the time for the chain-weighted GDP coincident index and 28% of the time for the fixed-weight GDP version. We can conclude that these coincident indices are still not perfect in indicating the start and ending point of a recession, however they do not miss any recession, also the average deviations for the start and ending points for these coincident indices are only 0.7 to 0.9 months compared to the 1.2 to 1.5 months for the monthly GDP method.

(4)

defined as: “a significant decline in the level of economic activity, spread across the economy of the

euro area, usually visible in two or more consecutive quarters of negative growth in GDP, employment and other measures of aggregate economic activity for the euro area as a

whole.” (cepr.org) The moment between peak and through is seen as a recession and the time span

between through and peak indicates that the economy is in a state of expansion.

2.2 Literature review

Several studies have been done about the stability of an individual’s Financial Risk Tolerance (FRT). Each of those studies used different time periods over which to test this stability or variability of the FRT. However the studies showed conflicting results, some of them claimed that people’s FRTs were influenced by mood or market sentiment, others reported that individuals have a rather stable FRT. One of the studies is from Grable, Lytton & O’Neill (2004), showing a significant relationship between the closing prices of the NASDAQ and individual’s FRT. It also showed similar results for the DOW JONES and the S&P 500. However this study has several drawbacks. First of all, it was conducted between September 2002 and December 2002, a period of just 4 months. Secondly, each respondent filled in the survey once, so they could never see if an individual’s FRT had changed over time. The only thing the researchers had done was categorize the respondents by the week they had filled in the survey and then look for a relationship between the FRT’s and the market closing prices. Yao and Curl (2011) did a study using a much longer period, 16 years. This study was done with the 1992 – 2006 waves of the Health and Retirement Study (HRS). The HRS interviews its respondents every two years, however three times the FRT question was not included, which only left four waves to use. Secondly, it only included one FRT question, which was related to salary uncertainty. This study also showed a significant positive relationship between S&P 500 returns and the FRT of the respondent. Another study researching the Australian risky retirement investments, by Bateman, Louviere, Thorp, Islam & Satchell (2011), showed that there was “a mild moderating of retirement investor risk

tolerance in 2008”. The study was conducted using two groups who were both representative of the

Australian population. The first group got a survey about different investment choices on march 2007, which was just before the recession started. The other group got the same survey in October 2008, when the recession was at its beginning.

(5)

Weber, Weber & Nosic (2012) showed in their study that there was barely a change in risk attitude during the great financial crisis. This study was conducted between the period of September 2008 and June 2009, which does not show a contrast between pré-crisis years and years of the crisis itself. This study used a survey to measure the 3-month expectations of the market of UK online-brokerage customers. It also asked for the 3-month expectations of their own portfolios and their self-reported risk attitudes. The surveys were filled in every three months, so the participants filled in the survey four times in total. The researchers attempted to get the four surveys filled in by the same investors, however the sample got smaller each time, so during the second round they searched for additional participators. The respondents were demographically not representative for the British population. The study found that the investors’ risk attitudes were fairly stable and that the changes in risk taking were due to changes in risk and return expectations. Hoffman, Post and Pennings (2013) did their research about almost the same period, April 2008 until march 2009. They however showed a conflicting result with that of Weber, Weber & Nosic (2012). They performed surveys each month which showed significant fluctuations in investors’ risk aversion, it was significantly increased during the worst months of the recession and recovered again towards the end of the research period. The most relevant article for this paper is that of Gerrans, Faff & Harnett (2015) For this research a group of investors was taken, who had filled in a certain survey at least twice. This survey was called the FinaMetrica FRT scale, in which 25 questions were used to determine the investor’s FRT. The survey also contained a question in which the respondents were asked to rate their own FRT, also known as the perceived financial risk tolerance (PFRT). The first survey had to be filled in before 2007 and the second survey had to be filled in either before 2007, or in the period 2007 until halfway 2009. The group who had filled in the second survey before 2007 formed the base group, because they had no experience with the recession during both of their surveys. The results showed that in a statistical sense the FRT scores had significantly decreased, this was however a rather minor

decrease. The PFRT showed no significant change at all due to the recession. These results support the view that FRT is a rather stable trait.

2.3 Hypotheses formulation

Previous studies mostly came to the conclusion that FRT is a rather stable trait. However most of these studies were conducted over a fairly short amount of time. Moreover investors were the only subjects for the surveys or the FRT was just measured with one question. This is why there is

(6)

is stable or positively correlated with the market indices, this study will only test for a negative relation between a recession and FRT. This gives us the following hypothesis:

H0: A financial crisis decreases the level of individual financial risk tolerance.

H1: A financial crisis does not affect the level of individual financial risk tolerance.

3. Data, methodology and variables

3.1 Data source

In this thesis the DNB household survey is being used.1_{The DNB household survey is held every year,}

starting from the year 1993, among approximately 2,000 Dutch household. These 2,000 households are specifically selected to form a good reflection of the complete Dutch population. Moreover, the survey contains questions about work, pensions, housing, mortgages, income, possessions, loans, health, economic and psychological concepts, and personal characteristic. The survey contains a section of six questions to measure how risk averse someone is on the area of financial decisions. The advantage of using such questions about risk aversion instead of using real investment data is that the answers are not directly dependent on a household’s income or wealth, they are purely

hypothetical. The data could be used from 1996 until 2015, because 1996 was the first year that the dependent and the control variables were all measured in the survey. In addition the questions have all remained the same until 2015, therefore the above mentioned period is used. Initially this period contained a total of 98.631 observations. More than 60% of the observations were dropped, because this fraction of respondents answered none of the six risk aversion questions. This could mean that the outcomes do not give a representative picture of the Dutch society anymore. It cannot be figured out how much less representative it became, since all respondents were anonymized and we have no way of ascertaining why respondents did not answer those questions. Also several other

observations were dropped because one or several of the control variables were not present for those observations. Table 1 shows the sample selection procedure.

Table1. Sample selection procedure

Initial sample 98,631 person-year observations

Minus: Missing data on risk aversion 58,380 person-year observations

Minus: Missing data on control variables 4,517 person-year observations

(7)

3.2 The measurement of variables

The independent variables

Three independent variables will be used, which will all be tested separately. All of these

independent variables are business cycle indices, however they are all composed in different ways. All three business cycles consist of a number, which by itself does not say anything, but has value when compared to other periods. The reason for this is that there is a so called ‘trend level’, which is indicated by the number zero. This trend level indicates the long term average of the state of the economy. However, if one would have an observation of just one month, there would not be a trend level and that month would automatically be indicated as zero. The indices are calculated for each month.

