Testing for drivers influencing the house prices in
Amsterdam over the past fifteen years
BSc Economics and Finance Thesis
Dorien van Ommeren
University of Amsterdam
26-06-2018
Supervisor: A. Akdeniz
Abstract
Over the past years, the house prices in Amsterdam are rapidly increasing. For a better understanding of why these prices are growing, a multiple regression is run based on the repayment model and long-term model of Boelhouwer et al. (2017, 2018) that is proposed for the Dutch market. In addition, more independent variables are included that were discussed in the report from the ING Economics Department (2017). The best-fitted regression is tested for an F-test, t-test and tested against other regressions. Population, income, economic growth, and inflation seem to have a positive impact on the Amsterdam house prices as mortgage interest rates, wealth, housing stock, unemployment, and rent seem to have a negative effect on the prices. Consumer trust and return on the stock market did not seem to influence the house prices at all.
Statement of Originality
This document is written by Student Dorien van Ommeren who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are 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.
Table of Content
1. INTRODUCTION ... 4
2. LITERATURE REVIEW ... 5
2.1 The Dutch housing market ... 5
2.2 Drivers of the rising house prices ... 6
2.3 Fundamental models ... 10 3. HYPOTHESES ... 11 4. RESEARCH DESIGN ... 12 4.1 Research method ... 12 4.2 Data ... 15 5. RESULTS ... 17
5.1 Testing the hypotheses ... 17
5.2 Analyses ... 22
6. CONCLUSION ... 24
7. DISCUSSION ... 24
1. Introduction
Amsterdam has not only the fastest growing housing prices of the Netherlands but also has the most rapidly growing house prices across whole Europe (Figure 1). Whereas other big cities, like Dublin and Paris, grow at a rate of respectively 12.3% and 8.3%, the prices of housing in Amsterdam are growing at a rate of respectively 21% (Het Parool, 2018). Possible factors stated are the growing economy, the decreasing interest rates and the shortage in supply of houses. The Netherlands has one of the fastest growing economies of Europe with a GDP growing at a 3.3% rate in the second quarter of 2017. The decreasing interest rates impact the house prices in two ways. First, they result in more investment in housing to get a higher return. Second, because of the low-interest rates, it is possible to get a higher mortgage. To cope with the increasing demand for houses in Amsterdam, the supply of 70,000 houses is needed but only 55,000 per year are being built (Sullivan, 2017).
Figure 1: House prices Amsterdam compared to the Netherlands (source: CBS)
So, the increasing prices can possibly be explained due to the changing economic and fiscal environment in the Netherlands in combination with a highly regulated market. Therefore the question arises, which factors have affected the increasing house prices in Amsterdam over the past fifteen years? This thesis will examine which drivers are causing this significant increase of house prices in Amsterdam. The factors discussed are income and affordability, borrowing capacity, buying versus renting, the influence of investors, influence of parents, influence of Airbnb, the supply of houses, demographic and sentiment. These factors are based on the report of the ING Economics Department (2017). Boelhouwer et al. (2017, 2018) proposed already a long-term relationship for house prices based on a multiple
regression analysis and a repayment model to find correlations. These formulas are the fundament of the proposed regression in this thesis.
The thesis is structured as follows. The following part, section 2, will discuss the theoretical background of the Dutch housing market and relevant drivers that possibly influence the increase. In section 3 the hypotheses will be formulated. Section 4 describes the research design. This section is divided into two parts: the research method and the discussion of the chosen data. Section 5 presents the obtained results and analysis of the results. Section 6 formulates the conclusion of this thesis. Section 7 is the discussion and will discuss the drawbacks of the thesis and proposes further research recommendations.
2. Literature review
2.1 The Dutch housing market
From 1990 until 2008, house prices in the Netherlands have been growing. In 2008 the financial crisis appeared and the housing market collapsed. As a result, the prices declined extremely. The difference between the Netherlands and other European countries is that the period of decline was longer in the Netherlands. It took until 2013 for the market to recover. Since then, the house prices are increasing and especially Amsterdam is suffering from significantly increasing house prices over the past years (ING Economics Department, 2017). To investigate the factors that caused this strong and persistent increase, relevant literature needs to be examined. First of all, it is important to know the distinction between the housing market of the Netherlands and other markets. Boelhouwer et al. (2018) stated that the Dutch house market reacts differently to economic shocks than other markets because the market is highly regulated. The Netherlands has a significantly inelastic supply sector, a social housing rental sector and a highly subsidized housing market. Consequently, regular asset-pricing models cannot be used to investigate the Dutch market. The asset-pricing models are made for markets like the US, which are characterized by almost no regulation. The underlying assumption of limited government intervention does not apply to the Netherlands.
