The Crowding Out Effect of Housing Wealth on Risky Financial Assets: Evidence from the Netherlands

(1)

The Crowding Out Effect of Housing Wealth on Risky

Financial Assets: Evidence from the Netherlands

Student name: Rongrong Tang Student number: S2364212 Study Program: MSc Finance

Supervisor: Dr. Plantinga

ABSTRACT

(2)

2

1. Introduction

For many individual investors, housing wealth and risky financial investment play crucial roles in their portfolio choices. Housing wealth is different from financial assets, since it also serves as a durable good which satisfies the basic needs of its owners. Compared to stocks, derivatives or other risky financial assets, housing wealth shows different characteristics because it is usually illiquid and financed with a lot of debt. Stocks, derivatives or other risky financial assets are normally liquid with limited leverage constraint for individual investors. The objective of this study is to investigate how housing wealth can crowds out the risky financial assets held by individual investors. In this paper, we examine this effect based on data from the Netherlands.

(4)

4

diversification effect. The crowding out effect is confirmed for individuals in the US, it is interesting to test it for individuals in the Netherlands.

Despite the growing literature regarding individual’s portfolio selection and house ownership1, many questions still remain unsolved. Previous studies only link risky financial assets to housing wealth. No psychological factor is involved. In this research, we use individual’s self-reported risk attitude as an explanatory variable. As to the dependent variable, we use the allocation of shares and options relative to total financial wealth. It can represent individual portfolio choice.

We also study other socio-economic factors as potential explanations for the allocation of shares and options relative to total financial wealth. These factors are included as control variables, such as age, gender, education, occupation, household size, region, net income and other personal debts. Various scientific studies have confirmed the impacts of those variables on the amount of risky financial assets, studies such as Miyata (2003), Binswanger (1980), Duasa and Yosuf (2013), Dohmen et al. (2011) and Davis et al. (2006). These socio-economic factors along with the housing wealth variable and the self-reported risk attitude tend to shed new light and get a comprehensive view on the relationship between housing wealth and risky financial assets. To achieve the research objective, we use data from the De Nederlandsche Bank (DNB) Household Survey2, which was launched in 1993 and contains information on work, pensions, housing, mortgages, income, assets, loans, health, economic and psychological concepts, and personal characteristics. The survey comprises data from over 2,000 households participating in the CentER panel.

The remaining of the paper is organized as follows. In Section 2, we present a review on the existing literature regarding the research topic. In Section 3, a brief data availability and description is given. In Section 4, we provide the methodology used in this paper. The test results and explanation are given in Section 5. In the final section, the conclusion is presented.

1_{Brueckner (1997), Cocco (2005), Faig and Pauline(2002), Flavin and Yamashita (2002), Henderson and} Ioannides (1983), Hochguertel and Van Soest (2001), Yao and Zhang (2005) etc.

2

(5)

5

2. Literature Review

Previous studies present various findings on individual portfolio choice in the existing of housing wealth. Brueckner (1997) investigates that the impact of housing wealth and investment motivations on individual portfolio choice. He uses the housing investment-consumption model of Henderson and Ioannides (1983) along with the mean-variance portfolio framework introduced by Fama and Miller (1972). The housing investment-consumption model contains an investment constraint of homeowners. The results show that when the investment constraint is not considered, the optimal portfolio for individual investors is a combination of the market portfolio and the risk-free assets. However, when the investment constraint is taken into consideration, housing investments distort an individual’s portfolio choice. To be specific, there is overinvestment in housing due to a rational balancing of the consumption benefits and portfolio distortion related to housing investment. One drawback is that the investment constraint does not apply to individuals who are renting a house.

(6)

6

A similar finding is provided by Faig and Shum (2002), whose model is associated to asset pricing models combing liquid with illiquid assets. They argue that illiquid projects have either entrepreneurial risk or home ownership risk. Entrepreneurial risk comes from individual investments in their own businesses. And home ownership risk is from housing investment. They find that those personal illiquid projects are a crucial determinant of individual portfolio choice and those individuals who are saving to invest in their own businesses or in their own houses have significantly safer financial portfolios. This suggests that they reduce risky assets holding due to the investments in illiquid projects.