Ortec Finance, “a global provider of technology and advisory services for risk and return

management”, has provided two business cycle indices, where the first one, to which we will refer in

thesis as ‘Ortec 1’, is an important driver of mid-term equity returns, growth and credit spreads. The second one is a driver of interest rates and inflations, we will refer to this varaible with ‘Ortec 2’. Both of these indices are based on the worldwide economy. (ortec-finance.com) For none of these two indices specific information is available on how they are exactly measured and/or composed. 1

The third business cycle index is the one from DNB, it is based on the Dutch monthly industrial output data. According to tradingecnomics.com this industrial output data “measures the output of

businesses integrated in industrial sector of the economy such as manufacturing, mining, and

utilities”. Since the industrial production is strongly oriented on the international market, it is a good

indicator for the Dutch business cycle, which is strongly influenced by its international environment. (dnb.nl)1

Graph 1 shows the course of all three business cycle indices from 1996 until the end of 2015.

1_{I would like to thank CantERdata, Ortec Finance and DNB and for providing me the data which were}

(8)

Graph 1. business cycle indices

The CEPR defined recession dates could also have been used as an independent variable, however this only tells us if the economy is in a state of recession or not. It misses all the moments when the economy is in near-recession, it misses the several gradations of the states of the economy. An index that would have shown those gradations of the state of the economy is the ECRI index. However, this index was not available for the Netherlands, while it does contain some highly country specific indicators like the unemployment rate. Therefore, it would have been inappropriate to use the worldwide ECRI world index. The indices of Ortec are also worldwide indices, but they contain less country specific indicators. Still, there are also some downsides to the Ortec business cycle indices of course. A number like inflation for example, which was used in the second index of Ortec, can differ greatly worldwide, but within the Eurozone the inflation does not differ that much.

Two other variables were created, based on the DNB business cycle. They reflect the change of the business cycle of the last six months. The first variable indicates the positive change in the business cycle index and has a value of zero when the change is negative. The second variable indicates the negative change and has a value of zero when the change is positive. These two variables combined are used as an independent variable. The reason for creating this fourth independent variable is that it shows the state of the economy compared to a short time ago, instead of compared to a long term average. Also, by splitting it into a variable for positive change and one for negative change, we can determine whether there is difference in the effect on risk aversion.

In general all observations contained data about when the respondent had filled in the survey. However in the years 2000 till 2003, that particular data was not available. The only information available for those years was during which months the complete survey, so for all respondents, was

(9)

conducted. However, the business cycle indices contain monthly data. To solve this problem, the average of the business cycle indices was taken for the months in which the survey was conducted. This could change the outcomes, certainly in rapidly changing times. The volatility of each of those periods can be calculated with the standard deviation. Graph 2 reflects the standard deviation of all three business cycle indices over a period of a year, six months before and six months after the stated month. We can see that the volatility was quite high during 2001 and 2002 for the DNB index and during 2003 it was quite high for both of the Ortec indices. Since the volatility of the DNB index is quite high during the period of 2000 until 2003, it was not possible to make an estimate of the six-months changes of the survey periods of those years.

Graph 2. Volatility of business cycle indices

The dependent variable

The dependent variable is the level of risk aversion with respect to financial decisions. This is measured by the following statements which use a seven-level Likert scale to quantify the levels of risk aversion (totally disagree to totally agree):

1. I think it is more important to have safe investments and guaranteed returns, than to take a risk to have a chance to get the highest possible returns.

2. I do not invest in shares, because I find this too risky.

3. If I think an investment will be profitable, I am prepared to borrow money to make this investment.

4. I want to be certain that my investments are safe.

5. If I want to improve my financial position, I should take financial risks.

(10)

6. I am prepared to take the risk to lose money, when there is also a chance to gain money. Not all answers with a high score correspond to level of risk aversion. Consequently, the answers of questions 3, 5 and 6 are converted so that a high level of risk aversion gives a high score and a low score corresponds to a high level of risk tolerance. This conversion is done using the following formula

Level of risk aversion = (Answer to the question – 8) * (-1) (1)

Some respondents have answered an ‘8’ or a ‘9’ on one or more of these questions, even though it was a seven-level Likert scale. Therefore these answers are excluded. The average of the six questions is then used as one dependent variable in the model, which we call the average risk

aversion.

The control variables

 Age: Ai indicates the age of a respondent. Age could influence someone’s risk aversion. According to Grable (2000) older people are less risk averse. This is why it is used as a control variable in this study. In the DNB household survey age is measured with

1. Year of birth of the respondent

This is then converted in the age. Four categories have been formed to categorize the respondents: 16 – 30 years, 31 – 45 years, 46 – 65 years and older than 65 years. Dummies are created for these four categories.

 Education: Ei indicates the highest level of education someone has attended. Those who have a higher level of education are more risk tolerant (Grable, 2000). Education is measured with the question:

2. Highest level of education attended (regardless of certificate/diploma):

o (Voortgezet) speciaal onderwijs / (continued) special education. o Kleuter-, lager- of basisonderwijs / kindergarten/primary education.

o Voorbereidend middelbaar beroepsonderwijs (VMBO) / pre-vocational education. o HAVO/VWO / pre-university education.

o MBO of het leerlingwezen / senior vocational training or training through apprentice

system.

(11)

o other sort of education/training.

The above mentioned are the categorized into three categories:

low: primary education, special education, no education and other education;

moderate: pre-vocational education, pre-university education and senior vocational training; and high: vocational colleges and university education.

‘Other education’ is an answer that badly describes the level of education. Therefore, every observation where this answer was given, the answers of other years of this person were checked. This was done in an attempt to find different answers that describe the education level more accurately. Dummies are created to reflect the levels of education.

Regarding this control variable, it should be noted that several respondents have filled in a lower level of education than the year before, which should not be possible. This could mean that also other respondents, which have only filled in the survey once or twice, filled in a wrong level of education. This could make this control variable unreliable.

 Gender: Gi indicates someone’s sex. Females are generally indicated as more risk averse. (Grable, 2000). Gender is measured with:

3. Gender of the respondent

With a ‘0’ indicating for a male and a ‘1’ for a female.

 Degree of future orientation, Fi. People who are more future oriented are more risk averse (Howlett et al, 2008)

This is measured by the following statements on a scale from 1 to 7 (extremely uncharacteristic to extremely characteristic) in the DNB household survey:

1. I think about how things can change in the future, and try to influence those things in my everyday life.

2. I often work on things that will only pay off in a couple of years.

3. I am only concerned about the present, because I trust that things will work themselves out in the future.

4. With everything I do, I am only concerned about the immediate consequences (say a period of a couple of days or weeks).

(12)

6. I am willing to sacrifice my well-being in the present to achieve certain goals in the future. 7. I think it is important to take warnings about negative consequences of my acts seriously,

even if these negative consequences would only occur in the distant future.

8. I think it is more important to work on things that have important consequences in the future, than to work on things that have immediate but less important consequences.

