The relation between homeownership and portfolio choice

The relation between homeownership and portfolio choice

Evidence from the Dutch market

Bogdan Bogdanov Student Number: s3238237 Supervisor: Prof. Dr. Robert Lensink

MSc Thesis Finance June, 2017 University of Groningen Faculty of Economics and Business

Abstract

This paper studies the impact of homeownership on the financial portfolios of the Dutch households. The relation between housing and portfolio choices is examined by estimating the distinct effects of home equity and property value on the risky share of liquid wealth using three econometric methods: (1) OLS Estimation, (2) Tobit Model, (3) Instrumental Variables approach. The result of the OLS regression shows that an increase in property value is related to an increase in the risky share of liquid wealth. The estimates of the Tobit Model support this effect and claim to be slightly stronger. The Instrumental Variables estimation predicts that an increase in home equity is associated with a decrease in the risky share of liquid wealth.

Additionally, I implement a research method which examines the changes in the households’

financial portfolios from the year before to the year after home purchase, but I do not find any

significant results.

1. Introduction

There are two basic human needs which should be satisfied in order to assure smooth physical survival, namely to procure enough food and to have a dwelling to live in. The food is a good which provision requires small and frequent monetary outlays mostly on a daily basis, it does not require lumpy investments and the purchase decisions are usually spontaneous and not a result of thorough analysis. On the other side, the housing-related decisions are much more complicated and require more deep and diligent analysis. The very first choice faced by the individuals is whether to rent or buy a house, both options have pros and cons and the optimal choice is not an easy decision. In both cases a large amount of the individuals' disposable income will be spend on housing, usually either in the form of rental or mortgage payments.

The large capital demand and complexity make housing-related decisions one of the most important and central financial issues for the households.

However, the contemporary life is more complicated and the individuals usually investment in more types of assets with different liquidity characteristics and risk, such as financial assets.

Nevertheless, for most people the investment in housing remains the largest one and its magnitude, risk and illiquid nature inevitably influence the investment decisions related to the other types of assets.

Current master thesis investigates empirically the relation between housing and financial portfolio choices. Specifically, how the households invest in the presence of owner-occupied housing and how the house value and outstanding mortgage debt affect the financial portfolios and attitude towards risk.

There are several basic theoretical models in the existing literature trying to predict the

influence of housing on the financial portfolios. Grossman and Laroque (1990) studied the

illiquid nature of housing and its influence on the portfolios, Flavin and Yamashita (2002)

showed the effects of the risk introduced by owning a house, Sinai and Souleles (2005)

emphasized on the hedging against house price risk effects. In another study Yao and Zhang

(2005) showed the role of housing in the overall portfolios and the diversification benefits

afforded by the low correlation between the returns of housing and financial assets. Chetty and

Szeidl (2010) developed a theoretical model which shows the distinct effects of property value

and home equity on the financial portfolios. Although the theoretical models have found relatively clear relation between housing and portfolio choices, the empirical studies have found mixed results and no conclusive and systematic relation.

The theoretical model developed by Chetty and Szeidl (2010) and also presented in Chetty, Sandor and Szeidl (2017) is supposedly the most comprehensive and robust in the existing literature, capturing the effects identified by several other models. For this reason, I use the predictions of this theoretical model to formulate the hypothesis which I test empirically using data for the Dutch homeowners’ behavior. In particular, I study the impact of property value and home equity on the risky share of liquid wealth. At first glance, this relation could be examined by simple OLS regression of property value and home equity on the risky share of liquid wealth, however, it does not result in a precise and reliable estimation. Basically, there are two main problems of the OLS estimation: (1) There is a problem originating from the fact that many households do not invest in risky financial assets, which means that the dependent variable is bounded by zero. (2) There is an endogeneity problem because there might be some unobserved factors which influence separately the dependent and the independent variable(s).

To address the first problem I use standard Tobit Model, while for the second I use instrumental variables (IV) approach. Additionally, I study the changes in the financial portfolios from the year before to the year after home purchase.

Considering that the existing empirical studies have found mixed results, it is difficult to say what could be the expected results. The OLS estimation shows that an increase in the mortgage debt is related to an increase in the risky share of liquid wealth. The outcome of the Tobit Model claims that 1,000 Euro increase in mortgage debt is associated with 0.1129 percentage points increase in the risky share of liquid wealth. The result of the instrumental variables approach shows that 1,000 Euro increase in home equity is related to 0.1136 percentage points decrease in the risky share of liquid wealth. In the last estimation, which studies the changes in the financial portfolios from the year before to the year after home purchase I do not find any significant results.

All estimates, however, are not in line with the predictions of the theoretical model developed

by Chetty, Sandor and Szeidl (2017). Later in the paper I discuss the results, the possible reasons

they are against the theoretical predictions and which estimation could be considered as the most accurate.

The reminder of the paper is organized into the following six sections: Section 2 outlines the most relevant existing literature. Section 3 describes the theoretical predictions. Section 4 characterizes the data. Section 5 presents the methodology. Section 6 shows the results and Section 7 concludes.

2. Literature review

In this section I review the most relevant literature related to the relationship between housing and portfolio choices. This particular topic attracted more interest during the past twenty years and most of the studies date from this period. The existing literature consists of theoretical models and empirical studies, both trying to explain how the investments in housing influence the financial portfolios of the individuals. Most of the empirical literature is from the US, though in the recent years some other studies were published, mostly for the European market.

Before the real identification of the specific relationship between housing and portfolio choices, Grossman and Laroque (1990) developed a model of optimal consumption and portfolio selection. In this model the consumption utility is generated by holding an illiquid good for which sale certain transaction cost must be paid. The outcome of the model shows that if the adjustment of the level of the illiquid consumption good is costly, the individuals tend to be more risk-averse after they acquire this particular consumption good. The agents should wait for a larger change in the wealth before they adjust their consumption. Although this model is not explicitly developed to study the relation between housing and portfolio choices, the selling costs and the illiquid nature of housing make the model very suitable for studying specifically the relation between housing and portfolio choices. Therefore, it has been successfully modified and implemented for that purpose by several academic studies, most notably by Yao and Zhang (2005).

The model developed by Yao and Zhang (2005) represents an extension of Grossman and

Laroque (1990) model by introducing: (1) a nondurable numeraire consumption good, (2)

collateral requirements, (3) housing price risk, (4) an uninsurable stochastic labor income. The

model examines the optimal dynamic portfolio decisions for the homeowners and the renters.