Hochguertel and Van Soest (2001) evaluate the interrelationship between financial and housing wealth from Dutch households. They argue that financial decisions and housing investment decisions essentially belong to one same decision process and therefore should be evaluated simultaneously. They define financial wealth as the sum of saving accounts, time deposit accounts, saving certificates and certificates of deposit, shares in domestic and foreign companies, shares in investment funds, options, bonds, and mortgage bonds. They treat both housing and financial wealth as dependent variables and estimate the joint demand of individual investors for financial and housing wealth. They consider both renters and homeowners and thereby make up the drawback of Brueckner (1997). The results show that demand for financial wealth for homeowners and renters is systematically different, while housing wealth is not affected by whether or not financial wealth is held. And higher house prices not only reduce house ownership but also decrease the financial wealth holding.

(7)

7

individuals who own a house hold less equity in their net worth (the substitution effect), on contrary, individuals who rent a house hold more equity in their net worth (the diversification effect). Renters and house owners’ portfolio choices have different determinants and their reactions to relevant variables, such as age and net income, are also different.

Cocco (2005) goes deeper into the analysis of the first effect addressed by Yao and Zhang (2005). He estimates individual portfolio choice and treats housing not only as an asset in individuals’ portfolios but also as consumption good. He finds that housing wealth crowds out stock holdings and this effect is larger for younger individuals with lower financial wealth. He also addresses the reasons why the crowding out effect exists. First, house price risk underlies housing investment. Second, housing investments reduce people’s willingness to pay the fixed cost for stock market participation. Third, shocks of net income are significantly positively correlated with shocks of house price. Poorer investors tend to further reduce their stock holdings. From the discussion of the literature, it seems fair to conclude that the investment in housing has crucial implications for individual portfolio choice.

The objective of this research is to investigate whether the crowding out effect defined by Cocco (2005) exists among individual investors in the Netherlands. Naturally, the hypothesis one is that the amount of housing wealth has negative effect on the allocation of shares and options relative to total financial wealth held by individual investors according to the DNB Household Survey data between 2002 and 2012.

3. The Data

3.1 Data availability and formation

(8)

8

survey was started in 1993 to collect information on work, pensions, housing, mortgages, income, assets, loans, health, economic and psychological concepts, and personal characteristics from over 2,000 Dutch households. The representativeness and comprehensiveness allow me to investigate the research question in detail. The survey is completed through the internet. One restriction of this data set is that not all respondents answered every question. The non-response rate is over 50% in certain variables regarding personal wealth and risk attitude. The data set is organized in nine files each year. Four files are chosen for each year to construct a panel data set used in the research, including the aggregate income file, the aggregate wealth file, the psychological concepts file and the household information file. For the entire research horizon from 2002 to 2012, 49,580 observations are included in the research. Each observation includes a unique identifier and information on participation year, age, gender, highest level of education completed, primary occupation, number of children, region, net income, mortgages, other personal debts, total value of housing wealth, total market value of risky financial assets and total market value of financial assets. A psychological question reflecting individual’s attitude towards risk in general is also included in each observation. The response to the question, “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”, ranges from “totally disagree” (score of 1) to “totally agree” (score of 7). Thus, a lower score implies a higher level of risk taking. The responses on several questions are missing in some cases. Over 50% of questions regarding income level and the amount of personal wealth are not completed by the respondents reflecting a self-defensive mechanism towards some personal information.

(9)

9 𝑅𝐹𝐴𝑅 =𝑇𝑜𝑡𝑎𝑙 𝑅𝑖𝑠𝑘𝑦 𝐹𝑖𝑛𝑎𝑛𝑐𝑖𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠 𝑇𝑜𝑡𝑎𝑙 𝐹𝑖𝑛𝑎𝑛𝑎𝑐𝑖𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠 = 𝑇𝑜𝑡𝑎𝑙 𝑅𝑖𝑠𝑘𝑦 𝐹𝑖𝑛𝑎𝑛𝑐𝑖𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠 𝑇𝑜𝑡𝑎𝑙 𝑅𝑖𝑠𝑘𝑦 𝐹𝑖𝑛𝑎𝑛𝑐𝑖𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠 + 𝑇𝑜𝑡𝑎𝑙 𝑁𝑜𝑛 − 𝑟𝑖𝑠𝑘𝑦 𝐹𝑖𝑛𝑎𝑛𝑐𝑖𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠

=

_{𝑆𝑡𝑜𝑐𝑘𝑠 + 𝐶𝑎𝑙𝑙 𝑂𝑝𝑡𝑖𝑜𝑛𝑠 +𝑃𝑢𝑡 𝑂𝑝𝑡𝑖𝑜𝑛𝑠 + 𝑇𝑜𝑡𝑎𝑙 𝑁𝑜𝑛−𝑟𝑖𝑠𝑘𝑦 𝐹𝑖𝑛𝑎𝑛𝑐𝑖𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠}𝑆𝑡𝑜𝑐𝑘𝑠 + 𝐶𝑎𝑙𝑙 𝑂𝑝𝑡𝑖𝑜𝑛𝑠 + 𝑃𝑢𝑡 𝑂𝑝𝑡𝑖𝑜𝑛𝑠 (1)

(10)

10

balance sheet such as redeeming mortgages or inheriting a fortune. Information relating to such events is not included in the survey, but fortunately these events are not systematic and therefore their impacts are limited.