9. In general, I ignore warnings about future problems because I think these problems will be solved before they get critical .

10. I think there is no need to sacrifice things now for problems that lie in the future, because it will always be possible to solve these future problems later.

11. I only respond to urgent problems, trusting that problems that come up later can be solved in a later stage.

12. I find it more important to do work that gives short-term results, than work where the consequences are not apparent until later.

Not for all questions a high score corresponds to a high level of future orientation. Therefore, the answers to questions 3, 4, 5, 9, 10, 11 and 12 are converted so that a high score corresponds to a high level of future orientation and a low score to a low level of future orientation. The conversion is done using the following formula

Degree of future orientation = (Answer to the question – 8) * (-1) (2)

Also according to the codebook of the survey, since 2010 these question are only answered once by each respondent, meaning that if they have answered these questions during one of the previous surveys, they are not asked again. However this does not appear to be true, since there are 1816 observations which contain both recent observations, but also the older responses. Sometimes those responses are the same, but approximately two-third of those observations show conflicting

answers. If this was the case, the newest answers to these questions were used. The average of all these twelve questions was then calculated and used as the control variable. In the year 2008, the questions were not included in the survey. However, this research would become less reliable if a whole year would have to be removed because of this. This is why for the year 2008, the average future orientation was replaced by that of the nearest year, preferably 2007. Since this was also the way it was done since the year 2010.

3.3 The empirical model

(13)

Risk aversioni,t = a0 +β1 Businesscyclei,t + β2 future orientationi,t + β3 Agei,t + β4 Educationi,t + β5 Genderi,t

+ εi,t (3)

On the left hand side of the equation is the dependent variable which is indicated by Risk aversioni,t, for person i at time t. The right hand side of the equation starts with a constant term, followed by the independent variable, which is one of the business cycle indices, Ortec 1, Ortec 2 or the DNB index. Since we test them all individually, we just have one term for this in the equation. The independent variable is then followed by the four control variables: degree of future orientation, age, education level and gender. Lastly there is an error term, which is indicated by εi,t. We assume the error term had a mean of zero and a constant variance.

3.4 Methodology

Considering that the date we use is panel data, we cannot use Ordinary Leased Squares (OLS). This is because OLS assumes that observations are independent from each other, which is not the case when a respondent takes part in the survey more than once. A different test which cannot be used is the seemingly unrelated regression (SUR). The SUR can only be used when the number of times series observations, T, per cross-sectional unit is larger than the total number of units. The panel dataset used is unbalanced and contains only 3.5 observations per respondent on average, with a total of 10,071 respondents, making the SUR framework impossible to use. The random and fixed effects models are more appropriate when using panel data. The Hausman test can determine which of the two models should be used. When Prob>chi2 had a value lower than 0.05, the fixed effects model should be used. In this case Prob>chi2 had a value of 0.0000, so the fixed effects model is more appropriate to use. The dependent variable is, because of its scale and the corresponding answers, an ordinal limited variable. This is why we basically should use an ordered logit or probit model. According to Brooks (2014) it does not make a difference to choose one or the other, due to the computational speed nowadays. However, the combination of a fixed effects model with a logit or probit model is not possible in the three main statistical programs (SPSS, Eviews and Stata). This is why the fixed effects model was used, while treating the dependent variable as continuous instead of ordinal. In practice this usually gives reasonable results, however ideally one should test for

(14)

4. Results

4.1 descriptive statistics

Table 2 shows the summery statistics of all the variables used in this study. DNB index, Ortec 1 and

Ortec 2, represent the three different business cycle indices which were used as the independent

(15)

Table 2. Descriptive statistics of all variables used

Variable

Mean

Std. Dev. Min

median

Max

N

Risk aversion

Average risk aversion 5.19997 1.052956 1 5.166667 7 35,734

Squared average risk aversion 28.14837 10.81239 1 26.69444 49 35,734 Logarithm of average risk aversion 1.62604 0.2197866 0 1.642228 1.94591 35,734

Risk 1 5.044019 1.788051 1 5 7 34,985 Risk 2 4.544951 2.054023 1 5 7 35,205 Risk 3 5.734197 1.580863 1 6 7 33,807 Risk 4 5.350725 1.475108 1 6 7 35,150 Risk 5 5.134372 1.659864 1 5 7 35,037 Risk 6 5.398024 1.548104 1 6 7 35,322

Business cycle indices

Ortec 1index 0.1171583 1.011536 -3.09189 0.599857 1.265344 35,734 Ortec 2 index -0.0374679 0.8490113 -2.40642 0.194783 1.549246 35,734 DNB index 0.0043306 1.00973 -2.17849 -0.26 2.60507 35,734 DNB change positive 0.2719988 0.402468 0 0.039592 1.411842 29,154 DNB change negative -0.2423385 0.3989809 -1.783841 0 0 29,155 gender 0.4397493 0.4963635 0 0 1 35,734 Age 16 – 30 years 0.0902502 0.2865439 0 0 1 35,734 31 – 45 years 0.2977836 0.4572903 0 0 1 35,734 46 – 65 years 0.4037891 0.490663 0 0 1 35,734

older than 65 years 0.2081771 0.4060099 0 0 1 35,734

Education

Low education level 0.0611742 0.239653 0 0 1 35,734

Moderate education level 0.4949628 0.4999816 0 0 1 35,734

High education level 0.443779 0.4968361 0 0 1 35,734

Average degree of future orientation

4.122467 0.77919 1 4.090909 7 35,734

(16)

Table 3 shows the means of the answers of the risk aversion questions from the survey, for all of the dummy variables. We can see that women show a higher level of risk aversion for each of the

questions and also for the average risk aversion. Also, risk aversion grows with age. For the education level it is harder to see a trend in the risk aversion level. However, we should take into account that only approximately 6% of the observations had a low education level, which could make the answers which this relatively small group gave unrepresentative.

Table 3. Mean risk aversion per risk aversion variable and the average risk aversion

Variables Average

risk aversion

Risk 1 Risk 2 Risk 3 Risk 4 Risk 5 Risk 6

Gender Men 5.004729 4.945872 4.203416 5.541551 5.289791 4.93234 5.115404 Women 5.43061 5.178624 4.938291 5.971019 5.430904 5.350958 5.713862 Age 16 – 30 years 4.954306 4.85517 4.440216 5.35197 5.047828 4.941394 5.089256 31 – 45 years 5.029888 4.970735 4.371316 5.527812 5.259444 4.901204 5.148817 46 – 65 years 5.255867 5.132407 4.535583 5.801161 5.438073 5.168296 5.459682 older than 65 years 5.395 5.075742 4.753433 6.039274 5.447069 5.390816 5.663665 Education level Low 5.155349 5.026287 4.568456 5.538335 5.445235 5.004929 5.34885 Moderate 5.281133 4.951829 4.814082 5.806627 5.313455 5.213418 5.587384 High 5.095759 5.153327 4.202098 5.667393 5.380492 5.021388 5.149854

Note: this table shows the means for each risk aversion question, for each dummy variable. It also shows the mean of the average risk aversion for each dummy variable.

