The effect of policy changes on
homeownership in the Dutch housing market
Lisa Dral – 10632115
Master Thesis
MSc Finance track: Finance & Real Estate Finance
July 2018
Statement of Originality
This document is written by Lisa Dral who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
Acknowledgements
First of all, I would like to thank my supervisor, Mr. D. W. van Dijk, for his guidance and support throughout the entire process and for providing me with valuable suggestions. Finally, in this paper, use is made of data of the DNB Household Survey, and therefore I would like to express my gratitude to CentERdata for providing me with this data.
Abstract
In response to the Global Financial Crisis, actions have been taken to prevent excessive lending in the Netherlands. Policy changes have been implemented in 2007, 2011 and 2013 regarding the issuance of mortgages. This paper examines the effect of these three policy changes on homeownership in the Dutch housing market. To do this, panel data of the years 1994 to 2017 from the DNB Household Survey is used. By means of a difference-in-difference method and a logistic regression model, the effects of the three policy changes can be tested. The results show that all three policy changes negatively impacted the probability of becoming a homeowner, indicating that it is likely that there are fewer new entrants in the Dutch owner-occupier sector. Therefore, the main finding of this research is that the policy changes had a negative effect on homeownership in the Dutch housing market.
Table of contents
1 Introduction ... 6
2 Literature review ... 8
2.1 The Dutch mortgage market ... 8
2.1.1 Before the Global Financial Crisis ... 8
2.1.2 The Global Financial Crisis and the aftermath ... 9
2.2 Recent policy changes impacting the Dutch mortgage market ... 10
2.2.1 What is the GHF ... 10 2.2.2 GHF Change in 2007 ... 10 2.2.3 GHF Change in 2011 ... 11 2.2.4 Law implementation in 2013 ... 11 2.3 Related literature ... 11 3 Methodology ... 16 3.1 Empirical method ... 16 3.2 Variables ... 17 3.2.1 Dependent variable ... 17 3.2.2 Variables of interest ... 18 3.2.3 Control variables ... 19
4 Data and descriptive statistics ... 21
5 Results ... 24
6 Robustness checks ... 28
7 Limitations and future research ... 30
8 Conclusion and discussion ... 31
9 References ... 32
10 Appendix ... 35
Appendix A ... 35
1 Introduction
In the Netherlands, homeownership has increased in importance. According to Figari et al. (2017), the owner-occupied sector makes up about 57 percent of all homes. Homeownership is associated with several external effects. According to DiPasquale and Glaeser (1999), homeownership has positive external effects compared to renting, because homeowners show more community engagement. Moreover, Haurin, Parcel and Haurin (2002) find evidence that homeownership is associated with better cognitive ability and behavioral outcomes of children. In contrast, residential mobility is lower among homeowners than renters because of the high transaction costs of moving, which results in more unemployment among homeowners than renters (Andrews and Caldera Sánchez, 2011).
Despite the ambiguous effect of homeownership, the Dutch government has been encouraging homeownership, for example through a favorable tax treatment for homeowners. However, after the Global Financial Crisis, policies regarding mortgage issuance have been changed to prevent high indebtedness. The results of Andrews and Caldera Sánchez (2011) show that a proportion of the trends in homeownership rates in many OECD countries remains unexplained by transformations in the characteristics of the population, such as age, income, and education. This outcome indicates that public policy could be one of the determinants of changes in homeownership rates. Therefore, this study will investigate whether the policy changes regarding mortgage issuance have had an effect on the probability of becoming a homeowner in the Dutch housing market. The policy changes that are examined are a change in the code of conduct in 2007, a change in the code of conduct in 2011, and the implementation of a law in 2013. To test the effect of the policy changes on the probability of becoming a homeowner, a difference-in-difference method is used with three interaction terms for each policy change, in combination with a logistic regression. Furthermore, use is made of panel data from the DNB Household Survey. The dataset contains data of the years 1994 to 2017 and the data is provided by CentERdata.
This research contributes to the literature by examining the effect of the mentioned policy changes on the probability of becoming a homeowner in the Netherlands. Since most of the existing literature studies the effect of borrowing constraints on house prices, the focus of this paper will be on the effect on homeownership. While the impact of borrowing constraints on homeownership has been revealed in previous studies, this research will provide the first results of the effect of the mentioned policy changes regarding mortgage issuance in the Dutch mortgage market on the probability of becoming a homeowner. Furthermore, the outcome of this research can be helpful in the decision-making process of
future policies, because the effect on homeownership can be taken into account by policy makers.
As these policy changes make it more difficult to get a mortgage, it is expected that the policy changes have negatively impacted the probability of becoming a homeowner. The results of this study support this expectation, as for each policy change a negative effect is found. These results hold after controlling for age, household structure, gross income, education, disability and fixed regional and time effects. More specifically, households are 3.17 percent less likely to move from a rented house to an owned house after the change in the code of conduct in 2007. After the change in the code of conduct in 2011, the probability of moving from a rented house to an owned house decreased by 4.5 percent. Lastly, after the law implementation of 2013, households are 4.49 percent less likely to become owner-occupier. As the probability of becoming a homeowner decreased as a result of the policy changes, it is likely that there are fewer first-time buyers per year, meaning that there are fewer households entering the Dutch owner-occupier sector. Therefore, the main finding of this study is that the policy changes have had a negative impact on homeownership in the Dutch housing market.
The remainder of this research is structured as follows. In the next section, the Dutch housing market and the policy changes are explained and related literature is discussed, leading to the hypotheses. Section three describes the empirical method that is used to test the hypotheses, and the variables included in the model. The data is discussed in the fourth section. After that, the results of this research are analyzed in section five, and robustness checks are displayed in section six. Section seven discusses the limitations of this research, while section eight concludes.
2 Literature review
This section starts with a description of the Dutch mortgage market before and after the crisis. After that, recent policy changes considered in this study are explained. Finally, a literature review of existing empirical research is provided.
2.1 The Dutch mortgage market
The Dutch mortgage market is significant for several reasons. One of these reasons is that housing is the biggest expenditure for most households. Furthermore, as can be concluded from the financial crisis, the housing market can have a deep impact on the macro economy. This section will discuss the characteristics of the Dutch mortgage market before and after the crisis.