Table 1 also shows the components of housing wealth and other personal debts. On average, total value of houses occupies 81.80% of individual’s total assets, specifically, the proportion of owner of first house as a percentage of individual’s total assets is 75.78%, and the proportion of owner of a second house is 1.79%. However, house ownership usually couples with mortgages. The corresponding mortgages on houses account for 78.85% of individual’s total liability in which 73.59% is attributed to the mortgages on first house. Thus, total value of housing wealth in each observation is the difference between total value of houses and mortgages. To be specific, total value of housing wealth is equal to the sum of first house, second house and real estate minuses the sum of mortgages on first house, second house and real estate. Because the previous papers find significant evidence that mortgages influence the individual stock holdings. For example, Becker and Shabani (2010) find that individuals with mortgages are 10% less likely to participate in the stock market and 37% less likely to purchase bonds compared to similar individuals with no mortgages. Chetty and Szeidl (2009) also point out that mortgages can reduce individual stock holdings due to home price risk. Therefore total value of housing wealth used in this research eliminates the impact of mortgages and represents the true housing surplus of individual investors. Besides, this research also includes other personal debts as a control variable. They account for 21.15% of individual’s total liability, including private loans, study loans and credit card debts etc.

Table 1

The decomposition of RFAR, housing wealth and other personal debts

This table indicates the components of RFAR, housing wealth and other personal debts and their mean allocations. Total market value of risky financial assets is made up of the sum of the absolute market value of shares, call-option bought, call-option written, put-option bought and put-option written. Total market value of financial assets comprises total risky financial assets and total non-risky financial assets. Total value of housing wealth is defined as total value of houses net of mortgages. Total amount of other personal debts is the sum of all personal debts excluding mortgages.

(11)

11

Total risky financial assets Total amount shares/companies 1.45%

(1.48%) Total amount put-options bought & written 0.01%

Total amount call-options bought & written 0.02% Total non-risky financial assets Total amount savings or deposit accounts 7.13%

(15.92%) Total amount checking accounts 1.04%

Total amount deposit books 0.16%

Total amount savings certificates 0.18%

Total amount empl.-sponsored sav. plan 1.30% Total amount single-premium annuity

insurance policies 1.79%

Total amount savings/endowment

insurance policies 1.15%

Total amount mutual funds/accounts 2.30%

Total amount bonds and/or mortgage bonds 0.88% Total amount money lent out family/friends 0.47% Total amount savings or investments not

mentioned before 0.33%

Total value of houses Total amount of real estate 4.23%

(81.80%) Total amount owner of house 75.78%

Total amount owner of a second house 1.79% Total amount of mortgages Total amount mortgages on real estate 3.27%

(78.85%) Total amount mortgages on the house 73.59%

Total amount mortgages second house 1.99% Total amount of other personal Total amount private loans 0.70% debts (21.15%) Total amount debts hire-purchase contract 0.07% Total amount debts with mail-order firms 0.77% Total amount loans from family/friends 0.68%

Total amount study loans 0.51%

Total amount credit card debts 0.21%

Total amount loans not mentioned before 18.21%

(12)

12

DNB Household Survey. Individuals with low education level are much less likely to invest in stock market. According to Davis et al. (2006), the amount of personal debts can either increase or decrease equity demand of individual investors, based on the cost of personal debts. Except factors mentioned before, Duasa and Yusof (2013) report several significant determinants of risk preference, including region and household size. Dohmen et al. (2011) find that height and parental background also play a role in individual’s risk preference. To gain a comprehensive view on the determinants of individual portfolio choice, this research includes relevant variables mentioned above as control variables. To be specific, age dummy takes a value of one for people who are younger than 52 years old and at their early stage of their career. Education dummy is equal to one for people who complete their highest education above the middle-level applied education (the so called Mbo in Dutch). Occupation dummy takes a value of one for people who are lack of working experience, including students and people who firstly enter the labor market. Children dummy takes a value of one for individuals with at least one child and zero otherwise. The west in the Netherlands are more developed than elsewhere within the country, as a result, it is reasonable to expect people living in the largest three cities which are Amsterdam, Rotterdam and the Hague will hold more financial wealth than those living elsewhere. So region dummy are created based on this expectation. The individual’s self-reported attitude towards risk, the risk dummy are created to separate risk-seeking individuals from risk-averse individuals. As to continuous control variables, they could induce heteroscedasticity which affects the statistical power of the test results. Table 2 indicates the White test results for pooled observations.

