The influence of the mortgage interest rate deduction on housing prices

The Influence of the Mortgage

Interest Rate Deduction on Housing

Prices

Student: Renske van Eijk (2659921)

Institution: Leiden University

Study: Public Administration:

Economics & Governance

Course: Master Thesis 2020

Instructor: E. Suari-Andreu

Second Reader: Dr. H. Vrijburg

Date: 2020, June 9

Content

1. Introduction………..………...3

2. Institutional Context………5

2.1. History of the Mortgage Interest Rate Deduction………5

2.2. The Influence on Housing Prices………...6

2.3. The Global Financial Crisis……….7

2.4. Reform of the Mortgage Interest Rate Deduction………9

3. Literature Review………..11 4. Research Design………13 4.1. Data Selection………13 4.2. Descriptive Statistics………. 14 4.3. Empirical Model……….18 5. Empirical Analysis………23

6. Discussion and Conclusion………...28

7. References……….30

8. Appendix………...33

Appendix Table 1. Basic FE Regression……….…..34

Appendix Table 2. OLS Regression……….……...……..35

Appendix Table 3. Comparison Renters and Home Owners………..…..36

Appendix Table 4. Comparison Different Ages………..……..36

Appendix Table 5. Comparison Youngest Age-group……….…………..37

1. Introduction

For years, the International Monetary Fund (IMF) has criticized the Mortgage Interest Deductibility (MID) in the Netherlands. In the years prior to the Global Financial Crisis (GFC), the Dutch government did not consider future reform regarding this policy. Whilst the housing market was booming, housing prices were also rising explosively, until the GFC hit in 2008, after which housing prices plummeted and many households came into problems with paying off their mortgages. It was in the aftermath of the crisis that several policies had to be reconsidered by the Dutch government. An important policy was the MID, as it concerns almost all landlords and potential home owners in the Netherlands. Therefore, this reform has gotten much attention, as it caused uncertainty amongst households in the Netherlands.

In 2012, the official reform was announced, and it came into force in 2013 (Lejour, 2016). However, in the years prior to 2013, the reform was highly debated for several years. The MID has made it increasingly possible for more people to buy a house, which caused an increase in the demand for housing. Since it takes time for the construction sector to respond to increasing demand by building more houses, this in turn results in the increase of housing prices. This, in turn, would also mean that a stricter MID causes demand and therefore housing prices to decrease. This is the theory that is tested in this research. The research is based on data of the DNB Household Survey (DHS), containing many questions about the characteristics of the households and mortgages. Since all potential new home owners are affected at the same time, the analysis focusses on what the effect is of the expectation of a limitation of the MID. To account for the demand, the analysis focusses on the intention to buy a house. Since the data are panel data, this allows for the Ordinary Least Squares (OLS) model, the Fixed Effects (FE) model or the Random Effects (RE) model to be applied. After performing several tests, the FE model is applied. The analysis focusses on household level and the years 2004 until 2013, since the reform came into force in that year.

The empirical results show that there is a significant effect of the expectation of a limitation on the intention to buy a house. However, this effect shows an increase of the intention to buy a house if one expects a limitation. For not expecting a limitation, there are no significant results in the basic FE regression. However, the theory is also tested for different subgroups. For the age-group of 34-40, and if the respondent followed vocational colleges, the expectation of a limitation also caused the intention to buy a house to increase, compared to not knowing if one should expect a limitation. There was no distinction between renters and home-owners and for the other age and education groups.

The thesis starts with explaining the institutional context. In doing so, the emergence of the MID, the effect on housing prices, the crisis and the reform are elaborated on. To understand how the policy reform came about it is important to understand the context in which the respondents answered the questions of the DHS. Several academics have discussed the effect of a possible limitation of the MID, which are discussed in the literature review.

Thereafter, in the second part of the thesis, the empirical analysis is elaborated. Firstly, the research design is discussed. This chapter contains the data selection, descriptive statistics and the explanation on the empirical model. After this, the empirical analysis is performed. Finally, the thesis concludes with a chapter discussing and concluding the empirical results.

2. Institutional Context

2.1. History of the Mortgage Interest Rate Deduction

One might argue that the MID has been introduced to encourage people to buy a house. However, according to Kromhout & Oving (2008), the MID initially was not created for this purpose. The main reason why the MID policy was created is because, in theory, the owner of a house is able to rent out the house to one another person, and therefore is able to earn money from the renting of their house. The MID remained unchanged over a long period of time: citizens included their homes into their income, after which citizens could deduct their mortgage interest from their taxes. According to Kromhout & Oving, it was only after the Second World War, that the view that the MID stimulating home ownership became more widely known. It was argued that the MID contributed to more people owning a home and therefore having more sense of responsibility, a stable family and an increasing savings account. Homeownership also stimulates the housing sector and the economy as a whole (Kromhut & Ovinig, 2008).

In the past decade, the number of citizens owning a home raised gradually: in the 50s the mortgage guarantee was introduced, which stimulated homeownership. In the 60s and 70s the percentage of housing for sale increased to 35% (Kromhout & Oving, 2008). According to Elsinga, de Jong-Tennekes & van der Heijden (2011), since the end of the economic crisis of the late 70s, house prices increased continuously until the crisis of 2008. Mostly in the 90s, housing prices rose vigorously. In this period there was a high confidence in the economy, which resulted in declining interest rates and rising incomes. Because the demand increased, this put pressure on the supply of houses, causing housing prices to increase. In reaction to this, the government broadened the provision criteria of mortgages. As a consequence, the amount of the mortgages that were taken out was higher than ever. Eventually, since 1997, the number of homeowners exceeded the number of renters (Elsinga, de Jong-Tennekes & van der Heijden, 2011).

Around the millennium change, almost every political party supported the idea of promoting home ownership by providing, among other policies, the MID (Kromhout & Oving, 2008). On the one hand, low interest and increasing housing prices, made it profitable for both buyers as for investors to invest in rental houses, for it yielded a high return. Municipalities on the other hand also profited of these times: investors and buyers had to pay money to the municipality for land, which again could be invested in society. Land yield was an important

source of income for municipalities in this time (Elsinga et al., 2011). Popularity of the MID kept on increasing among different actors.

