The impact of risk aversion on the savings
behaviour of Dutch households at different
stages of the business cycle
MSc Thesis, supervised by dr. A. Plantinga
Janiek Moes
ABSTRACT
1. Introduction
Based on the rising saving deposits, DNB (‘The Dutch Bank’, the central bank of the Netherlands) reports that it is tempting to conclude that the Dutch are thrifty savers (“Saving in the Netherlands”, 2014). According to DNB, the savings surplus was low for a long time compared to the 1980s. Since the 2008 recession, the savings surplus is relatively high again. In this study, we investigate whether the impact of risk aversion and the desire to decrease uncertainty causes Dutch households to exhibit different savings behaviour at different stages in the business cycle.
Savings behaviour is measured both as the level of savings (the savings rate) and the percentage allocated to risky assets (the risky share). According to Knothek and Khan (2011), a depressed economy is characterized by higher uncertainty. A typical response to higher uncertainty is increased savings (Caballero, 1990; Gourio, 2012; De Paoli and Zabczyk, 2013). Habit formation models suggest countercyclical risk aversion and increased precautionary savings (Abel, 1990; Campbell and Cochrane; Chan & Kogan, 2002; Brunnermeier and Nagel, 2008; Xiouros and Zapatero, 2010). The buffer-stock theory of savings closely relates to this. Carroll (1992) provides
a model based on that theory, and explains that an increase in uncertainty results in a higher target
buffer stock. In order to maintain the target, individuals increase savings. The availability heuristic
increases this perceived uncertainty in a depressed economy when media attention regarding
uncertainty is increased (Tversky and Kahneman, 1973; Folkes, 1988; Ackert and Deaves, 2009). Then, events that come easier to mind have a higher believed likelihood of occurring.
On the other hand, neoclassical economics assumes that individuals make rational decisions (Roy Weintraub, 1993). Following this view, risk aversion has a constant impact on the savings behaviour in different stages in the business cycle. Risk aversion negatively correlates with income, wealth and education level (Riley and Chow, 1992). Households with the highest risk aversion are typically the ones with the lowest income, value of wealth and education level. Such households are the ones who save the most for precautionary reasons. In periods with the highest
level of uncertainty, they could prefer to save the most, however, they may have the least to spend.
The largest part of total household savings belongs to households with the highest income, wealth and education level. Precautionary savings is not their primary reason to save, since they do not experience high level of uncertainty (Kimball, 1990).
income and wealth. Nevertheless, existing literature shows stronger support for increased savings
due to risk aversion. Therefore, the first hypothesis is:
Hypothesis 1: The impact of the self-assessed risk aversion of a household on the savings rate is stronger in depressed states of the economy compared to the impact in normal states of the economy
The second part of the study investigates if the percentage allocated to risky assets differs in different stages of the business cycle. A higher level of risk aversion increases the preference to decrease uncertainty and risk. Risk averse households will relatively allocate a lower percentage to risky assets compared to risk tolerant households. The ‘flight to safety’ hypothesis suggests a stronger influence of risk aversion on allocation of risky assets in depressed states of the economy compared to normal states of the economy. This stronger influence is caused by an excessive reaction to uncertainty by decreasing risky asset exposure (Callabero and Kurlat, 2008; Bertaut and Pounder, 2009; Apergis, 2014). Other studies that find a decreased percentage allocated to
risky assets due to increased risk aversion in a depressed economy are: Amromin and Sharpe
(2008), Chai et al. (2011); Weber et al. (2012); and Bayer (2013). Most of these studies find a
negative correlation between perceived risk and risky share.
However, there are also reasons to believe that the impact of risk aversion is stronger in
normal states of the economy instead of depressed states of the economy. Many households hold
very simple portfolios (Bertaut and Starr-Mccluer, 2002) and do not invest in risky assets at all,
especially households with a high level of risk aversion (Van Rooij et al., 2011). In depressed
states of the economy, stock prices are generally lower (Boubakari and Jin, 2010; Case et al., 2005), which directly influences the value of risky assets. The impact of bad stock prices on
households with a high level of risk aversion is small, since they typically have zero or a small
percentage invested in risky assets. (Von Gaudecker, 2015). In normal states of the economy,
stocks perform better in general, which causes households holding risky assets to end up
wealthier.
This point of view suggests a larger difference between the percentage allocated to risky assets of risk averse and risk tolerant households in normal states of the economy compared to depressed states of the economy.
Hypothesis 2: The impact of self-assessed risk aversion on the allocation of risky assets is
stronger in normal states of the economy compared to depressed states of the economy
We estimate the first model using a cross-sectional random effects regression. The dependent variable is the savings rate (savings per year divided by net income) and the explanatory variables are risk aversion and some control variables. We estimate all explanatory variables for both depressed states of the economy and normal states of the economy. Next, we perform Wald tests on the coefficient estimates, to test for their difference in different stages of the business cycle. Last, we perform some robustness checks.
We estimate the second model using a cross-sectional fixed effects regression. The dependent variable is the risky share (total risky assets divided by total asset value). The explanatory variables are risk aversion and some control variables. Again, we perform Wald tests to find if the coefficient estimates are different in normal states of the economy from the coefficient estimates in depressed states of the economy.
The main contribution of this paper is providing insights into the savings behaviour of
Dutch households at different points in the business cycle. In normal states of the economy, the
savings rate of risk averse households is higher than the savings rate of risk tolerant households. In depressed states of the economy, there is no significant difference between the savings rate of risk averse and risk tolerant households. If we include outliers, we find a significantly lower savings rate for risk averse households compared to risk tolerant households in depressed states of the economy. We conclude that all households will start to save more in depressed states of the economy, under the condition that they are financially able to. Households with a high level of
risk aversion typically have less income, wealth, and a lower education degree. While they prefer
to increase their savings rate, some of them are not financially able to. Furthermore, risk averse households always save for precautionary reasons, while risk tolerant households only start to save for precautionary reasons in depressed states of the economy.
The second contribution is to provide insights in the influence of risk aversion on the
relative allocation to risky assets at different points in the business cycle. This study shows that
2. Literature review
2.1 Savings rate
Theories that suggest a stronger influence of risk aversion on the savings rate in depressed states of the economy are related to precautionary savings, habit formation models, buffer-stock, and heuristics. On the other hand, theories suggesting a constant impact of risk aversion are neoclassical models, life-cycle hypothesis of saving, permanent income theories, and the general finding that risk aversion negatively correlates with income, wealth and education.
Mastrogiacomo and Alessie (2013) find that about 30% of the savings of Dutch households is for precautionary motives. For their calculations, they used self-reported measurements of uncertainty, as well as actual income uncertainty. Precautionary savings rise with income uncertainty, which is first shown by Leland (1968). He combines Pratt’s principle of decreasing absolute risk aversion in combination with a utility function including income and consumption to show this result. Zeldes (1989) confirms this result for constant relative risk aversion.