2.1.1 Before the Global Financial Crisis
Since the 1980s, the mortgage market in the Netherlands has been liberalized (Elsinga, Priemus, & Boelhouwer, 2016). The goal of the Dutch government was to encourage homeownership and one instrument to achieve this was the establishment of the National Mortgage Guarantee, or Nationale Hypotheek Garantie (NHG) in 1995 (Teye, De Haan, & Elsinga, 2015). This scheme is operated by a non-profit organization backed by the government called the Homeownership Guarantee Fund or Waarborgfonds Eigen Woningen (WEW). If a mortgage is NHG-backed, which can be achieved by paying a fee, and the mortgagor defaults due to unforeseen circumstances beyond the mortgagor’s control, the mortgagor can turn to the NHG for support (NHG, 2018). According to Teye et al. (2015), the main goal of the NHG was to improve access to mortgages for younger households and lower income groups. The support of the NHG made this possible, since these target groups could now obtain mortgages with loan-to-value ratios of up to 115 percent. The loan-to-value ratio is equal to the value of the mortgage divided by the value of the house. The improved access to mortgages contributed to the rise in the homeownership rate among these target groups. Next to the NHG scheme, the mortgage interest tax relief was another instrument to encourage homeownership. The deregulation in the Dutch mortgage market resulted in many mortgage suppliers and many mortgage products available to clients. Mortgage products with lower monthly payments were developed that allowed borrowers to maximally benefit from the income tax deductibility. Examples are the savings mortgage and the interest-only mortgage. These developments led to improved access to mortgages, lower monthly payments for owner-occupiers, and an increase in demand for owner-occupied housing (Teye et al.,
2015). This, in turn, resulted in a rise in house prices and in the period of 1985 to 2008, the average nominal house price increased by 313 percent and the average real house price increased by 153 percent.
Overall, it can be concluded that before the crisis the Dutch mortgage market was characterized by high loan-to-value ratios, a preference among households for interest-only mortgages, and a tax relief in the form of mortgage interest deduction (De Nederlandsche Bank, 2015). Because of the preference for interest-only mortgages, the outstanding mortgage balance in the Netherlands has been very high compared to other European countries.
2.1.2 The Global Financial Crisis and the aftermath
The Netherlands was hit by the financial crisis in 2008 and this had a severe impact on the housing and mortgage market. After a long period of rising house prices, house prices started to decrease in 2008 (Elsinga et al., 2016). In 2014, house prices had decreased by 21.5 percent compared to the highest level in 2008 (NVB, 2014). Furthermore, the number of transactions per year decreased with 55 percent from the highest level in 2006. As a result of the decline in house prices, many homeowners had negative equity, meaning that their outstanding mortgage balance was higher than their property value. This was possible because as mentioned before, the Dutch mortgage market was characterized by high loan-to-value ratios and high outstanding mortgage debt. According to De Nederlandsche Bank (2014), the share of mortgages that exceeded their underlying property value was thirty percent in 2014. Now that the market is recovering, this share has decreased to twenty percent in 2017, which is equal to 700,000 households (NVB, 2017). Furthermore, the number of households with payment problems decreased as well. From 2008 to 2013, the number of households with payment arrears over four months increased from 35,000 to 113,000. As of October 2016, this number was equal to 107,310.
The financial crisis made lenders more aware of the dangers of high mortgage debt and it led to a tightening of the rules and norms for granting mortgages to prevent high indebtedness, as will be further discussed in the next section. As a result of these changes, the average loan-to-value ratio at closing decreased to 89 percent in 2016 (NVB, 2017). The amount of mortgages with a loan-to-value ratio of 100 percent or higher is also coming down. This number was 55 percent in 2013 and declined to 50 percent in 2016. Moreover, Francke, Van Dijk and Mastrogiacomo (2017) show that across the country, the share of interest-only mortgages has decreased. According to De Nederlandsche Bank (2017), the total Dutch interest-only mortgage debt has come down by more than thirty billion euros since 2013, and
it is now equal to 340 billion euros. The number of mortgagors with interest-only mortgages has decreased by eight percent since 2013.
As can be concluded, the Dutch mortgage market was hit by the financial crisis. Measures have been taken to make the mortgage market less risky and as a result, the amount of non-amortizing loans and the average loan-to-value ratio have decreased.
2.2 Recent policy changes impacting the Dutch mortgage market 2.2.1 What is the GHF
The Gedragscode Hypothecaire Financieringen (GHF) is a code of conduct that has been in use since the 1990s (Homefinance, 2018). It contains requirements about conditions and information that have to be followed by the mortgagees. The goal of the GHF is to give consumers correct and comparable information when they are taking out a loan. In 2007, as a response to the Global Financial Crisis, the GHF was expanded and rules to prevent excessive lending were added. In 2011, these rules were tightened and limits about the level of the mortgage with respect to the value of the house were determined. As of 2013, some of these rules are incorporated into the Dutch law.
2.2.2 GHF Change in 2007
As of 2007, a new GHF is implemented because the government observed that mortgages were too high compared to the income. When households have high debt obligations, they are more vulnerable to rising interest rates, decreasing house prices or decreasing income (Homefinance, 2018). Therefore, this new GHF was composed to prevent high indebtedness. A first change is the fact that all issuers of mortgages have to use the same living quota or
woonquote when issuing a mortgage. This is the percentage of the income that a household
spends on their mortgage payments and this percentage is determined every year by Nibud (Homefinance, 2018). According to this woonquote, a household with a higher income and thus a higher living quota is allowed to get a higher mortgage. Another change in the GHF is the introduction of the toetsrente, which is the interest rate that is used by the mortgage issuer to determine the maximum loan amount of the household. Mortgages with a fixed interest rate term shorter than ten years have to employ this toetsrente, that is determined every quarter of a year.
Although these rules were meant to prevent too high indebtedness, the rules in the GHF of 2007 were not binding. Mortgage issuers were allowed to deviate from these rules and issue higher mortgages if, for example, the household had positive income prospects
(Homefinance, 2018). In this case, the mortgagee had the obligation to inform the borrower about the encroachment and the borrower had to sign that he had been informed and that he understands the risks associated with this (Rabobank, 2007).
2.2.3 GHF Change in 2011
To further protect consumers from excessive lending, the norms for getting a mortgage are tightened in the new code of conduct that is in effect since the 1st of August 2011
(Rijksoverheid, 2011). The financial crisis caused the government to realize that excessive lending can have big financial and economic consequences (Homefinance, 2018). Furthermore, the Netherlands turned out to be the leader in issuing mortgages that were worth more than the underlying value of the house. This led to pressure from the government to tighten the norms and the first most important change in the code of conduct is that the maximum loan-to-value ratio could be 104 percent, excluding the transfer tax. Moreover, the percentage of the mortgage without amortization (interest-only) that was allowed is 50 percent. For the other 50 percent of the mortgage balance, an amortization scheme has to be used. Finally, the code of conduct of 2011 became binding, so only in very exceptional cases the banks could deviate from these norms.