In that same period, according to Teye, de Haan, Elsings, Bodinuba & Gbadegesin (2017) also the Dutch perception on homeownership has shifted to wanting to pursue independency from the landlord, and therefore wanting to own a home. The authors argue that when people have a partner or if they get married, the couple wants to invest in a house together. From this perspective, homeownership also encourages deeper relationships with friends and family, and therefore is important (Teye, de Haan, Elsinga, Bodinuba & Gbadegesin, 2017). According to Scanlon & Elsinga (2014), homeownership also increases the total wealth of households. This has caused more people wanting to own a home and therefore using, amongst others, the MID to finance their house (Scanlon & Elsinga, 2014).

In the meantime, according to Kromhout & Oving (2008), the MID deduction imposed increasing costs to the Dutch government. For example, in 2001 the taxes on home ownership (the Rateable Value) brought in €5.1 billion, while the costs of the deduction were €7.3 billion. This means that the deductibility imposed €2.2 billion of costs on the Dutch government (Kromhout & Oveing, 2008). These high costs were mainly spent on the households that earn the most. In 2008, the taxes paid back by the government, due to the MID, was €10 billion. Almost half of this amount, was paid out to the 20% of households with the highest incomes. The higher the mortgage expenses of a household, the higher the amount that they are able to deduct (Central Agency for Statistics, 2010).

Also, people just took out full mortgages on second homes, sold their first home and spent the overvalue of this first home on consumption. That is why, since 2001, the deductibility was only possible for a maximum of 30 years. Also, it was no longer possible to deduct the interest of a second home. Additionally, in 2004 the government imposed a rule that also made it obligated to invest the overvalue of the old house into the new mortgage. As a consequence, households that move to another house were only able to have one mortgage with the deductibility. In 2005, the government made it more profitable to repay housing mortgages sooner by decreasing the Rateable Value (Kromhout & Oving, 2008).

2.2. The Influence on Housing Prices

The MID has made it possible for more people to own a home, since the rules before 2013 were less strict. This means that it was possible for more people with a lower income to borrow

(CBS), written by Frank Notten (2011), the interest-only mortgage caused households to take out higher mortgages, which were kept as high as possible for as long as possible, since this would benefit them in being able to deduct more interest. In other countries, where no MID or a stricter MID was offered, households chose to pay off their debt more quickly (Notten, 2011). In turn, according to Elsinga et al. (2011), the height of the available mortgages increases housing prices. The reason is that mortgage interests and mortgage conditions are the transmission channels between the financial market and the housing market: the higher the available mortgages, the higher the housing prices. This also works the other way around: mortgage lenders also adjust to the housing market, meaning that the higher the housing prices, the higher the mortgages will be, and therefore housing prices again will increase (Elsinga et al., 2011). Housing prices in the Netherlands increased with 228% in the period of 1985-2007, while inflation in that same period was only 56%. In turn, this has caused households in the Netherlands to have the highest long-term debts of Europe. Around 90% is caused by mortgage debts. The total long-term debt in the Netherlands, rose from 56% in 1985 of GDP, to 125% of GDP in 2007. The amount of the mortgage debts was increasing, as a consequence of increasing housing prices. Even tough in other countries housing prices were also rising, they did not rise as high as they did in the Netherlands (Notten, 2011).

Since it was easier for more people to lend money, this increases the demand for housing. However, housing supply responds to the demand for housing with a considerable delay. According to Elsinga et al. (2011), the duration of the building of a house is long, and due to this delayed response, housing prices increase to compensate for this delay (Elsinga et al, 2011). This means that the easier it is to deduct the mortgage interest; the higher housing prices will be.

2.3. The Global Financial Crisis

The GFC hit Europe in 2008, which put pressure on the government to cut spending, and as a consequence, amongst others, to limit the MID. Banks all over the industrialized countries came into trouble and national governments had to step in to help. Because governments had to intervene with these financial institutions, budget deficits and government debts rapidly increased. In the meantime, international trade declined, which had huge consequences for the national economy: there was less demand for Dutch products and many organizations had to fire employees. As a consequence, unemployment in the Netherlands increased (Elsinga et al., 2011).

The crisis also had its influence on the housing market in the Netherlands. Many households postponed moving to another house. This declining demand resulted in an increasing supply of housing. Due to this increasing supply, the value of many houses declined to an amount that was lower than the mortgage that was taken out on that house. Combined with an increasing unemployment rate, many households could not afford their mortgage anymore. Forced selling of houses increased, leaving these households with a residual debt due to the declining of the housing prices (Elsinga et al., 2011). In 2011, the National Mortgage Guarantee Institute in the Netherlands, reported 2,021 forced selling’s of houses, which was 51% more compared to 2010 (Nationale Hypotheek Garantie, 2012).

After the crisis, households in the Netherlands had a very high mortgage debt compared to other European countries, as measured by percentage of GDP. The ratio was even more than 100%, which was the highest in Europe (Scanlon & Elsings, 2014). Not only the MID contributed to these high mortgage debts, but also the loan-to-value measure. This is the ratio between the amount of the mortgage, and the market value of one’s house. This ratio determines the height of the mortgage one can take out. The higher the amount one can lend, the higher the housing prices and the higher the risk (Verbruggen, van der Molen, Kakes & Heeringa, 2015). Both the European Commission (Notten, 2011), and the International Monetary Fund (IMF) criticized the MID system in the Netherlands (IMF, 2011). After additional pressure from the Financial Markets Authority, De Nederlandsche Bank (DNB), the Ministry of Finance and other politicians themselves, Dutch banks changed the Code of Conduct for Mortgage Loans in 2011. Scanlon & Elsinga (2014) argue that during this period it became clear that consumers could make wrong decisions, especially in finance (Scanlon & Elsinga, 2014). Later, in 2012, an important report was published by four relevant social partners in the housing industry. These partners were the representative of home owners, housing associations, real estate agents and the tenant associations. In their rapport, called ‘Wonen 4.0’ (Living 4.0) the housing market of the future was illustrated, without the MID. Also, reports were written by the Social Economic Council (SER) in 2010, the CPB Netherlands Bureau for Economic Policy Analysis (CPB) in 2010, and warnings were presented by the DNB and the Workgroup on Wide Housing Reconsiderations that was organized by the central government. These reports made politicians and society become aware of the risks of the MID. These reports presented many alternatives to the current system, making a possible reform more tangible (Lejour, 2016; Mastrogiacomo, 2013). Lejour (2016) wrote an article for the CPB, in which he mentions two ways in which the crisis had an influence on the change of the MID. First of all, the government had to stimulate

the economic arguments that were once for the MID, were now changing, since the risks became clearer (Lejour, 2016).

