1
Household investment behavior: The effect of job security on Dutch household portfolios
in terms of financial asset allocation for the period 2002 – 2013.
G.P.L. Veerman Master thesis Finance June 2015
University of Groningen
Faculty of Economics and Business Master of Science (MSc.) Finance
Supervised by Dr. A. Plantinga
Abstract
This paper documents the effect of the level of job security on Dutch household portfolios in terms of financial asset allocation. By analyzing data from the Dutch Household Survey (DHS) for the period 2002 – 2013, I present three main results. First, I show that the population significantly differs in terms of household financials and household characteristics when it is observed from the perspective of the level of job security. Second, the level of job security has a significant effect on the allocation of financial assets among Dutch household portfolios. Third, a transition between the level of job security influences the allocation of financial assets among Dutch households, making households react more strongly to negative events than to positive events. Overall, a high level of job security leads to risk seeking, whereas a low level of job security leads to risk aversion.
Acknowledgements
2 1. Introduction
Due to the global economic crisis that hit the world in 2008, many businesses have been forced to reduce their number of employees. This happened across the globe in the West, Southeast Asia, China and India (Bhanot, 2010). In the Netherlands, the unemployment rate more than doubled. From approximately 4% in 2008, the level of unemployment rose to 8-9% in the subsequent years (CBS, 2014). Hand in hand with the increasing threat of unemployment goes a decrease in the level of job security. Figure 1 shows the development of job security among Dutch households for the period 2002-2013.
Figure 1. Job security development among Dutch households for the period 2002-2013
Panel A shows the development of households with a high level of job security, panel B shows the development of households with a low level of job security (DNB Household Survey).
Panel A. High job security Panel B. Low job security
Traditional finance theory makes several assumptions on how rational investors should make investment decisions. However, empirical literature shows an infinite set of variables affecting household portfolios, such as age, gender, marital status, net worth, and net income. Key in portfolio decision making in the life-cycle is labor income and the risk associated with it, as Cocco, Francisco, and Maenhout (2005) discuss. Although Cocco et al. (2005) find that labor income itself is crucial for optimal portfolio decisions, not so much the typical risk associated with it, Heaton and Lucas (2000), Angerer and Lam (2009), Benzoni and Chyruk (2009), and Guiso, Jappelli, and Terlizzese (1996) show that variability in labor income is an important factor in household portfolio decision making.
3 job security is only an implicit claim, I distinguish households based on their level of job security, e.g. high level of job security and low level of job security. By doing this I am able to compare the household financials and the household characteristics, primarily focusing on their portfolio of financial assets. By doing this I provide insight in the different investment strategies and decisions made by Dutch households in relation to their level of job security.
My research starts from the initial life-cycle hypothesis of household saving (Modigliani and Brumberg, 1954; Ando and Modigliani, 1963). This hypothesis assumes that households aim to maximize their utility of current and discounted future consumption over the course of their life subject to their budget constraint, which is represented by equation (1).
s.t.
(1)
where and are current and future consumption, respectively, and are current and future income, respectively, and r is the discount rate.
In the initial life-cycle hypothesis, the optimal choice of current and future consumption is not affected by the timing of income. However, if current income is expected to exceed future income, households are assumed to save. If future income is expected to exceed current income, households are expected to borrow. Where the initial life-cycle hypothesis assumes complete certainty, e.g. no precautionary saving motive, Browning and Lusardi (1996), for example, incorporate a vector in their model that accounts for risk in future income. They show that in case of certainty, timing of income does not affect saving and consumption. However, more uncertainty in future income leads to more current saving. In addition to levels of job security, I mention transitions in the level of job security which I relate to prospect theory. This theory evaluates certain outcomes relative to a reference point, suggesting that households react more strongly to negative events than to positive events and making households loss averse. This behavior can be observed in the value function which is presented by figure 2 (Kahneman and Tversky, 1979).
Figure 2. Prospect theory’s value function Value
4 The empirical literature on labor income risk shows similar results. Heaton and Lucas (2000) investigate the relationship between portfolio composition and labor income risk, mainly focusing on stockholdings. They find that entrepreneurial income risk has a significant influence on portfolio choice. Households with high and variable entrepreneurial income hold less wealth in stocks than other similar wealthy households. Heaton and Lucas (2000) explain this by the higher background income risk they face. Angerer and Lam (2009) investigate the relationship between portfolio composition and labor income risk. Their paper distinguishes permanent and transitory income risk and classifies risky and safe assets. They find that permanent income risk significantly shifts a household’s portfolio allocation away from risky assets, while transitory income risk has little effect. The effect of permanent income risk is about six times as large as the effect of transitory income risk. Benzoni and Chyruk (2009) show that long run labor income risk has a first order effect on optimal life-cycle asset allocation. Several researchers point towards the importance of precautionary saving estimates for policy design, especially for labor market, social insurance, and taxation policies which directly affect variance in households’ net income (Fossen and Rostam-Afscher, 2013; Alan, 2006; Dardanoni, 1991).
This paper contributes to existing literature by quantifying the level of job security with an alternative approach. Where empirical literature measures labor income risk mainly by its standard deviation (e.g. Angerer and Lam, 200; Heaton and Lucas, 2000), I use occupational and employment data in my research. The aim of this study is to determine the effect of job security on financial asset allocation among Dutch households since I find no existing literature on this topic for the Netherlands. According to the two period life-cycle hypothesis that allows uncertainty I expect households facing a low level of job security to balance their portfolio more towards safe financial assets, making households more risk averse. In my research I use data provided by the DNB Household Survey (DHS), containing extensive information about households, work and pension, accommodation, income, wealth, and psychological concepts. By performing several statistical tests and performing panel regressions I show that the level of job security has a significant influence in the allocation of financial assets among Dutch households, although the most prominent factors seem to be age and total net wealth.