The result shows that the determinants of the portfolio decisions of the renters and the homeowners are different and they react differently to some key variables like net worth-to- income ratio and investor's age. The main finding states that when the investors are indifferent between owning and renting a house, the homeowners tend to substitute home equity for risky stocks to make use of the diversification benefits afforded by the low correlation between stocks and housing returns. Moreover, the individuals prefer to rent a house when they face severe liquidity constraints or high mortality rate, when the liquidity constraint is not an issue the individuals prefer to buy a house. The empirical tests using PSID data partly support the predictions of the model and show again that the portfolio choices of the renters and the homeowners are different.

Another basic theoretical model is developed by Cocco (2005) to investigate the effects of house price risk and illiquidity on the stock market participation. It represents a model of optimal consumption and portfolio decisions of an individual with non-tradable human capital, where the owner-occupied housing is considered as an asset within the portfolio with specific utility in the form of consumption services. Therefore, the housing is viewed as an asset in the portfolio as well as a consumption good. After the parameterization of the model with PSID data, the results show that the housing investments explain to large extent the patterns of the cross-sectional variation in the composition of wealth and the stock investments. The younger and poorer investors who have already invested in housing have less financial wealth for investments in stocks, resulting in reduced benefits of the equity market participation.

Furthermore, house price risk generally crowds out the stock holdings both for high and low financial net-worth investors, but for the latter this effect is much larger. This model has been used by Chetty and Szeidl (2010) as a basic for the development of their model.

Flavin and Yamashita (2002) use similar to Cocco (2005) approach to study the optimal portfolio

choices of the homeowners. They also consider housing and financial investments as the two

constitutes of the overall portfolio of assets, the difference is that they use mean-variance

optimization to find the optimal portfolios. However, housing has also certain consumption

utility and its optimal value from the perspective of the households’ consumption needs for

housing services might not be the same as from the portfolio optimization perspective.

Therefore, after the calculation of the risk and the return of both housing and financial assets, they calculate a constrained by the housing consumption needs efficient frontier in the mean- variance space to estimate the optimal portfolio for a given housing constraint. The results show that the households with similar risk preferences and risk aversion invest in financial assets much differently, because each household optimizes its portfolio with respect to different housing constraint. The optimal stock allocation for the younger households is decreasing as the home-to-net worth ratio raises. When the individuals are still young, the housing-related investments represent very large part of the total net worth and as the house purchase is usually financed by a mortgage (the individuals are young and still have not accumulated enough financial wealth) this place the individuals in a situation of higher risk caused by the leverage. Therefore, these households should use their net worth to pay down the mortgage or to buy safety assets in order to decrease the risk they bear. On the opposite side, for the older households the optimal stock holdings are increasing when the home-to-net worth ratio is declining. This model has been empirically tested using SCF data by Yamashita (2002). The outcome shows mixed results, but generally claims that the home-to-net worth ratio has an impact on the portfolio choices and the households with higher ratio invest less in stocks, because owning a house introduces a certain level of housing risk. The differences between the model and the empirical study are explained by Yamashita (2002) with some biases in the empirical estimation caused by the inability to obtain the absolutely precise measure of some key variables such as household wealth.

The methodology proposed by Flavin and Yamashita (2002) was further tested by Waggle and Johnson (2008), they use the same model and assumptions to find the optimal portfolios

. The findings show that the optimal allocation to stocks is higher when home-to-net worth ratio is higher, which is valid for both younger and older individuals. This result is opposing to the results of Flavin and Yamashita (2002). The discrepancy is explained by Waggle and Johnson (2008) with some differences in the way the two papers consider the mortgage financings and not by some fundamental differences.

1There are some minor differences but they could be considered as negligible.

Amongst all models developed in the existing literature so far, the most recognizable one is the model developed by Chetty and Szeidl (2010), which captures the effects identified by some of the most important existing models. The theoretical model of Chetty and Szeidl (2010) and also presented in Chetty, Sandor and Szeidl (2017) estimates the effects of housing on the financial portfolios, more specifically the effects of property value and home equity on the stock share of liquid wealth. The main and the most important finding shows clearly the distinct effects of property value and home equity on the stock share of liquid wealth. It predicts that an exogenous increase in property value is related to a decrease in the stock share of liquid wealth by increasing liquidity, increasing exposure to risk, and reducing the present value of lifetime wealth. On the opposite side, in the case of an exogenous increase in home equity, the stock share of liquid wealth is also increasing with CRRA preferences through a wealth effect. These theoretical predictions are also tested empirically using three research designs: (1) mean house prices, (2) variation in housing supply elasticities, (3) changes in the financial portfolios around home purchase. The results of the empirical estimations support the predictions of the theoretical model and hence confirm its validity. The effects identified by Chetty and Szeidl (2010) are the same as Chetty, Sandor and Szeidl (2017), the latter uses more recent data for the empirical estimations.

Slightly different model has been developed by Sinai and Souleles (2005), which studies homeownership as a hedge against rental risk. The main proposition is that not only the homeowners are exposed to house price risk, but the renters are also considered to bear rental risk. The model predicts that the probability of buying a house increases with the time horizon, the individuals who are planning to live longer in certain accommodation are more likely to buy an own house than to rent. The results of the empirical estimations show that an increase in the rental risk results in an increase in the probability of buying a house.

The reviewed literature in this section up to now encompasses the most important and

comprehensive studies related to the relation between housing and portfolio choices. The most

important predictions could be summarized as follows: (1) Grossman and Laroque (1990)

emphasized on the illiquidity nature of housing, (2) Flavin and Yamashita (2002) showed the

importance of the risk introduced by housing, (3) Yao and Zhang (2005) presented the

diversification benefits of housing, (4) Sinai and Souleles (2005) showed the hedging effects of homeownership. Lastly and most importantly Chetty and Szeidl (2010) and Chetty, Sandor and Szeidl (2017) showed the distinct effects of property value and home equity on the stock share of liquid wealth in a theoretical model which captures the effects identified by all four models mentioned above. Furthermore, the empirical tests support the predictions of the theoretical model.

In the latter part of the current section I review some other important papers in the field, mostly empirical studies.

Another recognizable model is developed by Vestman (2012). It represents a life-cycle model of portfolio choice which endogenize the decision to rent or own a house and the decision to participate on the stock market or not. Two data sets are used to calibrate the model, one for the Swedish and one for the US renters and homeowners. The main finding claims that the homeowners participate on the stock market twice as much as the renters both in the US and Sweden.