2.2.4 Law implementation in 2013
A few norms from the code of conduct are registered in the Dutch law as of the 1st
of January 2013 (Homefinance, 2018). This has made these norms even more binding, but the code of conduct less important. The law is called tijdelijke regeling hypothecair krediet and it states that to qualify for the favorable tax treatment on mortgage interest, first-time buyers in the housing market need to have a completely amortized mortgage in 30 years. This means that a partly interest-only loan is not allowed anymore, as well as a savings mortgage. Next to that, the maximum loan-to-value ratio will be reduced to 100 percent (including 2 percent transfer tax) in 2018 with one percentage point every year, starting at 106 percent (including 2 percent transfer tax) in 2012.
2.3 Related literature
There is a lot of literature about homeownership, policy developments and the effects of these developments on the housing market. DiPasquale and Glaeser (1999) argue that homeownership has positive external effects compared to renting, because homeowners show more community engagement and are more likely to act in favor of the long-term, whereas
renters are incentivized to favor policies that lead to immediate benefits. They find that in Germany and the United States, there is correlation between homeownership and variables that strive to quantify good citizenship, such as being a member of a nonprofessional organization and being involved in local politics. For example, their results suggest that owners are on average members of 0.253 more nonprofessional organizations than renters and that owners are 9.4 percentage points more likely to know their school board. Furthermore, homeowners are 15.3 percent more likely to vote in local elections than renters, and 11.6 percent more likely to garden. Moreover, Haurin et al. (2002) find evidence of better cognitive ability and behavioral outcomes of children when they live in an owned house. Haurin et al. (2002) conclude that being a homeowner, compared to being a renter, results in a home environment with a higher quality of 13 to 23 percent. Math achievement is up to 9 percent higher for children living in owned homes and reading achievement is up to 7 percent higher, while behavioral problems are 1 to 3 percent lower. However, it is unclear whether the positive effects of homeownership on child outcomes is causal.
On the other hand, homeownership can also have negative external effects. Because of the high transaction costs of moving in the owner-occupied sector, it is more difficult to move for a homeowner than for a renter. In other words, residential mobility among homeowners is lower (Andrews & Caldera Sánchez, 2011). If the motive for a household to move is about better employee prospects, this can also decrease labor market efficiency. Therefore, labor market mobility is lower for homeowners and unemployment is higher among homeowners than renters.
Nevertheless, as mentioned before, the Dutch government has been encouraging homeownership. However, recent policy developments, as described in the previous section, might have an impact on homeownership in the Netherlands. In their research, Andrews and Caldera Sánchez (2011) find that a proportion of the trends in homeownership rates remains unexplained by transformations in the characteristics of the population, such as age, income, and education. This finding indicates that public policy could be one of the factors that explains changes in homeownership rates. As the policy developments described before are all meant to make the mortgage market less risky by decreasing access to homeownership and decreasing its benefits, the effects of this will be discussed using existing literature.
To qualify for the tax relief on mortgage interest, first-time buyers need to have a completely amortized mortgage in 30 years. Hence, it is more difficult to benefit from this favorable tax treatment. Andrews, Caldera Sánchez & Johansson (2011) prove that countries with a more generous tax relief have increased house prices as a result of more demand. In
other words, the favorable tax treatment tends to be capitalized into house prices, reallocating income from new entrants to insiders in the housing market. Therefore, the favorable tax treatment does have an effect on house prices, but there is no relationship between the tax relief and homeownership rates.
Next to policy changes regarding the mortgage tax relief, the maximum loan-to-value ratio is decreased with one percentage point per year until 100 percent in 2018. A higher loan-to-value ratio implies that a larger amount of the value of the house can be borrowed, decreasing the down payment for the house. This makes buying a house easier, especially for first-time buyers that have less time to save (Andrews et al., 2011). However, greater access to credit can also put a risk on macroeconomic stability, since a large outstanding mortgage debt balance in a housing bust can lead to negative equity for many households, as has been proven by the Global Financial Crisis. Furthermore, Almeida, Campello & Liu (2006) find that in countries where the maximum loan-to-value ratio is higher, house prices are more sensitive to income shocks. They test the effect of the maximum loan-to-value ratio on the log change in real house price index and on the number of new mortgages. Their results suggest a positive relationship between the maximum loan-to-value ratio and new mortgages and a negative (but mostly insignificant) relationship between the maximum loan-to-value ratio and the log change in the real house price index.
On the other hand, Duca, Muellbauer & Murphy (2011) find a positive relationship between loan-to-value ratios for first time buyers and house prices. They study the effect of credit constraints in the US using loan-to-value ratios for first time buyers as a measure for credit constraints and find that the loan-to-value ratio for first time buyers significantly and positively affects house prices. Moreover, Igan and Kang (2011) examine the effect of loan-to-value and debt-to-income limits on house prices and market activity in Korea. They find that in the three-month period followed by the tightening of the loan-to-value and debt-to-income regulations, transaction activity declined significantly. In the six-month period followed by the same tightening, house price appreciation also declined.
Francke, Van de Minne & Verbruggen (2015) find evidence of the sensitivity of house prices to credit conditions on first-time buyers in the Netherlands. They use a credit condition index based on data from the Dutch housing market in their regressions to proxy the credit conditions. From 2009 to 2012, house prices in real terms in the Netherlands declined with approximately 25 percent. Their results show that the decrease in credit supply after 2009 resulted in a house price decrease of 11 percent. Furthermore, De Nederlandsche Bank (2015) also studied credit supply in the Dutch housing market. They investigated the effect of a
further reduction in the loan-to-value limit to 90 percent and found that this will decrease the demand for owner-occupied housing, since younger households might not have enough funds for the down payment. This decrease in the amount of housing transactions can be related to the homeownership in the Dutch housing market.
Finally, Linneman, Megbolugbe, Wachter & Cho (1997) study the impact of policy changes regarding borrowing constraints on homeownership rates in the US. When the loan-to-value ratio is increased from 80 percent to 95 percent together with an increase in the income ratio from 28 to 33 percent, the homeownership rate increases with 3.02 percent, and the effect of changing the loan-to-value ratio is higher than the effect of changing the income ratio. Linneman et al. (1997) thus find a negative relationship between borrowing constraints and the homeownership rate in the US. This relationship is supported by the results of Acolin, Bricker, Calem and Wachter (2016). They also investigate the effect of borrowing constraints on homeownership in the US and conclude that a tightening of credit in the period 2000-2010 led to significant drops in homeownership rates. Furthermore, the results of Andrews and Caldera Sánchez (2011) suggest that a ten percentage point increase in the maximum loan-to-value ratio is associated with a 3.4 percent increase in the homeownership rate for all households in the second income quartile. However, for the households of age 25 to 34 in the second income quartile, a ten percentage point increase in the maximum loan-to-value ratio is associated with a 12.4 percent increase in the homeownership rate. This indicates that younger households are affected more by the down payment constraint associated with a lower loan-to-value ratio, since these households have had less time to save for this payment. Likewise, Chiuri and Jappelli (2003) investigate the distribution of homeownership rates across age groups, making use of an international dataset with almost 300,000 individuals out of 14 countries. They find that countries with down payment ratios of 40 percent have a proportion of young people being homeowners that is 5 to 8 percentage points less than countries with down payment ratios of 20 percent. Therefore, their results suggest that a relatively high down payment ratio induces young households to save longer, thereby postponing the purchase of a home and affecting the distribution of homeownership rates across age groups.