5 2. Literature review
Over time, households have experienced an expansion of financial opportunities. These opportunities can lead to great benefits, but also to great losses. With an increase in the scope for investment in risky assets, investors may end up with larger swings in wealth than they had anticipated. Dynan (2009) argues that since the mid-90’s many individual households, and ultimately the economy as a whole, destabilized.
One of the motives for households to save is to build up a reserve against unforeseen contingencies, the so called precautionary saving motive (Keynes, 1936). Where the initial life-cycle hypothesis assumes complete certainty, e.g. no precautionary saving motive, many researchers extended this hypothesis and incorporate risk components for such contingencies. Browning and Lusardi (1996) incorporate a vector in their model that accounts for risk in future income. They show that in case of certainty, timing of income does not affect saving and consumption. However, more uncertainty in future income leads to more current saving. Nagatani (1972) incorporates uncertainty concerning future income in his model of life-cycle saving. He shows that the lifetime consumption will be generally dependent on the income profile. For a given pattern of lifetime income, the discounted value of future income can be written as a function of current income. He shows that uncertainty about future income is translated into a risk premium which adds to the market interest rate in discounting future income, which may lead to an increase in savings. Menezes and Auten (1978) show that the effect of income risk on the marginal rate of time preference is both a necessary and sufficient condition to characterize how income risk affects saving. Their result concludes that income risk reduces the rate of time preference and hence increases current saving. Dardanoni (1991) estimates an equation for the structural relationship between savings and their determinants, containing a precautionary component that is directly affected by future income risk. Using cross-sectional data from the 1984 UK Family Expenditure Survey, he finds that precautionary savings are a significant proportion of total savings. Alan (2006) estimates the effect of labor income uncertainty on wealth accumulation in Canada. His empirical results suggest the presence of a strong precautionary saving motive among Canadian households for broad definitions of wealth. In addition, he finds that the level of precautionary savings significantly increases when households face liquidity constraints.
6 that the lack of portfolio diversification is party explained by non diversifiable income risk. They investigate the allocation of households’ financial assets by explicitly considering the role of income risk and borrowing constraints. Using the Bank of Italy’s Survey of Household Income and Wealth (SHIW), they argue that uninsurable, non diversifiable risks – such as income risk, which they measure as the variance of real income – may induce prudent investors to reduce holdings of risky assets in order to cut their overall exposure. Estimating a Tobit model, they find that when investors are confronted with uninsurable risk, they reduce overall exposure to risk by holding a lower proportion of risky assets. Their finding supports the proposition that background risk depresses the willingness to bear other avoidable risks. Jinguan (2012) shows that uncertainty about investment returns and labor income leads to positive precautionary saving under the condition that consumers are positioned above a certain prudence measure which is related to risk tolerance.
Deidda (2013) shows that portfolio diversification is not used by Italian households as a device to reduce total exposure to risk. Deidda (2013) investigates to what extent uncertainty about future contingencies affects the amount of desired precautionary wealth. She bases her research on a direct question about the desired amount of a household’s precautionary wealth using the Bank of Italy’s Survey of Household Income and Wealth. Performing Ordinary Least Squares (OLS) regressions, she shows that Italian households appear to desire to have precautionary holdings in amounts that are related to financial risk, ultimately affecting portfolio composition.
Heaton and Lucas (2000) investigate the relationship between portfolio composition and labor income risk, mainly focusing on stockholdings. They measure labor income risk by the standard deviation of wage income and entrepreneurial income. By using the 1989, 1992, and 1995 Survey of Consumer Finances, and the Panel of Individual Tax returns, they find that entrepreneurial income risk has a significant influence on portfolio choice. Households with high and variable entrepreneurial income hold less wealth in stocks than other similar wealthy households. Heaton and Lucas (2000) explain this by the higher background income risk they face.
7 Fossen and Rostam-Afscher (2013) contradict other researches by showing that no significant estimates of precautionary savings remain after controlling for entrepreneurship. The difference in the savings behavior of entrepreneurial versus non-entrepreneurial households may become especially pronounced in countries with an extensive social security system. Fossen and Rostam-Afscher (2013) provide evidence from Germany that higher income risk is associated with a portfolio shift from less liquid to more liquid assets. In addition, entrepreneurs with high income risk hold more wealth than employees, but this is not because of precautionary saving motives. These extra savings are likely to reflect the exclusion from the public pension system. Cocco et al. (2005) develop a quantitative and realistically calibrated model to determine the optimal consumption and portfolio decisions of a finitely lived individual who faces labor income uncertainty and can invest in either risky or riskless assets. They find that labor income itself is crucial for optimal portfolio decisions, not so much the typical risk associated with it.
Next to labor income and labor income risk, researchers point towards an infinite set of variables affecting households’ portfolio choice. Although traditional finance theory assumes that investors should have identical expectations with respect to the necessary inputs in making portfolio decisions and should base their investment decisions solely on risk-return considerations, I turn to empirical literature that contradicts this theory.
9 3. Data and methodology
From the initial life-cycle hypothesis (Modigliani and Brumberg, 1954; Ando and Modigliani, 1963), which assumes complete certainty, the timing of income does not affect a consumer’s optimal saving and consumption behavior. When relaxing the assumption of complete certainty, Browning and Lusardi (1996) show that uncertainty about future income does increase household savings. However, income certainty does not affect saving and consumption. In my research I focus on the effect of job security on the financial asset allocation among Dutch households. I assess this effect by making a distinction between high job security and low job security. According to empirical literature, I expect Dutch households facing low job security not only to have more savings, but especially balance their portfolio towards more safe financial assets due to risk aversion. Therefore, I formulate my main hypotheses as follows:
3.1 Data
Since 1993, CentERdata annually collects economic data via the DNB Household Survey (DHS). This survey is representative for the Dutch population aged 16 year and older, comprising extensive and detailed information about approximately 2,000 households. It contains data regarding household information, work and pension data, accommodation data, income data, wealth data, and psychological data.