Kullman and Siegel (2005) study empirically the effects of homeownership on the risky financial investments of the households. In the basic research they use Panel Study of Income Dynamics (PSID) data from 1984 to 1999 to analyze the effects of the following ratios on the risky investments: (1) house-to-net worth, (2) mortgage-to-net worth, (3) other real estate-to-net worth, (4) business-to-net worth. They find that the real estate exposure (measured by house- to-net worth ratio) is negatively associated with the amount invested in risky assets. On the opposite side, the mortgage debt is positively associated with the risky investments. Generally, the large housing investments are related to less risky financial holdings, but still the renters are less likely to participate on the stock market than the homeowners.

Another important research paper is Shum and Faig (2006) which studies the determinants of

the stock market participation using data from US Survey of Consumer Finances for the period

1992-2001. The results show large heterogeneity in the portfolio constructions amongst the

different households. Stock market participation is positively associated with some wealth

indicators like labor income and financial net worth, some household characteristics like age

and risk attitude and some saving motives. On the other hand, stock market participation is

negatively associated with the alternative risky investments like owning a house or private business equity.

In an empirical test of his own theoretical model Michael C. Fratantoni (1997) examines the proposition that the committed housing-related expenditures result in more safety investments. The results show that the households with larger committed expenditure uncertainty as indicated by the payment-to-income ratio invest less in risky assets. The empirical estimations use data from 1989 Survey of Consumer Finances.

Xiaoqing Hu (2005) investigates how the real constraints related to housing investments affect the individuals’ decision-making process. The study uses a model in which the households endogenously move from renting to owner-occupied housing financed by mortgage debt. The main finding is that owning a house crowds out to certain extent the investments in stocks. If a young or middle age individual does not own a house yet, but could afford it in the near future, it is more likely that this particular individual invests more in safety financial assets. In order to make sure that he/she will be able to finance the house purchase, the financial wealth is held in safety and liquid assets. For the individuals who already own a house the property is perceived as a risky asset and plays the role of a substitute for stocks.

One of the most recent empirical studies in the field is Michelsen, Mocking and van Veldhuizen (2015) which studies the effects of home equity and mortgage debt on the financial portfolio choices using Dutch data from Statistics Netherlands (CBS) for the period 2006-2012. More specifically, they estimate the effects of the house value and outstanding mortgage on the stock share of liquid wealth. The results of their preferred estimations show insignificant impact of home equity and mortgage debt on the stock share of liquid wealth.

In another study for the Dutch market Arthur van Soest (2001) examines jointly the financial

and housing-related decisions of the homeowners and renters. The outcome shows that the

financial wealth demand of the homeowners and the renters are systematically different and

housing wealth is not influenced by the decisions related to financial wealth. This result is

opposite to the perception that the individuals tend to accumulate financial wealth first and

then to buy an own house.

One of the most comprehensive empirical studies in the recent years is Insook Cho (2014) which investigates the relation between housing and portfolio choices across several European countries using data from 2004 Survey of Health, Ageing and Retirement in Europe (SHARE).

The empirical results show that the homeowners in the countries with bank-type economies generally participate on the stock market much less, while in the countries with better- developed stock market, homeownership has almost no impact on the participation rate.

Furthermore, the homeowners invest more in stocks than the renters.

Hwang and Quigley (2002) study empirically the investment behavior of the households which have their portfolios invested in stocks, bonds, T-bills and housing. They show that for short time horizons housing is not presented in the efficient portfolios, while for the longer time horizons the portfolios with lower risk have 15 to 50 percentage points invested in housing. The empirical study uses very rich data set for the Stockholm house prices, which makes it highly reliable in estimating the relation between housing and financial investments.

Brueckner (1997) studies the portfolio choices of the homeowners using a model implementing a constraint in which the investment in housing must be at least as big as the housing consumption. The outcome claims that in the mean-variance space when the constraint is binding the optimal portfolios of the homeowners are inefficient. However, this inefficiency is not driven by possible irrationality of the individuals, it is a consequence of balancing the consumption benefits and portfolio distortion related to the investment in housing.

To summarize, the theoretical models developed so far predict that housing has relatively clear effect on the portfolio choices and identify several mechanisms through which housing determines the portfolio choices. On the opposite side, the empirical literature has found mixed results and no systematic relation between housing and portfolio choices.

3. Theoretical Predictions and Hypothesis Theoretical Predictions

Many theoretical models trying to explain the effects of housing on portfolio choices have been

developed in the existing literature so far. Amongst the most important and recognizable are

these developed by (1) Grossman and Laroque (1990), (2) Flavin and Yamashita (2002), (3) Sinai

and Souleles (2005), (4) Yao and Zhang (2005), (5) Chetty and Szeidl (2010). Although it is difficult to conclude which is the most reliable and accurate, presumably the latter could be considered as the most comprehensive and robust as it captures the effects identified by the first four. Moreover, its theoretical predictions are supported by the numerical solution of the model as well as the empirical estimates of Chetty and Szeidl (2010) and Chetty, Sandor Szeidl (2017).

In the current master thesis I use the theoretical model developed by Chetty and Szeidl (2010) and also presented in Chetty, Sandor Szeidl (2017) as the theoretical predictions which I test empirically using data for the Dutch homeowners’ behavior. As a benchmark throughout the paper I use Chetty, Sandor Szeidl (2017). This particular model is one of the central and the most recognizable in the existing literature, the paper is published in one of the top financial journals (The Journal of Finance), which reassures its high quality and reliability.

Chetty, Sandor Szeidl (2017) developed a stylized two-period model of portfolio choice in the presence of housing in order to show the distinct effects of property value and home equity (treating them as exogenous) on the financial portfolios. Property value is the current market value of the owner-occupied dwelling, home equity equals property value minus the outstanding mortgage debt used to finance the home purchase. The model represents an updated version of Cocco’s (2005) model with some simplifying assumptions, the most important of which is that the households can move only at exogenous random dates to obtain an (approximate) analytic expression for portfolio shares.

The comparative statistics of the model show that an exogenous increase in property value

reduces the stock share of liquid wealth through three channels. First, for a certain amount of

total wealth, an increase in property value implies that a larger share of the wealth is blocked in

housing, which makes the marginal utility higher and more sensitive to shocks in the no-move

state. Second, when the covariance between housing prices growth and stock returns is more

than zero, a higher property value results in higher exposure to home price risk, which limits its

diversification and hedging effects. Third, higher property value leads to higher mortgage debt

as long as home equity is held fixed. Since the mortgage rate exceeds the risk-free rate, larger

mortgage payments decrease the lifetime wealth, resulting in less risky investments.