The Dutch market is strongly regulated and the government intervenes in almost every part of the market. Citizens with low incomes are subsidized through tax policy, namely through deduction of the monthly mortgage payments. When taking a mortgage, the interest can be deducted from the taxable income for a maximum of 30 years. Before the crisis, most
of the people had an interest-only loan and so the deduction of the taxable income was fully used. The principal needed to be repaid when the loan matured (Boelhouwer et al., 2018). However, after the crisis, the government changed this policy. Nowadays when you take out a new mortgage, you are obligated to repay your mortgage monthly. This repayment of the principal can be done either linear or annuity-based. As a consequence, the full advantage of the deduction on taxable income is not possible anymore (Van Ooijen & Van Rooij, 2016).
Another difference between the Netherlands and other countries is the tax legislation for companies. The Netherlands has favorable tax legislation for companies that are located in the Netherlands. It is one of the biggest channels of tax avoidance. Shaheen and Salmon (2017) investigated the tax havens of the world and concluded that five developed countries take up 47% of funneling multinationals offshore investments. The five biggest tax havens are the Netherlands (23%), the UK (14%), Switzerland (6%), Singapore (2%) and Ireland (1%). This favorable tax climate attracts big companies from outside of the Netherlands to locate them here, which causes the attraction of more people from abroad into the Netherlands. The people who move to the Netherlands constitute a significant part of the housing market as they increase the demand for housing.
2.2 Drivers of the rising house prices
The ING Economics Department (2017) published an article of possible relevant drivers influencing the increasing house prices in Amsterdam. The factors included by the ING are income and affordability, borrowing capacity, buying versus renting, the influence of investors, the influence of parents, influence of Airbnb, the supply of houses, demographic and sentiment.
First of all, the ING Economics Department (2017) discusses income and affordability as a driver for the increasing house prices in Amsterdam. The article stated that the affordability of properties in Amsterdam is decreasing. In the Netherlands, a buyer spends between 15% up to 30% of their household income on buying a house. However, in Amsterdam, this percentage is around 35%. The ‘price-to-income' ratio is used as an indicator of affordability, as it is easy to use. However, the ‘price-to-income' ratio is later rejected because it does not include the monthly expenses of a household, but only their income. The income is not the only driver that influences the monthly expenses; the interest rate is an important driver as well. Ayuso & Restoy (2006) also rejected the price-to-income ratio for
the Dutch market, as it does not consider the overall fiscal effect given the favorable tax deduction. The ING Economics Department (2017) discussed another indicator of affordability, the mortgage costs as a percentage of income. Since 2008, the affordability in the Netherlands is slightly improved because of the decreasing mortgage interest rates. Nevertheless, the total monthly mortgage cost decreased less than the interest rates, because households are obligated to repay their mortgage. Still, the increasing affordability has contributed to the increasing demand for owner-occupied properties and thus to the increasing prices. Unfortunately, the slightly improved affordability in the Netherlands does not apply to Amsterdam. The increasing house prices outweigh the decreased interest rates and as a result, the affordability of Amsterdam houses is decreasing. Especially for first house buyers defined as starters, the significant price increase and the new policy regarding mortgages make it unaffordable to buy a house. For non-starters, the situation is bearable, because they have more gain from the low interest. However, the affordability is still decreasing for them.
Another driver for the increasing house prices is the loan capacity. Before the crisis, high lending explained the rising house prices. After the crisis, the government decreased the loan capacity to lower the risk of failing to repay the mortgage (Braspenning, 2017). Even though the credit terms are stricter, the real maximum amount that buyers can borrow is not equally decreased. Since this year, this real amount is increasing again. Two main factors have caused this increasing amount. First of all, the mortgage interest rate is decreased since 2008. The monthly interest costs are lower, so buyers can borrow more money based on their monthly income. Second of all, the average income is increasing and so does the amount that can be borrowed (ING Economics Department, 2017). Altogether, the decreased loan capacity seems to outweigh the real increase in loan capacity, but it is difficult to examine the overall effect.