(17)

Table 4. Correlation matrix

Note: Significance levels: * = 5%; ** = 1%.

Ave rage ris k av ersi on Squa red aver age ris k av ersi on Logar ithm of ave rage ris k av e rsio n Ris k 1 Ris k 2 Ris k 3 Ris k 4 Ris k 5 Ris k 6 Orte c 1 inde x Orte c 2 inde x DN B inde x DN B c hang e po siti ve DN B c hang e ne gat ive gende r 16 – 3 0 ye ars 31 – 4 5 ye ars 46 – 6 5 ye ars olde r tha n 6 5 y ear s Low educ atio n le vel Mode rat e e duc atio n le vel Hig h educ atio n le vel Ave rage de gre e o f fut ure orie nta tio n

Average risk aversion 1

Squared average risk aversion 0,9917** 1

Logarithm of average risk aversion 0,9853** 0,9565** 1

Risk 1 0,587** 0,584** 0,5769** 1 Risk 2 0,6952** 0,6922** 0,6814** 0,2793** 1 Risk 3 0,5476** 0,5345** 0,5516 0,0632** 0,183** 1 Risk 4 0,5639** 0,5684** 0,5466** 0,541** 0,31** 0,0673** 1 Risk 5 0,5817** 0,5809** 0,572** 0,0535** 0,208** 0,3566** 0,0418** 1 Risk 6 0,7275** 0,7128** 0,7279** 0,1838** 0,4388** 0,4177** 0,1594** 0,5292** 1 Ortec 1 index -0,0323** -0,035** -0,0285** -0,0026 -0,0112* -0,0377** -0,0074 -0,0706** -0,0237** 1 Ortec 2 index -0,0097 -0,0118* -0,0065 0,0135* -0,0102 -0,0147** -0,0066 -0,0165** 0,0077 0,6888** 1 DNB index -0,0478** -0,048** -0,0458** -0,0033 -0,0337** -0,0385** -0,0119* -0,0379** -0,0358** -0,0322** 0,3925** 1 DNB change positive 0,0375** 0,0358** 0,039** 0,0202** -0,0163** 0,0442** -0,003 0,065** 0,0298** 0,2238** 0,5217** 0,4001** 1 DNB change negative -0,0253** -0,0275** -0,0226** 0,0018 -0,0288** -0,038** 0,0035 -0,0557** -0,0357** 0,8671** 0,5015** 0,1496** 0,4105** 1 gender 0,1969** 0,1921** 0,1962** 0,0634** 0,1755** 0,1325** 0,0491** 0,1218** 0,1901** 0,0068 0,01 -0,0012 0,0086 0,006 1 16 – 30 years -0,0705** -0,0721** -0,0661** -0,0318** -0,013** -0,076** -0,0626** -0,0353** -0,0593** 0,0004 0,0161** 0,0101 0,0054 0,0068 0,083** 1 31 – 45 years -0,0957** -0,0972** -0,0913** -0,0272** -0,0485** -0,0812** -0,0385** -0,0848** -0,0938** 0,0082 -0,0242** -0,0074 -0,0436** 0,0334** 0,062** -0,2051** 1 46 – 65 years 0,0486** 0,0512** 0,044** 0,038** 0,0049 0,0368** 0,0461** 0,0256** 0,0431** -0,0104 0,0104* 0,0002 0,0334** -0,0195** -0,0191** -0,2592** -0,5359** 1

older than 65 years 0,0988** 0,0984** 0,0963** 0,0071 0,0578** 0,1012** 0,0318** 0,0893** 0,0955** 0,003 0,0033 0,001 0,0041 -0,0176** -0,1053** -0,1615** -0,3339** -0,422** 1

Low education level -0,001 0,0001 -0,0026 -0,0045 0,0122* -0,0267** 0,0144** -0,0149** 0,0035 0,06** -0,0196** 0,0256** -0,0825** 0,0545** 0,0397** -0,0527** -0,0268** 0,0239** 0,0385** 1

Moderate education level 0,0806** 0,0797** 0,0796** -0,0553** 0,1376** 0,0478** -0,028** 0,0563** 0,131** -0,0186** 0,007 -0,0056 0,0376** -0,0201** 0,0749** -0,0217** -0,0597** 0,0559** 0,015** -0,2527** 1

High education level -0,0807** -0,0803** -0,0789** 0,0575** -0,1441** -0,0353** 0,0212** -0,0494** -0,1334** -0,0102 0,0023 -0,007 0,002 -0,0061 -0,0947** 0,0471** 0,0729** -0,0676** -0,0336** -0,228** -0,8843** 1

(18)

4.2 Regression results

For this study we have 4 different independent variables, which are substitutes for each other. This is why we test each of them separately. The four independent variables are: DNB index, Ortec 1, Ortec 2 and the change in the DNB index over the past 6 months, which is split into two variables, a positive change and a negative change. Since this last variable could not be retrieved for the years 2000 until 2003, we only test for this independent variable for the years 2004 until 2015. We repeat this for the other three independent variables, to give a good comparison. However those three independent variables are also tested for the complete sample period. For the dummy variables, age and

(19)

Table 5. Regression results for the years 1996-2015, with DNB index as independent variable.

1996 - 2015 Average risk

aversion

Fixed effects Fixed effects Fixed effects Fixed effects Fixed effects Fixed effects Fixed effects

explanatory variables

(robust st. error) (robust st. error) (robust st. error) (robust st. error) (robust st. error) (robust st. error) (robust st. error)

DNB index -0,044185** -0,001 -0,05669** -0,04971** -0,00913 -0,05785** -0,04382** (0,0037159) (0,007793) (0,008137) (0,007222) (0,00651) (0,007177) (0,005973) Female -0,214067 -0,54267 -0,20444 0,067827 -0,9689 0,292103 0,406192 (0,307212) (0,527065) (0,60736) (0,17097) (0,522695) (0,429125) (0,326078) Age: 31-45 0,0861418* -0,02305 0,27125** 0,326225** 0,070777 -0,16431* 0,054752 (0,038967) (0,066819) (0,078739) (0,060188) (0,057671) (0,065372) (0,057184) Age: 46-65 0,2315719** 0,054829 0,618404** 0,656623** -0,00261 -0,11939 0,293828** (0,047060) (0,081429) (0,098002) (0,075646) (0,069743) (0,080869) (0,0696)

Age: 65 and older 0,2923095** -0,19500* 0,932684** 0,974791** -0,19162* -0,08771 0,486071**