Table 2

The White test for heteroskedasticity

The dependent variable is residual^2. The method is Least Squares. And the included observations are 14249.

F-statistic 2.562 Prob. F(58,14190) 0.000

Obs*R-squared 147.652 Prob. Chi-Square(58) 0.000

Scaled explained SS 6374.103 Prob. Chi-Square(58) 0.000

(13)

13

logarithm for net income, housing wealth and other personal debts to scale those variables. Second, we use panel-corrected standard errors (PCSE) introduced by Beck and Katz (1995) in the regression. Beck and Katz (1995) find that the PCSE method produces accurate coefficient standard errors estimation at no, or little, loss in efficiency.

3.2 Summary Statistics

(14)

Table 3

Summary statistics for common and individual samples

This table shows the summary statistics including the mean, median, maximum, minimum, standard deviation, Jarque-Bera and the number of observations for various variables. The table does not include the summary statistics of dummy variables. Due to the missing data problem caused by high non-responses rates, the number of common observations is over a half less than that of individual observations.

Panel A: Summary statistics for common samples

Variables Mean Median Maximum Minimum Std. Dev. Jarque-Bera Observations

Total Risky Financial Assets 3877.597 0.000 1025950.000 0.000 31481.640 1.09E+08 14976

Total Financial Assets 43004.310 13023.030 3702125.000 -349621.000 115197.900 5.50E+07 14976

Risky Financial Assets Ratio 0.039 0.000 4.907 -1.223 0.161 4.37E+06 14976

Housing 152275.500 8877.500 5500000.000 -34500.000 232984.100 2.11E+06 14976

Mortgage 49472.550 0.000 2101102.000 0.000 99826.050 1.16E+06 14976

Housing Wealth 102802.900 25.000 5500000.000 -1126750.000 201533.500 5.46E+06 14976

Net Income 19453.050 17938.100 689704.100 -3243.000 22314.770 1.54E+07 14976

Other Personal Debts 14917.590 0.000 1303755 0.000 69213.220 1.92E+06 14976

(15)

15

Panel B: Summary statistics for individual samples

Variables Mean Median Maximum Minimum Std. Dev. Jarque-Bera Observations

Total Risky Financial Assets 1290.206 0.000 1025950.000 0.000 17831.900 3.27E+09 49580

Total Financial Assets 15867.170 0.000 3702125.000 -349621.000 70007.970 1.02E+09 49580

Risky Financial Assets Ratio 0.033 0.000 4.907 -2.395 0.151 6.59E+06 20356

Housing 71335.790 0.000 5500000.000 -34500.000 175145.500 1.38E+07 49580

Mortgage 23873.580 0.000 3502000.000 0.000 77004.310 8.85E+07 49580

Housing Wealth 47462.200 0.000 5500000.000 -3499500.000 146165.600 4.93E+07 49580

Net Income 8439.731 0.000 706429.700 -3243.000 17695.280 9.09E+07 49580

Other Personal Debts 6401.896 0.000 1303755.000 0.000 45941.320 3.47E+07 49580

Age 40 41 102 0 29.344 8.57E+09 48426 Gender 1 1 1 0 0.500 8256.333 49538 Education 5 5 9 1 1.991 2863.347 49388 Occupation 5 6 13 0 4.011 3488.824 49530 Children 1 1 7 0 1.256 3681.634 49538 Region 3 3 5 1 1.421 4075.338 49399 Risk Attitude 5 6 7 1 1.794 2814.553 20595

(16)

16

4. Methodology

This research applies Panel Least Squares method (panel regression) to deal with the panel data set. Various variables discussed in the previous section are used in a linear regression model, and the model is shown as follows:

𝑅𝐹𝐴𝑅_𝑖𝑡 = 𝛼 + 𝛽₁𝐻𝑂𝑈_𝑖𝑡+ 𝛽₂𝑁𝐼_𝑖𝑡+ 𝛽₃𝐷𝐸𝐵𝑇_𝑖𝑡+ 𝛽₄𝐴𝐺𝐸_𝑖𝑡+ 𝛽₅𝑆𝐸𝑋_𝑖𝑡+ 𝛽₆𝐸𝐷𝑈_𝑖𝑡+

𝛽₇𝐽𝑂𝐵_𝑖𝑡+ 𝛽₈𝐾𝐼𝐷_𝑖𝑡+ 𝛽₉𝑅𝐸𝐺_𝑖𝑡+ 𝛽₁₀𝐴𝑇𝑇_𝑖𝑡+ 𝜀_𝑖𝑡 (2)

(17)

17

(18)

18

Table 4

The Collinearity Diagnostics for independent variables

The table shows the eigenvalue, condition index and correlation matrix for every independent variable in 11 dimensions. The dependent variable is RFAR.

Model Dimension Eigenvalue Condition Index Variance Proportions (Intercept) Housing wealth (log) Net income (log) Other personal debts(log) Age (dummy) Gender (dummy) Education (dummy) Occupation (dummy) Children (dummy) Region (dummy) Risk attitude (dummy) 1 1 6.064 1.000 0.000 0.010 0.000 0.000 0.000 0.010 0.010 0.010 0.010 0.000 0.010 2 0.991 2.473 0.000 0.020 0.010 0.030 0.050 0.020 0.000 0.020 0.160 0.170 0.000 3 0.846 2.677 0.000 0.080 0.000 0.010 0.010 0.020 0.010 0.000 0.010 0.620 0.020 4 0.810 2.736 0.000 0.010 0.010 0.920 0.000 0.010 0.000 0.000 0.000 0.000 0.020 5 0.694 2.955 0.000 0.010 0.000 0.020 0.000 0.000 0.020 0.010 0.000 0.060 0.840 6 0.433 3.741 0.000 0.080 0.010 0.000 0.030 0.090 0.340 0.010 0.330 0.060 0.010 7 0.329 4.294 0.010 0.130 0.340 0.000 0.010 0.000 0.380 0.010 0.050 0.050 0.030 8 0.305 4.462 0.010 0.010 0.140 0.010 0.070 0.140 0.090 0.260 0.340 0.000 0.000 9 0.260 4.831 0.000 0.580 0.010 0.000 0.010 0.630 0.090 0.030 0.020 0.010 0.010 10 0.177 5.860 0.000 0.050 0.000 0.000 0.790 0.020 0.010 0.650 0.060 0.000 0.000 11 0.091 8.156 0.970 0.020 0.470 0.000 0.020 0.050 0.060 0.010 0.040 0.020 0.060

(19)

19

To determine the appropriate panel technique, we run the redundant test to check whether fixed effects panel regression is appropriate. Meanwhile, we run the Hausman test to check whether random effects panel regression is appropriate. The results are shown in Table 5. From Panel A, the F-statistic of cross-section/period is 8.054 with a p-value of 0.000 which means the null hypothesis should be rejected and thereby the cross-section and time period fixed effects panel regression should be used in the research. At the same time, from Panel B, the chi-square statistics is 52.728 with a p-value of 0.000 which means that the fixed effects test is favorable rather than cross-section random effects test. In addition, the period random effects test is also be rejected because the chi-square statistics is 38.065 with a p-value of 0.000. Hence, the appropriate approach used in this research is the Panel Least Squares with cross-section and time period fixed effects.

To conclude what stated above, this linear regression model contains socio-economic factors, the self-reported risk attitude, as well as the housing wealth and therefore gains a comprehensive view on individual portfolio choice. It allows attributing the change in individual’s risky financial assets to the change in several control variables as well as the change in the housing wealth. For more in-depth analysis, we apply the panel regression to evaluate whether those variables have impacts on the risky financial assets, especially the crowding out power of the housing wealth on the risky financial assets.

Table 5

The results of redundant test and Hausman test

The table indicates which panel regression approach is appropriate for the data set. Panel A: the redundant test results for cross-section and time fixed effects

Effects Test Statistic d.f. Prob.

Cross-section F 8.046 -383110397.000 0.000 Cross-section Chi-square 19627.350 3831.000 0.000 Period F 4.429 -1010397.000 0.000 Period Chi-square 60.574 10.000 0.000 Cross-Section/Period F 8.054 -384110397.000 0.000 Cross-Section/Period Chi-square 19665.420 3841.000 0.000

Panel B: the Hausman test results for cross-section random effects

(20)

20

Cross-section random 52.728 10.000 0.000

Panel C: Hausman test results for period random effects

Test Summary Chi-Sq. Statistic Chi-Sq. d.f. Prob.