Rent is also seen as an important driver for fluctuations in house prices. Gallin (2008) found that over a 4-year horizon rent and house prices correct each other. This means that when house prices are relatively high compared to rents (so the rent-to-price ratio is low), the growth of real rents is higher than normal and the growth rate of real house-prices will diminish compared to the period before. The response of the price effect is faster than the rent effect. An important indicator to compare renting with respect to buying is the ‘price-to-rent’ ratio. Ayuso and Restoy (2006) rejected this indicator, as it is based on a competitive market instead of a regulated market. The ING Economics Department (2017) makes use of both the ‘price-to-rent’ and the ‘monthly expenses-to-rent’ ratio. As the ‘price-to-rent’ ratio does not take into consideration the changing interest rates or changing policies, they prefer the
‘monthly expenses-to-rent’ ratio. When looking at the ‘monthly expenses-to-rent’ ratio, ING concluded that since 2013 buying is more attractive than renting. When including rent in an extensive model, the ratios do not need to be used because income and inflation will be included as separate variables in the model.
Moreover, investors are also an important driver of the increasing house prices. According to the International Monetary Fund (2018), the low-interest environment is causing the house prices to increase. Real-estate investments are more attractive because they have a higher return on investment. Therefore, the real-estate investments have increased, especially in urbanized areas like Amsterdam. The ING Economics Department (2017) also stated that the growing demand of investors is affecting the housing market. As a result, the housing market is getting more sensitive to shocks because the nature of demand is becoming more volatile. Investors are driven by return and if the return is higher elsewhere, the demand for houses will decline. Altogether, the housing market depends partly on the stock market. Kakes and Van den End (2004) found that house prices respond to shocks in the stock market in two ways. First of all, the stock market is an indicator of the investors' forward-looking behavior. Second of all, due to investments in the stock market, household’s income is related to the performance of the market. They conclude that equity is correlated with the house prices, but that it does not account for the overall movement in house prices. However, it is hard to compare stocks and houses with equal expected risk and return, because this is often unknown and hard to predict, especially for houses.
Another relevant driver discussed in multiple articles is Airbnb. Airbnb is a platform where you can rent your apartment. When you rent your apartment online through Airbnb, the income generated is much higher than if you rent it regularly for a long period. So a house has nowadays more functions than living: it is a source of income.This evidently will push up the price of houses (ING Economics Department, 2017). Horn and Merante (2017) also found that Airbnb drives up rents in Boston. And if rent and house prices are correlated, as stated above, this would have an impact on the house prices as well. However, since the new regulation concerning Airbnb in Amsterdam proposed by Alderman Ollongren and put into operation on November 22, 2016, it is harder to rent your apartment through Airbnb. Some examples of the rules are that there is a maximum of 60 days per year to rent your apartment; every time you rent your apartment, you have to report it; you need to be the resident of the apartment and a maximum of four people are allowed to rent your apartment. These rules seem to help to stabilize the prices; at the beginning of 2017, the number of apartments offered in Amsterdam through Airbnb decreased (Kraniotis, 2017). The ING Economic
Department (2017) also stated that the effect of Airbnb is decreasing and the overall effect is fading. Because Airbnb is a relatively new factor that already is restricted by the government, the effect will not be included in the regression.
Furthermore, the demographics of Amsterdam can also explain a lot about the increasing house prices. Mulder (2006) discusses that population and housing has a two-sided relation. When the population increases, the demand for housing rises. But, on the other hand, when the supply for housing changes, this changes the opportunities for population increase or decrease through immigration. So housing and population are inseparable. The ING Economic Department (2017) mentioned that the strong population growth in Amsterdam is mainly caused by immigration. The immigrants are mostly high-educated people with good financial situations. Nevertheless, it is not sure that the increasing migration keeps rising. Migration is volatile and depends on the economic situation and the legislation.
Moreover, the supply of houses is also an important driver for the price fluctuation in Amsterdam. The Netherlands already has a severe scarcity of land. For that reason, there are a lot of rules and legislation for building and renovating a house. As a result, the construction process is being delayed and makes it difficult to expand the housing stock. This sector was even described as fixed in size. The highly inelastic supply sector causes an extra positive pressure on the prices of houses (Boelhouwer et al., 2018). The ING Economic Department (2017) also stated that there is an upward pressure caused by the shortage of houses. Especially in the coming years, construction cannot keep up with the expected growth in the number of citizens and will result in a significant upward pressure on prices.