Knotek and Khan (2011) measure uncertainty in two ways: first, based on monthly
appearances of the words “uncertainty” or “uncertain” in articles in large newspapers. Second,
they measure uncertainty derived from monthly stock market volatility. They find that those two measures of uncertainty tend to move with each other. Episodes of economic and financial disruption are marked with an increase in uncertainty measures. Knothek and Khan (2011) conclude that, in general, high levels of uncertainty characterize depressed states of the economy, while a lower level of uncertainty characterizes normal states of the economy.
Caballero (1990) shows that uncertainty doubles the standard deviation of labour income, and triples precautionary savings. He finds that risk aversion increases precautionary savings. The studies of Caballero (1990) and Knotek and Khan (2011) combined suggest a stronger influence of risk aversion on precautionary savings in economic turndowns compared to normal states of the economy.
view using the habit formation model. They show that external habits generate cyclical swings in risk aversion due to uncertainty. Increased risk aversion results in increased precautionary savings.
If risk aversion increases in economic turndowns, this is being referred to as “countercyclical risk aversion”. Countercyclical risk aversion is a standard result of the habit
formation model, which is introduced by Abel (1990) (other work showing this result is that of
Campbell and Cochrane, 1999; Chan & Kogan, 2002; Brunnermeier and Nagel, 2008; and Xiouros and Zapatero, 2010). The habit formation model is a utility function, which incorporates past consumption, current consumption, relative risk aversion and an index of the importance of habits. This model explains a higher importance of the change in consumption rather than the absolute level of consumption. Time-varying risk aversion is generated by the difference between consumption and the habit. According to Campbell and Cochrane (1999), habit persistence can explain why economic recessions are so much feared: individuals are afraid of changes in their consumption.
The buffer-stock theory of saving is closely related to the idea of precautionary savings. In this model, consumers hold assets against a target, mainly with the purpose of shielding their consumption against uncertainty in income. Buffer stock savers have a target wealth-to-permanent income ratio. Permanent income is derived from the permanent income hypothesis (Friedman, 1957) and is equal to the long-term average income. If wealth is below the target, the individual
will start to save precautionary. If the wealth is above the target, the individual will start to dis
-save (Carroll, 1997). Carroll (1992) provides a model, which explains that consumers express a
greater desire to save in economic turndowns. He finds that an increase in the possibility of unemployment (or other uncertainty measurements) has a major impact on savings behaviour. When uncertainty increases, consumers increase the target buffer stock and thus save more for precautionary motives.
Other empirical work supporting a higher savings rate in depressed states of the economy due to risk aversion is that of Alan et al. (2012), Crossly et al. (2013), Adema and Pozzi (2015).
expect a stronger risk aversion since the media covers the topics related to uncertainty more often (Knotek and Khan, 2011).
However, other authors predict a constant influence of risk aversion in different economic situations. Neoclassical economics predict that individuals have rational preferences, and that they maximize utility and make independent decisions based on all relevant information (Roy Weintraub, 1993). According to that theory, risk aversion has a constant impact on financial decisions in different stages of the business cycle. From this perspective, media attention or increased perceived uncertainty alone will not cause an increase in the savings rate.
According to the life-cycle hypothesis of saving (developed by Modigliani and Brumberg, 1954), individuals plan to spread out their consumption and savings as smooth as possible over their life cycle. The theory assumes a homogeneous utility function with respect to consumption.
The permanent income hypothesis (developed by Friedman, 1957) is similar to the life-cycle hypothesis of saving, and predicts that the consumption of an individual is determined by the expected long-term average income. As with the life-cycle hypothesis of saving, individuals want to spread out their consumption. According to these theories, households will not quickly change their consumption/savings pattern.
As stated before, many authors prove that uncertainty increases precautionary savings. However, not all households experience uncertainty: the ones with high incomes or high values of wealth will not save primarily for precautionary motives (Kimball, 1990). A higher level of risk aversion increases precautionary savings, especially in times with higher uncertainty. Riley and Chow (1992) find that risk aversion negatively correlates with income, wealth and education level. Households with the highest risk aversion are typically the ones with the lowest income, wealth and education level. Households with the highest incomes generate the largest part of total savings, so a relatively small part of total household savings exists is driven by precautionary motives (Dardanoni, 1991). Therefore, the buffer-stock model and precautionary savings motive might not be a plausible description of the behaviour of wealthy households (Carroll, 1996). Concluding, households who are most likely to save for precautionary reasons seem to be the ones
with the lowest incomes and values of wealth. While they might prefer to increase the savings
rate, they are financially not able to save more in response to uncertainty.
2.2 Allocation to risky assets
The second part of this study is about the allocation to risky assets. We explore if risk aversion has
on the risky share in depressed states of the economy. In the following section, we discuss literature suggesting a countercyclical impact of risk aversion on the risky share. Second, we discuss literature suggesting a constant influence of risk aversion.
Risk aversion impacts the proportion of a household’s total wealth allocated to risky assets. Individuals with a higher level of risk aversion dislike risk and invest relatively less in risky assets. An increase in uncertainty causes stronger risk aversion in depressed states of the economy (Knotek and Khan, 2011). Media attention increases risk aversion through the availability heuristic (Tversky and Kahneman, 1973; Ackert and Deaves, 2009).
The ‘flight to safety’ hypothesis refers to unusual and unexpected events causing investors to experience extreme risk- and uncertainty aversion. An example of an unexpected event is increased uncertainty in depressed states of the economy. This results in a shift to safe investments and selling risky investments (Callabero and Kurlat, 2008; Bertaut and Pounder, 2009). Apergis
(2014) examines the influence of different stages of the business cycle on the risky share. He finds
that investors decrease risky investments faster during recessions than they increase them during economic booms. Thus, investor responses to uncertainty are not linear, but are stronger in recessions.
Amromin and Sharpe (2008) find that perceived risk and pessimism unduly rise during
recessions. Also, they find that expected return is pro-cyclical: expectations of better economic
performance are associated with higher expected stock return. They find that equity exposure negatively correlates with risk perception and positively correlates with self-reported expected return. This is supported by Weber et al. (2012) and Bayer (2013), who find that subjective feelings about future market risks and returns as a result of recent economic events causes changes in risk taking over time. Chai et al (2011) find that during immediate crisis, households reduce their equity exposure by more than 20%, on average, in favour of risk-free bonds. Lee et al. (2015) find a negative interaction between bad stock market expectations and risk aversion in determining whether households are entering the stock market. The chance of households who are risk averse and do not already own risky assets have an even lower chance of entering the stock market in economic turndowns. From this point of view, increased uncertainty causes a stronger impact of risk aversion. Therefore, the difference between the risky share of risk averse and risk tolerant
households seem to be larger in depressed states of the economy.
However, there are also reasons to believe that the difference in the risky share of risk
tolerant en risk averse households is larger in normal states of the economy than in depressed
correlated with national consumption (Case et al., 2005). Riley and Chow (1992) find that risk aversion negatively correlates with wealth, education, and income. Von Gaudecker (2015) finds that percentage allocation to risky assets rises with wealth. Therefore, the business cycle has a higher impact on total asset value of risk tolerant households, since stock prices are correlated to GDP growth (Boubakari and Jin, 2010).