Period random 38.065 10.000 0.000

5. Discussion of Results

5.1. Primary results

(21)

21

effect. First, the risk of house prices could cause individual investors to avoid stock market participation in case of a housing market collapse. Second, the illiquidity of housing investments further reduces stock holdings of poorer investors.

Furthermore, we observe the coefficient of Other Personal Debts (log) is -0.003 with a p-value of 0.004. The result implies that people with higher personal debts tend to reduce their risky financial assets. It is consistent with the findings from other literature on borrowing and the demand for equity (see, e.g.,Davis et al., 2006 and Becker and Shabani, 2010). In the presence of outstanding debts, individual investors tend to incorporate the debt repayment into their investment decisions. The reason behind this phenomenon is the costs of borrowing those investors are undertaking. Both papers present significant evidence that realistic debts with high interest rate dramatically decrease the equity participation of individual investors. In addition, it should be noted that no other control variables showing statistically significant impacts on the movement of RFAR, which is inconsistent with the existing literature thus far (see, e.g., Jagannathan and Kocherlakota, 1996; Brown et al., 1999; Carlsson et al., 2005; Dohmen et al., 2011 and Duasa and Yusof, 2013). One possible explanation is that, for simplicity, this model only includes one dummy variable for one control variable. It is more likely to obtain significant results if multiple dummy variables are created for one control variable. From this perspective, this problem is necessary to be addressed in further research. The other reasonable explanation is that these insignificant results are likely due to the missing data problem caused by high non-response rates. We will address this problem in the robustness test section.

Table 6

The primary results of Panel Least Squares with dependent variable of RFAR

The table shows the primary regression results of this research. The method is Panel Least Squares with cross-section and time period fixed effects. The research uses PCSE method to get accurate coefficient standard errors. The sample period ranges from 2002 to 2012. For Panel A and B, the number of cross-sections is 5,053 and 3,731 respectively. The total number of unbalanced panel observations is 19,606 and 13,552 respectively.

Panel A: the unconditional test results of Panel Least Squares with cross-section and time period fixed effects

(22)

22

Housing Wealth (log) -0.002 0.043

Intercept 0.038 0.000

R-squared 0.730

Adjusted R-squared 0.636

S.E. of regression 0.092

Sum squared resid 124.011

Log likelihood 21815.050

F-statistic 7.771

Durbin-Watson stat 1.579

Panel B: the conditional test results of Panel Least Squares with cross-section and time period fixed effects

Variable Coefficient Probability

Net Income (log) 0.000 0.462

Other Personal Debts (log) -0.003 0.004

Age (dummy) 0.002 0.703 Gender (dummy) 0.002 0.640 Education (dummy) -0.006 0.499 Occupation (dummy) -0.004 0.282 Children (dummy) 0.000 0.937 Region (dummy) -0.013 0.411

Risk Attitude (dummy) 0.004 0.191

R-squared 0.751

F-statistic 8.135

5.2. Robustness test results

(23)

23

enough to run the test due to the missing data problem. In Panel A of Table 7, the coefficient of Housing Wealth (log) is -0.002 with a p-value of 0.304 during 2002 to 2009. The result shows that housing wealth does not play a role in individual portfolio choice during 2002 to 2009. However, in Panel B of Table 7, the coefficient of Housing Wealth (log) is -0.002 with a p-value of 0.038 during 2005 to 2012. Besides, in Panel A, the p-values of Occupation (dummy), Children (dummy) and Region (dummy) are 0.022, 0.030 and 0.043 respectively. The results imply that during 2002 to 2009, many socio-economic factors influence individual portfolio choice. However, in panel B, the p-values of Other Personal Debts (log) and Risk Attitude (dummy) are 0.001 and 0.021 respectively. The results suggest that during 2005 to 2012, individual investors pay more attention to personal debt and risk attitude. Compared the results in Panel A to those in Panel B, the crowding out effect of individual investors becomes significant and the impact of other personal debts and risk attitude on the risky financial assets held by individuals also becomes significant during period 2005 to 2012. One plausible explanation is that after experiencing the severe financial crisis in 2008 and sovereign debt crisis in 2010, individuals with unfavorable net income expectations or huge loss of fortune become more prudent and risk averse when dealing with housing, personal debts and risky financial assets. Direct evidence comes from Gibson et al. (2013). They show that risk perception of individuals has an impact on financial risk preference. Furthermore, on average, individual investors have relatively lower risk preference scores in 2013 than those in 2011, suggesting that they perceive the stock market to be riskier after the financial crisis. Ragnarsdóttir et al. (2012) explain why there is a systematic reduction in individual’s risk preference. They argue that financial crisis evokes emotional distress through the deterioration of individual standard of livings, the situation of other people and the expected future incomes. This emotional distress then decreases their satisfaction with life. Because of the lower satisfaction with life along with the propensity for regret, individual investors are more risk averse and more likely to avoid stock market participation after a great financial crisis (see, Pan and Statman, 2012).

