The impact of debt on the investment decision of
Dutch households
by
Lena Weijer
1University of Groningen
Faculty of Economics and Business
MSc Finance
June 2014
Supervisor: Dr. A. Plantinga
Abstract
This paper examines the impact of debt on the investment decision of Dutch households. I use data from the DNB Household survey. Descriptive statistics indicate that households with high levels of debt have higher stock shares and lower bond shares compared to households with low levels of debt. However, performing a fixed effect panel regression gives no significant result for debt. This is due to differences in household characteristics. Households with a low level of debt are generally older, lower educated, more often female headed and have a lower net worth. These characteristics can explain the different investment decisions.
1
1
Preface
2
1. Introduction
Throughout life many households accumulate different types of debt. For example mortgages,
consumer credit and study loans. Over the past years, debt levels have increased substantially. Figure 1 shows the evolution of household debt levels and household debt relative to wealth over the period 2006-2012 according to data from CBS (Centraal Bureau voor de Statistiek). In comparison with other countries, household debts2 in the Netherlands are among the highest in the world (Economics in picture, 2013). It is interesting to know whether this affects the investment decision. In recent years, there is more interest in the study of household portfolios. There is a lot of evidence that households do not invest effectively and make investment mistakes (Campbell, 2006). For example, while theory predicts that all households should invest some amount of their wealth in stocks, in reality a large part of the households have no stocks at all. Another example is the evolution of the portfolio composition over the life cycle. While theory predicts that investors should shift their investment portfolio to safer assets when becoming older, empirical findings show the opposite. There are many factors influencing portfolio choice. Can the investment decision of Dutch household also be explained by debt? By comparing households with different levels of debt over time, we can investigate the effect of debt on the household portfolio, such that we can provide an understanding of the effects of debt on future assets demand.
Figure 1 - Evaluation of household debt in the Netherlands over the period 2006-2012
Panel A shows the average amount of household debt. Panel B shows the household debt as percentage of wealth. Source: CBS
A households’ balance sheet reveals a households’ assets, liabilities and net worth at a certain point in time. The debit side (asset portfolio) and the credit side (way of financing) of the balance sheet are clearly connected. However, standard portfolio choice models concentrate solely on the asset side of the balance sheet ignoring the other side (Samuelson 1969, Merton 1969). Although more researchers acknowledge the importance of debt in explaining household’s portfolio choice, there is no agreement
2
3 about the direction of the relationship. Becker and Shabani (2010) find that households with mortgage debt are 10% less likely to own stocks and 37% less likely to own bonds compared to similar
households without mortgage debt. In contrast, Flavin and Yamashita (2002) suggest that it is optimal for households to subsidize stock ownership using mortgage debt. These articles focus only on mortgage debt, while other kinds of debt, like consumer debt, may also be relevant in this context.
The aim of this study is to investigate the impact of debt on the share of liquid financial wealth invested in stocks and bonds. According to a portfolio choice model in the presence of debt (Becker and Shabani, 2010), I expect that higher debt leads to a higher stock share and a lower bonds share. I use data from the DNB Household Survey (DHS), which contains information on demographics, work, housing, health, income, wealth and psychological concepts for about 2000 households in the Netherlands. In particular, the data about wealth, which includes extensive information on ownership and amounts held in different asset and debt categories, is important for this research question. Descriptive statistics clearly indicate that households with a high level of debt have higher stock shares and lower bond shares compared to households with a low level of debt. However, the fixed effect panel regression, which controls for individual specific effects, gives no significant result for debt. These results hold for both mortgage debt and consumer credit. Apparently, households with a high level of debt have other characteristics than households with a low level of debt. These
underlying variables can explain the different investment decisions.
4
2. Literature review
Life cycle models that preclude borrowing can generate realistic equity holdings, but when borrowing is taken into account, a problem is presented. In the classic Merton-Mossin-Samuelson (MMS) model of portfolio choice, borrowing leads to huge highly levered equity positions. The fact that debt is widely available and used, drives researchers like Davis, Kubler and Willen (2006) and Becker and Shabani (2010) to construct a model which delivers realistic behaviour for both equity holdings and borrowing.
The relation of debt and portfolio choice is particularly relevant given the high and increasing levels of household debt in the Netherlands. Table 1 presents the debt characteristics of Dutch
households in the period 2003-2012 according to the DHS. It can be seen that households can hold many kinds of debt. While Davis, Kubler and Willen (2006) concentrate on unsecured borrowing, Becker and Shabani (2010) concentrate on mortgage debt (secured debt), the most common and largest form of debt on the household’s balance sheet.
Table 1
Characteristics of debt held by Dutch households
This table represents the characteristics of Dutch household debt for all sample members. The sample consists of 3773 households over the period 2003-2012. The first column shows the percentage of households who own a certain type of debt. The second and third column show respectively the median and mean balance of a certain type of debt, conditional on ownership of that kind of debt.