Since most of the existing literature is about the effect of borrowing constraints on house prices, this paper will examine what the effect is on homeownership. While the impact of borrowing constraints on homeownership has been shown in earlier studies, this research will provide the first results of the effect of the mentioned policy changes in the Dutch mortgage market on the probability of becoming a homeowner. On top of that, the outcome of
this study can be important for the decision-making process of policies, since the effect on homeownership can be taken into account in future policy implementations. As these policy changes make it more difficult to buy a home because obtaining a mortgage is more difficult and the maximum loan-to-value ratio is decreased, it can be expected that young households need more time to save for a down payment and postpone purchasing a house. This is also confirmed by the results of Andrews and Caldera Sánchez (2011) and Chiuri and Jappelli (2003). Therefore, this leads to the following hypothesis:
- Hypothesis 1: The policy changes will have a negative impact on the probability of becoming a homeowner.
Furthermore, because the change in the code of conduct in 2011 was binding, the impact of this policy change on homeownership is expected to be larger than the effect of change in the code of conduct in 2007. This leads to the following hypothesis:
- Hypothesis 2: The change in the code of conduct in 2011 has a larger impact than the change in the code of conduct in 2007.
The same is expected for the implementation of the law in 2013, as this has made the norms even more binding. Therefore, the last hypothesis is:
- Hypothesis 3: The implementation of the law in 2013 has a larger impact than the change in the code of conduct in 2011.
With these three hypotheses in mind, the effect of the policy changes on homeownership in the Dutch housing market will be investigated.
3 Methodology
This section explains the methodology that is used to test the hypotheses of this paper. The variables incorporated in the model will also be described. To employ this methodology, panel data on households in the Netherlands from the DNB household survey is used, as will be further discussed in section four.
3.1 Empirical method
To test the relationship between the policy changes and homeownership in the Dutch housing market, a difference-in-difference method will be used in combination with a logistic regression. The difference-in-difference method is used because it allows for estimation of the effect of a treatment, which in this research is a policy change. The first difference is the difference in the time period before and after the policy change, and the second difference is the difference between the treatment group and the control group. The treatment group consists of households that are likely to be affected by the policy changes, whereas the control group consists of households that are not likely to be affected by the policy changes. Since this paper focuses on three different policy changes, three time period differences are used, leading to three interaction terms in the model. The logistic regression is employed because this can estimate the probability that a dependent binary variable takes the value of one. In this paper, the dependent variable is the probability that a household goes from renter to homeowner. This means that the dependent variable is a binary variable that takes the value of one when a household moves from a rented house to an owned house in year t, and the value of zero otherwise. The model is inspired by the approach of Li (1977) and Andrews and Caldera Sánchez (2011), who use a logit method for the analysis of homeownership. Combining this method with the difference-in-difference method leads to the following model for this research:
1 Pr 𝑅𝑒𝑛𝑡 𝑡𝑜 𝑜𝑤𝑛 = 1
= 𝛽-+ 𝛽/𝑖𝑛𝑡071134+ 𝛽5𝑖𝑛𝑡1234+ 𝛽7𝑖𝑛𝑡1334+ 𝛽7𝑇𝑟𝑒𝑎𝑡34+ 𝛽𝑋34+ 𝛼3 + 𝜆4+ 𝜀34
In this model, the variables of interest are int0711it, int12it and int13it and Treatit is an
indicator of whether the household is part of the treatment group. The three variables of interest are interaction terms that measure the effect of the three policy changes on the
treatment group. Next to these variables, the model also contains control variables that are included in the vector X. These control variables are the age of the head of the household, household size, household structure, an indicator of disability of the head of the household, gross income of the household and education of the head of the household. Furthermore, regional fixed effects ai and time fixed effects lt are included in the model and standard
errors are clustered at the household level.
Finally, the identification strategy that is used to separate the treatment group from the control group has to be determined carefully. The reason for this is that during the time period of this sample, the Global Financial Crisis also happened, which could have led to changes in the behavior of households regarding homeownership. To distinguish the effect of the policy changes from other effects, a threshold for the loan-to-value ratio has been set. For households to be part of the treatment group, their loan-to-value ratio should be higher than 0.7. Households with a loan-to-value ratio of 0.7 or lower are part of the control group. This is because a loan-to-value ratio in this range requires a household to make a substantial down payment. Since these households were able to make this down payment, they are expected to not be impacted by the policy changes. Section six provides further information about the threshold that has been set. Furthermore, households that were already homeowners from the first year they participated in the survey are excluded from the sample, since they are not part of the treatment group and they also do not belong in the control group, as they were already homeowners from the beginning.
3.2 Variables
This section will provide more information about the dependent variable, the variables of interest and the control variables that are used in the model.
3.2.1 Dependent variable
In the logit model employed in this paper, the dependent binary variable is the probability that a household moves from a rented house to an owned house. Therefore, the binary variable takes the value of one in year t if a household changes from renter to owner in year t. For all other observations, which include remaining a renter in year t and moving from an owned house to a rented house in year t, the binary variable takes the value of zero in year t. The policy changes are likely to affect first-time buyers more than households who are already homeowners, as proven by Chiuri and Japelli (2003), since first-time buyers might need longer to save for the down payment of their first mortgage. Therefore, this binary variable is
used as the dependent variable to investigate what the effects of the policy changes are on homeownership in the Dutch housing market. When potential first-time buyers are negatively impacted by the policy changes and they have to postpone or even cancel the purchase of a house, there will be fewer first-time buyers, which can be related to homeownership in the Dutch housing market.