The last driver discussed by the ING Economic Department (2017) is the sentiment of buyers. Expectations and sentiment of buyers determine whether a buyer will invest. Almost 70% of the Dutch does not expect a decrease in prices for the next ten years. Even one out of seven thinks the prices will never decrease again. However, the ‘present bias' plays an important role here. The ‘present bias’ means that current developments influence your expectations about the prospect. These high expectations do not indicate a weakening of the price growth; however, they create an increased risk of a price correction (ING Economics Department, 2017). Altogether, the impact of consumer trust on the house prices is unclear, because of the ‘present bias' and the volatile nature of consumer trust.
2.3 Fundamental models
The multiple regression analysis (MRA) is the fundament for models including multiple independent variables. The multiple regression analysis makes use of data analysis. The regression can be linear or nonlinear. Furthermore, variables can be tested in relation with only the dependent variable or with other independent variables included (Candas, Bagdatli Kalkan, & Yomralioglu, 2015). The equation has the following form:
𝑦 = 𝛽!+ 𝛽!𝑥!+. . . +𝛽!!!𝑥!!!+ 𝜀 (1)
Where p represents the total number of variables in the model, y represents the dependent variable, the x’s represent the independent variables and the expectation of the error-term is equal to zero, assumed to be normally distributed and independently and identically distributed ε~ i. i. d. N(0, σ!) .
Boelhouwer et al. (2017) first examined which model would fit the regulated market of the Netherlands. The multiple regression analysis was used as a fundament for the repayment model. This model made the best fit with the regulated Dutch market. The formula for this model is:
log 𝑃! = 𝛼!log 𝜑! + 𝛼!log 𝑌! + 𝛼!log 𝐻𝑆! + 𝛼!log (𝑖!!) (2)
Where 𝑃! represents the house price, 𝜑! represents the maximum borrowing limit, 𝑌!
represents income from households, 𝑖!! represents the adjusted mortgage interest rate and the
𝛼!’s represent the parameters. Boelhouwer et al. (2018) also proposed a long-term relationship for house prices in the Netherlands based on the multiple regression analysis:
𝑙𝑛𝑃!= 𝑏!+ 𝑏!𝑙𝑛𝑌!+ 𝑏!𝑙𝑛𝑖!+ 𝑏!𝜋! (3)
Where bi’s are the parameters and 𝜋! represents the inflation rate. Equation (3) indicates that
house prices are determined by income, interest rates, and inflation over time. This equation is an extension of equation (2).
3. Hypotheses
I will be estimating the effect of different factors on the increasing house prices in Amsterdam. The independent variables being tested are population, wealth/savings, mortgage interest rate, income, housing stock, inflation, economic growth, unemployment, return on the stock market, consumer trust and rent. By reviewing section 2, the literature review, I expect that the tested variables will be correlated with the dependent variable, the increasing house prices in Amsterdam. Therefore, I expect that the overall regression model will explain the increasing house prices in Amsterdam. The general hypotheses that will be tested are:
1. Testing the regression models (a two-sided alternative hypothesis): 𝐻!: 𝛽! = 𝛽! =. . . = 𝛽!!! = 0
𝐻!: 𝑎𝑡 𝑙𝑒𝑎𝑠𝑡 𝑜𝑛𝑒 𝛽! ≠ 0
The null hypothesis says that the independent variables together cannot explain the dependent variable. The alternative hypothesis says that at least one of the independent variables has explanatory power on the dependent variable. This hypothesis will be tested against a significance level of 𝛼 = 0.05. I expect that the overall regression will explain the increase in house prices so that the null hypothesis will be rejected.