Type of debt % of households Median balance Mean balance
Private loans
Extended lines of credit Outstanding debt* Finance debts
Loans from family/friends Study loans
Credit card debts
Loans not mentioned before Mortgages on house
Mortgages on second house
Mortgages on other pieces of real estate Checking accounts with a negative balance
3,2 % 11,4 % 1,5 % 1,6 % 3,1 % 3,8 % 3,7 % 0,9 % 47,5 % 0,6 % 1,5 % 9,8 % € 6.000 € 3.877 € 700 € 7.100 € 5.000 € 7.325 € 980 € 30.000 € 107.000 € 79.000 € 106.535 € 600 € 15.252 € 9.387 € 1.249 € 34.188 € 17.031 € 11.121 € 1.775 € 74.404 € 127.477 € 121.383 € 163.149 € 2.818 Total debt 63.0 % € 84.287 € 109.170
5 A number of other studies show that debt should have an impact on the portfolio choice, though from a different perspective. For example, Sharpe and Tint (1990) have introduced an
approach in which liabilities are considered in the asset allocation strategy for dealing with liabilities. Their method allows the effect of liability and asset comovement to be seen as a utility benefit for the portfolio. In this way, investors can find the best asset mix, while taking existing liabilities into account.
2.1 Literature on limited equity market participation, housing and the life cycle
Until now, I emphasized the importance of debt for the investment decisions of households, but did not speak about the direction of this relationship. Next, a number of articles which concentrate on limited equity participation, the effect of housing and the life cycle will be discussed in order to look at the precise influence of debt on the portfolio choice.
First, this research is related to the literature for limited equity market participation. Financial theory dictates that it is optimal for all households to have some amount of stocks. Not only because it provides a high return, but also because households can benefit from risk diversification. However, in reality many households have no stocks in their portfolio. As an indication, 87.2% of households in the DHS panel do not own stocks. The equity participation puzzle provides a challenge to household finance. There are many explanations for limited equity participation; fixed costs of equity market participation (Vissing-Jorgensen, 2002), financial literacy (Van Rooij, Lusardi, Alessie, 2011), entrepreneurial risk (Heaton and Lucas, 2000) and according to Becker and Shabani (2010) debt is an explanation for limited equity market participation too.
Second, this research adds to the literature examining the effect of housing on portfolio choice. Housing differs from other financial assets in that it serves as both consumption good and an
6 although empirically there is no systematic relationship. Chetty and Szeidl (2009) mention that it is critical to separate the effects of mortgage debt and home equity. They find that mortgage debt induce substantial reductions in the share of liquid financial wealth held in stocks, while home equity wealth raises stock ownership.
Third, this research adds to the literature that focuses on portfolio choice over the life cycle. There is rich literature about how investors should choose consumption and portfolio allocation in order to maximize expected utility. Seminal references in this context are Samuelson (1969), Mossin (1968) and Merton (1969). The Merton-Mossin-Samuelson (MMS) models generate two sharp predictions. First, all investors should participate in the risky asset market at all ages. Second, the optimal fraction of wealth invested in risky assets is constant over the life cycle and independent of wealth and age. This conclusion is reached under the assumption of a frictionless market in the absence of labor income. However, in a realistic life cycle setting markets are incomplete and there is uncertain and tradable labor income. According to portfolio choice models in the presence of non-tradable income (Heaton and Lucas 1997, Vinceira 2001), equity shares should decline and bonds should increase over the life cycle. The explanation is that initially households choose an optimal share of wealth to invest in risky assets. Non-tradable future labour income is considered as an implicit safe asset holding. As the life-cycle progresses and future labour income is realized, the implicit holding of safe assets under the form of future labour income decreases. Investors compensate for this decrease by shifting portfolio allocation towards bonds, which are a tradable form of safe assets. Becker and Shabani (2010) incorporate the presence of mortgage debt in the portfolio choice model. The life cycle evolution of portfolio choice according to their model is shown in figure 2. At the start of the life cycle, households have only future labor income. Next to this, the purchase of a house is financed with mortgage debt. By paying off their mortgage debt, real estate will grow. As households realize labor income, they allocate it between risky assets and safe assets. Paying off mortgage debt is considered as investing in a safe asset. As long as a household has mortgage debt it prefers to repay rather than to invest in bonds, since the interest costs on mortgage debt is higher than the return on bonds. After the mortgage is paid off, the household begin to invest its realized labor income in bonds. We are interested in the share of liquid financial wealth invested in stocks and bonds. Liquid financial wealth is defined as the sum of stocks and bonds. As can be seen in figure 2, households invest their liquid portfolio fully in stocks in the beginning of life and shifting portfolio allocation towards bonds after the mortgage is paid off. Thus, stock share decreases and bond share increases over the life cycle in the presence of mortgage debt too.
7 equities equal to 100 minus the age. Obviously this rule is too simplistic. It ignores risk aversion and other individual specific characteristics, but it is correct in that equity shares should decrease over the life cycle.
Figure 2 – Life-cycle evolution of portfolio shares
This figure shows the life-cycle evolution of stocks share and bonds share in the presence of mortgage debt. The figure is based on a figure of Becker and Shabani (2010). The grey area indicates the liquid financial portfolio, which is defined by the sum of stocks and bonds.
Empirically, however the opposite is true. Cocco (2004) uses a life cycle model with housing and find that equity shares tend to increase over the life cycle, while bonds decrease over de life cycle. Due to investment in housing, younger and poorer investors have limited financial wealth to invest in stocks, which reduces the benefit to equity market participation. Flavin and Yamashati (2002) find the same results, though with a different explanation. Young household have a high holding in real estate relative to net worth. They are highly leveraged and therefore have a high portfolio risk. Households will respond to this by either paying down mortgage debt or buy bonds instead of stocks. In contrast, older households have accumulated greater wealth and have a lower ratio of housing to net worth. Therefore, stockholdings are more attractive for older households.