3.2.2 Variables of interest
The variables of interest are the interaction terms displayed in the model. These interaction terms are part of the difference-in-difference method and measure the effect of the policy changes. Since this research focuses on three policy changes, three interaction terms are included in the model. To create these interaction terms, dummy variables for the time periods are needed, as well as a dummy variable for the treatment group. The time period dummies indicate whether the observation is from after the policy change, and the treatment dummy indicates whether the household is part of the treatment group. As mentioned before, households are part of the treatment group when their loan-to-value ratio is higher than 0.7. Therefore, the treatment dummy takes the value of one when a household has a loan-to-value ratio higher than 0.7. The first interaction term, int0711it includes the time period between the
first and the second change in the code of conduct. Because the change in the code of conduct in 2011 happened in August, the year 2011 belongs to the time period between the first and the second change in the code of conduct (NVB, 2011). This results in a dummy variable taking the value of one if the observation is in year 2008, 2009, 2010 or 2011. If an observation is in this period and if the household is part of the treatment group, int0711it takes
the value of one. For the time period dummy of int12it, only year 2012 is included, because
the change in the code of conduct in 2011 happened in August and the implementation of the law in 2013 happened in January. The time period dummy of int13it contains the years 2013
and after.
When the effects of these interaction terms are negative and significant, it means that the policy changes have had a negative impact on the dependent variable, the probability that a household moves from a rented house to an owned house. This is what is expected, as stated in hypothesis 1. Moreover, hypothesis 2 and 3 state that it is expected that the change in the code of conduct in 2011 has a larger impact than the change in the code of conduct in 2007 and that the implementation of the law in 2013 has a larger impact than the change in the code of conduct in 2011. Therefore, the negative effect of int12it is expected to be stronger than the
negative effect of int0711it and the negative effect of int13it is expected to be stronger than the
negative effect of int12it.
3.2.3 Control variables
Besides the variables of interest, control variables are included in the model as well. The first control variable, which is also included in the model of Li (1977), is the age of the head of the household. This control variable is included in the model using dummy variables for different ranges of age, namely 30 or younger, 31-38, 39-47, 48-55, 56-64 and older than 65. As people get older, they are more likely to be homeowners because they have had more time to save for a house. Li (1977) and Andrews and Caldera Sánchez (2011) find positive relationships between age and the probability of being a homeowner. However, since the dependent variable of this paper is the probability of becoming a homeowner, that is the probability of moving from a rented house to an owned house, as opposed to the probability of being a homeowner, a negative relationship between age and the dependent variable is expected as the head of a household becomes older.
Next to age, the model also controls for household size and household structure. Li (1977) and Andrews and Caldera Sánchez (2011) include household size in their research, and Andrews and Caldera Sánchez add the structure of the household to their model. In this model, household structure can be living by himself/herself; couple with children living at home; couple without children living at home; and single with children living at home. Since Andrews and Caldera Sánchez (2011) find a significantly positive effect for household size on the probability of being a homeowner, this is also expected in this research.
Another control variable is the gross income of the household. Here, not only the gross income of the head of the household is used, but also the gross income of other members of the household are added to this. This is because the income of a woman, for example, can also determine the probability that a household moves to an owned house. To control for non-linearities, income is included in the model using dummy variables indicating different ranges of gross income. These ranges are 20000 euros or less, 20001-35000, 35001-50000, 50001-75000 and 75001 euros or more. Since having a higher income makes it easier to get a mortgage, because of the ability to pay for a down payment and the mortgage services, it is expected that higher ranges of gross income have a positive effect on the probability of becoming a homeowner (Andrews and Caldera Sánchez, 2011).
Apart from income, education is also controlled for in the model. Goodman (1988) argues that current income might not represent permanent income and that education could be
a determinant of permanent income. Therefore, education is included in this research as a dummy variable, indicating if a head of a household has completed education on a tertiary level. It is expected that education has a positive effect on the probability that a household moves from a rented house to an owned house.
Furthermore, Andrews and Caldera Sánchez (2011) also control for socio-economic disadvantages in their model by including indicators for disability, ethnic minority and immigrant status. Since the data used for this research only provides information about disability, a dummy variable is added to the model that takes the value of one if the head of the household is disabled. The effect of a disability of a household head is expected to be negative.
Finally, regional fixed effects and time fixed effects are added to the model. Regional fixed effects are incorporated using dummies that indicate the province a household lives, and dummies that indicate the level of urbanization in the area. Urbanization is divided into five categories, ranging from very low degree of urbanization to very high degree of urbanization. Year fixed effects are added to the model to control for economic trends.
4 Data and descriptive statistics
This section describes the data that is used in this research. Firstly, where the data is gathered from will be discussed. Secondly, data preparation is explained and at last descriptive statistics of key variables are presented.
For this study, personal data from households is needed, such as the income, the mortgage amount and the health status. Therefore, panel data from the DNB Household Survey is used. Since 1993, CentERdata has been collecting data annually through the DNB Household Survey (CentERdata, 2018). This survey consists of six questionnaires: general information on the household, household and work, accommodation and mortgages, health and income, assets and liabilities and economic and psychological concepts. The first five questionnaires are used in this paper. The DNB Household Survey is presented every year to approximately 2000 households that participate in the CentERpanel, to every person in a household that is older than sixteen. A household is replaced by another household with similar characteristics if it wants to leave the sample. The composition of the CentERpanel is being observed constantly to guarantee that the internet panel reflects the Dutch population (CentERdata, 2018).
In this research, survey data from the years 1994-2017 is used. For every year, the different questionnaires have been merged together based on household number and member of the household. After that the different years are appended together to get one file consisting of 121,576 observations. Gross income of all members of the household has been added up to get one number for gross income per household per year. This is done because only the head of the household is being kept in the sample, and the income of a partner can also have an impact on the probability of becoming a homeowner. The reason for only keeping the head of the household is that when someone moves, it is likely that the household moves, so only one observation per household per year is relevant. Furthermore, there were many missing values in the sample, and they have been removed. Households that were already homeowners from the first year they participated in the sample are removed as well, as explained earlier. Moreover, many outliers were found, which could indicate that people did not fill in the survey correctly. For example, the house price and mortgage amount had to be given in thousands of euros. However, sometimes these values were so high that it is expected that these values were not filled in in thousands of euros. This could mean that the sample is subject to what is called observational bias in this paper, which occurs when people tend to fill in false answers to the survey questions. To correct for these outliers, the mortgage amount has been divided by 1000 if it was higher than 2000, and house price has been divided
by 1000 if it was higher than 5000. Furthermore, gross income has been multiplied by twelve if it was below 2000, because it is assumed that those respondents did not fill in their yearly gross income and instead filled in their monthly gross income. Next to that, loan-to-value ratios higher than 1.5 have been removed from the sample. At last, to further correct for outliers, the numerical variables have been winsorized. This results in a dataset of 10,332 observations, of which descriptive statistics for key variables are presented in Table 1.