(0,054643) (0,099305) (0,110036) (0,088821) (0,082164) (0,09353) (0,079475) Moderate education level 0,2415876** 0,035087 0,428435** 0,296854** -0,03956 0,433899** 0,405495** (0,042925) (0,084297) (0,086092) (0,06999) (0,061816) (0,071464) (0,064606) High education level 0,3487845** 0,114591 0,546837** 0,395155** 0,012823 0,628796** 0,5236** (0,0536197) (0,097401) (0,106671) (0,090099) (0,077139) (0,087671) (0,080738) Average degree of future orientation -0,028700* 0,053551 -0,06801** -0,0736** 0,016946 -0,0226 -0,06051** (0,011933) (0,023319) (0,024086) (0,019895) (0,018596) (0,021284) (0,017001) R² 0,001900 0,0014 0,0014 0,0269 0,0028 0,0081 0,0493

(20)

Table 6. Regression results for the years 1996-2015, with Ortec 1 as independent variable.

1996 - 2015 Average risk

aversion

Ortec 1 -0,018789** -0,0015718 -0,01535 -0,02767** -0,00975 -0,09499** -0,02555** (0,003979) (0,0080644) (0,008749) (0,007552) (0,006918) (0,007431) (0,006225) Female -0,203506 -0,5424231 -0,19018 0,054716 -0,96645 0,320931 0,417614 (0,307705) (0,5269156) (0,603567) (0,179382) (0,522784) (0,433624) (0,33465) Age: 31-45 0,0889223* -0,0229236 0,273906** 0,331862** 0,071659 -0,15774* 0,057527 (0,038950) (0,066817) (0,078556) (0,06035) (0,057635) (0,065301) (0,057115) Age: 46-65 0,2354204** 0,0551135 0,621451** 0,667552** -0,00079 -0,10249 0,298498** (0,047119) (0,0814304) (0,097927) (0,07583) (0,069707) (0,080973) (0,069631)

Age: 65 and older 0,2996412** -0,1944687* 0,939102** 0,990593** -0,18819* -0,05521 0,495328**

(0,054735) (0,0993513) (0,110029) (0,089003) (0,082125) (0,093366) (0,079538) Moderate education level 0,2516928** 0,0349489 0,444453** 0,311144** -0,0395 0,424001** 0,413735** (0,042923) (0,0841482) (0,086277) (0,069769) (0,061915) (0,070999) (0,064712) High education level 0,358735** 0,1141288 0,56474** 0,407944** 0,011182 0,598886** 0,52949** (0,053747) (0,0973729) (0,106874) (0,090162) (0,077398) (0,087161) (0,081101) Average degree of future orientation -0,026470* 0,0537191* -0,06585** -0,06913** 0,01797 -0,0128 -0,05756** (0,011986) (0,0233585) (0,024155) (0,019984) (0,018581) (0,021302) (0,017061) R² 0,002200 0,0014 0,0000 0,0253 0,0028 0,0132 0,0488

(21)

1996 - 2015 Average risk

aversion

Ortec 2 -0,020834** 0,024414** -0,04057** -0,03573** -0,0012 -0,04917** -0,00188 (0,004628) (0,0095) (0,01029) (0,008187) (0,007823) (0,008696) (0,007469) Female -0,210297 -0,53379 -0,2006 0,050367 -0,9671 0,29859 0,416064 (0,306187) (0,526692) (0,603998) (0,176999) (0,521953) (0,428148) (0,330739) Age: 31-45 0,088386* -0,02361 0,274071** 0,330157** 0,071101 -0,16202* 0,056061 (0,038917) (0,066817) (0,078362) (0,060288) (0,057628) (0,065444) (0,057278) Age: 46-65 0,235374** 0,051182 0,625039** 0,665599** -0,00237 -0,11233 0,294098** (0,047110) (0,081415) (0,097756) (0,075837) (0,06971) (0,081119) (0,069735)

Age: 65 and older 0,297387** -0,1989* 0,941231** 0,985222** -0,19112* -0,07834 0,48736**

(0,054698) (0,099254) (0,109826) (0,088981) (0,082117) (0,093779) (0,079594) Moderate education level 0,260498** 0,032658 0,454374** 0,322726** -0,036 0,462398** 0,422872** (0,042837) (0,084192) (0,08618) (0,069748) (0,061752) (0,071186) (0,064522) High education level 0,368209** 0,117462 0,570203** 0,419117** 0,017113 0,65309** 0,545163** (0,053655) (0,097281) (0,106743) (0,090087) (0,077111) (0,087429) (0,08089) Average degree of future orientation -0,026592* 0,05169* -0,06418** -0,06945** 0,017156 -0,01808 -0,05985** (0,012001) (0,023346) (0,024196) (0,019974) (0,018592) (0,021346) (0,01706) R² 0,002800 0,0013 0,0001 0,0245 0,0029 0,0075 0,0474

(22)

Table 8. Regression results for the years 2004-2015, with DNB index as independent variable.

2004 - 2015 Average risk

aversion

DNB index -0.0248413** 0.012453 -0.03232** -0.03159** 0.004769 -0.03374** -0.01427* (0.0040869) (0.009039) (0.009245) (0.007977) (0.00738) (0.008056) (0.006635) Female -0.6789756 -1.79263** -0.78583 -0.01835 -1.94665** 0.026781 0.477222 (0.4796023) (0.452602) (0.795171) (0.130713) (0.522426) (0.765398) (0.646972) Age: 31-45 0.1159432* -0.04991 0.443827** 0.344581** 0.106741 -0.13308 0.01232 (0.0504310) (0.088976) (0.10386) (0.076789) (0.081063) (0.094557) (0.071513) Age: 46-65 0.2153088** -0.00486 0.768883** 0.57549** 0.035689 -0.17215 0.124257 (0.0590427) (0.106422) (0.128309) (0.097794) (0.095318) (0.110692) (0.086629)

Age: 65 and older 0.1864111** -0.27662 1.009585** 0.756766** -0.21956* -0.36226** 0.195788*

(0.0659234) (0.125647) (0.141937) (0.108686) (0.109129) (0.123596) (0.09708) Moderate education level 0.0155583 -0.55649* 0.209427 -0.17709 -1.27491** 1.033569 0.319241 (0.1988678) (0.38614) (0.33851) (0.409786) (0.404962) (0.489297) (0.502975) High education level 0.0155365 -0.70166 0.369312 -0.2689 -1.17709** 1.149867* 0.156567 (0.2123016) (0.399722) (0.369751) (0.431216) (0.417183) (0.500453) (0.522977) Average degree of future orientation 0.0072620 0.053646 -0.02843 -0.00444 0.022175 0.018576* -0.01042 (0.0158312) (0.033926) (0.034162) (0.027602) (0.027733) (0.031122) (0.021986) R² 0.027500 0.0031 0.0139 0.0193 0.0018 0.0041 0.0678