2.4. Reform of the Mortgage Interest Rate Deduction

In 2012, the government came to a new agreement about the MID, which would come into force the 1st _{of January of 2013. The following requirements have remained the same: (1)}

everyone is entitled to the MID for a maximum of 30 years. After 30 years, it is not possible to deduct the mortgage interest from the taxable income anymore; (2) the deductibility can only be applied to the main residence. It is therefore not possible to deduct the mortgage interest from, for example, a second house or a vacation house; (3) the mortgage only can be used to pay for the (maintenance of) the house; (4) If the house is sold with an overvalue, one is obligated to put this sum of money into the new mortgage. The value of the overvalue is not deductible (Redactie Rente.nl, 2019).

However, other requirements have changed. The first measure that changed is that it is only possible to take out a mortgage in form of a linear or annuity mortgage. Before 2013, it was also possible to take out a mortgage in form of an interest-only, investment or savings mortgage. Especially for first-time buyers, the abandoning of the interest-only mortgages has major consequences (Scanlon & Elsinga, 2016). The interest-only mortgage made it possible to only pay the monthly mortgage interest. It was not obligatory to pay off an amount of the mortgage loan itself, despite the fact that this was possible if one wanted to. In this way, the total mortgage debt did not change. At the end of the mortgage term, the full debt had to be paid. The responsibility lay with the person that took out the mortgage (ABN Amro, n.d.). Due to this policy change, a reduction in the demand for housing, and an increase in the demand for rental housing is expected (Scanlon & Elsinga, 2016). Also, instead of being able to deduct the mortgage interest for 30 years, from 2013 onwards one only receives MID if one takes out a mortgage that is fully paid back within 30 years. This measure is taken to lower the total mortgage debts households take out (Lejour, 2016).

Before 2013, it was also possible to deduct a mortgage interest of 52% of one’s taxable income (in scale 4). That was equal to the payroll tax rate one should pay in that scale. From 2013 onwards, there is a reduction of this interest of 0,5% each year. Eventually, the tax deduction will be lowered to a maximum of 38%, for the higher incomes. This reduction applies to all new mortgages that are taken out, but also to all mortgages that were taken out before 2013 (Scanlon & Elsinga, 2014; FDC, n.d.). However, the government that came into force in

2017 (Rutte III), choose to accelerate the reduction of the interest that could be deducted. From 2020 onwards, the deduction will fall 3% each year, meaning the maximum of 38% is reached in 2023 (Vereniging Eigen Huis, n.d.).

The final main measure states that from 2013 onwards, the loan-to-value decreases with 1% each year, until it reaches the maximum of 100% in 2018. This means that if one wants to take out a mortgage, an extra amount of money should also be paid, to cover the additional costs, which prior to the reform could partly be paid out of the mortgage. Following the recommendations of the IMF, the DNB has investigated the consequences of a decrease to a value of 90% in the future. They also propose to the government to lower the loan-to-value again, but no official decisions have been made so far (Verbruggen, et al., 2015).

3. Literature Review

In the previous chapter is has been made clear that changes in the MID system change several outcomes on the housing market, and therefore influence the way potential buyers respond to a reform. Several authors have done research on the subject of the change of the MID in 2013. In this chapter, these authors are discussed.

Mastrogiacomo (2013) has analyzed in what way the restrictions of 2013 influenced household’s behavior on the housing market. He used the DNB Household Survey (DHS), containing data between 1993 and 2012. For more specific data on households, the author has used an extension of the DHS in 2010. In this survey, respondents were asked if they would change behavior in response to the possible reform of the MID. Mastragiacomo found that 24% of the respondents were planning on changing their behavior after the reform: 6% of the tenants were planning on postponing the purchase of a house, compared to 3% of the home owners (Mastrogiacomo 2013).

The author mentions that in the years prior to the policy reform, there has been an ongoing debate about a potential policy change, in which even the fully abandoning of the MID has been proposed. This had a great influence on consumer confidence and therefore on the price risk, which is the possible loss of investment as a consequence of fluctuations in housing prices. People felt insecure about if the policy would change mortgage costs, which would leave people with a lower disposable income. In turn, this would cause less people to be able to buy a house, resulting in lower housing prices. For home-owners this is a negative effect, but for renters it might be profitable, depending on how much the housing prices will decrease compared to the rise in mortgage costs. Another risk of a changing MID, is that people that had already taken out a mortgage, were planning to have more wealth later in life, for example with the annuity mortgage, and the policy change might decrease this wealth. Mastrogiacomo (2013) found this uncertainty has increased savings of households with 6%. Even though the government presented the plans for reform, two years after the respondents answered the questions of the DHS (in 2012), this has merely caused more uncertainty, rather than that it decreased uncertainty, which leads to more savings (Mastrogiacomo 2013).

Damen, Vastmans & Buyst (2016) performed research on the effect of the MID on housing prices in 8 countries, amongst others in the Netherlands. They applied the Ability-To-Pay (ATP) model, which incorporates the effects of a decreasing interest rate, changes in the MID, and the characteristics of mortgages. By means of a cointegration test, granger causality test and applying the elasticity of house prices, they found that the MID influences the ATP, and

therefore housing prices. In the Netherlands, they find that housing prices are high compared to other countries, and this could be explained by the MID (Damen, Vastmans & Buyst, 2016). The research also focusses on how a possible abandoning of the MID would influence housing prices. They find a large effect for the abandoning of the MID in the Netherlands, namely that it would decrease housing prices by more than 39%. The authors prescribe this to the fact that the rate at which the mortgage is deductible is very high, and before 2013, many mortgages were in the form of an interest-only mortgage. Their conclusion about the reform of 2013 is that the largest effect will be due to the change to be obligated to take out a linear or annuity mortgage. However, it might be possible that this negative impact on housing prices, will have less drastic consequences since the reform is planned to cover a relatively long time period (Damen, Vastmans & Buyst, 2016).