The second statement of the comparative statistics claims that an exogenous increase in home equity is related to an increase in the risky share of liquid wealth. The individuals with CRRA utility aim to have a constant proportion of their wealth invested in risky assets. Each increase in home equity should result in a proportional increase in the risky share of liquid wealth to make use of the diversification benefits identified by Yao and Zhang (2005).

To confirm and complement the comparative statistics Chetty, Sandor and Szeidl (2017) calculate the model using exact numerical solution. The results confirm clearly the suggested predictions and show once again the distinct effects of property value and home equity on the financial portfolios.

To summarize, the findings of the model predict that an exogenous increase in property value results in a decrease in the stock share of liquid wealth by increasing liquidity, increasing exposure to risk, and reducing the present value of lifetime wealth. On the opposite side, an exogenous increase in home equity is related to an increase in the stock share of liquid wealth with CRRA preferences through a wealth effect.

Hypothesis

The theoretical model of Chetty, Sandor and Szeidl (2017) is used by current paper to formulate the following two hypothesis which are tested empirically:

1) Home equity is positively associated with the risky share of liquid wealth.

2) Mortgage debt is negatively associated with the risky share of liquid wealth.

4. Data

The main data provider is CentERdata via DNB Household Survey (DHS)

, which studies the economic and psychological determinants of the savings behavior of the Dutch households. For the first three estimation methods (OLS, Tobit and IV) I use only 2016 Data Wave, which was conducted over the period April 2016 - October 2016 by DHS. The wave consists of six data sets containing information about the following characteristics of the households: (1) general

2www.dhsdata.nl

information on the household, (2) household and work, (3) accommodation and mortgages, (4) health and income, (5) assets and liabilities, (6) economic and psychological concepts.

For the purposes of the current research the data is processed in order to form a single data set containing all information needed. First, a unique personal index is calculated which serves as an identification number by which the observations (household members) from the different data sets are connected. Not all observations presented in one data set are presented in the others, hence after the connection I ended up with a sample of 1654 observations.

Subsequently, the following observations were excluded: (1) the non-homeowners, (2) the observations with missing data on property value or mortgage debt, (3) the households which bought their houses before 1995

, (4) the households with negative risky investments. The basic characteristics of the households within the final sample used in the first three estimation methods (OLS, Tobit and IV) are presented in Table 1.

Table 1. Basic characteristics of the households within the sample

Mean Median Standard Deviation

Age (years) 50.93 48 14.88

Number of Household members 2.68 2 1.30

Number of Children 0.86 0 1.12

Net Household Income (Euro) 33823 31396 22066

Gross Household Income (Euro) 43800 41744 24174

Housing:

Property Value (Euro) 268085 240000 137128

Mortgage (Euro) 127118 102500 135604

Home Tenure (years) 11.26 10 7.71

Wealth:

Total Wealth (Euro) 319153 256708 238151

3The households which bought their houses before 1995 are excluded, because the data for the instrumental

Liquid Wealth (Euro) 40693 15937 141359

Home Equity (Euro) 114031 107500 140055

Equity In Other Real Estate (Euro) 15111 0 97926

Vehicle Equity (Euro) 10533 6000 18589

Portfolio Allocation:

Percent of Households Holding Risky

Assets 21%

Risky Share (% of liquid wealth) 9.5%

Safe Assets Share (% of liquid wealth) 91%

Number of observations 516

Table 2: displays the basic characteristics of the households within the data sample used in the first

three estimation methods (OLS, Tobit and IV).

In the current paper I define “Net Household Income” as the income of all household members combined after deduction of taxes and social security benefits. “Property Value” as the most recently determined WOZ-value of the owner-occupied house. I use the WOZ-values because they are the most accurate indicators of the real values of the houses. The number shows the price at which each house would have sold for if it was sold on 1 January of the previous year.

“Mortgage” is the amount of the outstanding debt on all mortgages used to finance the

purchase of the current accommodation. “Home Equity” is defined as property value minus the

outstanding mortgage debt used to finance the purchase of the house. “Home Tenure” is

defined as the years living in current accommodation. “Liquid Wealth” is the sum of the

following financial assets held by the households: (1) total amount checking accounts, (2) total

amount savings or deposit accounts, (3) total amount mutual funds, (4) total amount bonds, (5)

total amount stocks. “Total Wealth” is defined as the amount of liquid wealth plus the following

assets held by the households: (1) other real estate excluding current accommodation, (2)

amount of vehicle equity, (3) savings or investments not mentioned before, (4) the WOZ-value

of the current accommodation. Note that the debt obligations are not subtracted from the

amount of total wealth, as proposed by Chetty, Sandor and Szeidl (2017). “Vehicle Equity” is the

amount invested in: (1) cars, (2) motorbikes, (3) boats, (4) caravans/trailers. In the current

master thesis only the investments in stocks and mutual funds are considered as risky financial

investments.

In the data sample 21 % of all households have positive risky financial investments, while for 79

% the sum of the investments in stocks and mutual funds equals zero. The average percentage of liquid wealth invested in risky assets is 9.55, but it should be noted that this percentage is generated from 21% of the households within the sample. If only the households with positive risky investments are considered, the risky share of liquid wealth is 38.5 %.

The implementation of instrumental variables approach requires some additional information about the development of the housing prices over time. The data for that purpose is obtained from Statistics Netherlands (CBS)

, in particular I use “Price index purchase price, 2010=100” for the period 1995-2016. This index shows the price development of the existing own homes located on the territory of the Netherlands over the years. Nonetheless, the price development is not the same across the different parts of the Netherlands, hence for further precision of the estimation I use separate index data for each of the twelve Dutch provinces. Table 2 presents the location distribution across the provinces for the homeowners within the sample.

Table 2. Location distribution across the Dutch provinces

Province % of the total

sample Province % of the total

sample

Groningen 5.23% Zuid-Holland 17.83%

Friesland 5.81% Noord-Brabant 16.28%

Drenthe 4.26% Limburg 7.75%

Overijssel 5.81% Utrecht 6.40%

Flevoland 2.33% Noord-Holland 13.57%

Gelderland 11.63% Zeeland 3.10%

Table 2: displays the location distribution of the owner-occupied houses across the Dutch provinces of

the households within the data sample used in the first three estimation methods (OLS, Tobit and IV).