Table 1: Descriptive statistics
Mean Median Std. Dev. Minimum Maximum Observations
Age 50.58 50 16.25 19 94 10,332 Gross income 39,716.31 33,000 29,000.46 3,299 227,893 10,332 House price 191,445.20 185,000 95.16 11,000 590,000 1,186 Mortgage amount 182,527.20 177,000 93.41 4,800 500,000 1,186 LTV 0.968 1.033 0.235 0.0086 1.5 1,186 Household size 1.98 2 1.12 1 8 10,332 Household structure - Single
- Couple with children - Couple without children
- Single with children
0.405 0.185 0.348 0.061 0 0 0 0 0.491 0.388 0.476 0.240 0 0 0 0 1 1 1 1 10,332 10,332 10,332 10,332 Education 0.373 0 0.484 0 1 10,332 Disability 0.069 0 0.253 0 1 10,332
Table 1 shows the mean, median, standard deviation, minimum and maximum of the variables included in the regression. The number of observations of each variable in the dataset is also displayed. Household structure, education and disability are dummy variables, which can only take the value of zero or one. The average age of the heads of the household in the sample is 50.58 years, with heads ranging from 19 years old to 94 years old. The average gross income per household in this sample is 39,716 euros, while the median is 33,000 euros. House price, mortgage amount and loan-to-value ratio have less observations because not every household owns a house and because loan-to-value ratios above 1.5 have been removed from the sample. House price has a mean of 191,445 euros, and mortgage amount has a mean of 182,527 euros. These observations result in an average loan-to-value ratio in the sample of 0.968, since the loan-to-value ratio in this dataset is calculated by dividing the mortgage amount by the house price. The mean of the loan-to-value ratio is quite
high, and a high loan-to-value ratio in the Netherlands is one of the reasons for the policy changes discussed in this paper. The mean of the household size is 1.98, and the majority of the sample, namely 40.5 percent, are living by himself/herself. Only 6.1 percent of the households in the sample are singles living with children at home. Lastly, 37.3 percent of the heads of the household have completed education on a tertiary level, and 6.9 percent of the heads of the household are disabled.
Next to the descriptive statistics of the key variables, the dependent binary variable is presented in a graph. This is the dummy variable that takes the value of one in year t if a household moves from a rented house to an owned house in year t. Therefore, Figure 1 shows the number of households that became homeowners every year.
Figure 1: Number of households moving from rented to owned house each year
It can be seen that in the run up to the Global Financial Crisis, many households moved from a rented house to an owned house, whereas this number decreased after the Global Financial Crisis. In the latest years, the number slightly increased again, but is also slightly decreasing after that. This could indicate that households are affected by the policy changes, especially the law implementation in 2013. However, it can also be seen that there are not many observations of households moving from a rented to an owned house. There are 187 of these observations in the whole sample, which could make it difficult to draw conclusions. 0 0 0 0 0 5 0 3 2 5 28 18 20 23 11 9 8 4 7 5 10 12 9 8 0 5 10 15 20 25 30
5 Results
In this section, the effects of the described policy changes on the probability that a household moves from a rented house to an owned house will be discussed. To do this, the results of the empirical model explained before will be interpreted.
The marginal effects of the logistic regression are shown in Table 2. The coefficients of the logistic regression are reported in table A.1 of Appendix A. In the first column of both tables, a basic logistic regression is presented, including only the variables of interest. When looking at Table 2, it can be seen that the three marginal effects of the interaction terms are all significantly negative at the 1% level, indicating negative effects of the three policy changes on the probability that a household moves from a rented house to an owned house. In column two, the marginal effects of the logistic regression are presented, without controlling for fixed effects. Again, all three marginal effects coefficients are negative and significant at the 1% level. When looking at the third column, which is the logistic regression with control variables and fixed effects, all three interaction terms still have negative and significant marginal effects coefficients. This is in line with hypothesis one of this research, which stated that all three policy changes have had a negative impact on the probability of becoming a homeowner. The first interaction term, Int0711, indicates the first policy change. This is the change in the code of conduct in 2007. The marginal effects coefficient of this interaction term is equal to -0.0317 and significant at the 5% level. This means that the probability that a household moves from a rented house to an owned house decreased by 3.17 percentage points after the change in the code of conduct in 2007, holding all else equal. The marginal effects coefficient of Int12 implies that after the change in the code of conduct in 2011, the probability of moving to an owned house decreased by 4.5 percentage points compared to the period between the first two policy changes. This effect is also significant at the 5% level. The effect of the change in the code of conduct in 2011 is bigger than the effect of the change in the code of conduct in 2007, since the marginal effects coefficient of Int12 is more negative than the marginal effects coefficient of 2007. This is consistent with the expectation described in hypothesis two. The third hypothesis stated that the implementation of the law in 2013 has had a larger impact on the probability of becoming a homeowner than the change in the code of conduct in 2011. This hypothesis is not supported by this research. The marginal effects coefficient of Int13 is equal to -0.0449 and significant at the 1% level, which means that after the law implementation in 2013 the probability that a household moves to an owned house decreased by 4.49 percentage points, compared to the period between the change in the code of conduct in 2011 and the law implementation in 2013. The decrease in the probability by
Table 2: Marginal effects of logistic regression coefficients Dependent variable: probability of renter to homeowner
Note: *, **, *** denote significance at the 10%, 5% and 1% level, respectively. Robust standard errors clustered at the
household level are reported in parentheses. The benchmark against which themarginal effects are interpreted is a household with a head younger than 31, living by himself/herself, with an income lower than 20001 euros per year and a relatively low
education and no disability.
(1) (2) (3) Int0711 -0.0176*** -0.0172*** -0.0317** (0.00277) (0.00267) (0.0148) Int12 -0.0255*** -0.0214*** -0.0450** (0.00651) (0.00629) (0.0225) Int13 -0.0228*** -0.0199*** -0.0449*** (0.00305) (0.00295) (0.0124) Treat 0.0803*** 0.0757*** 0.103*** Age of head (0.00463) (0.00450) (0.0103) 31-38 years -0.00255 -0.00468 (0.00331) (0.00415) 39-47 years -0.00768** -0.0103** (0.00372) (0.00461) 48-55 years -0.0115** -0.0175*** (0.00485) (0.00646) 56-64 years -0.0151*** -0.0208*** (0.00578) (0.00735) >65 years -0.0195*** -0.0277*** Household structure (0.00508) (0.00680) Household size -0.000660 -0.00114 (0.00261) (0.00341)
Couple with children at home -0.000230 -0.000278
(0.00806) (0.0104)
Couple without children at home -0.000115 -0.00266
(0.00380) (0.00521)
Single with children at home 0.00380 0.00344
Gross income of household
(0.00740) (0.0103) 20001-35000 euros 0.0103* 0.0144** (0.00536) (0.00633) 35001-50000 euros 0.00322 0.00642 (0.00556) (0.00672) 50001-75000 euros -0.000658 0.00212 (0.00578) (0.00715) >75001 euros -0.000425 0.00364 Education of head (0.00621) (0.00803) Tertiary -0.000335 -0.00208
Indicator of disability of head
(0.00256) (0.00357)
Disability -0.0225* -0.0278*
(0.0129) (0.0155)
Observations 10,332 10,332 8,057
Regional fixed effects No No Yes
4.49 percentage points is a slightly smaller impact than the decrease in the probability by 4.5 percentage points, and therefore hypothesis three is not true.