Risk tolerant households hold relatively more risky assets compared to risk averse
households, which causes them to end up wealthier in normal states of the economy in comparison
to depressed states of the economy (Xiouros and Zapatero, 2010). Furthermore, many households, especially the ones with a high level of risk aversion, do not participate in risky asset markets at all (Calvet et al., 2006; Von Gaudecker, 2015). Then, the risky share of risk tolerant households is the only driver of the difference between the risky share of risk tolerant and risk averse households. Therefore, if risk averse households do not participate in the risky asset market, the difference between the risky share of risk tolerant and risk averse household is larger in normal states of the economy.
2.3 Control variables
Individual characteristics affect households’ savings behaviour (Riley and Chow, 1992; Barber and Odean, 2001). This section discusses the influence of individual characteristics on both the savings rate and risky share.
A standard empirical result is that females have a higher level of risk aversion compared to
males (Riley and Chow, 1992; Barber and Odean, 2000). This suggests that females have a higher savings rate but lower risky share (Sapienza et al., 2009; Hibbert et al., 2013; etc.). However, Hibbert et al. (2013) find that the gender-differences in risk aversion is absent when individuals have the same level of financial education, after controlling for age, income, and some other characteristics.
The education level is one of the most important determinants of risk aversion. Typically, college-educated households are less risk averse (Bertout and Star-McCluer, 2000), invest more in stocks (van Rooij et al., 2012), and own more assets in all categories (transaction accounts, vehicles, residential real estate, business and non-residential real estate, retirement accounts, stocks/bonds/mutual funds, other assets)(Merry and Thomas 2014).
background risk, which reduces demand for risky assets. Older households have built up their assets, and can allocate higher shares of their financial assets to risky assets.
On the other hand, Body and Crane (1997) find that age does not have a significant effect
on the amount of risky assets, but rather the amount of wealth, which is usually larger at a higher
age. Also, the elderly have a higher chance of long-term unemployment in economic turndown periods relative to the young workforce (de Graaf-Zijl, 2015). In case of unemployment, they have less time to adjust their total saving (Chai et al., 2011). This can cause a stronger response to increased uncertainty at a higher age.
3. Data and Methodology
3.1 Dataset
This study analyses the savings behaviour of Dutch households using the Dutch Household survey (DHS). This survey consists of a questionnaire filled in by about 1,500 Dutch households each year from the period 1994-2014. The DHS is sponsored by the Central Bank and administered by CentERdata, which is associated with Tilburg University, the Netherlands. This survey obtains economic and psychological data through five sub-questionnaires released throughout the year. The goal of the survey is to study psychological and economic aspects of financial behaviour. The survey includes information about work and pensions, housing and mortgages, income, assets and debts, health, as well as individual characteristics. Most households only participate for a few years. Also, not all respondents complete the survey, or do not respond to some questions. This causes an unbalanced panel.
The unrestricted dataset includes 95,921 observations, consisting of 89,953 households. All monetary variables, such as the asset values and net income, are aggregated within the household (following Dimmock and Kouwenberg, 2010). The monetary variables measured until 2002 are corrected for the guilders-euro exchange rate. For all other variables, we use responses of the person responsible for the household finances. We assume that the respondent is responsible for the households finances if he/she answers “yes” to the following question: “Are you the person
who is most involved with the financial administration of the household?”. We exclude
respondents under the age of 18. Also, we exclude households with negative total asset values. The restricted sample includes 22,149 observations.
3.2 Savings rate
We examine the impact of risk aversion on the savings rate, given a number of demographic and economic variables.
In particular, the model is given in Eq. (1) as follows:
𝑆𝑎𝑣𝑖𝑛𝑔𝑠 𝑟𝑎𝑡𝑒!" = 𝛼 + 𝐷! 𝛽!𝑅𝐴!"+ 𝛽!𝐴𝐺!"+ 𝛽!𝐺𝐸!"+ 𝛽!𝐸𝐷!"+ 𝛽!𝐸𝐶!" +
𝐷! 𝛽!𝑅𝐴!"+ 𝛽!𝐴𝐺!"+ 𝛽!𝐺𝐸!"+ 𝛽!𝐸𝐷!"+ 𝛽!𝐸𝐶!" + 𝑢!"
With:
- 𝑠𝑎𝑣𝑖𝑛𝑔𝑠 𝑟𝑎𝑡𝑒: the amount of savings per year divided by total net income. Savings is derived from the answer to the question: “About how much money has your household put
aside in the past 12 months?”. The answer is given on a discrete scale with seven tiers
between 0 and more than 75,000 euros. We create a continuous variable using the central range of each of the seven options, following Bucciol and Veronesi (2014).
Savings as a percentage of income is a better estimator of differences in the savings behaviour compared to the absolute value of savings, because savings are driven by net income (Belke et al., 2012)
- 𝐷!: normal state of the economy, with 𝐷!= 1 for the years 1996-2001 and 2004-2007 and
0 otherwise;
- 𝐷!: depressed state of the economy, with 𝐷! = 1 for the years 2002, 2003 and 2008-2014
and 0 otherwise;
- 𝑥!" : level of risk aversion. Following Alessi et al. (2002), Dimmock and Kouwenberg
(2010), Bucciol and Miniaci (2013) and Von Gaudecker (2015), we derive this variable from a set of five questions, which reflect whether the respondent agrees with the statement that it is more important to have save returns to financials investments than to take some risk to get more return. We use principle component factor analysis to combine the questions to one variable. The scaling is set to normalize loadings so the scores have variances equal to the eigenvalues of the decomposition. A high level of risk aversion is represented by a dummy variable, which takes a value of one for all respondents with higher than average risk aversion. The questions are presented in figure 1 in the appendix.
- 𝑥!" : gender of the respondent. We create a dummy variable which takes a value of 1 if the
respondent is male and 0 if she is female (following the findings of Riley and Chow, 1992; Barber and Odean, 2001l Sapienza et al., 2009 and Hibbert et al., 2013; Yilmazer and Lich, 2013);
- 𝑥!": age of the respondent (following Alessie et al., 2002; Dimmock and Kouwenberg,
2010);
- 𝑥!": education of the respondent; we create dummy variables representing high and low
levels of education. (Bertout and Star-McCluer, 2000; Van Rooij et al., 2012; Merry and Thomas, 2014);
- 𝑥!": economic situation of the household, which is the level in which the household can
easily pay the bills. We base this variable on the answer on the following question: “How
well can you manage on the total income of your household?”. We create two dummy
variables that represent a good financial situation and a bad financial situation.
We estimate the coefficient estimates for risk aversion and the control variables separately for normal and depressed states of the economy. Prior to the estimation of the model, we construct a correlation matrix to test for multicollinearity. We conclude that none of the explanatory variables are highly correlated.