Teye et al. (2017) have done research on the risks in homeownership in the Netherlands. They have applied a broad literature review in which they discuss different concepts on risk in homeownership. These concepts are combined into a framework, that gives an overview of the risk perceptions. They argue that before 2013, the payment risk, which is the risk of not being able to pay back the mortgages, was quite low compared to other European countries. The authors prescribe this to the social security system of the Netherlands, which is generous and guarantees a continuous income. These benefits have had a positive influence on households after the crisis, causing most of them to be able to continue their mortgage payments. However, the price risk is much higher in the Netherlands. Since the reform of 2013, the costs of taking out a mortgage have increased, mainly for first-time buyers. Together with changing risk perceptions after the crisis, the authors argue that there is a drop in the confidence of consumers when it comes to buying a new house. This in turn influences demand for housing and therefore housing prices (Teye, et al, 2017).

4. Research Design

4.1. Data Selection

For this research, the panel data of the DNB Household Survey (DHS) is used. In 1993 the questions of the DHS were asked for the first time. Every year approximately 2,000 households participate in this panel survey. The questions consist of different subjects: general information, household and work, accommodation and mortgages, health and income, assets and liabilities and psychological concepts (CentER Data, n.d.). For this research, the data about the general information, income, and accommodation and mortgages are used. Since the data about accommodation was not available in 2003, the year 2004 is used as the starting point. The research focusses on the years prior to the reform of 2013. The expectation of a reform already influences people’s behavior. Since the reform came into force in 2013, that is the last year that is considered in this research.

After combining the three datasets, the raw data consists of 49,351 observations in total. However, this research focusses on the household level. People that are living together, will also make their decisions about home ownership together. Therefore, the household head represents each household. After transforming the data to households, 18,517 observations are left. Both homeowners as renters are considered, since both groups are potential buyers of a new house.

The data of these households are collected over several years. This means that it is a panel dataset. The year is specified as time variable, and the household as the panel variable. However, the data is unbalanced, meaning that not every household has answered the questions of the DHS every year. The total number of households in the dataset is 4950, where 14,12% of these households participated every year.

Amongst others the reliability and the validity of the DHS are considered in the article by Teppa & Vis (2012). According to their article, the recruitment process is random, since it is based on a sample of private postal addresses. In their selection, the DHS tries to have a sample that is comparable to the characteristic’s Dutch population, presented by the CBS. In this way the DHS tries to account for validity. To account for reliability, for example, responses of questions that should measure the same variable are compared (Teppa & Vis, 2012).

4.2. Descriptive Statistics

This following section discusses the descriptive statistics. First, Table 1 shows the characteristics of the households. These are presented as percentage of the number of respondents in that year.

The number of households that are left in the analysis varies around 1,850 each year. In the dataset the majority of the respondents is male. This might be explained by the fact that only household heads are considered. More than half of the respondents has already bought their own home, and this percentage is increasing every year. Logically, the number of respondents that rent decreases every year. Sub rented flats/houses are also included in the ‘renter’ category. Another answer respondent could give, was that they lived in free accommodation. However,

the number of respondents that gave this answer was below 20 in these 10 years. Therefore, this category is not presented in this table. Every year, the percentage of household heads that live together with a partner varies around 70%, and the percentage of households with kids varies around 35% each year.

The majority of the respondents did not follow any higher education. Thereafter, most respondents followed vocational colleges, which is called Hbo in the Netherlands. The senior vocational training, or Mbo in the Netherlands, then has the highest percentage. The smallest group of respondents followed education at a university. Education might influence one’s future income, which also might influence being able to buy a house or not. In this dataset, approximately 85% of all respondents answered to earn below €50,000 each year. What is important to keep in mind, is that these incomes are self-reported. In the dataset, it seems like some respondents reported their incomes wrong. For example, a respondent reported to earn €3054.542, but possibly means to earn €3054,542 each year. Since it is not possible to determine which people have reported their true yearly income and which people made mistakes, there are no changes being made. Therefore, some incomes may be higher or lower in real life.

Another important characteristic is the primary occupation of respondents. As can be seen in Table 1, most respondents work on a contractual basis. Thereafter, the dataset consists of a large group of retired respondents. Similar to education and income, the primary occupation might influence the willingness to buy a house. Someone that is retired, for example, might be less willing to buy a (new) house since one’s future is shorter. Also, being a freelancer, self-employed or looking for a job might influence one’s prospect on income in the future, and this in turn might influence the willingness to buy a (new) house. Other answers respondents could give were that they did not have any occupation yet, worked in their own business, were looking for work, performed unpaid work and kept benefits, or worked as volunteers. The answer ‘other occupation’ is transformed into a missing value, since this answer does not say anything about the primary occupation. Only the categories with highest percentages are displayed in the table. The intention of the DHS is to collect data of respondent’s multiple years in a row. As can be seen, the age of respondents is increasing until 2011/2012 and then decreases again to 53 in 2013. The percentage of people that earns below €20,000 each year decreases until 2012, and then increases again in 2013. This might be explained by the fact that most respondents become older each year, and therefore also earn more each year. The number of retired respondents in 2013 is lower than the preliminary years. However, compared to 2004 there are still participating more retired people in 2013.

In the next figures, the number of people that expect a limitation of the MID in the upcoming 10 years is presented in Figure 1. The answer ‘Otherwise’ is reported as a missing value. From 2012 onwards, the respondents were also asked if they expected a limitation of the MID in the upcoming 3 and 5 years. These questions are presented in Figure 2.

In the years prior to the crisis, the number of people that expected a reform of the MID within 10 years, decreased a few percentage points. However, from 2008 onwards this percentage increased from 65.54% to 84.6% in 2012. The government presented the reform in 2012. Since

Figure 1. Source: DHS (2004-2013) Figure 2. Source: DHS (2004-2013) 0 10 20 30 40 50 60 70 80 90 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013

Do you expect a limitation of the MID in the

upcoming 10 years?