For the last estimation method, which studies the portfolio changes from the year before to the

year after home purchase I use DNB household surveys for the period 2004-2016 for the

households which bought a house in the period 2005-2015. This means that I observe the

households’ behavior in the year before they bought the house and in the following year

directly from the DHS data waves from 2004 to 2016. The information in which year the house was bought is obtained from DHS Data Wave 2016. Unfortunately, relatively small number of households participate in both the previous and next year surveys and which home purchase year is observable in the later data waves (particularly the 2016 wave). After the data procession I ended up with a sample of 75 observations (homeowners). For 40 of them the observed house purchase is a move from one to another accommodation, for 35 it is the first house purchase and represents a transition from renters to homeowners.

5. Methodology

In this section I describe the methodology used to test the theoretical predictions of the model developed by Chetty, Sandor and Szeidl (2017) and to estimate empirically the impact of housing on the financial portfolios. Based on the model of Chetty, Sandor and Szeidl (2017) throughout the current paper I use property value and home equity as two separate variables which have distinct impact on the risky share of liquid wealth. The main equation for studying the relation between housing and portfolio choices as proposed by Chetty, Sandor and Szeidl (2017) is given by:

Risky Share

= α + β

Property Value

+ β

Home Equity

+γ X

+ ε

(1)

This equation will be referred as “main equation” in the following pages. In current master

thesis I define “Risky Share” as the percentage of liquid wealth invested in mutual funds and/or

stocks, “Property Value” is the most recent WOZ-value of the owner-occupied dwelling, “Home

Equity” is property value minus the outstanding mortgage debt used to finance the home

purchase, X

denotes a set of control variables. As already mentioned earlier in the paper, the

outcome of the simple OLS regression as given in the main equation will be a biased and

inaccurate estimation, which basically suffers from two shortcomings. First, there is a problem

originating from the fact that many households do not invest in risky financial assets (risky

share equals zero), which means that the dependent variable is bounded by zero. I use Tobit

Model which introduces a latent variable to address this problem, as proposed by Michelsen,

Mocking and van Veldhuizen (2015). Second, the portfolio and housing choices are

endogenous and influenced by unobserved factors like future labor income and preferences as suggested by Cocco (2005) and Vestman (2012). Hence, there is an endogeneity problem which makes the OLS estimation unable to identify properly the causal effects of housing on portfolio choices. To account for this problem I use Instrumental variables (IV) approach as proposed by Chetty, Sandor and Szeidl (2017).

Additionally, I observe the changes in the financial portfolios from the year before to the year after home purchase. This method is intended to study the investment behavior of the households around the time of home purchase.

The empirical methodology implemented in the current paper is separated in the following three subsections: (1) OLS Estimation & Tobit Model, (2) Instrumental variables (IV) approach, (3) Changes in the financial portfolios around home purchase.

5.1. OLS Estimation & Tobit Model OLS Estimation

As a starting point I estimate the main equation using OLS without any control variables, the coefficients β

and β

are simply from the regression of risky share on property value and home equity. Although the result could be impaired by potential omitted-variable biases, it is important to have an indication for the impact of housing on the risky share of liquid wealth and to verify the validity of the data.

Presumably, the portfolio choices could be better explained by some observable household

characteristics. Therefore, I estimate the main equation again, where the following set of

controls is included: household income, province of residence, household head’s age, year of

house purchase and number of children. The inclusion of the control variables allows me to

estimate the relation between housing and portfolio choices for the households with similar

characteristics, home tenures and reside locations. Generally, they account for some of the

potential omitted-variable biases.

Tobit Model

As already mentioned earlier, the simple OLS regression faces the problem that many households do not have any risky investments, hence the dependent variable is bounded by zero. In current data sample 79% of the households do not have any risky investments. To account for this problem I use standard Tobit Model as proposed by Michelsen, Mocking, van Veldhuizen (2015), given by:

Risky Share

*= α + β

Property Value

+ β

Home Equity

+γ X

+ ε

(2) I use a latent variable Risky Share* defined as:

Risky Share = Risky Share* IF Risky Share* > 0

=0 IF Risky Share*

0

The estimation also includes controls for number of children, household head’s age, household net income, purchase year and province of residence.

5.2 Instrumental Variables approach

In this subsection I describe the methodology used to address the endogeneity problem. I follow the methodology proposed by Chetty, Sandor and Szeidl (2017) and estimate the main equation using 2SLS, where property value and home equity are instrumented using the average housing prices in the current year and in the year of home purchase. The purpose of the two instruments is to generate variation in property value, home equity and implicitly mortgage debt (as property value is the sum of home equity and mortgage debt) in order to identify the causal effects of mortgage debt and home equity on portfolio choices. The housing price level in the current year predicts rather precisely the current property values, but it creates variation in the household wealth, as an increase in the average housing prices raises home equity wealth as well. For this reason, the housing prices in the year of purchase are introduced as the second instrument, which accounts for the causal effect of owning a more expensive house, while holding household wealth fixed.

The intuition behind this methodology could be explained by a hypothetical experiment, as

proposed by Chetty, Sandor and Szeidl (2017): “Imagine we have two individuals buying the same houses and paying only interest on their mortgages. Individual A – buys a house in 1995 whose portfolio is observed in 2010 and individual B – buys the same house in 2005 and whose portfolio is observed also in 2010. Both individuals have the same current property values, as they own the same house. However, housing prices (measured by CBS Price Index Purchase Prices) in 2005 were higher than in 1995, hence individual B paid more for the same house. As a result his mortgage is larger and the home equity in 2010 is lower. Next, consider individual C whose house was bought in the same year as individual A, but his portfolio is observed in 2015 instead of 2010. Individual C has the same mortgage as individual A (we assume that both pay only interest) but individual C has more home equity and wealth.”

In the current paper the two instruments (average housing prices in current year and in the year of home purchase) are calculated using “Price Index Purchase Prices, 2010 = 100” obtained from Statistics Netherlands (CBS)

. This particular housing index is used because it shows the price movements of the stock of existing own homes over time, which is exactly the required information for the instruments. All owner-occupied houses in the data sample are located on the territory of the Netherlands and bought in the period 1995-2015. However, the price development in the given period is different in the different regions of the Netherlands. To tackle this problem I use additional information showing the province in which the house is located. Then, I use separate index data for each province to build the instrumental variables.

The first instrument represents the “Price Index Purchase Prices, 2010 = 100” of the province where the property is located in the year in which it was bought, referred in the following pages as “price index purchase year”. The second instrument represents “Price Index Purchase Prices, 2010 = 100” of the corresponding province in the current year, referred as “price index current year”.