To estimate our model, we check whether we should use a fixed or random effects regression. If we run a redundant fixed effects test, we find that the p-values associated with test statistics are zero, indicating that a pooled regression cannot be employed. Fixed effects have the attractive feature of controlling for all stable characteristics of the individuals, whether measured or not. However, a drawback of using fixed effects is time-invariant variables dropping out of the model. An alternative is the random effects model, which is sometimes known as the error components model. We perform a Hausman test on the estimation of Eq. (1) with cross-section random effects, which does not show significant results. This indicates that using a cross-section random effects regression (OLS) is appropriate.
To test if the coefficient estimates are significantly different among the different stages of the business cycle, we perform Wald tests. More specifically, we perform a Wald test on the following null and alternative hypotheses:
H0: 𝐷!𝛽! = 𝐷!𝛽!
If the p-value of the Wald test is equal to zero, we conclude that self-assessed risk aversion has a significantly different impact on the savings rate in normal and depressed states of the economy. To test for other causes of a different savings rate, we perform similar Wald tests on the control variables.
For robust outcomes, we also estimate some variations of Eq. (1). First, we estimate Eq. (1) by excluding households with the 2.5% highest and 2.5% lowest savings rates. By excluding those households, we control for outliers in the savings rate. Next, we only include risk aversion, gender, age, and education. Then, we estimate Eq. (1) while only including the years with the highest and lowest GDP growth (the first and fourth quartile). We only include the years 1997-2000, 2002, 2003, 2007, 2009 and 2012-2012 in the sample.
Last, we estimate the model without the business cycle dummy variables, using a cross-section random effects regression. In particular, we estimate Eq. (1b):
𝑆𝑎𝑣𝑖𝑛𝑔𝑠 𝑟𝑎𝑡𝑒!" = 𝛼 + 𝛽!𝑅𝐴!"+ 𝛽!𝐴𝐺!"+ 𝛽!𝐺𝐸!"+ 𝛽!𝐸𝐷!"+ 𝛽!𝐸𝐶!"+ 𝑢!"
3.3 Risky share
The estimation and analysis of Eq. (2) is similar to the estimation and analysis of Eq. (1). We examine the impact of risk aversion and some control variables on the risky share. In particular, the model is given in Eq. (2) as follows:
𝑟𝑖𝑠𝑘𝑦 𝑠ℎ𝑎𝑟𝑒!" = 𝛼 + 𝐷! 𝛽!𝑅𝐴!"+ 𝛽!𝐴𝐺!"+ 𝛽!𝐺𝐸!"+ 𝛽!𝐸𝐷!"+ 𝛽!𝐸𝐶!" +
𝐷! 𝛽!𝑅𝐴!"+ 𝛽!𝐴𝐺!"+ 𝛽!𝐺𝐸!"+ 𝛽!𝐸𝐷!"+ 𝛽!𝐸𝐶!" + 𝑢!"
With:
- 𝑟𝑖𝑠𝑘𝑦 𝑠ℎ𝑎𝑟𝑒!" ∶ risky assets as a share of total asset value. We divide the total value of
shares, mutual funds and options by total asset value. The total asset value sums up the balance of checking accounts, saving accounts, deposits or certificates, net real estate value, luxury assets such as cars or boats, and business accounts as well as balances of stocks, bonds, and mutual funds, plus some the other assets.
We estimate the coefficient estimates for risk aversion and the control variables separately for normal and depressed states of the economy. Prior to the estimation of the model, we construct a
correlation matrix to test for multicollinearity. We conclude that none of the explanatory variables are highly correlated.
To estimate our model, we check whether we should use a fixed or random effects regression. If we run a redundant fixed effects test, we find that the p-values associated with test statistics are equal to zero. This indicates that a pooled regression cannot be employed. Next we estimate the model by employing random effects. If we perform a Hausman test, we gain a p-value equal to zero. We conclude that using a cross-section fixed effects regression (OLS) is preferred.
After estimating (2), we perform Wald tests on coefficient estimates of risk aversion and the control variables. The outcomes of the Wald test will show is the coefficient estimates are equal in depressed states of the economy and normal states of the economy.
More specifically, we test the following null and alternative hypotheses:
H0: 𝐷!𝛽! = 𝐷!𝛽!
H1: 𝐷!𝛽! ≠ 𝐷!𝛽!
Again, we estimate some variants for Eq. (2) as a robustness check. First, we exclude households with the 2.5% highest and 2.5% lowest risky shares. By excluding households with the highest and lowest risky shares, we control for outliers. Next, we estimate Eq. (2) by only including risk aversion, age, gender, and education. Then, we estimate Eq. (2) while only including the years with the highest and lowest GDP growth (the first and fourth quartile).
Last, we estimate the model without the business cycle dummy variables, using a cross-section random effects regression. In particular, we estimate Eq. (2b):
4. Results
4.1 Descriptive statistics
Table 2 shows the demographics of the sample. The sample includes individuals who are responsible for the household finance only. We exclude respondents with an age under 18, and respondents with negative asset values. The average age of the respondents is 52 years old and 67.3% is male. 47.0% of the respondents have followed a high education, and 19.3% of the respondents have followed a low education. About half of the households can manage their financials very well, and 15.3% of the households have troubles managing on the household income.
Table 2: Descriptive statistics of participants’ characteristics
Observations* Mean Age 22,149 52.221 Risk aversion 18,430 high 12,041 65.3% low 5,129 27.8% Gender 41,753 male 28,084 67.3% female 13,669 32.7% Education 22,131 high 10,394 47.0% medium 5,284 23.9% low 4,275 19.3% other 2,178 9.8% Economic situation 19,785 good 9,809 49.6% normal 6,942 35.1% bad 3,034 15.3%
*the panel is unbalanced, because some participants do not participate in every survey or do not respond to some of the questions.
Table 3 shows the summary statistics of the monetary variables after controlling for age and negative asset values. None of the variables are normally distributed. The mean net income of the sample is €26,399.69 per year, with mean savings of €3,080.77 per year. The mean savings rate equals 14.3%. The mean total asset value equals €112,993.80. The mean risky asset value equals €11,579.54, so the mean percentage allocated to risky assets is 6.1%.
Table 3: Summary statistics monetary variables
Mean Maximum Minimum Std. Dev. Observations
4.2 Savings rate
In this section, we show main results from the estimation of Eq. (1). We explore if and how the savings rate differs in different stages in the business cycle. The coefficients estimates of Eq. (1) corresponding outcomes of the Wald tests are presented in table 4.
We first have a look at the coefficients of risk aversion in normal and depressed states of the economy. We see that being risk averse in a period of high economic growth has no significant effect on the savings rate. However, risk averse household has a significantly lower savings rate in depressed states of the economy compared to a risk tolerant household. If we perform a Wald test on the two coefficients for risk aversion, we gain highly significant results. This corresponds to our first hypothesis:
Hypothesis 1: The impact of the self-assessed risk aversion of a household on the savings rate is stronger in depressed states of the economy compared to normal states of the economy.
We conclude that self-assessed risk aversion has a stronger impact on the savings rate in depressed states of the economy compared to normal states of the economy. However, we find that this impact is negative, which is in contrast with our predictions. The precautionary savings motive, the habit formation model and buffer stock model predict a stronger positive influence of risk aversion on the savings rate in depressed states of the economy.