Yes No I don't know

0 10 20 30 40 50 60 70 80 90 Yes No Yes No 2012 2013

Do you expect a limitation of the MID in the

upcoming ...?

decreased again until 75.6%. This percentage is still higher than the expectation prior to the crisis. Figure 2 also presents that in 212, more people expect a reform in the upcoming 3 or 5 years. However, in 2013 still 49.69% expects a reform within 3 years and 74.7% expects a reform within 5 years. Overall it might be argued that the number of people expecting a limitation on the MID, in general, is very high. The percentage of people expecting a limitation in 10 years has not been below 62.2%.

The next figure displays the willingness to buy a house. The question could be answered with ‘yes’ and ‘no’. In the dataset, the answers also contained a reason of why the respondent answered that way, for example ‘No, I prefer renting’. This variable is designated as the dependent variable, but this means it has to be transformed to a binary variable. That is why the answers that start with ‘Yes’ and the answers that start with ‘No’ are clustered together. The answers ‘Otherwise’ and ‘I don’t know’ are reported as missing values. See Figure 3 below for the results.

Since the crisis in 2008, the number of people willing to buy a house decreased severely. In 2007, still 59.8% of the respondents answered to wanting to buy a (another) house eventually, while in 2008 only 21.38% of the respondents answered they were (still) willing. This number decreased until 2012 to 14.29%. In 2013, after the MID reform came into force, this percentage increased a little to 16.18%. This means that the crisis had a severe influence on people’s future perception on buying a house.

The DHS also contains a question about if the respondent favors a limitation of the MID or not. From 2004-2011 the DHS asked respondents if they were ‘for a limitation’. However,

0 10 20 30 40 50 60 70 80 90 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013

Do you intend to buy a (another) house

eventually?

Yes No

since 2012 the question changed to ‘Do you favor a limitation’. Although the question changed the answers are interpreted the same. The answers ‘Otherwise’ and ‘I don’t know’ are reported as missing values. Figure 3 below presents the results.

Since 2008, the number of people favoring a limitation increases until it’s highest percentage in 2012. Although less people were willing to buy a house in this period, 73.24% of the respondents answered they favor a limitation in 2012, compared to 54.29% in 2004. The number of people that do not know the answer, also decreases from 2009 onwards. As mentioned in the institutional context, the issue got more attention by politicians in the years prior to the reform. This might explain why more people are able to give their opinion. Since the reform is announced in 2012, the number of people favoring a limitation decreased again until 59.3% in 2013.

4.3. Empirical Model

The dataset consists of panel data, which means that it is possibly to apply the Ordinary Least Squares model (OLS), a Fixed Effects model (FE) or a Random Effects model (RE). The OLS model is used for doing a simple linear regression. In a simple linear regression, the relationship between one or more independent variables between a dependent variable is measured. According to Wooldrige (2010), OLS is most suitable when a different sample is selected for each year, since it exploits variation within and between households. However, the DHS tries

Figure 4. Source: DHS (2004-2013) 0 10 20 30 40 50 60 70 80 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013

Do you favor a limitation of the MID?

the households is most interesting. This means that the FE or RE model are most suitable (Wooldridge, 2010). The FE model assumes that the characteristics of each household are unique, and this in turn also influence the dependent variable. These characteristics are ‘fixed’, like for example one’s gender, meaning it remains the same over time. The differences over time are incorporated in the random error. The error term and the constant are not allowed to be correlated with other individual characteristics, since each household is unique. However, it allows for the individual and time-fixed effects to correlate with the independent variables (Torres-Reyna, 2007; Hsiao, 2007).

If the error term and the constant are correlated, the RE model is more suitable. The RE model assumes that heterogeneity over time cannot be called random error. It assumes that the error term is not allowed to be correlated with the independent variables. This means that the time-effect also plays an important role in explaining the dependent variable. Consequently, the RE model considers both the differences between the households, as the differences in time within these households (Torres-Reyna, 2007). This means that with the RE model, the dataset is partial pooled. A disadvantage is that because of pooling, the results possibly are biased (Clark & Linzer, 2012).

The FE model also has some disadvantages. Since the FE-model includes an intercept for each household specific, it does not allow to estimate coefficients for characteristics that remain the same over time. Also, when the number of sample observations increases, the number of specific unobserved household intercepts increase. When the number of observations is finite, this can cause a classical incidental parameter problem. It means that the coefficients will be inconsistent (Hsiao, 2007). This only is a problem if non-linear methods are applied. However, in this analysis the dependent variable is binary, meaning that a linear probability model is applied.

The basic equation is presented as followed:

Y

= β

β

X

+ …+ β

X

In this equation (1), Yhy is the dependent variable, where h is household, and y is the time

variable of year. The Xk,hy stands for the main independent variable and control variables. In

addition, βk is the coefficient for the independent variables. Lastly, uhy is theerror term

(Torres-Reyna, 2007).

To determine whether the OLS, FE or RE model applies, several tests are executed. First, the F-test is performed to choose between the OLS and FE model. In this case, the F-test shows that FE model is more suitable. Then, by performing the Breusch and Pagan Langrangian

multiplier test, the outcomes suggest the RE model is more suitable compared to the OLS model. Thereafter, the Hausman test is executed to choose between the FE or RE model. The model shows whether the error terms of the households are correlated with the independent variables. The FE model is applied when the intercept of the household αh is fixed or random,

while the RE model is applied when the intercept of the household is uncorrelated with the independent variables (Hsiao, 2007; Torres-Reyna, 2007). The outcomes suggest that the FE model is most suitable. This means that the equation changes:

Y

= β

β

X

+ …+ β

X

+ u

In equation 2, the intercept for each specific household αh is added. This intercept is part of the

error term, since it explains variation across households. However, it remains the same over time. This means that changes in the dependent variable, are to prescribe to the characteristics that do change over time (Torres-Reyna, 2007).

In the FE model it is usual to incorporate time-fixed effects in the equation and the regression. To determine of this is also the case in this regression, this is tested by creating dummies for each year and see whether jointly they are equal to 0. If they are not, this means that each year has a different effect on the outcome (Torres-Reyna, 2007). In this case, the test shows that time-effects are needed in this regression. Therefore, the equation changes as followed:

Y

= β

β

X

+ …+ β

X

+

s

T

+…+

s

T

+ u

In equation 3, time as dummy variable Ty is added. This variable is binary, which means that

the variable takes on the number 0 or 1. If the variable takes on number 1, it means that the respondent answered the survey in that specific year. It is then multiplied with the coefficient of the dummy variable sy. Since it is a dummy variable, there are y-1 time periods. The

interpretation of the model is as followed: for given household h, when Xk,hy varies across time

y by one unit, Yhy increases or decreases by βk units. Thereafter, the applicable time coefficients

sy, the intercept for the specific household αh and the error term uhy are added up. This gives

outcome Yhy for household h (Torres-Reyna, 2007).