For the key variable relating to the level of job security I use the work and pension data from the DHS. To determine the effect of the level of job security I distinguish a high and low level of job security. High job security refers to households that are employed on a contractual and permanent basis, or perform military services. Low job security refers to households that are employed on a temporary basis, perform stand by work, work in a family business, are temping, work as a free-lance, perform a free profession, are self-employed, are looking for work after losing job or are looking for first time work.
10 the number of wealth components by creating three levels: the initial level as formulated by the DHS, the aggregation level in which I aggregate different wealth components, and the categorization level in which I categorize the different aggregation levels as financial assets, non-financial assets and debt. A detailed overview of these levels is presented in appendix A. In my research I focus on the financial assets mutual and grow funds, shares, checking and saving accounts, value of insurances, and bonds and options, which in sum makes up for a household’s financial asset portfolio.
11 Using the unique household index that is allocated to each household in the DHS I am able to connect each data module for every household individual. By connecting all data modules for the period 2002 – 2013 I end up with a sample size of 54,503 individual observations. Since my research focuses on household data I transform the data modules from an individual level to a household level. I aggregate the household financials by summing up every individuals financial belongings to the same household index. I determine the household characteristics by selecting the characteristics from the person who is most involved with the financial information of the household. This approach leads to a total of 22,412 household observations. Finally, I select only those households that hold total financial assets with a value of at least equal to or larger than one hundred euro. This leads to a full sample size of 17,074 observations, a reduction of 37,429 observations from the original sample size. The summary statistics of the full sample size are presented in table 1.
Since my research focuses on the impact of job security on a household’s financial asset allocation I make a distinction in my full sample based on high and low job security. The two sub samples comprise 6,186 and 6,003 observations, respectively. The summary statistics of both the sub samples are presented in table 2.
Out of the full sample, referring to table 1, approximately one third is considered to have a high level of job security, and one third is considered to have a low level of job security. Table 1 shows that the two sub samples especially differ in terms of age. Where the sub sample relating to high job security is distributed among the ages 1 to 64, the majority of the sub sample relating to low job security is distributed for the age interval 65 or older. Table 1 shows that, on average, 41% (4%) of the Dutch households is high (low) educated, 8% (39%) has a high (low) risk tolerance, and 21% (11%) has high (low) financial knowledge. Interesting is that, on average, 11% of the Dutch households are looking for a(nother) job, while 36% has a precautionary saving motive relating to job security. This result may imply that job security is indeed an implicit claim, as argued by Cornell and Shapiro (1987).
12 Table 1. Summary statistics, full sample
This table presents the summary statistics (mean, median, standard deviation, minimum, maximum) for the relevant variables used in this paper. The data consists of the full sample comprising 17,074 observations. “(%)” indicates the percentage of total financial assets in a household’s portfolio.
Mean Median Stdev Min Max
Household financials
Total net income 26,452 24,429 26,655 -4,580 1,185,983
Total net wealth 198,967 126,071 294,093 -4,722,945 5,882,050
Total assets 275,523 229,798 316,885 100 5,882,050
Total liabilities 76,556 22,252 125,244 0 5,000,000
Total financial assets 53,389 21,503 118,108 100 3,702,525
Checking and saving accounts 26,078 9,879 71,855 -202,442 3,701,225
Value of insurances 13,775 875 46,457 0 2,304,544
Mutual and grow funds 7,403 0 39,299 0 1,300,000
Shares 3,818 0 30,241 0 1,025,950
Bonds and options 2,316 0 23,394 0 1,134,554
(%) Checking and saving accounts C&S(%) 63.24 75.65 46.28 (%) Value of insurances INS(%) 26.23 4.54 41.46 (%) Mutual and grow funds M&G(%) 6.45 0.00 18.34
(%) Shares Shares(%) 2.71 0.00 11.09
(%) Bonds and options B&O(%) 1.38 0.00 12.86
Household characteristics
High job security JobSecH 0.36 0.00 0.48
Low job security JobSecL 0.35 0.00 0.48
Age ˂ 35 Age1 0.15 0.00 0.36 35 ≤ age ˂ 44 Age2 0.19 0.00 0.39 45 ≤ age ˂ 54 Age3 0.21 0.00 0.41 55 ≤ age ˂ 64 Age4 0.21 0.00 0.41 Age ≥ 65 Age5 0.24 0.00 0.43 Male Male 0.55 1.00 0.50 Children Children 0.34 0.00 0.47 Married Married 0.48 0.00 0.50 West West 0.45 0.00 0.50
High education EduH 0.41 0.00 0.49
Low education EduL 0.04 0.00 0.19
High risk tolerance RiskH 0.08 0.00 0.27
Low risk tolerance RiskL 0.39 0.00 0.49
High financial knowledge KnowH 0.21 0.00 0.41
Low financial knowledge KnowL 0.11 0.00 0.31
Looking for a(nother) job Looking 0.11 0.00 0.31
13 Table 2. Summary statistics, sub samples
This table presents the summary statistics (mean, median) for the relevant variables used in this paper. The data consists of the sub sample high job security (JobSecH) and low job security (JobSecL), comprising 6,186 and 6,003 observations, respectively. “(%)” indicates the percentage of total financial assets in a household’s portfolio. The Z score refers to the Wilcoxon Rank Sum test. With this test I compare the two sub samples, rejecting the null hypothesis for Z scores greater than 1.96 or smaller than -1.96 (Keller, 2009).