If we look at the control variables, we see a significant difference in the impact of age and gender on the savings rate in different stages of the business cycle. A higher age increases the savings rate in normal states of the economy, but has no significant impact in depressed states of the economy. We conclude the same for gender: males have a significantly lower saving rate than females in normal states of the economy. In depressed states of the economy, there is no difference in the savings rate among males and females. Education does not seem to have a strong significant influence on the savings rate. Furthermore, the savings rate significantly increases if a household has much to spend (on the 1% level) in all stages of the business cycle. Having troubles to pay
bills does not seem to influence the savings rate.
If we look at the regression diagnostics, we see that the Durbin-Watson statistic is equal to
1.262, so there is no prove of autocorrelation. A Jarque-Bera test shows that the residuals are not
normally distributed (JB=1.400*109, p=0.000). Since we have a sufficient number of observations,
Table 4: Cross-section random regression coefficients for Eq. (1), with R2=0.499%
Dependent variable: savings rate Normal economy Depressed economy Wald t-stat Risk Aversion (1=high) 0.003 -0.112*** 2.786***
Age -0.002 -0.003*** 1.797*
Gender (1=male) -0.200*** -0.023 -3.962***
Education (1=high) -0.026 0.025 -1.109
Education (1=low) 0.010 0.021 -0.189
Ability to pay bills (1=easy) 0.048* 0.076** -0.674 Ability to pay bills (1=hard) -0.105** -0.075 -0.411
Constant 0.369***
* significant at 10% level ** significant at 5% level *** significant at 1% level
Next, we run some robustness tests in order to check the validity of the results. First, we estimate Eq. (1) by excluding the households with the 2.5% highest and 2.5% lowest savings rate.
Next, we estimate Eq. (1) by first excluding the variables related to the ease in which a household can pay its bills. Next, Finally, we estimate Eq. (1) by excluding the years with the most average GDP growth.
If we estimate Eq. (1) by excluding households with the 2.5% highest and 2.5% lowest savings rate, we find slightly different results. First, risk averse households have significantly higher savings rates compared to risk tolerant households in normal states of the economy. The results are presented in table 5. In depressed states of the economy, we do not observe significant differences between risk averse and risk tolerant households. A Wald test on the coefficients is highly significant. Therefore, if we exclude outliers, the difference in savings rate between risk averse and risk tolerant households is significantly larger in normal states of the economy compared to depressed states of the economy. In table 4, we observe other results. There, risk averse households have a significantly lower savings rate compared to risk averse households in depressed states of the economy. It is plausible to assume that the households with the 2.5% lowest savings rates are mainly risk averse, causing different coefficient estimates in table 5.
financial situation and the rest is larger in depressed states of the economy compared to normal states of the economy. This can either indicate that households with a good financial situation save relatively more, or that all other households save less in depressed states in the economy. However, having a bad financial situation in depressed states of the economy does not significantly impact the savings rate. Therefore, we conclude that households with a good financial situation have a significantly higher savings rate in depressed states of the economy compared to normal states of the economy.
Table 5: Cross-section random effects regression coefficients for Eq. (1), with R2=4.491%, respondents with 2.5% highest and lowest savings rate allocation excluded
Dependent variable: savings rate Normal economy Depressed economy Wald t-stat
Risk Aversion (1=high) 0.007* 0.004 12.040***
Age 0.000 0.000 -0.085
Gender (1=male) -0.056*** -0.022*** -10.709***
Education (1=high) 0.012** 0.014** -0.343
Education (1=low) 0.018*** 0.013 0.517
Ability to pay bills (1=easy) 0.032*** 0.055*** -3.559*** Ability to pay bills (1=hard) -0.027*** -0.013 -0.943
Constant 0.136***
* significant at 10% level ** significant at 5% level *** significant at 1% level
We now estimate Eq. (1), which only includes risk aversion, age, gender, and education. If we do not exclude households with the 2.5% highest and lowest savings rates, we gain similar coefficient estimations and Wald test results compared to the full model. The results are shown in table 6 in the appendix.
If we only include the years with the highest and lowest GDP growth (the first and fourth
quartile) we gain different results, which are presented in table 7. Risk aversion does not have a
If we estimate Eq. (1b), which excludes the dummy variables for the business cycle, we find that risk averse households have a higher savings rate compared to risk tolerant households. Also, the savings rate is higher for females, households with a higher age, and a good financial situation. Having a poor financial situation decreases the savings rate. The results are shown in table 8 in the appendix.
4.3 Risky share
In this section, we provide and discuss the main results of Eq. (2) in order to explore if and how risk aversion impacts the percentage allocated to risky assets. We use a cross section fixed effects regression to estimate Eq. (2). The results are shown in table 8. We find that risk aversion has a negative, significant impact on the percentage allocated to risky assets in normal states of the economy. In depressed states of the economy, the risky share is not significantly different among risk averse and risk tolerant households. A Wald test does not prove that the impact of risk
aversion on the percentage allocated to risky assets is significantly different among different
stages of the business cycle. We can neither reject nor accept the second hypothesis:
Hypothesis 2: The impact of self-assessed risk aversion on the allocation of risky assets is stronger in normal states of the economy compared to depressed states of the economy
In depressed states of the economy, we do not find any significant difference between the risky shares of risk averse and risk tolerant households. This can be the result of the insignificant Wald test.
We find that a higher age decreases the risky share in general, which is in contrast with
Riley and Chow (1992), Alessie et al. (2002) and Dimmock and Kouwenberg. Males own relatively less risky assets than females in both states of the economy, which also is inconsistent with Riley and Chow (1992), Barber and Odean (2001), and Sapienza et al., (2009).
Respondents with a both a high and a low education relatively invest less in risky assets in
normal states of the economy compared to respondents with an average education. Again, this is in contrast with previous research, which suggests that households with a higher education relatively invest more in risky assets (Riley and Chow, 1992; Bertaut and Star-McCluer, 2000; Van Rooij, 2012).
In depressed states of the economy, we do not find any significant difference between the risky share of risk averse and risk tolerant households.
p=0.000). Since our sample is sufficiently large, non-normality is not considered to be a problem. The Durbin Watson statistic equals 1.635, which means that there are no signs of autocorrelation. If we estimate Eq. (2) using White consistent standard errors, we gain similar results as provided in table 9.
Table 9: Regression coefficients for Eq. (2), with R2=72.610%
Dependent variable: risky share Normal economy Depressed economy Wald t-stat
Risk Aversion (1=high) -0.007** -0.003 -1.000
Age -0.002*** -0.002*** 1.759*
Gender (1=male) -0.185*** -0.185*** -0.026
Education (1=high) -0.009** -0.006 -0.711
Education (1=low) -0.015** -0.009 -0.920
Ability to pay bills (1=easy) -0.004 -0.002 -0.409 Ability to pay bills (1=hard) 0.004 0.002 0.249
Constant 0.309***
* significant at 10% level ** significant at 5% level *** significant at 1% level
As robustness check, we again estimate Eq. (2) by excluding households with 2.5% highest and 2.5% lowest percentage allocated to risky assets. The results are shown in table 10.