(23)

2004 - 2015 Average risk

aversion

Ortec 1 -0.018850** 0.0019057 -0.03137** -0.03404** -0.00591 -0.09788** -0.03855** (0.004340) (0.0089626) (0.009684) (0.008141) (0.007685) (0.007964) (0.00682) Female -0.650548 -1.802637** -0.74495 -0.03476 -1.94675** 0.10591 0.509053 (0.491031) (0.4531762) (0.811219) (0.135419) (0.525004) (0.797309) (0.666649) Age: 31-45 0.106015* -0.0478171 0.428252** 0.324343** 0.104944 -0.17505 -0.00417 (0.050610) (0.089096) (0.104278) (0.077374) (0.081392) (0.094133) (0.071262) Age: 46-65 0.199173** -0.00087 0.74406** 0.541900** 0.033533 -0.23402* 0.099809 (0.059182) (0.1065275) (0.128864) (0.098279) (0.095751) (0.110384) (0.086353)

Age: 65 and older 0.174412** -0.2735036* 0.991163** 0.72237** -0.22104 -0.40731* 0.178008

(0.066002) (0.1256386) (0.142287) (0.109219) (0.109397) (0.123206) (0.096775) Moderate education level 0.024064 -0.5595591 0.2215 -0.17622 -1.27515** 1.055627* 0.328149 (0.202778) (0.386397) (0.338811) (0.417197) (0.405728) (0.488404) (0.505179) High education level 0.020355 -0.7041858 0.375414 -0.27194 -1.17827** 1.154438* 0.158561 (0.216065) (0.3998673) (0.369541) (0.438423) (0.41786) (0.499372) (0.524672) Average degree of future orientation 0.008651 0.0531901 -0.02635 -0.00068 0.022219 0.022862 -0.00877 (0.015880) (0.033933) (0.034207) (0.027596) (0.027731) (0.03098) (0.021978) R² 0.027400 0.0031 0.0132 0.0182 0.0018 0,00000 0.0669

(24)

2004 - 2015 Average risk

aversion

(robust st. error) (robust st. error)

(robust st. error) (robust st. error) (robust st. error) (robust st. error) (robust st. error)

Ortec 2 -0.028398** 0.045702** -0.08999** -0.05949** 0.015242 -0.06666** -0.03288** (0.005421) (0.011638) (0.012335) (0.009487) (0.009576) (0.010157) (0.008743) Female -0.659115 -1.80476** -0.75618 -0.06099 -1.95117** 0.055716 0.489782 (0.488815) (0.447527) (0.821843) (0.135475) (0.517508) (0.779491) (0.656641) Age: 31-45 0.097102* -0.02237 0.388979** 0.303102** 0.115955 -0.17484 -0.00798 (0.050426) (0.089291) (0.103797) (0.077286) (0.081352) (0.094061) (0.071547) Age: 46-65 0.179841** 0.04588 0.667213** 0.498268** 0.052761 -0.24973* 0.086577 (0.059036) (0.107091) (0.128613) (0.098568) (0.0963) (0.110265) (0.086751)

Age: 65 and older 0.139610* -0.20627 0.869594** 0.649658** -0.19598 -0.46765** 0.144272

(0.065982) (0.126645) (0.142259) (0.110114) (0.110533) (0.123683) (0.097393) Moderate education level 0.013821 -0.54701 0.193174 -0.19044 -1.27199** 1.023784* 0.313862 (0.200855) (0.386451) (0.335986) (0.412664) (0.404091) (0.487817) (0.503959) High education level 0.005355 -0.67908 0.32712 -0.30016 -1.16979** 1.12069* 0.14166 (0.214215) (0.400113) (0.366245) (0.433935) (0.416124) (0.499234) (0.523734) Average degree of future orientation 0.008263 0.052897 -0.02663 -0.00235 0.021912 0.020163 -0.00974 (0.015856) (0.033867) (0.034107) (0.027517) (0.027712) (0.031087) (0.021983) R² 0.029400 0.0029 0.0131 0.0154 0.0017 0.0037 0.0646

Note: Significance levels: * = 5%; ** = 1%. For each dummy variable a reference category is excluded: gender (male); Age (16-30); education (low).

(25)

2004 - 2015 Average risk

aversion

(robust st. error) (robust st. error) (robust st. error) (robust st. error) (robust st. error) (robust st. error) (robust st. error) Positive change in DNB index 0.012917 0.083949** -0.25935** -0.06238* 0.009299 0.333251** 0.020429 (0.013621) (0.028372) (0.030476) (0.024524) (0.02476) (0.028372) (0.021202) Negative change in DNB index -0.027618* -0.04355 0.055028 -0.08034** -0.01015 -0.34328** -0.0828** (0.014031) (0.028411) (0.029932) (0.027314) (0.024238) (0.026292) (0.021599) female -0.654822 -1.79627** -0.75785 -0.04355 -1.94808** 0.116585 0.507677 (0.486346) (0.45068) (0.816184) (0.136697) (0.523185) (0.777698) (0.663938) Age: 31-45 0.112974* -0.04211 0.415347** 0.324762** 0.107606 -0.12247 0.007683 (0.050547) (0.089249) (0.102476) (0.076922) (0.081125) (0.094295) (0.0713) Age: 46-65 0.210425** 0.016878 0.695517** 0.533676** 0.03836 -0.1218 0.118662* (0.059151) (0.106741) (0.127107) (0.098221) (0.095395) (0.11063) (0.08663)

Age: 65 and older 0.187958** -0.23154 0.869629** 0.695687** -0.21408 -0.20291 0.201991

(0.066334) (0.126955) (0.141001) (0.109503) (0.109642) (0.123407) (0.097549) Moderate education level 0.024640 -0.54874 0.191858 -0.18438 -1.27443** 1.096299* 0.331052 (0.203094) (0.379947) (0.318904) (0.409502) (0.405633) (0.512902) (0.5056) High education level 0.024002 -0.6874 0.327352 -0.28409 -1.17612** 1.227622* 0.165673 (0.216631) (0.393901) (0.351157) (0.430833) (0.417722) (0.524711) (0.525307) Average degree of future orientation 0.008242 0.054972 -0.03284 -0.00311 0.022204 0.025994 -0.00978 (0.015879) (0.033922) (0.034127) (0.027606) (0.027753) (0.031021) (0.022005) R² 0.027400 0.003 0.0123 0.0172 0.0018 0.0031 0.0678

(26)