2. Testing the unrestricted model against the restricted model:
𝐻!: 𝐴𝑙𝑙 𝑡ℎ𝑒 𝑒𝑥𝑡𝑟𝑎 𝑝𝑎𝑟𝑎𝑚𝑒𝑡𝑒𝑟𝑠 𝑖𝑛𝑐𝑙𝑢𝑑𝑒𝑑 𝑖𝑛 𝑡ℎ𝑒 𝑢𝑛𝑟𝑒𝑠𝑡𝑟𝑖𝑐𝑡𝑒𝑑 𝑚𝑜𝑑𝑒𝑙 𝑎𝑟𝑒 𝑒𝑞𝑢𝑎𝑙 𝑡𝑜 𝑧𝑒𝑟𝑜 𝐻!: 𝐴𝑙𝑙 𝑡ℎ𝑒 𝑝𝑎𝑟𝑎𝑚𝑒𝑡𝑒𝑟𝑠 𝑒𝑥𝑡𝑟𝑎 𝑖𝑛𝑐𝑙𝑢𝑑𝑒𝑑 𝑖𝑛 𝑡ℎ𝑒 𝑢𝑛𝑟𝑒𝑠𝑡𝑟𝑖𝑐𝑡𝑒𝑑 𝑚𝑜𝑑𝑒𝑙 𝑎𝑟𝑒 𝑛𝑜𝑡 𝑒𝑞𝑢𝑎𝑙 𝑡𝑜 𝑧𝑒𝑟𝑜 The restricted and unrestricted model become clear in the upcoming sections. The null
hypothesis is rejected if the obtained p-value from the F-statistics is smaller than the significance level of 0.05. I expect that the null hypothesis will be rejected so that the parameters included in the unrestricted model are not equal to zero.
3. Testing the contribution of a single independent variable: 𝐻!: 𝛽! = 0
𝐻!: 𝛽! ≠ 0
The null hypothesis says that the independent variable i cannot explain the dependent variable. The alternative hypothesis says that the independent variable has explanatory power on the dependent variable. The hypothesis will be tested against a significance level of 𝛼 = 0.05. I expect that every single independent variable that is tested will be correlated with the dependent variable so that the null hypothesis is rejected for every independent variable. This implies that the variables population, wealth/savings, mortgage interest rate, income, housing stock, inflation, economic growth, unemployment, return on the stock market, consumer trust, and rent have explanatory power.
4. Research design
4.1 Research method
To find the drivers that influence the housing market in Amsterdam, a different approach is used than that of the ING Economics Department. Instead of testing each factor separately against the rising house prices, I will take an overall regression including all relevant factors. The benefit of taking an extensive regression model instead of testing variables separately is that no information from the dataset gets lost. When you test variables separately, you might overestimate some of the coefficients because they measure the combined effect of included and excluded variables if these are correlated. The goal is to set up a regression where the increasing house prices will be the dependent variable, and the independent variables will be the factors that might be correlated with the increasing prices. The drivers that I will be investigating are income, savings, population, mortgage interest rates, housing stock, inflation, economic growth, rent unemployment, return on the stock market and consumer trust. Lastly, an error term needs to be included in the overall equation.
The models proposed by Boelhouwer et al. (2017,2018) in section 1.3 (equations 2 and 3) are used as a foundation for a new regression that will include more variables:
ln 𝑃!= 𝑏!+ 𝑏!ln 𝑃𝑜𝑝!+𝑏!𝑙𝑛𝑊!+ 𝑏!𝑙𝑛𝑖!+ 𝑏!𝑙𝑛𝑌!+ 𝑏!𝑙𝑛𝐻𝑆!+ 𝑏!𝜋!+ 𝑏!𝑔!+ 𝑏!𝑈!+ 𝑏!𝑆!+ 𝑏!"𝐶!+ 𝑏!!𝑟!+ 𝜀 (4)
Where 𝑃𝑜𝑝! is the population of Amsterdam, 𝑊! is the wealth/savings of households, 𝐻𝑆! is the housing stock, 𝑔! is the economic growth of Amsterdam, 𝑈! is the unemployment rate of Amsterdam, 𝑆! is the average return in the stock market, 𝐶! is consumer trust and 𝑟! is the
relative rent increase. Because it is a time series regression, all variables are measured at multiple points over time. The overall goal is to find the best-fitted regression, so the regression that explains the rising house prices best. This overall model is going to be tested against two different models. First of all, the second model a restricted version of the equation 4 however without the variables consumer trust and stock market return because their effect on the dependent variable is ambiguous. Second of all, the third model is again a restricted model of the last two models proposed to check whether the long-term relationship proposed by Boelhouwer et all. (2018) explains the dependent variable better (see Equation 3).