High job security Low job security
Mean Median Mean Median Z score
Household financials
Total net income 31,431 28,682 25,093 23,463 3.39
Total net wealth 162,643 106,621 263,531 182,486 -17.90
Total assets 261,619 230,018 325,175 258,638 -7.16
Total liabilities 98,976 62,784 61,644 5,000 -7.84
Total financial assets 47,251 22,724 68,509 27,297 -5.37
Checking and saving accounts 21,782 9,296 34,714 14,775 -14.53
Value of insurances 15,930 4,799 12,249 0 -17.96
Mutual and grow funds 6,344 0 10,638 0 -111.18
Shares 2,187 0 6,904 0 -139.09
Bonds and options 1,008 0 4,004 0 -167.40
(%) Checking and saving accounts C&S(%) 56.14 58.17 70.98 96.41 -57.22
(%) Value of insurances INS(%) 34.50 24.29 15.57 0.00 -14.67
(%) Mutual and grow funds M&G(%) 6.17 0.00 7.84 0.00 -110.76
(%) Shares Shares(%) 2.47 0.00 3.63 0.00 -138.95
(%) Bonds and options B&O(%) 0.72 0.00 1.98 0.00 -167.37
Household characteristics
High job security JobSecH 1.00 1.00 0.00 0.00 -
Low job security JobSecL 0.00 0.00 1.00 1.00 -
Age ˂ 35 Age1 0.20 0.00 0.08 0.00 -139.56 35 ≤ age ˂ 44 Age2 0.28 0.00 0.08 0.00 -125.81 45 ≤ age ˂ 54 Age3 0.31 0.00 0.09 0.00 -119.60 55 ≤ age ˂ 64 Age4 0.20 0.00 0.18 0.00 -128.71 Age ≥ 65 Age5 0.00 0.00 0.58 1.00 -142.64 Male Male 0.61 1.00 0.65 1.00 -104.81 Children Children 0.44 0.00 0.14 0.00 -93.14 Married Married 0.57 1.00 0.58 1.00 -97.06 West West 0.46 0.00 0.45 0.00 -94.20
High education EduH 0.49 0.00 0.42 0.00 -88.82
Low education EduL 0.02 0.00 0.05 0.00 -176.09
High risk tolerance RiskH 0.08 0.00 0.09 0.00 -158.87
Low risk tolerance RiskL 0.42 0.00 0.45 0.00 -97.95
High financial knowledge KnowH 0.27 0.00 0.23 0.00 -115.79
Low financial knowledge KnowL 0.11 0.00 0.12 0.00 -151.19
Looking for a(nother) job Looking 0.17 0.00 0.11 0.00 -141.18
14
3.2 Methodology
When analyzing the summary statistics in table 2 for the two sub samples high and low job security, I observe differences in the household financials as well as in the household characteristics. To test for statistical differences I use the summary statistics in table 2. I perform the Wilcoxon Rank Sum test, which tests for equality of population medians of two independent samples (Keller, 2009). This non-parametric test is formulated in equation (1).
(1)
where is the rank sum of the sub sample with the fewest number of observations, is the number of events in the subsample with the fewest number of observations, and is the number of events in the subsample with largest amount of observations. Performing a two-sided test, and using a 5% significance level, I reject the null hypothesis for Z-values greater than 1.96 or smaller than -1.96 (Keller, 2009). The test results, along with the summary statistics, are presented as the Z score in table 2. These results show that the two sub samples indeed differ significantly both in terms of household financials and household characteristics. Confirming this statistical difference triggers me to do further research on the level of job security. Therefore, I now turn to the full sample.
To assess the effect of the level of job security on the financial asset allocation of Dutch household portfolios, I perform five regressions on the full sample, corresponding to the five categorical financial assets mutual and grow funds, shares, checking and saving accounts, value of insurances, and bonds and options. Since some of the observations are missing or incomplete I use the unbalanced panel technique. In order to select the appropriate panel model, I perform the Redundant Fixed Effects test and the Hausman test (Brooks, 2008). Significant test results show me that the time and cross-sectional fixed panel techniques are appropriate to estimate the regressions.
15 The regression used in this research is presented by equation 2.The explanatory variables of interest are the dummy variables and . Control variables refer to age, gender, marital status, children, geographics, education, financial knowledge, risk tolerance, job seeking behavior, positive precautionary saving, total net income, and total net wealth. The control variables are further discussed below.
where is the dependent variable that refers to one of the five financial assets calculated as the percentage of total financial assets. , , , and are dummy variables that represent the age intervals “<35”, “[35, 44]”, “[45, 54]”, and “[55, 64]”, respectively. is a dummy variable that takes on value 1 when the respondent is a male, and 0 otherwise. is a dummy variable that takes on value 1 when the respondent is married, and 0 otherwise. is a dummy variable that takes on value 1 when the respondent has one or more children, and 0 otherwise. is a dummy variable that takes on value 1 if the respondent is residing in the west of the Netherlands, and 0 otherwise. The dummy variables , , , , , , and denote high and low education, high and low financial knowledge, high and low risk tolerance, and high and low job security, respectively. These dummy variables take on value 1 if a household is labeled to the specified category, and 0 otherwise. is a dummy variable that takes on value 1 when the respondent is looking for a(nother) job, and 0 otherwise. is a dummy variable that takes on value 1 when the respondent finds precautionary saving for unemployment important, and 0 otherwise. and are the logarithms of a household’s total net income and total net wealth, respectively. I use logarithms to reduce the disparity in the distribution of total income and total net wealth.
Next to performing fixed panel regressions on the full sample, I check whether a shock in the level of job security, e.g. transition between high and low job security, leads to a change in investment behavior among Dutch households. To perform such an analysis I distinguish negative and positive events. Negative events are defined as transitions from a high to a low level of job
17 4. Results
4.1 Regression results
The regression results are presented in table 3 and are discussed in five parts: mutual and grow funds, shares, checking and saving accounts, value of insurances, and bonds and options. Appendix B presents a cross-sectional fixed unbalanced panel regression which serves as a robustness check for the variables relating to the level of job security. The appendix shows that the coefficient significance does not change when the unbalanced panel regression is not time fixed.