In the first stage estimation - the OLS regression of property value and home equity on price index purchase year and price index current year is intended to estimate to what extent the average price level of certain province in the purchase year and the average price level in the current year predict current property values, home equities and mortgage debt. It also shows

5The idea behind this experiment is taken from Chetty, Sandor and Szeidl (2017)

the impact of buying a house when the prices are higher/lower on the current values of home equity and mortgage debt. Moreover, the estimation serves as an indication what kind of properties the households purchase during periods of higher/lower price levels.

In the second stage estimation - the coefficients β

and β

which show the effects of property value and home equity on the risky share of liquid wealth are now estimated by the changes in the provincial housing prices since the year of purchase. The estimation shows the causal effect of mortgage debt and home equity on the risky share of liquid wealth. This approach isolates the effects of possibly endogenous decisions regarding home equity and the outstanding mortgage debt like investing in property improvements, refinancing the mortgage, paying down the mortgage or taking out a second mortgage.

However, the introduction of the IV approach creates concern about the validity and reliability of the instrumental variables. The validity assumption of the instruments in current estimation requires that the changes in the housing prices (measured by the instruments) are orthogonal to all other possible (unobserved) determinants of the portfolio choices. Unfortunately, the two instruments might influence portfolio choices through other unobserved determinants as long as they are correlated with some other omitted factors which affect directly the portfolio choices. To eliminate some of the omitted-variable biases I include controls for household income, household head's age, children number, province of residence and purchase year.

Hence, I am able to estimate the coefficients β

and β

from the differential changes in the housing prices across provinces for the households with similar characteristics and home tenures

.

5.3 Changes in the financial portfolios around home purchase

In this subsection I present the estimation method which studies the changes in the households financial portfolios from the year before to the year after home purchase. This approach addresses the question whether the households which buy more expensive houses reduce their risky investments more from the year before to the year after home purchase. To answer this

7Home tenure is defined as the years living in current accommodation.

question I use the following modification of the main equation, proposed by Chetty, Sandor and Szeidl (2017):

Δ Risky Share

= α + β

Δ Property Value

+β

Δ Total Wealth

+ χ

+ ε

(3) Where

X

X

X

for a household which buys a new house in year t. To control for the endogeneity of the size of the house I estimate the equation using 2SLS, as proposed by Chetty, Sandor and Szeidl (2017). Δ Property value is instrumented using the housing price index of the province where the property is located in the purchase year

. The estimation also includes controls for number of children, household head's age and province of residence.

6. Results

In this section I present the empirical results related to the effects of homeownership on the financial portfolios, specifically the relation between housing and risky investments. The estimations are carried out implementing the methodology described in the previous section and using data for the Dutch homeowners’ behavior.

6.1. OLS results

Table 3 presents the results of the OLS regression of the main equation.

Table 3. Results of the OLS Regression

Dependent variable: Risky Share

(1) Property Value

(x1000)

0.0369 ****

(0.010) Home Equity

(x1000)

0.0012

(0.008)

(2) Property Value

(x1000)

Home Equity

0.0227 **

(0.011) 0.0035 (0.008)

8For the instrumental variable I use "Price Index Purchase Prices, 2010=100" obtained from Statistics Netherlands

Table 3:

displays the OLS regression of property value and home equity on the risky share of liquid wealth. The first estimation (1) does not include any control variables, while (2) includes controls for household income, household head's age, children number, province of residence and purchase year.

Asterisks indicate significance at the 10% (*), 5% (**), 1% (***) and 0.1%(****) level. Standard errors are in parenthesis.

The first estimation shows that an increase in property value is related to an increase in the risky share of liquid wealth. This effect remains after the inclusion of controls for number of children, household head's age, household income, purchase year and province of residence.

Although the controls decrease the coefficient of property value, it remains positive and statistically significant. The coefficient of 0.0227 means that an increase in property value by 1,000 Euros is associated with an increase in the risky share of liquid wealth by 0.0227 percentage points (as property value is measured in thousands of Euros). In this regression the increase in property value can be interpreted as an increase in mortgage debt, given that home equity is held fixed. Therefore, if the households experience 1,000 Euro increase in their mortgage debt, they increase the risky share of liquid wealth by 0.0227 percentage points.

This result contradicts the predictions of the theoretical model developed by Chetty, Sandor and Szeidl (2017), which states that an increase in property value is negatively associated with the risky share of liquid wealth (holding home equity fixed). Nevertheless, it is in line with the empirical results of Cocco (2005), Yao and Zhang (2005) and Chetty, Sandor and Szeidl (2017).

The latter explains this result by the fact that the individuals who own larger houses are wealthier or experience less background risk. However, as already mentioned before, the OLS regression does not give us a precise estimation of the relation between housing and portfolio choices, mostly because of the presence of the endogeneity problem described earlier in the paper.

6.2. Tobit results

Table 4 shows the impact of property value and home equity on the risky share of liquid wealth

in an estimation using standard Tobit Model.

Table 4. Tobit Estimation

Dependent variable: Risky Share

(1) Property Value

(x1000)

0.1560 ****

(0.043)

Home Equity

(x1000)

0.0026

(0.032)

(2) Property Value

(x1000)

Home Equity

0.1129 ***

(0.042)

-0.0074 (0.033)

Table 4:

displays the Tobit regression of property value and home equity on the risky share of liquid wealth. The first estimation (1) does not include any control variables, while (2) includes controls for household income, household head's age, children number, province of residence and purchase year.

Asterisks indicate significance at the 10% (*), 5% (**), 1% (***) and 0.1%(****) level. Standard errors are in parenthesis.

The coefficients of property value in both estimations (1) and (2) are higher than the OLS estimates. The property value coefficient is slightly decreased by the inclusion of the controls for number of children, household head's age, household net income, purchase year and province of residence. The coefficient of 0.1129 implies that 1,000 Euro increase in property value is associated with 0.1129 percentage points increase in the risky share of liquid wealth.

Considering that home equity is held fixed the increase in property value is generated form an

increase in mortgage debt, which means that an increase in mortgage debt is associated with

an increase in the risky share of liquid wealth. This result is still against the predictions of the

theoretical model, which could be explained by the fact that the endogeneity problem is not

addressed yet. The coefficient of home equity is slightly negative, but still statistically

insignificant.

6.3. Instrumental Variables (IV) approach

This subsection presents the results of the 2SLS estimation of the main equation, where property value and home equity are instrumented using housing price index in the current year and in the year of home purchase. First, I report the first stage results and then the second stage estimates.