First of all, the proposed models need to be tested for their ability to explain the dependent variable. Hence, the multiple regression models were tested against a significance level of 5% (𝛼 = 0.05). An F-test was used to test the joint significance of all the independent variables included in the model. The test statistic is:
𝐹 =𝑆𝑆𝑅/(𝑝 − 1) 𝑆𝑆𝐸/(𝑛 − 𝑝)=
𝑀𝑆𝑅
𝑀𝑆𝐸 ~𝐹(𝑘, 𝑛 − 𝑘 − 1)
When the significance level is greater than the p-value of the test statistic than the null-hypothesis can be rejected. This means that the betas are not equal to zero with a certainty of 95% and the null hypothesis is rejected in favor of the alternative. Moreover, when testing the overall regression it is important to also look at the R-squared (R2) and adjusted R-squared (adjusted R2) for the goodness-of-fit. The R-squared is the multiple coefficient of determination and lies between 0,1 where one means a perfect fit and zero means no fit at all. The downside of the R2 is that it always increases when you add a variable. The adjusted R2 penalizes you for adding a value and takes the total amount of variables into account. So, when you have more independent variables, the adjusted R-squared weights errors more heavily. Lastly, to check whether the error-term is unpredicted, random and in expectation equal to zero I took a residual plot.
Second of all, we need to test whether the added parameters to the unrestricted model are helping to increase the explanatory power in comparison with the restricted model. To compare an unrestricted model to a restricted model we use the F-test to compare nested models. The formula for the obtained F-statistics that includes the unrestricted model and the restricted model is:
𝐹!"# = 𝑆𝑆𝐸! − 𝑆𝑆𝐸! 𝑝! − 𝑝! 𝑆𝑆𝑅! 𝑛 − 𝑘 ~𝐹!!!!!,!!!
Where 𝑆𝑆𝐸! is the Explained Sum of Squares of the restricted model, 𝑆𝑆𝐸! is the Explained Sum of Squares of the unrestricted model, 𝑆𝑆𝑅! is the Residual Sum of Squares of the unrestricted model, n is the number of observations and k is the number of independent variables in the unrestricted model (Keller, 2008). The obtained F-value was tested against a significance level of 5% (𝛼 = 0.05). If the null hypothesis cannot be rejected there is a possibility that the added independent variables in the unrestricted model do not contribute explaining the dependent variable. As a result, the restricted model is preferred over the unrestricted model.
Thirdly, the contribution of a single independent variable was tested against a significance level of 5% (𝛼 = 0.05). The t-test statistic was used to test the contribution of an independent variable. The test statistic is:
𝑡 =𝑏!− 0
𝑠!! ~ 𝑡(𝑛 − 𝑘 − 1)
Again, when the significance level is greater than the p-value of the test statistic, the null-hypothesis can be rejected. This means that the beta of the independent variable is not equal to zero with a certainty of 95%. There is evidence to support that the independent variable explains the dependent variable. If the null hypothesis cannot be rejected there is a possibility that the independent variable does not contribute to the explanation of the dependent variable. However, you do not remove the variable out of the regression because it can contribute by affecting other variables and explaining the dependent variable jointly. Removing an independent variable is only allowed when the adjusted R-squared increases when an independent variable is removed (Keller, 2008).
4.2 Data
To investigate the rapidly increasing house prices in Amsterdam, several data sources need to be used. First of all, monthly data about the house prices in Amsterdam can be found in the Central Bureau Statistics (CBS) Database that they collected together with the Kadaster.
For the income, the average yearly income of households in Amsterdam is used and converted to monthly figures. There is only data available until 2015; a trend line therefore estimated the other values (see Figure 2). The price-to-income ratio discussed by the ING Economics Department (2017) and by Ayuso and Restoy (2006) was rejected because it did not explain the actual spendable income. Because this is a multiple regression, factors such as inflation and the mortgage interest rate can be included to reflect a more realistic representation. The inflation is retrieved from the CBS Database. For inflation, the monthly CPI, Consumer price index, is used. The monthly CPI shows the average price change with respect to the month before. This monthly CPI is only available for the Netherlands. The mortgage interest rates in the Netherlands can be found in the database of the Dutch Bank, De Nederlandsche Bank (DNB). When the mortgage interest rate increases, you will expect fewer people to take a mortgage and evidently a decrease in the house prices.
Besides the income of households, the amount of assets at the Bank is considered to provide more information about the wealth of the citizens. The Dutch National Bank (DNB) provided information about these amounts. Deposits with a variable duration were taken into account because it is easier to withdraw money at any moment in time.
Internationalization and urbanization can be derived from migration and population differences over time. Migration difference is the difference between immigration and emigration into and out of Amsterdam. This variable is not taken into account in the regression, because it is correlated with the total population growth of Amsterdam. Not only is there an increasing number of foreign people moving into Amsterdam, but also an increasing number of Dutch citizens who move to the cities. By taking the total population growth of Amsterdam, both groups of people are taken into account. The changes in the population of Amsterdam can be found in the Database of CBS in collaboration with the municipality of Amsterdam.