It is not possible to measure the effect of the MID directly on housing prices since it is a time effect, which means that it affects everyone at the same time. Also, it is not possible to link the change in MID directly to an increase or decrease in housing prices, since there are many other factors involved. Therefore, this research focusses on the expectations of the

respondents regarding a possible reform of the MID on the intention to buy a house, since these play a decisive role in the fluctuations of housing prices. Both Mastrogiacomo (2013) and Teye et al. (2017) have done research about uncertainty on the housing market, which could directly influence supply and demand, and this in turn influences housing prices. According to Teye et al. the first effect due to expectations is called the payment of default risk. It means that households are concerned with their ability to pay their monthly mortgage. This risk depends on the future of employment, demographic developments, like accidents or divorce, personal mismanagement and the way households manage their finances. Additionally, according to Mastrogiacomo this also includes the effect the crisis had on people’s general uncertainty for the future. The crisis has shown that, although people were able to take out a mortgage, it still was possible to get into financial difficulties. Ever since, people are wondering if they will have enough money to pay their mortgage in the future (Mastrogiacomo, 2013).

The second effect of expectations, mentioned by Teye et al. (2017), is called the property price risk. It means that there is a risk that housing prices will decrease, which might result in a loss of investment capital. Here government regulations play an important role: the more generous the regulations are, the higher housing prices will be. If the government imposes restrictions in the future, housing prices possibly will decrease (Teye et al., 2017). In addition, if people expect these restrictions, they will feel more insecure about taking out a mortgage, since they are afraid they will not get the same tax refund after the reform, as when they took out the mortgage (Mastrogiacomo, 2013). The more people are expecting restrictions and are concerned with their ability to pay of their mortgage, the fewer people eventually will take out a mortgage. This decreases demand for housing, which lowers housing prices. If people do not feel uncertain, demand for housing increases, which in turn increases housing prices (Teye et al, 2017).

The main independent variable X1,hy therefore is the expectation of a limitation within 10

years, which might influence the dependent variable Yhy: the intention to buy a house eventually.

Based on the theory, it is expected that the beta β1 of the independent variable X1,hy is negative:

expecting a limitation of the MID lowers the intention to buy a house. Since the independent variable X1,hy is a categorical variable, it is added as a dummy variable in the regression. The

answer ‘I don’t know’ is designated as the reference category. Thereafter, the first control variable X2,hy is age, since it might be the case that the older, the less willing one is to buy a

house. The second control variable X3,hy is the net disposable income of the respondent. Income

might also influence one’s ability to buy a house, and therefore the willingness. The third control variable X4,hy is the respondents residency, by which is meant whether someone rents

or is already a home owner. If someone already owns a home, one might be less willing to buy a (new) home. Therefore, being a renter might increase one’s willingness to buy a house eventually. Another option is having free accommodation. This variable is also added as dummy variable, and ‘free accommodation’ is designated as reference category. The fourth control variable X5,hy is the highest level of education completed, since this might influence

one’s ability to buy a house and to judge whether this will ever be a possibility. This variable categorical and therefore is also included as a dummy in the regression, where the answer ‘(Continued) special education’ is designated as the reference category. The fifth control variable X6,hy is the number of kids. If this number changes, this might influence one’s ability

to buy a (new) home. The sixth control variable X7,hy is whether the respondent is having a

partner or not. If one has a partner, this might cause more financial stability and an increasing willingness to buy a house. The last control variable X8,hy is the main occupation of the

respondent. For example, if someone works on contractual basis, is retired, or has his/her own business, might influences one’s future financial stability, and therefore might influence the willingness to buy a house. Since this variable also is categorical, it is included as a dummy in the regression. The answer ‘Employed on contractual basis’ is designated as the reference category. Lastly, year is added as the time-fixed variable Ty. Since this variable is also added

as a dummy variable to the regression, the year 2004 is designated as the reference category. The year 2005 therefore is T2, the year 2006 is T3 etcetera.

5. Empirical Analysis

Table 2 below presents the empirical results of the first regression. The first column shows the output without control variables. The second column shows the output with control variables added. Next to the coefficient, the standard error is reported in parentheses. In this table the coefficients of the control variables are not presented. For the extensive results, see the Appendix.

It was expected that after adding the control variables, the coefficient of expecting a limitation would be negative. However, this result show that expecting a limitation increases the intention to buy a house with 0.050, compared to not knowing if one should expect a limitation. This coefficient shows a stronger effect compared to the ‘No’ answer. The coefficient of ‘Yes’ also is highly significant, since the p-value is below 0.01. The coefficient of not expecting a limitation shows that this would result in a 0.035 higher intention to buy a house compared to not knowing. Not expecting a limitation is significant in the first column, however, after adding the control variables it is not anymore. This means that although people expect a limitation, demand does not decrease. In fact, it could be argued that demand increases, since the intention to buy a house increases. Therefore, housing prices might increase.

In Appendix Table 1, the extended output is presented with control variables included. Age has a significant influence on the intention to buy a house. The older someone is, the lower the intention to buy a house. Additionally, only having attended kindergarten lowers the intention to buy a house. All year dummies, except for 2005 and 2013, also have a significant

effect. The strongest effect could be found for the year 2008, which lowered the intention to buy a house severely.

In Table 3 below, the results are compared to an OLS regression analysis. In this equation, gender is also added as control variable. In the FE model this control variable cannot be added, since it is constant over time. However, in the OLS model it is important to also take this variable into account.

This table shows highly significant results. With and without the control variables, the output shows a high significant effect of the expectation of a limitation on the intention to buy a house. Appendix Table 2 shows that also almost all control variables contribute significantly this this effect. It is remarkable that the coefficients of the expectation of a limitation, both are positive. Expecting a limitation (or not) causes the intention to buy a house to increase, compared to not knowing. The effect of expecting a limitation is stronger compared to not expecting a limitation. According to this model, there is a large effect of the MID on the demand for housing. The consequence is that housing prices might increase. However, this model is more likely to be biased, since it does not consider the individual characteristics of the households over time.