Table 10: Cross-section fixed effects regression coefficients for Eq. (2), with R2=76.589%, respondents with 2.5% highest and lowest savings rate allocation excluded
Dependent variable: risky share Normal economy Depressed economy Wald t-stat
Risk Aversion (1=high) -0.008** -0.001 -1.389
Age -0.001*** -0.002*** 1.528
Gender (1=male) -0.005 -0.003 -0.351
Education (1=high) -0.025*** -0.022** -0.507
Education (1=low) -0.016 -0.019 0.324
Ability to pay bills (1=easy) -0.004 0.002 -1.000 Ability to pay bills (1=hard) -0.001 0.006 -0.427
Constant 0.166* * significant at 10% level ** significant at 5% level *** significant at 1% level
5. Discussion
This section will discuss the results gained from Eq. (1) and Eq. (2). Furthermore, we will discuss possible explanations for our findings.
5.1. Savings rate
The precautionary savings motive (Caballero, 1990; Knotek and Khan, 2011; Gourio, 2012), habit formation model (Abel, 1990; Campbell and Cochrane, 1999; Chan and Kogan, 2002; Brunnermeier and Nagel, 2008; Xiouros and Zapatero, 2010; De Paoli and Zabczyk, 2013), buffer-stock theory of saving (Carroll, 1997), and availability heuristic (Tversky and Kahneman, 1973; Folkes, 1988; Ackert and Deaves, 2009) all predict higher savings in depressed states of the economy due to increased uncertainty. However, we find that risk averse households have a significantly lower savings rate in depressed states of the economy. In the following section, we provide two possible explanations for our finding. First, risk aversion is correlated to wealth (Riley and Chow, 1992), so risk averse households have less income and wealth compared to risk
tolerant households in general. Financial difficulties could be a possible explanation for risk
averse households to have lower saving rates in depressed states of the economy, which causes a lower savings rate. The second explanation is a relatively stronger response to uncertainty by risk tolerant households compared to risk averse households. Risk averse households always have a high savings rate for precautionary reasons, while risk tolerant households only save for precautionary reasons times with high uncertainty.
5.1.1 Risk aversion is correlated to wealth
A negative correlation between risk aversion and wealth can explain a decreased savings rate by risk averse households (Riley and Chow, 1992). Households with lower levels of income and wealth can suffer from depressed states of the economy by having less to spend, forcing their savings rate down. If we estimate Eq. (1) by excluding households with the 2.5% highest and 2.5% lowest savings rates, our results change (see table 6). In normal states of the economy,
households with a high level of risk aversion do have a significantly higher savings rates
because they have less to spend compared to normal states of the economy. That can be the cause
of the significant negative difference between the savings rate of risk averse and risk tolerant
households. Risk aversion itself does not cause a lower savings rate.
5.1.2 Risk tolerant households respond stronger to increased uncertainty
A second explanation for our finding is a stronger response by risk tolerant households to increased uncertainty, compared to risk tolerant households. It is possible that our theoretical framework is valid, but has a stronger influence on risk tolerant households. In that case, risk tolerant households increase their savings rate in response to increased uncertainty, while risk averse households keep a stable savings rate. In other words, risk tolerant households only save for precautionary reasons in depressed states in the economy, while risk averse households always do. The change in savings behaviour can be explained by the availability heuristic (Tversky &
Kahneman, 1973; Ackert & Deaves, 2009). Increased perceived uncertainty affects the savings
behaviour of risk tolerant households unconsciously, while risk tolerant households always save as much as they can.
This explanation is supported by the results in table 6, which excludes households with the 2.5% highest and 2.5% lowest savings rates. In depressed states of the economy, there is no significant difference in the savings rate of risk averse and risk tolerant households. However, in normal states of the economy, risk averse households have a significantly higher savings rate. The result of a Wald test is significant, which proves a significant difference in the savings rate of risk averse and risk tolerant households between different stages of the business cycle. Furthermore, males are typically more risk averse than females (Riley and Chow, 1992; Barber and Odean, 2001). If we look at the coefficient estimates and outcomes of the Wald test for gender (provided in table 6), we see that the outcomes are similar to the outcomes of risk aversion.
If we take a closer look to the results in table 6, we see that a good financial situation of the household significantly increases the savings rate. A Wald test on the two coefficient estimates is significant on the 1% level. This indicates that households with a good financial situation save more in depressed state of the economy compared to a normal state of the economy.
by the habit formation model (Campbell and Cochrane, 1999; Chan and Kogan, 2002; Brunnermeier and Nagel, 2008; and Xiouros and Zapatero, 2010).
5.2 Risky share
We find that risk averse households have a significantly lower risky share compared to risk tolerant households in normal states of the economy. However, in depressed states of the economy, we do not find significant differences between the risky share of risk averse and risk tolerant households. Also, a Wald test does not support a significant difference in the impact of risk aversion on the risky share. In this section, we provide three explanations for this finding. First, nonparticipation in the risky asset market can cause a lack of significant results. Second, households liquidating their risky assets can cause insignificant results in depressed states of the economy. Then, all households have a very small risky share, which explains the insignificant difference of the risky share of risk averse and risk tolerant households. Third, a fixed effects specification may not be the most suitable model, because the intercept captures too much of the variance in the risky share.
5.2.1 Undiversified portfolios and nonparticipation in the risky asset market
Economic theory suggests that households should invest their wealth in a combination of cash and a well-diversified portfolio. However, many households hold a very undiversified portfolio with a few assets, or do not participate in risky asset markets at all (Calvet et al., 2006; Von Gaudecker, 2015). In other words, most households, especially risk averse households, do not respond to uncertainty changes at all, since they are not very involved in the risky asset market.
5.2.2 Liquidation of risky assets
The ‘flight to safety’ hypothesis can explain insignificant difference between the risky share of risk averse and risk tolerant households in depressed states of the economy. Flight to safety refers to a shift to safe investments, due to unusual and unexpected events such as an economic recession (Callabero & Kurlat, 2008; Bertaut & Pounder, 2009).
20% in response to immediate crisis. A very small risky share among all households then causes the insignificant difference between the risky share of risk averse and risk tolerant households.
5.2.3 Fixed effects model is not optimal estimation method
We estimate Eq. (2) using a cross-section fixed effects model, because the results of both a redundant fixed effects test and Hausman test are highly significant. Significant outcomes on the redundant fixed effects test and Hausman tests suggest that the fixed effects specification is to be preferred. Fixed effects impose time independent effects for each entity that are possibly correlated with the regressors. These fixed effects greatly reduce the chance that a relationship is driven by an omitted variable. Statistically, we should use the fixed effects methods in estimating Eq. (2), since the redundant fixed effects test and Hausman test both show significant results. However, the results on the control variables are contradicting to previous research. The results estimated by a pooled regression model are more consistent with previous research, and are provided in table 14 in the appendix. We expect that the intercept of the fixed effects model captures too much of the variation in the risky share.