The regression results show that for the majority of the models, the different independent variables are significantly negatively related to the different dependent variables which were tested. Only ‘Risk 4’ does not show any significant results for any of the business cycle indices. Risk 4 belongs to the statement: ‘I want to be sure my investments are safe’. Also the business cycle indices are always significantly negatively related to the average risk aversion. Another import result to notice is that the R2_{are noticeably higher for the period 2004 until 2015, than for the period 1996 until 2015. This}

means that the created models are a better fit for this time period. However, the most important thing to notice is that values of the coefficients of the business cycle indices are rather low. For example, the coefficients of the age dummy variables have much higher values. Meaning that the business cycle indices have very little predicting value of the individual’s risk aversion level. This is in line with hypothesis H1, since a recession has little to no effect on people’s individual risk aversion. When paying attention to the control variables, we can see that gender generally does not give a significant result, even though we earlier observed that for all the risk variables and the average risk aversion, the mean value of risk aversion for women was higher than for men. When we look at the age dummy variables we can see they all show significant results for the average risk aversion, compared to the reference category (age: 16 – 30). When focusing on the education variables, we can observe that they are only significant in the time period of 1996 – 2015 and not for the period 2004-2015. This is also the case for the average degree of future orientation.

5. Further analysis

The sample was also tested for autocorrelation. The hypothesis for this test was:

H0: No first-order autocorrelation

H1: first-order autocorrelation is present

(27)

approximately the same. The same test was also conducted for the squared average risk aversion, which gave approximately the same results, regarding to significance and R-square, as the Log average risk aversion, therefore the results of these regression results are not included in this thesis. This squared average risk aversion was calculated in two different ways. The first method squared the variable average risk aversion. The second method first squared each of the six separate risk variables and then the average of those squared variables was taken. Both squared risk aversions were tested separately, but gave approximately the same results. This method of calculating was also done for the Log average risk aversion, which did not change the results given in table 12 either.

Note: Significance levels: * = 5%; ** = 1%. For each dummy variable a reference category is excluded: gender

(male); Age (16-30); education (low).

Table 12. Regression results for the years 2004-2015, with the logarithm of the average risk aversion as the dependent

variable

combined

2004 - 2015

independent

variable DNB index Ortec 1 Ortec 2

Positive change in DNB index Negative change in DNB index dependent variable = Log

average risk aversion Fixed effects Fixed effects Fixed effects Fixed effects Fixed effects

explanatory variables

(robust st. error) (robust st. error) (robust st. error) (robust st. error) (robust st. error) Independent variable -0,00528** -0,00302** -0,0053** 0,00179 -0,00334 (0,000832) (0,000874) (0,001089) (0,002739) (0,002853) Gender -0,11348 -0,10799 -0,10931 0,095419 (0,094337) (0,096348) (0,096118) (0,095419) Age: 31-45 0,023248* 0,021522* 0,019667 0,010113* (0,010085) (0,010126) (0,010091) (0,010113) Age: 46-65 0,042731** 0,039843** 0,035967** 0,011923** (0,011895) (0,011927) (0,011906) (0,011923)

Age: 65 and older 0,039028** 0,036867** 0,030182* 0,01334**

(0,013261) (0,013281) (0,013279) (0,01334)

Moderate education level 0,008273 0,009936 0,008099 0,041339

(0,040554) (0,041327) (0,040994) (0,041339)

High education level 0,009602 0,010652 0,007842 0,04398

(0,04316) (0,043924) (0,043594) (0,04398) Average degree of future

orientation 0,002401 0,002663 0,002607 0,00327

(0,003261) (0,003271) (0,003266) (0,00327)

(28)

6. Discussion

In this section the results of the regression analyses will be discussed. Also possible explanations for those results will be given.

R-square

When we look at the regression results with the average level of risk aversion as dependent variable, that for all the models the independent variable shows a significant result. However for the models in the period 1996-2015, all the R-squares are below one percent. Meaning that the models basically are not able to predict anything about the average level of risk aversion. Though, the results of the models with time period 2004-2015 all have R-squares of two to three percent. A possible

explanation for this difference could maybe be found if we look at the per year mean of the average risk aversion, which can be found in graph 3.

Graph 3. Mean per year of the average risk aversion

In graph 3 we can see that in the year 2000, there is a relatively large decrease in the level of risk aversion. It is also notable that 1999 and 2001 have almost the same level of risk aversion and that the year 2000 is an outlier. When we look at the period of 2004-2015, we can see that the

fluctuations stay within an area of just 0,2, on the initial Likert-scale of 1 till 7. While the period of 1996 until 2004 has fluctuations within the area of 0,6. While, when we look at graph 1 and 2, we can

(29)

see that the fluctuations in the business cycle indices are larger for the period 2004-2015, than for the period 1996-2004. This might give an explanation for why the models seem to be a better fit for the period 2004-2015, instead of the whole sample period.

A different explanation for the differences in R-squares could be the fact that the exact dates on when the respondents had filled in the surveys were unknown during the period of 2000 until 2003. This has led to the fact that there was not an exact fit anymore between the business cycle data and the risk aversion. This may have meant that there was a mismatch between a potential cause and consequence relationship of the business cycle and risk aversion levels of individuals.

Some other R-squares that stand out compared to that the other models, are the R-squares of the models with ‘risk 6’ as dependent variable. The models in the period 1996-2015 have R-squares of four to five percent. The models with the time period of 2004-2015, have an R-square of six to seven percent. The ‘risk 6’ variable represents the statement: ‘I am prepared to take the risk to lose money,

when there is also a chance to gain money’. The answers were reversed to make a high score

represent a high level of risk aversion. Control variables

When we look at the control variables, we can see that gender only gives significant results for ‘risk1’ and ‘risk 4’ and only for the period 2004-2015. Those were both statements about safe investments. If we look at the coefficients for the models of ‘risk 1’ and ‘risk 4’ we can see they all have a negative sign, meaning that being female has a negative effect on the level of risk aversion. This is contrary to the study of Grable (2000).

The age dummy variables give many significant results for all the average risk aversion models and also for all the models with ‘risk 2’ and ‘risk 3’. Since the coefficients are all positive for those models, we can conclude from this that risk aversion increases with age, which is the opposite of what the study of Grable(2000) found.