2.2 Hypothesis
According to the discussed literature debt has an impact on the investment decision of households, although there is no agreement about the direction of the relationship. I base my
hypothesis on the paper of Becker and Shabani (2010), who consider a simple portfolio choice model in the presence of debt.
8 investment. According to standard portfolio choice models, households without a mortgage will choose an equity share equal to the ratio of the expected excess return to the price and quantity of risk. The equity share of households with the option to retire their mortgage debt depends on the mortgage interest rate. Becker and Shabani (2010) discuss two implications of the model. First, households having mortgage debt have less benefit from stock ownership, so having mortgage debt decreases the probability of stock market participation. Second, households having mortgage debt will forgo bond ownership in favour of paying down mortgage debt. Having a mortgage decrease the probability of bond market participation by a greater amount than the decline in the probability of equity market participation.
Until now, we only look at ownership of stocks and bonds. Now we focus on stocks and bonds relative to liquid financial portfolio. According to above reasoning, having mortgage debt causes stock holdings to decline and bond holdings to decline even more. When liquid wealth is the sum of stocks and bonds, this will result in a higher equity share and a lower bond share. Therefore my hypotheses are:
H0: Households having higher debt have a higher share of stocks in their liquid financial portfolio.
H0: Households having higher debt have a lower share of bonds in their liquid financial portfolio.
9
3. Methodology
I have performed two panel regressions in order to assess the impact of debt on the investment decision of households. The dependent variables are the share of financial wealth invested in stocks (%Stocks) and the share of financial wealth invested in bonds (%Bonds). The independent variable of interest is in both cases total debt. Control variables include the household characteristics gender, age, number of children, education, risk tolerance, net income and net worth.
Most of the control variables are categorical. For these I created dummy variables. In order to prevent the dummy trap, I excluded one of the categories in the regression. The reference household is a male, aged between 35 and 45, has mediate education, mediate risk tolerance and no children. The distribution of the financial variables net income, net worth and total debt are skewed, so the natural logarithms of these variables are included in the regression analysis. The estimated equation for investment in stocks and bonds is shown respectively in equation (1) and (2).
( ) ( ) ( ) (1) ( ) ( ) ( ) (2)
is the percentage of the liquid financial portfolio invested in stocks. is the percentage of the liquid financial portfolio invested in bonds. is a dummy variable which takes value 1 if the head of the household is a woman and 0 otherwise. The variables , , and denote the dummy variables for the intervals, age < 35, 45 ≤ age < 55, 55 ≤ age < 55 and age ≥ 65 respectively. The dummy variables and denote respectively low and high education. The dummy variables and denote respectively low and high self-reported risk tolerance. For the precise way in which risk tolerance is measured, see appendix A. Each dummy variable takes on the value 1 if household i belongs to the specific category at time t and 0 otherwise. The variables , and are respectively the number of children, net income, total debt and net worth (defined by total assets minus total debt) for household i at time t.
10 matter in investment decisions is acknowledged by Barder and Odean (2001). In addition, Cocco (2004) use the presence of housing and household composition in his research.
11
4. Data
Positive household finance, exploring what households actually do, requires high quality data that are hard to obtain. According to Campbell (2006) the ideal data set for positive household finance has the following characteristics: It should be a representative sample of the entire population, it measures total wealth and provided an exhaustive breakdown into relevant asset classes, it should have a high level of accuracy and it should follow households over time (panel data). The data used in this paper, the DNB Household Survey (DHS), satisfy the characteristics mentioned by Campbell (2006), although the level of accuracy is questionable. Like any survey, the DHS relies on willingness of households to participate and the accuracy of the answers they give when they do participate.
The DHS is conducted among households in the CentERpanel, a panel which comprises around 2,000 households and is representative for the Dutch population aged 16 year and older. The DHS consist of six distinct questionnaires and collects information on demographic, work, housing, health, income, wealth and psychological concepts. All databases use surveys over the period 2003-2012. The wealth questionnaire is of great importance for this research. It contains an extensive breakdown of household wealth holding. DHS asks details with respect to ownership and value of 40 assets and liabilities categories.The other questionnaires are mainly used for the control variables, such as gender, age, education, household composition, net income and risk attitude.
The investment decision is often not made individually, but takes the whole household into account. By converting the individual data to household level data, I had to make a distinction between financial and non-financial information. For financial information I have added up the amounts. This is possible because DHS prevents double reporting (CentERdata 2013). Joint assets are mentioned by only one member of the household, the ‘head of the household’. This person has been selected to report not only personal assets but also joint assets of the households. For the other variables, which are often categorical, I use the information of the ‘head of the household’, because it is assumed that this person does the financial administration, makes financial decisions and has the largest influence on the portfolio choice.
12 The original sample consist of 44,598 observations (18,367 households), while the main analysis contains only 8,943 observations (2,256 households). This is a quite large loss of
observations. The major reason for this is the conversion of the data set from individual level data to household level data, 26,231 observations are lost. In addition, 4,247 observations are lost because of missing values in the dependent variables; stocks and bond relative to liquid financial wealth. Finally, adding independent variables with missing values to the model leads to the following consecutive losses:net income (1,698), net worth (935), risk attitude (2,539), education (5).