6. Conclusion
In this study, we investigate if self-assessed risk aversion causes differences in savings behaviour
at different stages of the business cycle. In depressed states of the economy, GDP growth is low
while savings are high (“Saving in the Netherlands.”). The goal of this study is to investigate if households save more in depressed states of the economy in response to increased uncertainty. The second part of this study investigates if risk aversion has a different impact on the risky share in different stages of the business cycle.
We find that risk aversion does have a significant influence on the savings rate in different
stages in the business cycle. We expected risk averse households to respond stronger to increased uncertainty in depressed states of the economy. However, we find that the difference between the savings rates of risk averse and risk tolerant households is larger in normal states of the economy. If we pool the years together and exclude households with the 2.5% highest and 2.5% lowest savings rates, we find that risk averse households do have a significantly higher savings rate compared to risk tolerant households.
Therefore, we conclude that risk tolerant households respond stronger to increased
uncertainty by increasing their savings rate. Our findings can be explained by the correlation
between risk aversion and wealth. On average, risk averse households have a lower income and
total asset value compared to risk tolerant households (Riley and Chow, 1992). In depressed states of the economy, many households with lower incomes and wealth will have less to spend. While they might prefer to increase their savings rate in response to increased uncertainty, they are financially not able to.
On the other hand, risk tolerant households may not save for precautionary reasons in normal states of the economy. When uncertainty increases, they can decide to increase their buffer stock by saving more (Carroll, 1992; Carroll, 1997), because they are afraid of changes in their consumption pattern (see literature on habit formation models: Campbell and Cochrane, 1999; Chan & Kogan, 2002; Brunnermeier & Nagel, 2008; and Xiouros & Zapatero, 2010). Then, the
difference between the savings rate of risk averse and risk tolerant households has decreased. We
find that households with a good financial situation significantly increase their savings rate in
We conclude that all households will start to save more in depressed states of the economy, under the condition that they are financially able to.
As expected, we find that risk averse households have a lower risky share compared to risk
tolerant households. We do not find evidence that the difference in percentage allocated to risky assets between risk averse and risk tolerant households is higher in normal states of the economy
compared to the percentage allocated in depressed states of the economy. When we employ a
fixed effects regression, we find coefficient estimates for the control variables that are not in line with previous research. We expect that this is due to the intercept capturing too much of the variation in the risky share. When we estimate Eq. (2) using a pooled regression, we find results that are more consistent with previous research. We find that the risky share is higher for males,
and for respondents with a higher age, education and good financial situation.
There are some limitations regarding the dataset of this study. First, most of the households only participate for a few years. This makes the panel regression less powerful. Furthermore, the
dependent variable of Eq. (1) is the savings rate. This variable is calculated through dividing the
amount of savings per year by the net income of the household. However, the question regarding the amount of savings is answered on a discrete scale with seven tiers. We converted this variable to a continuous scale, which makes the data less precise. Small changes in the savings rate cannot be measured.
We provide evidence that the difference between the savings rate of risk averse and risk
tolerant households is larger in normal states of the economy compared to the difference in
7. References
Abel, A. B., 1990. Asset prices under habit formation and catching up with the Joneses. American Economic Review, 80(2), 38.
Ackert, L., Deaves, R., 2009. Behavioral finance: Psychology, decision-making, and markets. Cengage Learning.
Alessie, R., Lusardi, A., 1997. Consumption, saving and habit formation. Economics letters, 55(1), 103-108.
Alessie, R. H., Hochgürtel, S. S., Van Soest, A., 2002. Household portfolios in the Netherlands. Household Portfolios, 341-388.
Amromin, G., Sharpe, S. A., 2008. Expectations of Risk and Return Among Household Investors: Are Their Sharpe Ratios Countercyclical?. Working Papers -- U.S. Federal Reserve Board's Finance & Economic Discussion Series, 1-33.
Ando, A., Modigliani, F., 1963. The" life cycle" hypothesis of saving: Aggregate implications and tests. The American economic review, 55-84.
Bayar, Y., 2013. The Effects of Global Financial Crisis on Behavior of Individual Investors: Evidence from Borsa Istanbul. Journal of Economic and Social Research, 15(1), 55-72.
Barber, B. M., Odean, T., 2000. Trading Is Hazardous to Your Wealth: The Common Stock Investment Performance of Individual Investors. Journal Of Finance, 55(2), 773-806.
Belke, A., Dreger, C., Ochmann, R., 2012. Do wealthier households save more? The impact of the demographic factor. International Economics and Economic Policy, 1-11.
Bertaut, C. C., Starr, M., 2000. Household portfolios in the United States.
Bloemen, H. G., 2002. The relation between wealth and labour market transitions: an empirical study for the Netherlands. Journal of Applied Econometrics 17.3 249-268.
Bodie, Z., Crane, D. B., 1997. Personal investing: Advice, theory, and evidence. Financial Analysts Journal, 53(6), 13-23.
Boubakari, A., Jin, D., 2010. The Role of Stock Market Development in Economic Growth: Evidence from Some Euronext Countries. International Journal Of Financial Research, 1(1), 14-20.
Bucciol, A., Miniaci, R., 2013. Financial Risk Attitude, Business Cycles and Perceived Risk Exposure.
Brumberg, R. E., 1956. An approximation to the aggregate saving function. Economic Journal, 6666-72.
Caballero, R. Krishnamurthy, A., 2008. Collective risk management in a Flight to quality episode. Journal of Finance 63, 2195-2230
Caballero, R. J., 1990. Consumption puzzles and precautionary savings. Journal of monetary economics, 25(1), 113-136.
Caballero, R. J., & Kurlat, P. D., 2008. Flight to quality and bailouts: Policy remarks and a literature review. Working paper
Calvet, L. E., Campbell, J. Y., & Sodini, P., 2006. Down or out: Assessing the welfare costs of household investment mistakes (No. 12030). National Bureau of Economic Research.
Carroll, C. D., 1997. Buffer-Stock Saving and the Life Cycle/Permanent Income Hypothesis. Quarterly Journal Of Economics, 112(1), 1-55.
Carroll, C. D., Weil, D. N., 1994. Saving and growth: a reinterpretation. In Carnegie-Rochester Conference Series on Public Policy (Vol. 40, pp. 133-192). North-Holland.
Carroll, C. D., Overland, J., Weil, D. N., 2000. Saving and growth with habit formation. American Economic Review, 341-355.
Case, K. E., Quigley, J. M., Shiller, R. J., 2005. Comparing wealth effects: the stock market versus the housing market. Advances in macroeconomics, 5(1).
Chai, J., Maurer, R., Mitchell, O. S., & Rogalla, R., 2011. Lifecycle impacts of the financial and economic crisis on household optimal consumption, portfolio choice, and labor supply (No. w17134). National Bureau of Economic Research.
Cronqvist, H., Siegel, S., 2013. The origins of savings behavior. In AFA 2011 Denver Meetings Paper.
Dardanoni, V., 1991. Precautionary savings under income uncertainty: a cross-sectional analysis. Applied Economics, 23(1B), 153-160.