(30)

Graph 4. Distribution of education level per year

For the last control variable, average degree of future orientation, there are again clear differences between the models of the different time periods. None of the models for the time period 2004-2015 give a significant result. But for the 1996- 2015 period the models for the average risk aversion all give significant results, also some of the models for the separate risk factors give significant results. Independent variables

This study used three independent variables, DNB index, Ortec 1 and Ortec 2. These are all

substitutes for each other and were thus tested separately from each other. A fourth independent variable was created with the DNB index. This variable reflects the change of the last 6 months in the DNB index. It is split into two separate variables, one that reflects the positive change and one that reflects the negative change. The results for all four independent variables were all significant for the average risk aversion. In the case of the change variable, this was only significant for the negative change. Also all the coefficients of the independent variables had a relatively low value with a negative sign, meaning that when the business cycle index decreases, the level of risk aversion slightly increases. This supports the findings of Gerrans, Faff & Harnett (2015) that there is a

(31)

Main results

The results of this thesis regarding the influence of a recession on financial risk aversion are in line with the findings of Gerrans, Faff & Harnett (2015). The influence of a recession on people’s

individual risk aversion show statistically significant results, however the influence is negligibly small. This suggests that risk aversion is a rather stable trait. This is also in line with the study of Bateman, Louviere, Thorp, Islam & Satchell (2011), which showed a mild moderating effect on people’s risk tolerance. Weber, Weber & Nosic (2012) showed that the same effect applies to British investors. When we look at graph 3 and focus on the difference in average risk aversion in 2008, we can see that this has increased from 5.2 to 5.4, an increase of 0.2 on a scale of 1 to 7. This is an increase of just 3,33% (₍₇₋₁₎0.2 ∗ 100%). Graph 5 shows that the risk aversion is rather stable. The exact same data were used as in graph 3, however in this graph the complete 1 to 7 scale was used as Y-axis.

Graph 5. Mean per year of the average risk aversion

All in all, we can conclude that most studies found that there were statistically significant increases in the level of risk aversion due to a recession, but that these increases are rather small. This study agrees with these conclusions.

7. Conclusion

This thesis started by explaining the disposition effect, the phenomenon that investors sell their winning stocks too fast and that they hold on to their losing stocks too long. Though this was not the case during recessions, investors sold their stocks as soon as possible when the market dropped. However in this thesis we researched the effects of a recession on the risk aversion of the complete

(32)

population and not the behavior of investors. The research question of this thesis was: Do individuals

become more risk averse during a recession? To help us answer this question we analyzed the DNB

household survey for the years 1996 until 2015. We also used three different business cycle indices to reflect the state of the economy. Some major differences were found between the two different time periods tested, 1996-2015 and 2004-2015. Overall for this second time period the model fitted better. Still the R-squares were rather low, meaning that the model was a bad fit. Though the coefficients of the independent variables were always significant, the coefficients had a rather low value, meaning that the business cycle indices predict very little about the average risk aversion. This means that we can reject the null-hypothesis and that we can conclude that a recession does not affect an individual’s financial risk tolerance.

This thesis contributes to the behavioral finance literature by measuring the influence of the state of the economy on risk aversion while focusing on a complete population over a relatively long period of time. Also it used four different independent variables to represent the economy, which were all tested separately from each other. Like most studies on this subject it concluded that a recession has little to no influence on people’s individual risk tolerance.

Limitations & recommendations

The complete sample with which we started, was supposed to represent the complete Dutch population. As mentioned in section three, approximately 60% of the person-year observations had to be removed from the sample. Mainly because the respondent had not filled in any of the risk aversion questions, or because one or more of the control variables were missing. Removing such a considerable part of the sample may have a substantial effect on the representativeness. When we look at the age distribution per year of the used sample, in graph 6, it proves that the sample is not representative anymore.

In graph 6 we can see relatively large fluctuations within the distributions of the different age

categories. It is not uncommon that the proportion of a certain age category rises or falls 10% within a period of just one or two years. This is not possible in the real population of course. From the distribution of the education level, in graph 4, we could also already conclude that the sample was not representative, since such a sudden rise in education level is also not possible in a real

(33)

Graph 6. Age distribution per year

Another representativeness issue is that of the indices used to measure the state of the economy. Do those indices reflect people’s feelings about the economy well enough? The ECRI index used by Layton and Banerji (2003), is a relatively good index and is composed of several different and important indicators for the state of the economy. However this index was not available for the Netherlands. Since the ECRI index contains quite country specific numbers like unemployment rate, it would have been inappropriate to use in a worldwide or European form. Instead we used the two indices of Ortec and the DNB index. One issue with the indices of Ortec is that they are based on the worldwide economy, while this study was done with only respondents living in the Netherlands. However in contrast to the ECRI index, the indicators used in these indices are less country specific, such as interest rates and credit spreads. Also the DNB index might not give a good indication for people’s mood towards the economy, since it is solely based on the Dutch industrial production. In this thesis we merely focused on the Dutch population, because of cultural and demographic reasons these results cannot be generalized. Therefore additional research is needed in order to give a clear picture on whether the situation worldwide gives the same results. Secondly, future research is also needed to generate knowledge about whether there is a difference between the influence of a recession on the risk aversion of investors and on non-investors. Lastly, an aspect that could be improved in further research is the business cycle index, it should be able to reflect the feelings of the population about the state of the economy.

I would like to thank DNB, Ortec Finance and Centerdata for providing me the data which were needed for this thesis.

(34)

References

Ackert, L. F., Deaves, R., 2010. Behavioural finance, psychology, decision-making, and markets. South-western cengage leaming, Mason, OH, USA.

Bateman, H., Louviere, J., Thorp, S., Islam, T., Satchell, S., 2011. Retirement investor risk tolerance in tranquil and crisis periods: experimental survey evidence, Journal of Behavioral Finance 12, 201–218.

Bellofatto, A., De Winne, R., D’Hondt, C., 2014. Beyond the disposition effect: evidence from the 1999-2012 period. Unpublished working paper. Louvain school of management.

Brooks, C., 2014. Introductory econometrics for finance. Cambridge university press, Cambridge. Gerrans, P., Faff, R., Hartnett, N., 2015. Individual financial risk tolerance and the global financial crisis, Accounting & Finance v55 (1), 165 – 185.

Grable, J.E., 2000. Financial risk tolerance and additional factors that affect risk taking in everyday money matters, Journal of business and psychology, Vol. 14, No. 4, 625-630.

Grable, J., Lytton, R., O’Neill, B., 2004. Projection bias and financial risk tolerance, Journal of Behavioral Finance 5, 142–147.

Hoffmann, A. O., Post, T., Pennings, J. M. 2013. Individual investor perceptions and behavior during the financial crisis. Journal of Banking & Finance, 37(1), 60-74.

Huei-Wen, L., 2011. Does the disposition effect exhibit during financial crisis? International Conference on Economics and Finance Research, vol. 4, 6-10.

(35)

Weber, M., Weber, E., Nosic, A., 2012. Who takes risks when and why: determinants of changes in investor risk taking, Review of Finance 17, 847–883.

Yao, R., Curl, A., 2011. Do market returns influence risk tolerance? Evidence from panel data, Journal of Family and Economic Issues 32, 532–544.

Zweig, J. (2007) Your Money and Your Brain: How the New Science of Neuroeconomics Can Help Make You Rich, Simon & Schuster, New York.