Table 7

(24)

24

The table shows the test results of panel regressions with cross-section and period fixed effects. PSCE is used to eliminate heteroscedasticity.

Panel A: the test results from 2002 to 2009 with 3,080 cross-sections and 10,164 observations

Age (dummy) 0.000 0.925 Gender (dummy) 0.000 0.863 Education (dummy) -0.009 0.509 Occupation (dummy) -0.010 0.022 Children (dummy) -0.011 0.030 Region (dummy) -0.013 0.043

R-squared 0.764

F-statistic 7.407

Panel B: the test results from 2005 to 2012 with 3,193 cross-sections and 10,624 observations

Age (dummy) 0.011 0.095 Gender (dummy) 0.006 0.271 Education (dummy) 0.002 0.798 Occupation (dummy) -0.001 0.815 Children (dummy) 0.005 0.519 Region (dummy) -0.005 0.875

R-squared 0.812

(25)

25

F-statistic 9.956

(26)

26

B decreases below 50% after the linear interpolation. However, through this robustness test, one cannot deny that there is higher possibility of significant test results if those missing data points are interpolated.

Table 8

The robustness test results after the linear interpolation in all raw variables

The dependent variable is RAFR. The method is Panel Least Squares with cross-section and time period fixed effects. Cross-section weights (PCSE) standard errors & covariance are used to eliminate heteroscedasticity. The sample period ranges from 2002 to 2012.

Panel A: the unconditional test results of the linear interpolation method. The number of cross-sections is 10,398. And the total number of unbalanced panel observations is 37,709.

R-squared 0.507

F-statistic 2.702

Panel B: the conditional test results of the linear interpolation method. The number of cross-sections is 10,386. And the total number of unbalanced panel observations is 37,558.

Other Personal Debts (log) 0.000 0.115

Age (dummy) -0.004 0.393 Gender (dummy) -0.009 0.783 Education (dummy) 0.008 0.044 Occupation (dummy) -0.007 0.068 Children (dummy) -0.008 0.021 Region (dummy) 0.034 0.007

R-squared 0.509

(27)

27

F-statistic 2.708

6. Conclusions

This paper examines individual portfolio choice between housing wealth and risky financial assets in the Netherlands. To derive the results, we construct a panel data set from the DNB Household Survey. The data set contains 49,580 observations and an 11-year research horizon. There is no multicollinearity between those independent variables according to the collinearity diagnostics table. Panel Least Squares with cross-section and time period fixed effects is the primary regression method applied in this research. To eliminate heteroscedasticity, we use panel-correlated standard errors (PCSE). The primary test results show that individual investors in the Netherlands exhibit the crowding out effect; that is, even after eliminating the impact of mortgages, individual investors tend to reduce their risky financial assets in the presence of housing wealth. In addition, we also find that people with more personal debts tend to hold risky financial assets even less. There is significant evidence that personal debts and risky financial assets are negatively correlated.

(28)

28

stocks, derivatives or other risky instruments. This effect is so called the crowding out effect (see, Cocco, 2005). Third, Hochguertel and Van Soest (2001) find that house prices and restrictions can affect housing wealth as well as financial wealth in individuals in many ways, such as down payment constraints can have an impact on the individuals’ saving behavior. Last, Yao and Zhang (2005) find that owning a house causes individuals to reduce financial wealth (the substitution effect) and renting a house causes them to increasing financial wealth (the diversification effect).

(29)

29

significant and robust for individual investors in the Netherlands. As to the reasons behind this effect, Cocco (2005) argues that the risks associated with house prices and the illiquidity of housing investments lead individual investors to avoid stock market participation in case of a meltdown in housing market.

(30)

30

Appendix

This table shows supplementary information for Table 3. Education, Occupation, Region and Risk attitude are explanatory variables. The DNB Household Survey records the answers to these variables in scores. This table indicates the scores and the corresponding answers for those variables.