Table 2
Household’s balance sheet
This table describes the household’s balance sheet for all sample members from the DHS panel with complete information about their wealth holdings. The sample consist of 14,786 observations over 3,430 households in the period 2003-2012. The first column show the percentage of household who own a certain asset or liability. The second column shows the mean balance of a certain asset or liability over all households. The third and fourth column show respectively the composition of the liquid financial wealth and the total wealth.
Asset/liabilities % own Mean
balance
% liquid fin. wealth
% total wealth
Checking and saving accounts 93.8% € 24,317 87.5% 22.5%
Bonds 4.4% € 2,179 1.4% 0.4%
Stocks 12.8% € 3,790 3.8% 0.9%
Mutual funds 19.3% € 6,044 7.3% 2.0%
Total liquid financial wealth 94.2% € 36,330 100.0% 25.8%
Defined contribution plans 23.3% € 4,662 2.2%
Cash value of life insurance 23.7% € 5,808 2.3%
Employer-sponsored saving plan 33.1% € 1,370 1.9%
Real estate
House 64.0% € 181,496 52.8%
Other real estate 3.7% € 10,723 1.4%
Business equity 3.8% € 2,841 1.0%
Stocks of durable goods 78.3% € 8,799 11.6%
Other financial assets 10.5% € 2,129 1.0%
Total wealth 98.5% € 254,158 100.0
Mortgage and real estate debt 48.6% € 64,271
Consumer credit 18.3% € 2,411
Study loans 4.2% € 465
Negative balance checking account 9.8% € 277
Other debt 4.3% € 1,315
Total debt 63.0% € 68,739
Net worth (total assets less total debt) 91.5% € 185,419
Total wealth 98.5% € 254,158
13 necessary for my research question. I use the same aggregation of wealth components as Alessie, Hochguertel, Van Soest (2002), resulting in 12 asset components and 5 debt components3. One exception is the mutual funds. Alessie et al. (2002) defines mutual funds by mutual funds plus growth funds. However, the DHS contains a lot of missing values for growth funds, and since 3 percent of the households own growth funds, I did not include grow funds in my research.
As shown in table 2, checking and saving accounts is the most common asset in the household portfolio. With 87.5% of the liquid financial portfolio it is the most important asset in the investment portfolio of Dutch household, indicating that Dutch households hold safe investment portfolios. Concentrating on stocks and bonds, we see that 4.4% of the households participate in the bond market and 12.8% of the households participate in the stock market. This indicated that limited equity market participation is also present in the Netherlands. For my research, I am interested in the share of stocks and bonds as percentage of the liquid financial portfolio. Liquid financial portfolio is defined as the sum of checking and saving accounts, bonds, stocks and mutual funds. With a share of respectively 1.4% and 3.8%, bonds and stocks does not seem to be important in the liquid financial portfolio. However, the standard deviation of the bond share is more than 6 times the mean bond share value and the stock share is more than 3 times the mean stock share value (Table 3). This indicates that there is large variation in bond and stock shares among Dutch households. For the debt section on the balance sheet we see again that mortgage debt is the most common and important kind of debt for household. Almost half of the Dutch households have mortgage debt. Subsequent, the risky consumer credit is with an ownership rate of 18.3% used most often.
Table 3 shows the summary statistics of the variables used in the empirical analysis. I use a lot of dummy variables for the demographic characteristics like gender, age, marital status, household composition, region and education. In prevention of the dummy trap, one category of each variable is leaved out. Besides this we have three kinds of psychological variables; risk tolerance, time preference and financial knowledge. The precise way in which we measure these abstract concepts can be found in appendix A. Finally, there are financial variables which measure income and wealth.
3
14
Table 3 Summary statistics
This table presents the summary statistics (mean, median and standard deviation) for all variables used in this paper. The data consist of all sample members of the DHS panel, which consist of 18,367 observations over 4,162 households. However, due to missing values, the number of observations is for some variables smaller.
Variable Obs Mean Median Std. Dev.
Households investment
Stock share (%Stocks) 14,120 3.84 0 14.21
Bond share (%Bonds) 14,120 1.43 0 8.86
Demographic characteristics Woman (DWoman) 18,367 .24 0 .43 Age < 35 (DAge1) 18,367 .14 0 .35 45 ≤ Age < 55 (DAge2) 18,367 .22 0 .41 55 ≤ Age < 65 (DAge3) 18,367 .21 0 .41 Age ≥ 65 (DAge4) 18,367 .23 0 .42 Nr. of children (NChild) 18,367 .70 0 1.06
Nr. of household members (NMem) 18,367 2.43 2 1.28
Married (DMar) 15,786 .59 1 .49
North region (DNorth) 18,342 .11 0 .32
South region (DSouth) 18,342 .23 0 .42
East region (DEast) 18,342 .20 0 .40
Low education (DEduL) 18,352 .04 0 .19
High education (DeduH) 18,352 .53 1 .50
Psychological characteristics
Low risk tolerance (DRiskL) 12,181 .51 1 .50
High risk tolerance (DRiskH) 12,181 .11 0 .31
Short time preference (DTimeS) 13,056 .32 0 .47
Long time preference (DTimeL) 13,056 .16 0 .36
Low financial knowledge (DKnowL) 12,936 .16 0 .37
High financial knowledge (DKnowH) 12,936 .03 0 .18
Financial/wealth characteristics
Net income (Income) 14,587 24,043.36 23,130.40 23,909.27
Net worth (NetWorth) 14,786 185,419.00 118,252.00 272,115.10
House ownership (DHouse) 16,678 .64 1 .48
Total surplus of house (Surplus) 16,678 126,711.70 67,000.00 198,817.40
Total debt (Debt) 14,786 68,738.86 16,051.00 106,640.20
Total mortgages (Mortgage) 16,665 63,701.45 0 100,958.20
15
5. Results
I am interested in the relationship between debt and the share of financial liquid wealth invested in stocks and bonds. As a starting point, I therefore summarized the stocks and bonds shares for different levels and kind of debts. Table 4 shows the average share of financial liquid wealth invested in stocks and bonds for different kinds and levels of debt over the period 2003-2012. They are respectively 3.84% and 1.43%. For all measures of debt, households with low levels of debt (less than the mean value) have a lower stock share than households with high levels of debt (more than mean value). The differences are quite large, indicating that debt has a positive effect of the stock share. This is consistent with the hypothesis. For bonds the results are less clear. In accordance with the
hypothesis, for mortgage debt and consumer credit, households with low levels of debt (less than the mean value) have a higher bond share than households with high levels of debt (more than mean value). However, total debt shows the opposite.