Overall, regarding the variables of interest, the results of the logistic regression indicate negative impacts of the three policy changes on the probability of moving from a rented to an owned house. As this probability decreased, it is likely that there are fewer first-time buyers per year. This means that there are fewer new entrants in the owner-occupier sector, indicating a negative effect of the policy changes on homeownership in the Dutch housing market. This effect is equal to the effect that was expected, based on the negative relationships between borrowing constraints and homeownership that were found in previous literature (Acolin et al., 2016; Andrews and Caldera Sánchez, 2011; Linneman et al., 1997)
With regard to the control variables, the dummy variables for age are partly in line with the expectations. The age range of 31-38 years has an insignificant marginal effects coefficient, and the marginal effects coefficients for the other age ranges are significantly negative, meaning that as heads of households get older, they are less likely to move from a rented house to an owned house. This is because as the age of the head of the household increases, the household is less likely to become a first-time buyer, because they might already be homeowners or have other reasons for not wanting to become a homeowner. Apart from age, the model also controls for household structure and household size. Household size was expected to have a positive relationship with the probability of moving from a rented house to an owned house, but in this research no statistically significant effect is found. For the dummies for household structure there are also no significant marginal effects coefficients, and therefore no conclusion can be drawn regarding the effect of these variables on the probability of becoming a homeowner.
When looking at the dummies for gross income, it can be seen that a gross income between 20001 and 35000 euros increases the probability of moving from a rented house to an owned house by 1.44 percentage points, compared to an income lower than 20001 euros. The other income ranges all have positive marginal effects coefficients as well in column three, but they are all insignificant. A reason for this could be that the positive effect of gross income on the probability of becoming a homeowner decreases as gross income increases. However, another explanation could be that the sample is subject to observation bias, which is explained earlier in section four.
Furthermore, a head with a higher level of completed education is not more likely to become a homeowner, as the marginal effects coefficient of this variable is statistically insignificant. Andrews and Caldera Sánchez (2011) find insignificant effects in some
countries as well, and they mention that the reason for that could be the possibility of a high level of correlation between education and income.
Finally, the marginal effects coefficient of the indicator of a disability of the head of the household is negative and significant at the 10% level. The effect can be interpreted as a decrease of 2.78 percentage points in the probability of moving from a rented to an owned house if the head of the household is disabled, holding all else equal. This is equal to the effect that was expected, based on the findings of Andrews and Caldera Sánchez (2011).
6 Robustness checks
This section discusses robustness checks that are performed to assess the results of this research. Since this research makes an important assumption when determining the treatment group, it is crucial to investigate how the results change when this assumption is being changed. The assumption used for the treatment group is that a household has a loan-to-value ratio that is higher than 0.7. This threshold is set at lower and higher values, to see what this does to the outcome of the logistic regression. In Table 3, these results are reported. Table 3 presents the marginal effects coefficients of the variables of interest, while the coefficients of all variables of the logistic regression are presented in table B.1 of appendix B.
From Table 3 it can be seen that for a loan-to-value ratio of 0.6 as the threshold, the marginal effects coefficients barely change compared to a loan-to-value ratio of 0.7. When using a threshold of a loan-to-value ratio equal to 0.8, however, the variable Int12 becomes insignificant, which results in hypothesis one and hypothesis two being not true in this research. A reason for this could be that when the threshold is set at this level, many people are in the treatment group and thus affected by the policy change, leading to an inadequate distribution between the treatment and the control group. The same happens when a loan-to-value ratio of 0.5 is used as the threshold for the treatment group, possibly because of a limited amount of observations that are present in the treatment group. For the same reason, probably, the model has trouble converging for thresholds with loan-to-value ratios lower than 0.5.
From these findings, it can be concluded that setting a loan-to-value ratio as the threshold that is above 0.8 or below 0.6, results in slightly different outcomes. Since households with loan-to-value ratios lower than 0.7 are still required to make a substantial down payment, it is assumed that they are not affected by the policy changes and that they belong in the control group.
Next to the analysis for the threshold of the treatment group, another robustness check is presented in Table 4. Here, all households with a head older than 50 are deleted from the sample, since they are not likely to be first-time buyers. Table 4 displays marginal effects coefficients of the variables of interest, and the coefficients of the logistic regression are reported in Table B.2 of Appendix B. From Table 4, it can be concluded that removing these observations from the sample does not have an impact on the marginal effects and the hypotheses. However, when removing heads of households with ages higher than 35 years from the sample, the logistic regression model has trouble converging. The same happens for removing the heads of households that are older than 40 and 45, and a reason could be an
inadequate distribution of the treatment and the control group across the dependent binary variable. As can be seen in the descriptive statistics in Table 1, the mean of age is 50.58 and the median is 50. From table 4 it can be seen that almost half of the sample is removed if heads older than 50 are dropped. Dropping heads older than 35, 40, and 45 leaves even less observations in the sample, meaning the loss of many observations of households becoming a homeowner. Since there were not many observations of this in the first place, as is shown in Figure 1, this could be the reason that the model does not converge. Nonetheless, the results discussed in section fiveare still robust to removing the heads of the household older than 50 from the sample.
Table 3: Marginal effects of logistic regression coefficients of robustness checks Dependent variable: probability of becoming a homeowner
(1) (2) (3) (4) LTV 0.5 LTV 0.6 LTV 0.7 LTV 0.8 Int0711 -0.0436** -0.0370** -0.0317** -0.0330*** (0.0178) (0.0153) (0.0148) (0.0108) Int12 0.155*** -0.0504** -0.0450** -0.0276 (0.0215) (0.0228) (0.0225) (0.0221) Int13 -0.0449*** -0.0428*** -0.0449*** -0.0355*** (0.0146) (0.0132) (0.0124) (0.0105) Treat 0.119*** 0.1072*** 0.103*** 0.0864*** (0.0131) (0.0112) (0.0103) (0.00838) Observations 8,057 8,057 8,057 8,057
Controls Yes Yes Yes Yes
Regional FE Yes Yes Yes Yes
Year FE Yes Yes Yes Yes
Note: *, **, *** denote significance at the 10%, 5% and 1% level, respectively.
Robust standard errors clustered at the household level are reported in parentheses.