In the following tables, this research focusses on comparison between different subgroups. The first comparison is made between renters and home-owners. See Table 4 for the results.

Table 4 shows that when the dataset is separated into renters and home-owners, the results are not significant for both groups. In column 1, again the regression is presented without control variables. Here only the coefficients of the home owners are significant. However, after adding the control variables this is not the case anymore. For renters, almost none of the control variables contribute significantly to the outcome, but for home-owners education, age and the year dummies do have a significant effect. However, since the independent variable is not significant, it still means that no distinction can be made between renters and home-owners.

The following table focusses on differences between age-groups. The first category consists of respondents of 40 and younger, which accounts for 25.74% of the dataset. The second group consists of respondents between the age of 41 and 60. This group accounts for 41.61% of the dataset. The last group consists of respondents above the age of 61. This group accounts for 30.44% of the dataset. See Table 5 below for the results.

Table 5 shows significant results only for the lowest age-group: 18-40 years old. In this case there is a stronger effect for not expecting a limitation, compared to do expecting a limitation. Again, both coefficients are positive, meaning that (not) expecting a limitation, will cause the intention to buy a house to increase, compared to not knowing if there will be a limitation. For the age-group of 18-40, only the year dummies contributed significantly to this strong effect. For the age-group 41-60, the year dummies and being a renter contributed significantly. The oldest age-group has more significant control variables: education, being retired or looking for work after having lost job and all year dummies. However, the effect of the independent variable of two oldest age-groups still is not significant after adding the control variables. To conclude: only for the age-group of 18-40 years old there is a significant effect that both the answer ‘Yes’ and ‘No’. It increases demand for housing and therefore will increase housing prices. See Appendix Table 4 for the extended results.

In the following table, the results for Table 5 are exploited. Since it mainly affects the youngest age-group, next table will divide this age-group into two smaller groups. The first group consists of respondents between the age of 18-33, which accounts for 11,84% of the total dataset. The second group therefore consists of respondents between the age of 34-40, which accounts for 13,9% of the dataset. See Table 6 for the results.

The output presents significant results for the age-group 34-40 and no significant results for the age-group 18-33. In this case, the expectation of a limitation again does not mean that the intention to buy a house decreases. To the contrary, both the answers ‘Yes’ and ‘No’ cause the age-group of 34-40 to have a greater intention to buy, compared to not knowing the answer to the question. There is a stronger effect for not expecting a limitation, compared to do expecting a limitation. This mean that for this group causes demand to increase, and therefore housing

prices might increase. Again, for both groups only the year dummies contribute significantly to this effect. See Appendix Table 5 for these results.

In the following table the effect of the expectation of a limitation is presented for different education groups. Again, this regression follows the categories of the descriptive statistics. See Table 7 and 8 below for the results.

Table 10 shows that only at the vocational colleges level, the expectation of a limitation is significant. Again, when the respondent expects a limitation, the intention to buy a house increases with 0.083, compared to not knowing the answer. For all groups ‘age’ is a significant control variable, and again almost every year dummy is significant. However, since the outcomes are not significant except for the vocational colleges group, it can only be concluded that if one falls in this group, the expectation of a limitation causes the intention to buy a house to increase. Therefore, for this group the demand increases, which might increase housing price.

6. Discussion and Conclusion

This analysis has focused on the effect the expectation of a policy reform of the MID has on the willingness to buy a house. In the years 2004-2013 there has been a prevailing uncertainty on the housing market. The crisis of 2008 has had a great influence on the way housing policy is shaped, as it changed people’s perception on financial risk. Since expectations regarding the housing market have a great influence on demand for housing, this research has focused on the willingness to buy a house. The expectation of a policy reform, causes people to think about their ability to pay their mortgage in the future, and about if they expect to sell their house at the same value as when they bought it. If people become aware of the policy reform, it was expected they might become less willing to buy a house.

The data consisted of questions conducted by the DNB Household Survey (DHS) and focused on the household level. After applying the FE model, the null hypothesis, in which the expectation has no effect on the willingness to buy a house, in some cases could be rejected, but not in the way it was expected. Following the theory, it was expected that if the respondent expects a limitation, the intention to buy a house would decrease. However, the results show the opposite: expecting a limitation, increases the intention to buy a house. In almost all cases, the effect of the ‘Yes’ answer was stronger and/or more significant, compared to the ‘No’ answer of expecting a limitation.

There is no distinction between home-owners and renters. The rationale is that there must be a large group of home-owners that does not intend to buy a new home, and there also must be a large group of renters’ that are never going to be willing to buy a house. However, in both groups there must be respondents that want to buy their first or second home. Therefore, this explains why the model is more significant when both groups are combined. The results also showed high significant results first for the age-group of 18-40, and after exploiting this finding for 34-40 years old. This might be explained by the fact that younger groups are less able to buy a house, and older groups already have bought a house or already choose to prefer renting. Additionally, for education groups, the only significant effect was found for the respondents that graduated from vocational college. Also, only the ‘Yes’ answer showed significant results. To conclude, the expectation of a limitation does not affect the intention to buy a house in a negative manner. To the contrary, the expectation of a limitation (or not) increases the intention to buy a house. This might have several explanations.

Although the limitation might make it harder for respondents to buy a house, this does not mean they immediately change to not wanting to buy a house ever again. Respondents might also want to buy a house within 10 years, and therefore care less about the limitation of the MID on the short-term. This could also result in people that are wishing to buy a house, answering ‘Yes’. They might intend to hopefully do so, for example when their income increases in the future or when they find a partner. This influences the validity of the analysis.

Another possibility might be that most people that intend to buy a house, also are proponents of the limitation, and therefore their intention will not be affected. Figure 4 shows that from 2008 onwards, a large percentage of the respondents became a proponent of the limitation. It might be that they simply do not care if the MID changes, but another possibility is that they feel more secure buying a house when the MID is limited. Eventually this limitation was not only implemented to save money, but also to make the housing market more sustainable and less risky. The last possibility is that there exists reverse causality: the people that were already willing to buy a house, are probably also informed about a possible limitation of the MID. Therefore, it might be that in this regression the effect for expecting a limitation is so strong due this reverse causality.