First stage results

Table 5. First stage OLS results

Dependent variable: Property Value

(x1000)

Home Equity

(x1000)

Mortgage Debt

(x1000)

Price index current year 4.1538 ****

(1.234) 1.9754

(1.591) 2.1784

(1.590)

Price index purchase year -0.2858

(0.378) -1.8379****

(0.487) 1.5521***

(0.487)

Table 5:

displays the first stage (OLS) regression of property value, home equity and mortgage debt on the two instrumental variables (price index current year and price index purchase year). All regressions include controls for household income, household head's age, children number, province of residence and purchase year. Asterisks indicate significance at the 10% (*), 5% (**), 1% (***) and 0.1%(****) level.

Standard errors are in parenthesis.

Note that "Mortgage debt" is not included in the 2SLS estimation, it is reported separately in Table 5 only for the sake of interpretation, as property value is the sum of home equity and mortgage debt.

In the estimation of property value - the coefficient of the price index in current year shows that higher current housing prices in given province (measured by the index) predict higher current property values in the same province. This is an obvious effect confirming the validity of the data.

In the home equity estimation - the coefficient of the price index in the purchase year shows

that higher prices at the time of home purchase predict lower home equities at the current

moment. Chetty, Sandor Szeidl (2017) explains this outcome by the individuals' behavior to buy

smaller houses when the price level is higher.

In the mortgage debt estimation - the coefficient of house price index in purchase year shows that if the prices at the time of purchase were higher, the mortgage debt at the current moment is also higher. Apparently, when housing prices at the time of purchase are higher, the households need more capital and hence larger mortgages to finance the house purchase.

Second stage results

Table 6 reports the results of the second stage estimation of the main equation using IV approach, where property value and home equity are instrumented using housing price index in the current year and in the year of purchase. With the introduction of the IV approach the endogeneity problem is addressed and the two coefficients are estimated by the changes in provincial house prices since the year of purchase. The two coefficients β

and β

measure the causal effect of mortgage debt and home equity on the risky share of liquid wealth.

Table 6. Second stage estimates

Dependent variable: Risky Share

Property Value

(x1000)

0.0665

(0.086)

Home Equity

(x1000)

-0.1136**

(0.582)

Table 6:

displays the 2SLS estimates of the coefficients β

and β

in the equation: Risky Share

= α + β

Property Value

+ β

Home Equity

+γ X

+ ε

price index in current year and housing price index in the year of home purchase. The estimation includes controls for household income, household head's age, children number, province of residence and purchase year. Asterisks indicate significance at the 10% (*), 5% (**), 1% (***) and 0.1%(****) level.

Standard errors are in parenthesis.

The coefficient of property value is positive and therefore opposing to the predictions of the

theory, but it is statistically insignificant and cannot be considered as a real estimate. The

coefficient of home equity is negative and statistically significant at 5 % level. The result

predicts that an increase in home equity is associated with a decrease in the risky share of

liquid wealth, for every 1,000 Euro increase in home equity the risky share of liquid wealth is

decreased by 0.1136 percentage points. In current data sample the mean risky share of liquid

wealth is 9.5 %, which is generated only from the households with positive risky investments (21 % of the sample). Therefore, taking into consideration the mean stock share of liquid wealth, the result shows that 1,000 Euro increase in home equity results in 1.195 percentage points or 46.23 Euro decrease in the risky share of liquid wealth.

This result is still contradicting the predictions of the theoretical model and hence there should be some sources of bias in the estimation. As mentioned before the main requirement of the IV approach is that the changes in the housing prices (measured by the two instruments) are orthogonal to all other factors which affect the portfolio choices.

To assess the reliability of the instruments I calculate the correlation coefficient between current housing prices (price index current year) and current property values

. The higher the correlation coefficient, the better the instrumental variables. In current data sample it equals 0.192, which looks relatively low, but one might find it difficult to say what correlation could be considered as sufficient to infer that the instruments are good. Michelsen, Mocking and van Veldhuizen (2015) implemented almost the same IV approach to study the portfolio choices of the Dutch households and calculated the same correlation coefficient at 0.119, which could be used as an indication what should be expected.

Presumably the IV estimation might be still significantly biased if the decision to purchase a house at certain price level is influenced by some unobserved household characteristics or risk preferences, as suggested by Cocco (2005) and Vestman (2012). In this regard, Chetty, Sandor and Szeidl (2017) argue that the unobserved risk preferences usually bias the results against the theoretical predictions. Although the controls for household income, household head's age, children number, province of residence and purchase year eliminate some of the omitted- variable biases, there is still possibility that the instruments are correlated with the portfolio choices through other factors than property value and home equity. Moreover, the inability to apply a proper test for weak instruments creates a significant concern about the validity of the IV estimation.

9Unfortunately, I cannot use Sargan Test, because I need over-identifying restrictions and hence more instrumental that endogenous variables.

6.4. Changes in the financial portfolios around home purchase

Unfortunately, all results of the last estimation method are statistically insignificant (the outcome is reported in the Appendix). The main downside of the estimation is the small sample size. I observe the financial portfolio changes from the year before to the year after home purchase of only 75 households, which constitute the whole data sample. The validity of the estimation is further limited by the fact that the observed home purchase represents a transition from renters to homeowners only for 35 of the households within the sample. This is important because if the house purchase is actually a move from one accommodation to another, the observed change in property value is relatively small and it could be either positive or negative. I have also estimated the equation only with the households which transit from renters to homeowners, but the results remain insignificant.

Results summary and discussion

To summarize, the results of all estimations are against the theoretical predictions of the model developed by Chetty, Sandor and Szeidl (2017). However, the sign of the OLS estimate (also the Tobit estimate) of property value is in line with the estimates of Cocco (2005), Yao and Zhang (2005) and Chetty, Sandor and Szeidl (2017). Unlike the OLS estimation, the IV result does not have any support from the existing literature and is exactly opposite to the empirical findings of Chetty, Sandor and Szeidl (2017).

The main concern in the IV estimation is related to the validity of the instrumental variables.

The validity assumption of the instruments requires that the changes in the housing prices (measured by the instruments) are orthogonal to all other possible (unobserved) determinants of the portfolio choices. In the 2SLS estimation the number of the endogenous variables is equal to the number of the instruments, which means that I cannot assess the validity of the instrumental variables using Sargan Test. I have calculated the correlation coefficient between current house prices (price index current year) and current property values, but it is difficult to say what correlation coefficient could be sufficient to assume that the instruments are acceptable.