Table 3 shows that only a high level of job security has a significant influence on the percentage of mutual and grow funds that households hold in their portfolio of financial assets. Households that face a higher level of job security balance their portfolio with +1.33% more towards mutual and grow funds. When considering mutual and grow funds as a more risky financial asset, this result may imply that a higher level of job security leads households to be more risk seeking, which is in line with Guiso et al. (1996). The control variables relating to the age intervals “45≤age<54” and “55≤age<64”, (high) education and total net wealth also have a significant influence on the percentage of mutual and grow funds that households hold in their portfolio of financial assets. The effects of aging, education and total net wealth on portfolio choice is discussed by Coile and Miligen (2009) and Calvet and Campbell (2009). Overall, these results are in line with empirical literature which will be further elaborated on in section 4.3.
Table 3 shows that only a low level of job security has a significant influence on the percentage of shares that households hold in their portfolio of financial assets. Households that face a lower level of job security balance their portfolio with -0.79% less towards shares. When considering shares as a more risky financial asset, this result may imply that a lower level of job security leads households to be more risk averse, which is in line with Guiso et al. (1996). The control variables relating to the age interval “55≤age<64”, gender and total net wealth have a significant influence on the percentage of shares that households hold in their portfolio of financial assets. The effects of aging, gender and total net wealth is discussed by Coile and Miligen (2009), Halko et al. (2012), and Calvet and Campbell (2009). Overall, these results are in line with empirical literature which will be further elaborated on in section 4.3.
18 saving accounts. When considering checking and saving accounts as a relatively safe financial asset, this result may imply that a lower level of job security leads households to be more risk averse, which is in line with Guiso et al. (1996). The control variables relating to age, gender, children, (high) education and total net wealth have a significant influence on the percentage of checking and saving accounts that households hold in their portfolio of financial assets. Most interestingly, from the regression results for checking and saving accounts, age shows a U shaped function over the lifetime of a household, which may imply an age-cohort effect. The effects of aging, gender, children, education and total net wealth are discussed by Coile and Miligen (2009), Halko et al. (2012), Love (2010), and Calvet and Campbell (2009). Overall, these results are in line with empirical literature which will be further elaborated on in section 4.3.
Table 3 shows that both a high and low level of job security have a significant influence on the percentage of value of insurances that households hold in their portfolio of financial assets. Households that face a higher level of job security balance their portfolio with -2.56% less towards value of insurances. Households that face a lower level of job security balance their portfolio with -2.85% less towards value of insurances. Although value of insurances is considered to be a relatively safe financial asset, and a lower level of job security seems to have the biggest impact in asset allocation, no unambiguous implications can be presented with regard to households’ risk preferences. However, considering the significantly positive effect of low job security on the percentage held in checking and saving accounts, low job security leads to a shift from less liquid to more liquid assets, which is in line with Fossen and Rostam-Afscher (2013). The control variables relating to age, children, region, education and total net wealth have a significant influence on the percentage of value of insurances that households hold in their portfolio of financial assets. Most interestingly, from the regression results for value of insurances, age shows an inverted U shaped function over the lifetime of a household, which may imply an age-cohort effect. In addition, relative to high education, low education remarkably seems to have a larger effect on the percentage of value of insurances that households hold in their portfolio of financial assets. The effects of aging, children, geographics, education, and total net wealth are discussed by Coile and Miligen (2009), Christelis et al. (2013), and Calvet and Campbell (2009). Overall, these results are in line with empirical literature which will be further elaborated on in section 4.3.
19 Table 3. Regression results
This table presents the results for the fixed effect unbalanced panel regression based on equation 2, where the top row indicates to the five dependent variables. Each column presents the coefficient and the t-statistic (between brackets) per variable. The data used in this regression originates from DHS for the period 2002 – 2013. The coefficient significance is denoted by * at 10%, ** at 5%, and *** at 1%.
Variable M&G(%) Shares(%) C&S(%) INS(%) B&O(%)
20
4.2 Transition effects
Table 4a and 4b present the summary statistics for the household financials from households that made a transition from high to low job security (negative event), and from low to high job security (positive event), respectively. In addition, both tables 4a and 4b present the test results for the non-parametric Wilcoxon Rank Sum test, rejecting the null hypothesis for Z scores greater than 1.96 or smaller than -1.96 (Keller, 2009). First I discuss the test results for the negative events, followed by the test results for the positive events.
In the full sample I use in this research I observe 108 households that made a transition from high to low job security. Because my selection criteria require me to select four sequential years per household, the full negative event sample in this analysis comprises 432 observations. The summary statistics are presented in table 4a.
Table 4a. Summary statistics, transition effects for negative events
This table presents the financial household means and medians for households that made a transition from high to low job security (negative event). With a total number of 108 households, each containing 4 years of data due to selection criteria, the full sample size in this analysis comprises 432 observations. “(%)” indicates the percentage of total financial assets in a household’s portfolio. The Z score refers to the Wilcoxon Rank Sum test. With this test I compare the two sub samples, rejecting the null hypothesis for Z scores greater than 1.96 or smaller than -1.96 (Keller, 2009).