This result is not influenced by time. Table 6 shows the same information, but now for the period 2003-2007 and 2008-2012 separately. This table yields the same conclusion. In both periods the results indicate that a higher debt level leads to a higher stocks share. For bonds the result is dependent on the way debt is measured.
Table 4
Stocks and bonds shares for different levels of debt
This table reports the share of liquid financial wealth invested in stocks and bonds for different levels of total debt, mortgage debt and consumer credit. Low debt means below the average debt value and high debt means above the average debt value. The overview is based on data from the DHS panel in the period 2003-2012.
%Stocks %Bonds
All households 3.84% 1.43%
Low total debt < 68,739 3.06% 1.33%
High total debt > 68,739 5.22% 1.59%
Low mortgages < 63,701 3.12% 1.46%
High mortgages > 63,701 5.15% 1.37%
Low consumer credit < 2,187 3.79% 1.50%
16
Table 5
Stocks and bonds shares for different levels of debt in two time periods
This table reports the share of liquid financial wealth invested in stocks and bonds for different levels of total debt, mortgage debt and consumer credit in two distinct time periods. Low debt means below the average debt value and high debt means above the average debt value. The overview is based on data from the DHS panel in the period 2003-2012.
Period 2003-2007 Period 2008-2012
%Stocks %Bonds %Stocks %Bonds
All households 4.50% 1.44% 3.17% 1.41%
Low total debt < 68,739 3.53% 1.42% 2.57% 1.24%
High total debt > 68,739 6.20% 1.48% 4.21% 1.70%
Low mortgages < 63,701 3.66% 1.46% 2.56% 1.45%
High mortgages > 63,701 6.04% 1.41% 4.24% 1.34%
Low consumer credit < 2,187 4.48% 1.51% 3.10% 1.49%
High consumer credit > 2,187 4.66% 0.96% 4.75% 0.74%
Table 4 and 5 give an indication of the relationship between debt and portfolio composition, although it is not very precise. There is not taken into account that some of the data points are related because they are from the same household or the same year. Something within the household or within the time period may impact the portfolio composition, so we need to control for this.
17
Table 6
Fixed effect regression
This table reports the results of two fixed effect panel regressions explaining the percentage of stocks and bonds in the liquid financial portfolio. The first column shows for each variable the estimated coefficients and standard errors (between brackets) for the model with percentage of stocks in the liquid financial portfolio (%Stocks) as a dependent variable. The second column shows for each variable the coefficients and standard errors (between brackets) for the model with percentage of bonds in the liquid financial portfolio (%Bonds) as a dependent variable. The regression is based on data from the DHS panel in the period 2003-2012.The reference household is a male aged between 35 and 45 with mediate education and mediate risk tolerance. The coefficient significant by 5% are denoted by *, at the 1% level by ** and at the 0.1% level by ***.
Variable %Stocks %Bonds
Woman -2.6989* -0.0194 (1.2210) (0.7939) Age< 35 0.7438 -0.5033 (0.7940) (0.5162) 45 ≤ Age < 55 -3.3407*** 0.5183 (0.6998) (0.4550) 55 ≤ Age < 65 -5.4546*** 0.1366 (0.9179) (0.5968) Age ≥ 65 -6.5292*** -0.4908 (1.0805) (0.7025) Nr. of children 1.1731** 0.0275 (0.3988) (0.2593) Low education -5.7213* 2.2304 (2.7550) (1.7912) High education -2.0565 -0.1688 (1.1870) (0.7718)
Low risk tolerance -0.3840 -0.0660
(0.3138) (0.2040)
High risk tolerance 0.2948 -0.1804
(0.4862) (0.3161)
LOG (Net income) 0.0320 -0.0050
(0.0411) (0.0267)
LOG (Total debt) 0.0464 0.0023
(0.0437) (0.0284)
LOG (Net worth) 0.5403** -0.0631
18 It results that gender, age, number of children, education and net worth have a significant influence on the stock share. The most interesting result is that older households invest significantly less in stocks. This result is consistent for all categories with age above 45 years. Apparently, Dutch households behave like life cycle theory subscribes as explained in section 2. In contrast to many other empirical findings, like Cocco (2004) and Flavin and Yamashati (2002), who find that equity shares tend to increase over the life cycle. The result that woman invest a lower share of the portfolio in stocks is consistent with the literature about gender and investing. Barber and Odean (2001) find that because women are less overconfident than men, they trade less. Less trading will result in a liquid financial portfolio with a higher share of checking and saving accounts and a lower stock share. The increase in stockholdings with net worth also reconciles empirical findings, like Heaton and Lucas (2000). Surprisingly, risk tolerance is not significant. According to classic theory about optimal portfolio allocation (Tobin 1958), allocations across assets will vary across individuals due to
differences in risk attitude. It is well known that risk tolerant investors invest more in stocks than risk averse investors. However, Kapteyn and Teppa (2002) also noted that risk tolerance appears to have little explanatory power in portfolio choice.