It is difficult to infer which estimation is more in line with the reality. On the one hand, the OLS

estimation could suffer from self-selection bias caused by the fact that the individuals with certain characteristics invest in risky assets. On the other hand, the IV estimation could be significantly impaired by potential weak instruments bias. Assuming the instruments are good, the IV result is definitely better than OLS because it measures the causal effect of home equity on the risky share of liquid wealth. Unfortunately, I cannot test properly for weak instruments and I have no reliable indication whether the instrumental variables are good or not.

However, the fact that the OLS result is similar to Chetty, Sandor and Szeidl (2017) means that any differences in the IV estimations generally should not be driven by possible differences in the data. The differences, however, might be driven by the instruments and the presence of weak instruments bias, which seems to be the most probable explanation for the results of the IV estimation. Although it is difficult to draw a clear conclusion, I would assume that the instruments are not good and there is a significant bias in the IV estimation caused by them, meaning that the OLS estimation should be better.

7. Conclusion

This paper studies empirically the relationship between homeownership and households’

portfolio choices. Although the theoretical literature has found relatively systematic relation

between housing and portfolio choices, the empirical studies have found mixed and

inconclusive results. I use the theoretical predictions of the model developed by Chetty, Sandor

and Szeidl (2017) to formulate the hypothesis which I test empirically using Dutch data. The

empirical methodology of current master thesis mostly follows the methodology proposed by

Chetty, Sandor and Szeidl (2017). Specifically, this paper investigates the impact of property

value (mortgage debt) and home equity on the risky share of liquid wealth. Property value is

defined as the most recent WOZ-value, while home equity equals property value minus the

outstanding mortgage debt used to finance the purchase of the house. The empirical study is

conducted using four estimation methods: (1) OLS Regression, (2) Tobit Model, (3) IV approach,

(4) Changes in the financial portfolios around home purchase. The first method represents an

OLS regression of property value and home equity on the risky share of liquid wealth including

different sets of control variables. The second uses Tobit Model to account for the fact that the

dependent variable is bounded by zero. The third method implements instrumental variables approach to account for the endogeneity problem arising from the fact that there might be some unobserved factors which influence separately the dependent and the independent variable(s). The last method studies the changes in the households’ financial portfolios from the year before to the year after home purchase.

The outcome of the OLS regression shows that an increase in property value is related to an

increase in the risky share of liquid wealth. This result is against the theoretical predictions,

though it was expected to a certain extent as the estimation could suffer from several already

identified biases. The result of the Tobit Model supports the effect presented in the OLS

regression and even claims that it is slightly stronger. This outcome is also against the

theoretical predictions, but the estimation is still biased by the endogeneity problem. The IV

estimation predicts that an increase in home equity is associated with a decrease in the risky

share of liquid wealth. Although the endogeneity problem is addressed, the result is still not in

line with the theoretical predictions. This outcome could be explained by the suggestion that

the IV estimation does not capture all potential sources of bias, especially as I cannot verify

clearly the validity of the instrumental variables. However, the result could be also driven by

the specifics of the Dutch households within the sample. The last approach which examines the

changes in the households' financial portfolios from the year before to the year after home

purchase does not find any significant results. Although it is difficult to infer clearly, I would

assume that the OLS (and Tobit) estimations are better that the IV estimation.

References

Grossman, S. and Laroque, G., 1990. Asset pricing and optimal portfolio choice in the presence of illiquid durable consumption goods. Econometrica, 58(1), 25–51.

Flavin, M. and Yamashita, T., 2002. Owner-occupied housing and the composition of the houshold portfolio. The American Economic Review, 92(1), 345–362.

Yao, R. and Zhang, H., 2005. Optimal consumption and portfolio choices with risky housing and borrowing constraints. Review of Financial Studies, 18(1), 197–239.

Cocco, J., 2005. Portfolio choice in the presence of housing. Review of Financial Studies, 18(2), 491–533.

Chetty, R. and Szeidl, A., 2010. The effect of housing on portfolio choice. NBER Working Paper 15998.

Chetty, R., Sandor, L. and Szeidl, A., (2017). The effect of housing on portfolio choice. The Journal of Finance, doi:10.1111/jofi.12500

Fratantoni, M., 1998. Homeownership and investment in risky assets. Journal of Urban Economics 44, 27–42.

Sinai, T. and Souleles, N., 2005. Owner-occupied housing as a hedge against rent risk. Quarterly Journal of Economics 120, 763–789.

Vestman, R., 2012. Limited stock market participation among renters and home owners.

Stockholm Institute for Financial Research Working Paper.

Yamashita, T., 2003. Owner-occupied housing and investment in stocks: An empirical test.

Journal of Urban Economics 53, 220–237.

Hochguertel, S. and van Soest, A., 2001. The relation between financial and housing wealth:

Evidence from dutch household. Journal of Urban Economics, 49(2), 374–403.

Kullman, C. and Siegel, S., 2005. Real estate and its role in household portfolio choice. EFA 2003 Annual Conference Paper No. 918.

Shum, P. and Faig, M., 2006. What explains househld stock holdings? Journal of Banking and Finance, 30(9), 2579–2597.

Englund, P., Hwang, M., and Quigley, J., 2002. Hedging housing risk. Journal of Real Estate

Finance and Economics, 24(1-2), 167–200.

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

Waggle, D. and Johnson, D., 2009. Homeownership and mixed-asset portfolio allocations. The Quarterly Review of Economics and Finance, 49 (2009) 484–500

Hu, X., 2005. Portfolio choices for homeowners. Journal of Urban Economics, 58(1):114-136 Cho, I., 2014. Homeownership and Investment in Risky Assets in Europe. Review of European Studies, Vol. 6, No. 4; 2014

Michelsen, T., Mocking, R. and van Veldhuizen, S., 2015. Home Ownership and Household

Portfolio Choice. CESIFO working paper no. 5705

Appendix

Table 7. Results of “Changes in the financial portfolios around home purchase” estimation.

First stage

estimates

Dependent variable:

Property Value

Price index purchase year -764.20 (1683.209)

Second stage

estimates

Dependent variable:

Risky Share

Δ Property Value

Δ Total Wealth

-2.08 (5.631)

1.82 (4.610)

Table 7:

displays the first and second stage results of the 2SLS estimation of the equation:

Risky Share = α + β1

Property Value +β2

Total Wealth+ χi+ εi.

Property Value is instrumented using price index purchase year. The estimation includes controls for province of residence, purchase year and age.

Asterisks indicate significance at the 10% (*), 5% (**), 1% (***) and 0.1%(****) level. Standard errors

are in parenthesis.