Before transition After transition
(High job security) (Low job security)
Mean Median Mean Median Z score
Total net income 35,298 30,997 29,422 28,144 -6.55
Total net wealth 241,479 200,216 265,825 219,163 0.57
Total assets 315,892 293,755 342,188 299,126 0.51
Total liabilities 74,413 42,613 76,362 29,500 -4.93
Total financial assets 67,629 38,560 79,490 36,531 0.71
Checking and saving accounts 30,553 14,095 38,718 15,297 -1.39
Value of insurances 23,589 7,391 20,624 4,076 -0.86
Mutual and grow funds 8,968 0.00 11,685 0.00 -19.07
Shares 2,854 0.00 5,018 0.00 -25.72
Bonds and options 1,663 0.00 3,442 0.00 -31.08
(%) Checking and saving accounts 56.66 63.07 52.21 66.83 -3.21
(%) Value of insurances 32.94 26.02 36.77 15.89 -0.46
(%) Mutual and grow funds 6.90 0.00 6.77 0.00 -18.91
(%) Shares 2.52 0.00 2.37 0.00 -25.68
21 Table 4a shows that, in terms of percentages held in a portfolio of financial assets, checking and saving accounts, mutual and grow funds, shares, and bonds and options significantly change after a negative event occurs. However, value of insurances does not significantly change due to a negative event. This result is in line with Benzoni and Chyruk (2009) who find that long run labor income risk has a first order effect on optimal life-cycle asset allocation. This result may also be explained by the significant change in total net income and total liabilities I observe from table 4a. However, performing a pooled panel regression which is presented in appendix C.1, I find no significant effects on the interaction between a change in total net income and a dummy variable that controls for negative events on the change in portfolio compositions.
In the full sample I use in this research I observe 34 households that made a transition from low to high job security. Because my selection criteria require me to select four sequential years per household, the full positive event sample in this analysis comprises 136 observations. The summary statistics are presented in table 4b.
Table 4b. Summary statistics, transition effects for positive events
This table presents the financial household means and medians for households that made a transition from low to high job security (positive event). With a total number of 34 households, each containing 4 years of data due to selection criteria, the full sample size in this analysis comprises 136 observations. “(%)” indicates the percentage of total financial assets in a household’s portfolio. The Z score refers to the Wilcoxon Rank Sum test. With this test I compare the two sub samples, rejecting the null hypothesis for Z scores greater than 1.96 or smaller than -1.96 (Keller, 2009).
Before transition After transition
(Low job security) (High job security)
Mean Median Mean Median Z score
Total net income 26,897 25,437 24,174 25,322 -3.27
Total net wealth 95,225 50,363 89,901 54,960 -0.45
Total assets 195,689 186,175 212,311 212,951 0.58
Total liabilities 100,463 97,200 122,410 125,500 -0.39
Total financial assets 35,290 27,158 35,611 20,252 -1.07
Checking and saving accounts 14,817 8,021 21,676 10,531 -0.17
Value of insurances 16,964 5,115 12,662 5,590 -1.01
Mutual and grow funds 2,542 0.00 859 0.00 -13.04
Shares 651 0.00 331 0.00 -14.39
Bonds and options 313 0.00 81 0.00 -18.52
(%) Checking and saving accounts 55.61 51.18 59.42 58.70 -1.66
(%) Value of insurances 36.75 36.19 36.02 33.58 -1.27
(%) Mutual and grow funds 5.97 0.00 2.26 0.00 -13.04
(%) Shares 1.16 0.00 2.06 0.00 -14.46
22 Table 4b shows that, in terms of percentages held in a portfolio of financial assets, mutual and grow funds, shares, and bonds and options do significantly change after a positive event occurs. However, value of insurances and checking and saving accounts do not significantly change due to a positive event. This result is in line with Benzoni and Chyruk (2009) who find that long run labor income risk has a first order effect on optimal life-cycle asset allocation. This result may also be explained by the change in total net income I observe from table 4b. However, performing a pooled panel regression which is presented in appendix C.2, I find no significant effects on the interaction between a change in total net income and a dummy variable that controls for positive events on the change in portfolio compositions.
4.3 Overall results in perspective
The overall results for the five regressions performed in this research show that the level of job security has a significant influence on the allocation of financial assets among the portfolios of Dutch households. In line with Alan (2006) and Dardanoni (1991), I show that precautionary saving motives exist and are a significant proportion of total savings.
In particular interest for my research, it can be shown that my findings are in line with Guiso et al. (1996). They find that when investors are confronted with uninsurable risk, they reduce overall exposure to risk by behaving in a more risk averse way. In line with Guiso et al. (1996), my findings support the proposition that background risk depresses the willingness to bear other avoidable risks. Overall, my findings are in line with Benzoni and Chyruk (2009) and Deidda (2013) who show that labor income risk affects portfolio compositions.
There seems to be a distinction in the magnitude of the effect of a higher and a lower level of job security, which is in line with Angerer and Lam (2009), Heaton and Lucas (2000), and Benzoni and Chyruk (2009), and Guiso et al. (1996). Overall, a higher level of job security seems to direct households towards a more risk seeking behavior, whereas a lower level of job security seems to direct households towards a more risk averse behavior. By purely focusing on the transition effects, it can be shown that households react more strongly to negative events than to positive events.
23 The overall results in my research show that, although the level of job security has some significant influence on the allocation of financial assets among Dutch household portfolios, the most prominent factors seem to be age and total net wealth, primarily balancing between checking and saving accounts and value of insurances. This results is in line with Coile and Miligen (2009) who find significant effects of aging on a household’s portfolio composition, and Calvet and Campbell (2009), among others, who find significant effects of wealth on a household’s portfolio composition.
24 5. Conclusion and discussion
In this paper I document my research on the effect of the level of job security on Dutch household portfolios in terms of financial asset allocation. I contribute to existing literature by quantifying the level of job security with an alternative approach. Where empirical literature measures labor income risk mainly by its standard deviation, I use occupational and employment data to determine the level of job security. By using data provided by the DHS for the period 2002 – 2013 I am able to make a clear distinction between a high and a low level of job security and to determine how these different levels affect financial asset allocation among Dutch household portfolios, which has not been done before in the Netherlands.