None of the variables have a significant influence on the bond share. For the model explaining bond share the F statistic is not significant, meaning that there is no evidence that any of the
coefficients are different from zero. In this model, none of the explanatory variables helps explaining the investment in bonds. Therefore, I will shift my focus more towards explaining the investment in stocks.
The variable of interest, total debt, does not have a significant influence on stock and bond share. So, a change in debt does not significantly effects the portfolio composition within a household. I performed the same analysis with extra control variables. In turns out that changing the specification still leads to the conclusion that debt does not have a significant influence of stocks and bonds shares. The results are in appendix table B1. In the first specification two extra household characteristics are included; marital status and the region in which the household live. The same variables as in the original specification have a significant influence. Besides this, households living in the east region of the Netherlands invest significantly less in stocks. In the second specification two more psychological variables are included; time horizon and financial knowledge. The same variables as in the original specification have a significant influence. Moreover, households with a low self-reported financial knowledge invest significantly more in bonds.
19 variable which measures the surplus value of the house to the model instead of net worth. Almost the same variables as in the original specification have a significant influence. Not surprisingly, in this case surplus value of the house instead of net worth is significant. In the second specification I included consumer credit instead of total debt. Almost the same variables as in the original
specification have a significant influence. In turns out that changing the measure of debt still leads to the conclusion that debt does not have a significant influence on stocks and bonds shares.
The result from the fixed effect regression that debt does not have a significant influence on portfolio choice is surprising. Apparently, the significant control variables for household
characteristics can explain the results from table 4 and 5. This is the case when households with high levels of debt have other characteristics than households with low levels of debt. To investigate this, I compare the distribution of households over different characteristics for households with low total debt and households with high total debt (Table 7). In order to make a good comparison, I use the same categories of household characteristics as in table 4 and 6.
The first characteristic is age. According to the fixed effect panel regression in table 6, age has a significant negative effect on stock share. This means that when the household becomes older, the stock share will decline. In table 7 we see that households with a lower debt balance have a higher age
Table 7
Relation between total debt and the significant control variables.
Each household belongs to a certain category for the household characteristics age, education, gender and net worth. This table shows the distribution of households over the different characteristics for households with low total debt (column 1) and households with high total debt (column 2). Column 3 displays the mean level of total debt for all household characteristics.
% of households within this class if total debt < 68,739 % of households within this class if total debt > 68,739
Mean total debt
Overall 100% 100% 68,739 Age classes Age < 35 11% 15% 87,713 35 ≤ Age < 45 16% 24% 89,937 45 ≤ Age < 55 20% 25% 79,601 55 ≤ Age < 65 23% 20% 59,084 Age ≥ 65 30% 15% 42,043 Education classes Low Education 4% 2% 37,343 Mediate Education 48% 36% 50,932 High Education 48% 63% 85,398 Gender classes Woman 26% 20% 58,083 Man 74% 80% 72,042
Net worth classes
Net worth < 18,419 32.6% 13.0% 49,856
20 and vice versa. 53% of the households with low debt are aged 55 years and older compared to 35% of the households with high debt. So while it seems to be that households with less debt have a lower stocks share (table 4), age is one of the underlying variables which accounts for this relation. Households with low debt are characterized by a higher age and a higher age leads to a significant lower stock share.
The second characteristic is education. According to the fixed effect panel regression in table 6, having low education has a significant negative effect on stock share. Meaning that lower educated households have a lower stock share. In table 7 we see that lower educated households have less debt. 52% of the households with low debt have low or mediate education compared to 38% of the
households with high debt. While it seems to be that households with less debt have a lower stock share (table 4), education is one of the underlying variable which accounts for this relation. Households with low debt are characterized by a lower education and a lower education leads to a lower stock share.
The third characteristic is gender. According to the fixed effect panel regression in table 6, women have significant negative effect on stock share. Meaning that households with a female head have a lower stock share. In table 7 we see that households with female head have less debt. 26% of the households with low debt are female headed, compared to 20% of the households with high debt. While it seems to be that households with less debt have a lower stocks share (table 4), gender is one of the underlying variables which account for this relation. Households with low debt are have more often a female head and having a female head leads to a significantly lower stock share.
The fourth characteristic is net worth. According to the fixed effect panel regression in table 6, net worth has a significant positive effect on stock share. Meaning that households with higher net worth have a higher stock share. In table 7 we see that households with more net worth have more debt. 67.4% of the households with low debt have high net worth, compared to 87% of the households with high debt. While it seems to be that households with less debt have a lower stocks share (table 4), net worth is one of the underlying variables which account for this relation. Households with low debt are characterized by a lower net worth a lower net worth leads to a significantly lower stock share.