The strength of this study is that the hypothesis has been tested with the data that reflects the research question well. Also, it focusses on a large sample, that is intended to be generalizable to the overall population of the Netherlands. However, a weakness of the research may be that the answers are provided by people themselves, and they therefore do not reflect people’s real behavior. There is a possibility that respondents behave exactly in opposite of what their original statements in the dataset are.

For further research, a recommendation it to look into data that show if the intention to buy a house changes from the short-term to the long-term. In this way, it might be clearer if the MID alters behavior directly. Another recommendation is to look into concrete data, that prove if indeed less or more households after all bought a house after the reform of 2013. Also, it might be interesting to ask respondents 5-10 years after the policy reform, if it did influence their behavior or if it did not.

7. References

ABN Amro (n.d.). Interest-Only Mortgage. ABN Amro Bank. Retrieved at April 7, 2020, from

https://www.abnamro.nl/en/personal/mortgages/applying-for-mortgage/mortgage-types/interest-only-mortgage/index.html CentERdata (n.d.). DHS data. Retrieved at May 12, 2020, from

https://www.centerdata.nl/nl/databank/dhs-data

Central Agency for Statistics (2010). Helft belastingvoordeel hypotheekrenteaftrek naar rijkste huishoudens. Central Agency for Statistics (CBS). Retrieved from

https://www.cbs.nl/nl-nl/nieuws/2010/19/helft-belastingvoordeel-hypotheekrenteaftrek-naar-rijkste-huishoudens

Clark, T., S. & Linzer, D., A. (2012). Should I Use Fixed of Random Effects? Political Science Research and Methods 3:2, 399-408.

Damen, S., Vastmans, F. & Buyst, E. (2016). The Effect of Mortgage Interest Deduction and Mortgage Characterisitcs on House Prices. Journal of Housing Economic 34, 15-29 Elsinga, M., Jong-Tennekes, de, M. & Heijden, van der, H. (2011). Crisis en Woningmarkt.

Research Institute OTB, Technical University Delft. Retrieved from

https://www.rijksoverheid.nl/binaries/rijksoverheid/documenten/rapporten/2011/09/12 /crisis-en-woningmarkt-eindrapport/eindrapport-crisis-en-woningmarkt.pdf

FDC (n.d.). Beperking Hypotheekrenteaftrek. FDC Financieel Planners & Adviseurs. Retrieved at April 6, 2020, from https://www.fdc.nl/beperking-hypotheekrenteaftrek/ Hsiao, C. (2007). Panel Data Analysis – Advantages and Challenges. Test 16, 1-22.

International Monetary Fund (2011). Kingdom of the Netherlands – The Netherlands 2011 Article IV Consultation: Preliminary Conclusions. International Monetary Fund (IMF) Retrieved from https://www.imf.org/en/News/Articles/2015/09/28/04/52/mcs032811

Kromhout, B. & Oving, R. (2008, July). Geschiedenis van de Hypotheekrenteaftrek; Niet meer dan een belastingtechnische maatregel. Historisch Nieuwsblad. Retrieved from

https://www.historischnieuwsblad.nl/nl/artikel/10741/geschiedenis-van-de-hypotheekrenteaftrek.html

Lejour, A. (2016). Een politiek-economische analyse van de groei en beperking van de hypotheekrenteaftrek. CPB Netherlands Bureau for Economic Policy Analysis (CPB) Retrieved from

https://www.cpb.nl/sites/default/files/omnidownload/CPB-Achtergronddocument-

30juni2016-Politiek-economische-analyse-van-groei-en-beperking-hypotheekrenteaftrek.pdf

Nationale Hypotheek Garantie (2012). Jaarverslag 2011. Retrieved from

https://www.nhg.nl/Portals/0/Documenten/Publicaties/Jaarverslag_2011_en_Liquiditei tsprognose_2012-2017.pdf?ver=2015-10-19-091550-957

Mastrogiacomo, M. (2013). Reform of the Mortgage Interest Tax Relief System, Policy Uncertainty and Precautionary Savings in the Netherlands. DNB Working Paper 380, 1-28.

Notten, F. (2011). Hypotheekschuld in Nederland. Central Agency for Statistics (CBS). Retrieved from

https://www.cbs.nl//media/imported/documents/2012/35/hypotheekschuld%20in%20n ederland.pdf

Redactie Rente.nl (2019, May 7). Wat is hypotheekrenteaftrek. Rente.nl Kennisbank. Retrieved from https://informatie.rente.nl/hypotheken/wat-is-hypotheekrenteaftrek Scanlon, K. & Elsinga, M. (2014). Policy Changes Affecting Housing and Mortgage Markets:

How Government in the UK and the Netherlands Responded to the GFC. Journal of Housing and the Built Environment 29:335, 335-360.

Teppa, F. & Vis, C. (2012). The CentERpanel and the DNB Household Survey: Methodological Aspects. DNB Occasional Studies, vol.10, no.4, 1-53.

Teye, A.L., de Haan J., Elsinga M., Bondinuba F.K., & Gbadegesin, T (2017). Risk in

Homeownership: A Perspective on The Netherlands. International Journal of Housing Market and Analysis, 10(4), 472-488.

Torres-Reyna, O. (2007). Panel Data Analysis Fixed and Random Effects Using Stata. Princeton University. Retrieved from

https://www.princeton.edu/~otorres/Panel101.pdf

Verbruggen, J., Molen, van der, R., Jonk, S., Kakes, J. & Heeringa, W. (2015). Effecten van een verdere verlaging van de LTV-limiet. De Nederlandsche Bank. Retrieved from

https://www.dnb.nl/binaries/Effecten%20van%20een%20verdere%20verlaging%20va n%20de%20LTV-limiet_tcm46-322569.pdf

Vereniging Eigen Huis (n.d.). Hypotheekrenteaftrek. Vereniging Eigen Huis. Retrieved at April 7, 2020, from https://www.eigenhuis.nl/hypotheken/verhuizen-naar-een-volgende-woning/hypotheekrenteaftrek#/

Wooldride, J., M. (2010). Econometric Analysis of Cross Section and Panel Data. Cambridge Massachusetts & Londen, England: The MIT Press.

8. Appendix