This study presents three main results in particular. First, by comparing the summary statistics for both high and low job security and performing the Wilcoxon Rank Sum test, I show that the Dutch population significantly differs in terms of household financials and household characteristics when it is observed from the perspective of the level of job security. Second, by performing five cross-sectional and time fixed unbalanced panel regression on the full sample, I show that the level of job security has a significant effect in the allocation of financial assets among Dutch household portfolios. Overall however, age and total net wealth seem to be the most prominent factors influencing the allocation of financial assets among Dutch household portfolios, balancing between checking and saving accounts and value of insurances. Although this effect is probably due to the fact that these financial assets account for approximately 90 percent of Dutch financial household portfolios, this result may imply that checking and saving accounts and value of insurances are substitutes. These results hold when performing only cross-sectional fixed unbalanced panel regressions as a robustness test. Third, by comparing the summary statistics for households that made a transition in the level of job security and performing the Wilcoxon Rank Sum test, I show that a transition between the level of job security significantly influences the allocation of financial assets among Dutch households. When such a transition occurs, negative events seem to have a greater impact on the allocation of financial assets among Dutch household portfolios when compared to positive events. However, one must interpret these results with caution, as total net income significantly changes after a transition in the level of job security occurs, either negative or positive.
I conclude my research by relating the level of job security to risk behavior among Dutch households. Overall, a higher level of job security seems to direct households towards a more risk seeking behavior, whereas a lower level of job security seems to direct households towards a more risk averse behavior.
26 References
Alan, S., 2006, Precautionary wealth accumulation: evidence from Canadian micro data, Canadian Journal of Economics 39, 1105-1124.
Ando, A., and F. Modigliani, 1963, The ‘life cycle’ hypothesis of saving: aggregate implications and tests, American Economic review 53, 55-85.
Angerer, X., and Lam, P.S., 2009, Income risk and portfolio choice: An empirical study, Journal of Finance 64, 1037-1055.
Barber, B.M., and Odean, T., 2000, Trading is hazardous to your wealth: The common stock investment performance of individual investors, Journal of finance 55, 773-806.
Benzoni, L., Chyruk, O., 2009, Investing over the life cycle with long run labor income risk, Economic perspectives 33, 29-43.
Bhanot, S., 2010, The global economic crisis: A perspective, SIES Journal of Management 6, 47-52.
Blume, M.B., and Friend, I., 1975, The asset structure of individual portfolios and some implications for utility functions, Journal of Finance 30, 585-603.
Brooks, C., 2008, Introductory Econometrics for Finance, second edition.
Browning, M., and Lusardi, A., 1996, Household saving: Micro theories and micro facts, Journal of Economic literature 34, 1797-1855.
Bucciol, A., and Miniaci, R., 2011, Household portfolios and implicit risk preference, Review of Economics & Statistics 93, 1235-1250.
Bucciol, A., and Stuefer, M., 2012, Measuring household financial risk, Journal of Wealth Management 15, 20-29.
Calvet, L.E., and Campbell, J.Y., 2009, Fight or Flight? Portfolio rebalancing by individual investors, Quarterly Journal of Economics 124, 301-348.
CBS, 2014, Unemployed and employed labor force per month, Available from;
http://statline.cbs.nl/Statweb/publication/?DM=SLEN&PA=80479ENG&D1=4-5,10-13&D2=0&D3=0&D4=26-37,39-50,52-152&LA=EN&HDR=T&STB=G1,G2,G3&VW=T [Accessed December 10, 2014].
Christelis, D., Georgarakos, D., and Haliassos, M., 2013, Differences in portfolios across countries: Economic environment versus household characteristics, Review of Economics & Statistics 95, 220-236.
Cocco, J.F., Francisco, J.G., and Maenhout, P.J., 2005, Consumption and portfolio choice of the life cycle, The Review of Financial Studies 18, 491-533.
27 Cornell, B., and Shapiro, A.C., 1987, Corporate stakeholders and corporate finance, Financial management 16, 5-14.
Dardanoni, V., 1991, Precautionary savings under income uncertainty: a cross-sectional analysis, Applied economics 23, 153-160.
Deidda, M., 2013, Precautionary saving, financial risk, and portfolio choice, Review of Income & Wealth 59, 133-156.
Dynan, K.E., 2009, Changing household financial opportunities and economic security, Journal of Economic Perspectives 23, 49-68.
Fossen, F.M., Rostam-Afschar, D., 2013, Precautionary and entrepreneurial savings: New evidence from German households, Oxford bulletin of economics & statistics 75, 528-555.
Guiso, L., Jappelli, T., and Terlizzese, D., 1996, Income risk, borrowing constraints, and portfolio choice, American Economic Review 86, 158-172.
Halko, M.J., Kaustia, M., Alanko, E., 2012, The gender effect in risky asset holdings, Journal of Economic Behavior and Organization 83, 66-81.
Heaton, J., and Lucas, D., 2000, Portfolio choices and asset prices: The importance of entrepreneurial risk, Journal of Finance 55, 1163-1198.
Ivkovic, Z., Sialm, C., and Weisbenner, S., 2008, Portfolio performance and the performance of individual investors, Journal of Financial & Quantitative Analysis 43, 613-655.
Jianakoplos, N.A., and Bernasek, A., 2006, Financial risk taking by age and birth cohort, Southern Economic Journal 72, 981-1001.
Jinguan, L., 2012, Precautionary saving in the presence of labor income and interest rate risks, Journal of Economics 106, 251-266.
Kahneman, D., and Tversky, A., 1979, Prospect theory: An analysis of decision under risk, Econometrica 47, 263-291.
Keller, G., 2009, Managerial statistics, eight edition, South-Western Cengage Learning, Mason.