21
6. Conclusion
In this paper the impact of debt on the investment decision of Dutch households is
investigated. Researchers acknowledge that debt should have an impact on portfolio choice, but there is no agreement about the direction of the relationship. I use data from the DNB Household survey (DHS), which contains information on demographics, work, housing, health, income, wealth and psychological concepts for about 2000 households in the Netherlands. Descriptive statistics clearly indicate that households with a high level of debt have a higher stock share and a lower bond share compared to households with a low level of debt. This holds for both mortgage debt and consumer credit. However, the fixed effect panel regression, which controls for individual specific effects, gives no significant result for debt. So, a change in debt does not significantly effects the portfolio
composition within a household.Different model specifications and different measures of debt do not change the result. There seems to be a clear relation between debt and investment decision, but that is due to underlying control variables. Households with different levels of debt have different
characteristics. Specifically, households with low levels of debt are generally older, lower educated, more often female headed and have a lower net worth than households with a high level of debt. These underlying variables can explain the different investment decisions. An interesting finding is that Dutch households seem to behave in accordance with life cycle theory. In contrast to earlier empirical findings, who find that older household invest more in stocks, my results show that when a household becomes older it shifts their portfolio away from stocks.
22
References
Alessie, R., Hochguertel, S., Van Soest, A., 2002. Household portfolio in the Netherlands. In Guiso L., Haliassos, M., and Jappelli, T. (Eds.), Household portfolios, Cambridge, 340-388.
Barber, B. M., Odean, T., 2001. Boys will be boys: Gender, Overconfidence, and Common Stock Investment. Quarterly Journal of Economics 116 (1), 261-292.
Becker, T.A., Shabani, R., 2010. Outstanding debt and the household portfolio. Review of Financial Studies 23, 2900-2934.
Campbell J.Y., 2006. Household finance. Journal of finance 61, 1553-1604. CentERdata, 2013. DNB Household Survey 2012 Documentation English.
Chetty, R., Szeidl, A., 2010. The effect of housing on portfolio choice. Unpublished working paper. National bureau of economic research, Cambridge.
Cocco, J.F., 2004. Portfolio choice in the presence of housing. Review of Financial Studies, 18, 535 - 567.
Cocco, J.F., Gomes, F.J., Maenhout P.J., 2005. Consumption and Portfolio Choice over the Life Cycle. Review of Financial Studies 18(2), 491-533.
Davis, S.J., Kubler, F., Willen, P., 2006.Borrowing costs and the demand for equity over the life cycle. Review of Economics & Statistics 88, 348-362.
Economics in picture, 2013. Household debt versus disposable income. Available at:
http://www.economicsinpictures.com/2013/01/household-debt-versus-disposable-income.html (Accessed: 1 April 2013)
Flavin, M.,Yamashita, T., 2002. Owner-occupied housing and the composition of the household portfolio. American Economic Review 92, 345-362.
Heaton, J., Lucas, D., 2000. Portfolio choice and asset prices: The importance of entrepreneurial risk. Journal of Finance 55(3). 1163-1198.
Kapteyn, A., Teppa, F., 2001. Subjective measures of risk aversion, fixed costs, and portfolio choice. Journal of Economic Psychology 32(4), 564-580.
23 Merton, R.C., 1969. Lifetime Portfolio Selection under Uncertainty: The continuous-Time Case. Review of Economics and Statistics 51(3), 245-257.
Mossin, J., 1968. Optimal multiperiod portfolio policies. Journal of Business 41(2), 215-229. Samuelson, P., 1969. Lifetime portfolio selection by dynamic stochastic programming. Review of Economics and Statistics 51(3), 239-246.
Sharpe, W.F., Tint, L.G., 1990. Liabilities: a new approach. Journal of Portfolio Management, 16, 5-10.
Tobin, J., 1958. Liquidity Preference as Behavior Towards Risk. Review of Economic Studies 67, 65-86.
Van Rooij, M., Lusardi, A., and Alessie, R., 2011. Financial literacy and stock market participation. Journal of Financial Economics, 101 (2). 449-472
Vinceira, L.M., 2001. Optimal Portfolio Choice for Long-Horizon Investors with Nontradable Labor Income. Journal of Finance 56, 433-470.
Vissing-Jorgensen, A., 2002. Towards an explanation of household portfolio choice heterogeneity: Nonfinancial income and participation cost structures. Unpublished working paper. National bureau of economic research, Cambridge.
24
Appendix A – Psychological variables
For the variables risk tolerance, time preference and financial knowledge I use questions from the “Economic and Psychological Concept” survey. The exact wording of the questions is as follows:
Risk Tolerance
Please indicate on a scale from 1 to 7 to what extend you agree with the statement. “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.”
Totally disagree 1 2 3 4 5 6 7 totally agree
Households who answer 1 or 2 are considered to have a high risk tolerance, households who answers 6 or 7 are considered to have a low risk tolerance.
Time preference
People use different time-horizons when they decide about what part of income to spend, and what part to save. Which of the time-horizons mentioned below is in your household most important with regard to planning expenditures and savings?
1. the next couple of months 2. the next year
3. The next couple of years 4. The next 5 to 10 years 5. More than 10 years from now
Households who answer 1 are considered to have a short time horizon, households who answers 4 or 5 are considered to have a long time horizon.