De Graaf-Zijl, M., van der Horst, A., van Vuuren, D., Erken, H., Luginbuhl, R., 2015. Long-term unemployment and the Great Recession in the Netherlands: Economic mechanisms and policy implications. De Economist, 1-20.
De Paoli, B., Zabczyk, P. 2013. Cyclical Risk Aversion, Precautionary Saving, and Monetary Policy. Journal Of Money, Credit & Banking (Wiley-Blackwell), 45(1), 1-36.
Dvorsky, S., Scheiber, T., Stix, H., 2009. CESEE Households amid the Financial Crisis: Euro Survey Shows Darkened Economic Sentiment and Changes in Savings Behavior. Focus on European Economic Integration Q, 4, 71-83.
Roy Weintraub, E., 1993. Neoclassical Economics. The Concise Encyclopedia of Economics. Library of Economics and Liberty. Retrieved December 29, 2015 from the World Wide Web:
http://www.econlib.org/library/Enc1/NeoclassicalEconomics.html
Echevarría, C. A., 2012. Income tax progressivity, physical capital, aggregate uncertainty and long-run growth in an OLG economy. Journal of Macroeconomics, 34(4), 955-974.
Friedman, M., 1957. The permanent income hypothesis. In A theory of the consumption function (pp. 20-37). Princeton University Press.
Gibson, R., Michayluk, D., Van de Venter, G., 2013. Financial risk tolerance: An analysis of unexplored factors. Financial Services Review, 22(1), 23-50.
Gourio, F., 2009. Disasters risk and business cycles (No. 15399). National Bureau of Economic Research.
Gouskova, E., Juster, F. T., Stafford, F. P., 2006. 11. Trends and turbulence: allocations and dynamics of American family portfolios, 1984–2001. International Perspectives on Household Wealth, 365.
Hilgert, M. A., Hogarth, J. M., Beverly, S. G., 2003. Household Financial Management: The Connection between Knowledge and Behavior. Federal Reserve Bulletin, 89(7), 7
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.
Kimball, M. S., 1990. Precautionary Saving in the Small and in the Large. Econometrica: Journal of the Econometric Society, 53-73.
Knotek, E. I., Khan, S., 2011. How Do Households Respond to Uncertainty Shocks? Federal Reserve Bank Of Kansas City Economic Review, 96(2), 5-34.
Leland, H. E., 1968. Saving and uncertainty: the precautionary demand for saving. Quarterly Journal Of Economics, 82(3), 465-473.
Levine, R., Renelt, D., 1992. A sensitivity analysis of cross-country growth regressions. The American economic review, 942-963.
Lusardi, A., 2008. Household Saving Behaviour: The Role of Financial Literacy, Information, and Financial Education Programs. Working paper
Mastrogiacomo, M., Alessie, R., 2014. The precautionary savings motive and household savings. Oxford Economic Papers, 66(1), 164-187.
Merry, E. A., Thomas, L., 2014. Asset Holdings of Young Households: Trends and Patterns. Federal Reserve Bank of St. Louis Review, 96(4Q), 391-411.
Modigliani, F., 1970. The life cycle hypothesis of saving and intercountry differences in the saving ratio. Induction, growth and trade, 197-225.
Modigliani, F., Brumberg, R., 1954. Utility analysis and the consumption function: An interpretation of cross-section data. Franco Modigliani, 1.
Riley Jr., W. B., Chow, K. V., 1992. Asset Allocation and Individual Risk Aversion. Financial Analysts Journal, 48(6), 32-37.
Saving in the Netherlands. 2014, July 24. Retrieved from http://www.dnb.nl/en/news/news-and-archive/dnbulletin-2014/dnb309964.jsp
Sheshinski, E., 1989. A simple model of optimum life-cycle consumption with earnings uncertainty. Journal of Economic Theory, 49(1), 169-178.
Tversky, A., Kahneman, D., 1973. Availability: A heuristic for judging frequency and probability. Cognitive psychology, 5(2), 207-232.
Van Rooij, M., Lusardi, A., Alessie, R., 2011. Financial literacy and stock market participation. Journal of Financial Economics, 101(2), 449-472.
Von Gaudecker, H. M., 2015. How does household portfolio diversification vary with financial sophistication and advice? The Journal of Finance, 70(2), 489-507
Weber, M., Weber, E. U., & Nosić, A., 2012. Who takes Risks When and Why: Determinants of Changes in Investor Risk Taking. Review of Finance, rfs024.
Yilmazer, T., & Lich, S., 2013. Portfolio choice and risk attitudes: a household bargaining approach. Review of Economics of the Household, 13(2), 219-241.
8. Appendix
Figure 1: Questions regarding risk aversion
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. I want to be certain that my investments are safe
4. If I want to improve my financial position, I should take financial risks
5. I am prepared to take the risk to lose money, when there is also a chance to gain money
Table 1: GDP growth Year GDP growth in % 1996 3.6 1997 4.3 1998 4.5 1999 5.1 2000 4.2 2001 2.1 2002 0.1 2003 0.3 2004 2.0 2005 2.2 2006 3.5 2007 3.7 2008 1.7 2009 -3.8 2010 1.4 2011 1.7 2012 -1.1 2013 -0.5 2014 1.0 Average GDP 1.89
Table 6: Cross-section random effects regression for Eq. (1) (R2=0.038%)
Dependent variable: savings rate Normal economy Depressed economy Wald t stat Risk Aversion (1=high) 0.006 -0.108*** 2.794***
** significant at 5% level *** significant at 1% level
Table 7: Cross-section random effects regression coefficients for Eq. (1) (R2=0.557%) for 1997-2000, 2002, 2003, 2007, 2009 and 2012-2014
Dependent variable: savings rate Normal economy Depressed economy Wald t-stat
Risk Aversion (1=high) 0.0064 -0.024 0.471
Age -0.001 -0.003 0.664
Gender (1=male) -0.280*** -0.240*** -0.573
Education (1=high) -0.016 -0.009 -0.091
Education (1=low) 0.015 0.031 -0.158
Ability to pay bills (1=easy) 0.046 0.190*** -2.185** Ability to pay bills (1=hard) -0.127 -0.140* 0.112
Constant 0.417***
* significant at 10% level ** significant at 5% level *** significant at 1% level
Table 8: Cross-section random effects regression coefficients for Eq. (1b), with R2=2.570%
Dependent variable: savings rate Coefficient estimate
Risk Aversion (1=high) 0.006*
Age 0.001*** Gender (1=male) -0.046*** Education (1=high) 0.020*** Education (1=low) 0.018***
Ability to pay bills (1=easy) 0.038***
Ability to pay bills (1=hard) -0.021***
Constant 0.112*** * significant at 10% level ** significant at 5% level *** significant at 1% level
Table 11: Cross-section fixed effects regression coefficients for Eq. (2) (R2=72.589%)
Dependent variable: risky share Normal economy Depressed economy Wald t stat
Risk Aversion (1=high) -0.007** -0.003 -1.105