Scores Variables

Education Occupation Region Risk

Attitude

1 (Secondary) special education

(low) Employed on a contractual basis Three big cities Totally disagree

2 Nursery, primary or primary

education (low) Works in own business Other west Disagree

3 Secondary education (VMBO)

(mediate)

Free profession, freelance,

self-employed North

Slightly disagree

4 HAVO/VWO (high) Looking for work after

having lost job East Neutral

5 Middle-level applied

education (Mbo)

Looking for first-time

work South

Slightly agree

6 HBO (first or second stage)

(high) Student Agree

7 University education WO

(high) Works in own household

Totally agree

8 Did not have education (yet) Retired (pre-retired, AOW, VUT)

9 Other sort of

education/training Partly disabled

10 Unpaid word, keeping

benefit payments

11 Works as a volunteer

12 Other occupation

13 Too young, has no

(31)

31

References

Beck, N., Katz, J., 1995. What to do (and not to do) with time-series cross-section data. American Political Science Review 89, 634-647.

Becker, T., Shabani, R., 2010. Outstanding debt and the household portfolio. Review of Financial Studies 23, 2900-2934.

Bem, D., 1972. Self-perception theory. In: L. Berkowitz (Ed.), Advances in Experimental Social Psychology. the United States, New York, pp. 1–62.

Benartzi, S., Thaler, R., 2002. How much is investor autonomy worth? Journal of Finance 57, 1593-1616.

Binswanger, H., 1980. Attitudes toward risk: experimental measurement in rural India. American Journal of Agricultural Economics 62, 395-408.

Brueckner, J., 1997. Consumption and investment motives and the portfolio choices of homeowners. Journal of Real Estate Finance and Economics 15, 159-180.

Carlsson, F., Daruvala, D., Johansson-Stenman, O., 2005. Are people inequality-averse, or just risk-averse? Economica 72, 375-396.

Charness, G., Gneezy, U., 2012. Strong Evidence for Gender Differences in Risk Taking. Journal of Economic Behavior & Organization 83, 50-58.

Cocco, J., 2005. Portfolio choice in the presence of housing. Review of Financial Studies 18, 535-567.

Davis, S., Kubler, F., Willen, P., 2006. Borrowing costs and the demand for equity over the life cycle. Review of Economics & Statistics 88, 348-362.

Dohmen, T., Falk, A., Huffman, D., Schupp, J., Sunde, U., Wagner, G., 2011. Individual risk attitudes: measurement, determinants, and behavioral consequences. Journal of the European Economic Association 9, 522-550.

Duasa, J., Yusof, S., 2013. Determinants of risk tolerance on financial asset ownership: a case of Malaysia. International Journal of Business and Society 14, 1-16.

Faig, M., Shum, P., 2002. Portfolio choice in the presence of personal illiquid projects. Journal of Finance 57, 303-328.

(32)

32

Flavin, M., Yamashita, T., 2002. Owner-occupied housing and the composition of the household portfolio. American Economic Review 92, 345-362.

Gibson, R., Michayluk, D., Van der Venter, G., 2013. Financial risk tolerance: an analysis of unexplored factors. Financial Services Review 22, 23-50.

Henderson, J., Ioannides, Y., 1983. A model of housing tenure choice. American Economic Review 73, 98-114.

Hochguertel, S., Van Soest, A., 2001. The relation between financial and housing wealth: evidence from Dutch households. Journal of Urban Economics 49, 374–403.

Jagannathan, R., Kocherlakota, N., 1996. Why should older people invest less in stocks than younger people? Federal Reserve Bank of Minneapolis Quarterly Review 20, 11-23. Pan, C., Statman, M., 2012. Questionnaires of risk tolerance, regret, overconfidence, and

other investor propensities. Unpublished working paper. SCU Leavey School of Business, available at SSRN: http://ssrn.com/abstract=1549912.

Ragnarsdóttir, B., Bernburg, J., Ólafsdóttir, S., 2013. The global financial crisis and individual distress: the role of subjective comparisons after the collapse of the Icelandic economy. Sociology 47, 755-775.

Roszkowski, M., Davey, G., Grable, J., 2005. Insights from psychology and psychometrics on measuring risk tolerance. Journal of Financial Planning April, 66-77.

Schubert, R., Brown, M., Gysler, M., Brachinger, H., 1999. Financial decision-making: are women really more risk-averse? American Economic Review 89, 381-385.

Van Rooij, M., Lusardi, A., Alessie, R., 2001. Financial literacy and stock market participation. Journal of Financial Economics 101, 449-472.

The Crowding Out Effect of Housing Wealth on Risky Financial Assets: Evidence from the Netherlands