Table 4: Marginal effects of logistic regression coefficients of robustness check 2 Dependent variable: probability of becoming a homeowner
(1) (1)
Whole sample Only heads younger than 50
Int0711 -0.0317** -0.0759** (0.0148) (0.0311) Int12 -0.0450** -0.105*** (0.0225) (0.0402) Int13 -0.0449*** -0.0945*** (0.0124) (0.0284) Treat 0.103*** 0.186*** (0.0103) (0.0237) Observations 8,057 4,091
Controls Yes Yes
Regional FE Yes Yes
Year FE Yes Yes
Note: *, **, *** denote significance at the 10%, 5% and 1% level, respectively.
7 Limitations and future research
Although section six shows that the results still hold after various robustness checks, this research has several limitations. First of all, as can be seen in Figure 1 in section four, there are only 187 observations of households moving from a rented house to an owned house in the years 1994 to 2017 in this dataset. Even though Figure 1 displays increases and decreases in the number of households moving from a rented house to an owned house per year, the number of observations might be too low to draw valid conclusions. Moreover, all observations in the dataset could be subject to observation bias, since answers might not be filled in correctly. Although it is tried to correct for outliers and incorrect answers, this bias might still be present. Next to observation bias, the survey could be subject to attrition bias, because households have left the sample throughout the study (Stock and Watson, 2015). Although these households are replaced by other households with the same characteristics, it could be possible that the households are not perfectly the same, so that attrition bias might occur. Finally, a strong assumption is made about the threshold for the treatment group and although robustness checks are presented that show the reasoning behind this threshold, one could argue that a different threshold needs to be set to be certain that households are not affected by the policy changes and are thus part of the control group. However, as shown by the robustness checks, this is not feasible when using this dataset.
For future research about this topic or to improve the quality of this research, it might therefore be better to use a different dataset. Despite the lavishness of the DNB Household Survey, other data that is not subject to the biases mentioned above might be able to provide more valid results and conclusions. Furthermore, since the results of this research indicate that the probability of becoming a homeowner has decreased, it is likely that there are fewer new entrants in the Dutch owner-occupier sector after the policy changes. More research could be done to confirm the relationship between fewer new entrants and homeownership in the Netherlands. Lastly, since there is little research about these relatively recent policy changes in the Netherlands, more research could be done to endorse the effects of the policy changes on this topic but also on other topics, such as what the effect of the policy changes is on house prices in the Netherlands.
8 Conclusion and discussion
This study investigated the effects of three policy changes regarding mortgage issuance on the probability of becoming a homeowner in the Netherlands. The three policy changes are the change in the code of conduct in 2007, the change in the code of conduct in 2011 and the implementation of a law in 2013. This has been investigated using a difference-in-difference method in combination with a logistic regression. Use is made of panel data for the years 1994 to 2017 from the DNB Household Survey provided by CentERdata. The main finding of this paper is that the three policy changes all have had a negative effect on the probability of becoming a homeowner. Following this decrease in probability of becoming a homeowner, it is likely that there are fewer first-time buyers entering the Dutch owner-occupier sector. Therefore, this paper provides evidence of a negative effect of the examined policy changes on homeownership in the Dutch housing market.
Since the policy changes make it more difficult to get a mortgage and to purchase a house, the findings of this paper are in line with the results of Linneman et al. (1997), Acolin et al. (2016) and Andrews and Caldera Sánchez (2011), which show a negative relationship between borrowing constraints and homeownership. This paper extends the literature by providing the first results of the effects of the mentioned policy changes on the probability of becoming a homeowner in the Netherlands. Besides this contribution to the literature, this research is also of practical relevance. Policy makers can take the found effect on homeownership into account in the decision-making process of future policies regarding mortgage issuance.
Nonetheless, this research also has its limitations. The data might be subject to observation bias and attrition bias and there might be too few observations to draw valid conclusions. Next to that, a strong assumption about the distinction between the treatment group and the control group has been made. These limitations could be improved in future research by the use of a different dataset. Lastly, since the effects of the mentioned policy changes have not been studied often before, it could be interesting to further examine the effects on homeownership in the Netherlands or on other topics such as house prices in the Netherlands.
9 References
Acolin, A., Bricker, J., Calem, P., & Wachter, S. (2016). Borrowing Constraints and Homeownership, American Economic Review, 106(5), 625-629.
Almeida, H., Campello, M., & Liu, C. (2006). The Financial Accelerator: Evidence from International Housing Markets. Review of Finance, 10(3), 321-352.
Andrews, D., & Caldera Sánchez, A. (2011), The Evolution of Homeownership Rates in Selected OECD Countries: Demographic and Public Policy Influences. OECD
Journal: Economic Studies, 2011(1), 1-37.
Andrews, D., Caldera Sánchez, A., & Johansson, A. (2011). Housing Markets and Structural Policies in OECD Countries. OECD Economics Department Working Papers, No. 836, OECD Publishing, Paris.
CentERdata, (2018). DNB Household Survey (DHS). Retrieved from
https://www.centerdata.nl/en/projects-by-centerdata/dnb-household-survey-dhs CentERdata, (2018). The CentER panel. Retrieved from
https://www.centerdata.nl/en/projects-by-centerdata/the-center-panel
Chiuri, M., & Jappelli, T. (2003). Financial Market Imperfections and Homeownership: A Comparative Study. European Economic Review, 47(5), 857-875.
De Nederlandsche Bank, (2014). Overview of Financial Stability in the Netherlands. Retrieved from https://www.dnb.nl/en/binaries/OFSnajaarUK_tcm47-312971.pdf De Nederlandsche Bank, (2015). Dutch mortgages in the DNB loan level data. Occasional
Studies, 13(2). Retrieved from https://www.dnb.nl/en/news/dnb-publications/dnb-
occasional-studies/dnb332480.jsp
De Nederlandsche Bank (2015). Effects of further reductions in the LTV limit. Occasional
Studies, 13(2). Retrieved from https://www.dnb.nl/en/binaries/OS13%20uk_tcm47-
322569.pdf
De Nederlandsche Bank, (2017). Financial Stability Report. Retrieved from https://www.dnb.nl/en/binaries/OFS_Autumn%202017_tcm47-363954.pdf
DiPasquale, D., & Glaeser, E. (1999). Incentives and Social Capital: Are Homeowners Better Citizens? Journal of Urban Economics, 45(2), 354-384.
Duca, J. V., Muellbauer, J. & Murphy, A. (2011). House Prices and Credit Constraints: Making Sense of the US Experience*. The Economic Journal, 121(552), 533-551. Elsinga, M., Priemus, H., & Boelhouwer, P. (2016). Milestones in housing finance in the Netherlands, 1988–2013. Milestones in European Housing Finance, 255.