Financial knowledge
“How knowledgeable do you consider yourself with respect to financial matters?” 1. Not knowledgeable
2. More or less knowledgeable 3. Knowledgeable
4. Very knowledgeable
25
Appendix B - Supplementary tables
Table B1 - Fixed effect regression (other control variables)
This table reports the results of two fixed effect panel regressions explaining the percentages of stocks and bonds in the liquid financial portfolio using extra control variables. For each independent variable the estimated coefficients and the standard errors (between brackets) are listed. The regression is based on data from the DHS panel in the period 2003-2012.The reference household is a male aged between 35 and 45 with mediate education and mediate risk tolerance. The coefficient significant by 5% are denoted by *, at the 1% level by ** and at the 0.1% level by ***.
Variable %Stocks %Bonds %Stocks %Bonds
Woman -3.0999* -0.0262 -3.1470* -0.2283 (1.2595) (0.8193) (1.2727) (0.8238) Age< 35 0.8185 -0.4793 0.8612 -0.3804 (0.7998) (0.5203) (0.8252) (0.5342) 45 ≤ Age < 55 -3.2955*** 0.5134 -3.3794 0.4440 (0.7046) (0.4583) (0.7246) (0.4690) 55 ≤ Age < 65 -5.4182*** 0.1310 -5.5205*** 0.0545 (0.9240) (0.6010) (0.9467) (0.6128) Age ≥ 65 -6.4729*** -0.4901 -6.5453*** -0.6991 (1.0868) (0.7070) (1.1125) (0.7201) Married -1.1103 -0.0269 (0.7141) (0.4645) Nr. of children 1.2248** 0.0296 1.2386** 0.0269 (0.4015) (0.2612) (0.4104) (0.2656) North region -3.0062 -0.7549 (3.2332) (2.1032) South region -1.6453 -0.8383 (3.0202) (1.9647) East region -4.8215* -1.4943 (2.2328) (1.4525) Low education -5.9144* 2.2251 -6.1235* 2.4678 (2.7609) (1.7960) (2.9145) (1.8865) High education -2.0865 -0.1749 -2.1026 -0.1655 (1.1907) (0.7746) (1.2381) (0.8014)
Low risk tolerance -0.3726 -0.0572 -0.3946 -0.0409
(0.3149) (0.2049) (0.3229) (0.2090)
High risk tolerance 0.2818 -0.1767 0.3376 -0.1576
(0.4883) (0.3177) (0.5030) (0.3256)
Short time preference -0.0800 -0.2066
(0.3174) (0.2055)
Long time preference -0.6007 0.2251
(0.3887) (0.2516)
Low financial knowledge 0.2110 0.8235**
(0.4890) (0.3165)
High financial knowledgde 1.3343 0.7270
(0.8978) (0.5811)
LOG (Net income) 0.0359 -0.0056 0.0365 -0.0028
(0.0412) (0.0268) (0.0424) (0.0274)
LOG(Total debt) 0.0465 0.0024 0.0414 -0.0023
(0.0439) (0.0286) (0.0452) (0.0293)
LOG (Net worth) 0.5701** -0.0604 0.5694** -0.0609
26 Table B2 - Fixed effect regression (other measures for debt)
This table reports the results of two fixed effect panel regressions explaining the percentages of stocks and bonds in the liquid financial portfolio using different specifications for debt. For each independent variable the estimated coefficients and the standard errors (between brackets) are listed. The regression is based on data from the DHS panel in the period 2003-2012.The reference household is a male aged between 35 and 45 with mediate education and mediate risk tolerance. The coefficient significant by 5% are denoted by *, at the 1% level by ** and at the 0.1% level by ***.
Variable %Stocks %Bonds %Stocks %Bonds
Woman -4..2132*** 0.0158 -2.7189* -0.0258 (1.1467) (0.7528) (1.2204) (0.7935) Age< 35 0.8131 -0.6119 0.7243 -0.4998 (0.7104) (0.4663) (0.7943) (0.5164) 45 ≤ Age < 55 -3.0475*** 0.5076 -3.3092*** 0.5125 (0.6691) (0.4392) (0.7007) (0.4556) 55 ≤ Age < 65 -4.8434*** 0.1557 -5.4316*** 0.1279 (0.8783) (0.5766) (0.9188) (0.5974) Age ≥ 65 -5.9408*** -0.4549 -6.5146*** -0.5018 (1.0416) (0.6837) (1.0811) (0.7029) Nr. of children 1.1427** 0.0586 1.1725** 0.0289 (0.3746) (0.2459) (0.3988) (0.2593) Low education -4.7153 21.098 -5.7200* 2.2421 (2.5642) (16.832) (2.7549) (1.7912) High education -1.7083 -0.1349 -2.0723 -0.1661 (1.1298) (0.7416) (1.1872) (0.7719)
Low risk tolerance -0.5854 -0.0803 -0.3870 -0.0661
(0.2993) (0.1965) (0.3138) (0.2040)
High risk tolerance 0.0146 -0.2090 0.2797 -0.1801
(0.4596) (0.3017) (0.4862) (0.3161)
LOG (Net income) 0.0233 -0.0147 0.0339 -0.0050
(0.0388) (0.0254) (0.0410) (0.0267)
LOG (Mortgage debt) -0.0066 0.0340
(0.0432) (0.0284)
LOG(Surplus value of house) -0.2682** -0.1580**
(0.0843) (0.0554)
LOG (Consumer credit) 0.0640 -0.0081
(0.0586) (0.0381)
LOG (Net worth) 0.5048* -0.0669