Keynes, J.M., 1936, The general theory of employment, interest and money, London: MacMillen.
Love, D.A., 2010, The effects of marital status and children on savings and portfolio choice, Review of Financial Studies 23, 385-432.
Menezes, C.F., Auten, G.E., 1978, The theory of optimal saving decisions under income risk, International economic review 19, 253-259.
Modigliani, F., and Brumberg, R., 1954, Utility analysis and the consumption function, Post-Keynesian Economics, edt. K. Kurihara, New Brunswick, N.J.: Rugers University Press, 338-436.
28 Neelakantan, U., 2010, Estimation and impact of gender differences in risk tolerance, Economic Inquiry 48, 228-233.
Odean, T., 1998, Volume, volatility, price and profit when all traders are above average, Journal of finance 53, 1887-1934.
Riley Jr., W.B., and Chow, K.V., 1992, Asset allocation and individual risk aversion, Financial Analysts Journal 48, 32-37.
29 Appendix A. Portfolio variables, descriptions
This appendix presents the wealth components from the DNB Household Survey (DHS). Level 0 presents the initial level, level 1 presents the aggregation level, and level 2 presents the categorization level. This approach is similar to Von Gaudecker (2013).
Level 0 Level 1 Level 2
Grow funds Mutual and Grow funds Financial assets
Mutual funds Mutual and Grow funds Financial assets
Shares Shares Financial assets
Bonds Bonds and options Financial assets
Options Bonds and options Financial assets
Checking accounts Checking and saving accounts Financial assets
Savings and/or deposit accounts Checking and saving accounts Financial assets
Deposit books Checking and saving accounts Financial assets
Savings certificates Checking and saving accounts Financial assets
Employer-sponsored savings plan Value of insurances Financial assets
Single-premium annuity insurance policies Value of insurances Financial assets
Savings/endowment insurance policies Value of insurances Financial assets
Life insurance real estate Value of insurances Financial assets
Life insurance mortgage primary real estate Value of insurances Financial assets
Life insurance mortgage secondary real estate Value of insurances Financial assets
Cars Durables Non-financial assets
Motorbikes Durables Non-financial assets
Boats Durables Non-financial assets
(Site-)caravans/trailers Durables Non-financial assets
Money lent out to family/friends Other non-financial assets Non-financial assets
Savings or investments not mentioned before Other non-financial assets Non-financial assets
Stocks from substantial holdings Other non-financial assets Non-financial assets
Business equity Other non-financial assets Non-financial assets
Business equity self-employed Other non-financial assets Non-financial assets
Primary housing Primary housing Non-financial assets
Real estate Secondary real estate Non-financial assets
Secondary housing Secondary real estate Non-financial assets
Private loans Total consumer credit Debt
Extended lines of credit Total consumer credit Debt
Study loans Total consumer credit Debt
Credit card debts Total consumer credit Debt
Loans not mentioned before Total consumer credit Debt
Outstanding debt not mentioned before Other debt Debt
Loans from family/friends Other debt Debt
Finance debts Other debt Debt
Mortgage primary housing Mortgage primary housing Debt
Mortgage real estate Mortgage secondary real estate Debt
30 Appedix B. Robustness check
This appendix presents the results for the cross-sectional fixed effect unbalanced regression based on equation 2, where the top row indicates the five dependent variables. Each column presents the coefficient and the t-statistic (between brackets) per variable. The data used in this regression originates from DHS for the period 2002 – 2013. The coefficient significance is denoted by * at 10%, ** at 5%, and *** at 1%.
Variable M&G (%) Shares (%) C&S (%) INS (%) B&O (%)
31 Appendix C.1. Pooled panel regression, transition effects for negative events
This appendix presents the results for the pooled panel regression based on the transition effects for negative events. The dependent variable in the regression is a change in the percentage of financial assets. I control for changes in net total income and the interaction between changes in net total income and a dummy (DNEG) that takes on value 1 for negative events, and zero otherwise. Each column presents the coefficient and the t-statistic (between brackets) per variable. The data used in this regression originates from DHS for the period 2002 – 2013. The coefficient significance is denoted by * at 10%, ** at 5%, and *** at 1%.
Variable ∆M&G (%) ∆Shares (%) ∆C&S (%) ∆INS (%) ∆B&O (%)
C 1.4217 ** 10.9237 ** -26.6250 3.7727 *** 0.1111 (2.8967) (2.1000) (-1.3789) (2.7727) (0.4065) ∆TOTINC -0.0606 -11.3657 19.7657 -1.8144 0.3712 ** (-0.1616) (-1.6940) (1.0875) (-1.4440) (2.9915) DNEG*∆TOTINC -0.1907 9.7420 -13.0665 1.0850 0.1663 (-0.4672) (1.5956) (-0.6951) (0.8151) (0.4536)
Appendix C.2. Pooled panel regression, transition effects for positive events
This appendix presents the results for the pooled panel regression based on the transition effects for negative events. The dependent variable in the regression is a change in the percentage of financial assets. I control for changes in net total income and the interaction between changes in net total income and a dummy (DPOS) that takes on value 1 for positive events, and zero otherwise. Each column presents the coefficient and the t-statistic (between brackets) per variable. The data used in this regression originates from DHS for the period 2002 – 2013. The coefficient significance is denoted by * at 10%, ** at 5%, and *** at 1%.
Variable ∆M&G (%) ∆Shares (%) ∆C&S (%) ∆INS (%) ∆B&O (%)
32 Appendix D. Correlation matrix
This appendix presents the correlation matrix for the explanatory variables used in this research. This correlation matrix serves as a check for multicollinearity between the explanatory variables in the full sample of 17,074 observations. With correlations within the range [-1.00, +1.00], -1.00 indicates perfect negative correlation and +1.00 indicates perfect positive correlation.