Consumer trust is also retrieved from the Consumer Statistics Bureau (CBS). This indicator of consumer trust is corrected for seasonal changes. The value lies between -100 and +100 where -100 means no trust at all, 0 means that the amount of negatives is equal to the number of positive persons and +100 means full confidence. The change relative to the month before is used in the regression. The supply of houses is also an important factor to enable the population growth of Amsterdam. For the number of houses available in Amsterdam, the total housing stock is used. The housing stock implies all available houses if you add the constructed houses and subtract the demolished houses. For the housing stock of Amsterdam, data from CBS in cooperation with the municipality of Amsterdam is used. The unemployment rate is the overall unemployment rate of Amsterdam. The unemployment rate is corrected for the seasonal changes and retrieved from the CBS. It is important here to take the unemployment rate from Amsterdam instead of the unemployment rate from the Netherlands because the unemployment rate of Amsterdam is much higher than the overall Dutch unemployment rate.
The monthly average return on investment form the Dutch stock market, AEX, is used to provide information about other investments that investors would consider besides buying a house. This data is retrieved from Euronext Amsterdam. Rent is also seen as an important driver and is retrieved from Pararius. The data retrieved is the price per square meter. There is only data available from 2010 until 2017. The missing data is estimated by taking a linear trend line through the known data (see Figure 3). When using the data, the relative rent change is calculated and used. The economic growth of Amsterdam is retrieved from the
CBS. This data presents the change relative to the year before and converted to monthly values.
Figure 3: Trend line rent
5. Results
5.1 Testing the hypotheses
There are three multiple regression models to be analyzed. The first model includes all independent variables. The second model excludes the stock market return and the consumer trust because the influence of both independent variables seemed ambiguous. The third model represents the long-term equation as stated by Boelhouwer et al. (2018). All models were tested for explanatory power making use of an F-test. After that, the best regression is determined by making use of the adjusted R-squared. This measurement gives insight independently from the number of variables. Thereafter, the unrestricted models were tested against the restricted models to again check which model explained the dependent variable best. Only the best regression was used for the t-test.
(1) (2) (3)
VARIABLES Model 1 Model 2 Model 3
lnPop 2.651*** 2.671*** (0.742) (0.739) lnY 0.237 0.264* 0.826*** (0.147) (0.144) (0.0740) lni -0.169*** -0.171*** -0.226*** (0.0347) (0.0336) (0.0413) lnHS -0.919 -1.002 (0.847) (0.842) lnW -0.125** -0.127*** (0.0484) (0.0455) Economic growth 3.899* 3.863* (2.353) (2.255) Unemployment -6.571*** -6.474*** (0.452) (0.442) Inflation 0.295 0.286 0.646 (0.704) (0.702) (1.318) Relative rent -2.152*** -2.212*** (0.624) (0.618) Consumer trust -0.00390 (0.00368) Stock 0.0123 (0.0685) Constant -12.55*** -12.02*** 3.318*** (4.277) (4.204) (0.689) Observations 180 180 180 R-squared Adjusted R-squared F-test model P-value of F-test 0.918 0.912 170.38 0 0.917 0.913 209.09 0 0.692 0.687 131.80 0 Standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Table 1: Multiple regression models
5.1.1 Hypothesis 1
First, Model 1 is discussed. When looking at Table 1, you see that the p-value for the F-test is equal to zero. The significance level that we had chosen was α= 0.05. This means that the p-value is smaller than the significance level (𝛼 > 𝑝 → 0.05 > 0.00), so we can reject the null-hypothesis. Furthermore, looking at the R-squared, the value is close to 1. This means the regression is a good fit. When looking at the adjusted R-squared and compare it with the
adjusted R-squared from Model 2, you see that it is smaller. This means that removing the variables ‘Stock return’ and ‘Consumer trust’ would give a better regression to explain the increased house prices. The first multiple regression model is:
ln 𝑃! =(0.004) +−12.55 (0.00) ∗ ln 𝑃𝑜𝑝2.651 !− 0.011 ∗ 𝑙𝑛𝑊0.125 !− 0.1690.00 ∗ 𝑙𝑛𝑖!+ 0.2640.108 ∗ 𝑙𝑛𝑌!− 0.919 0.280 ∗ 𝑙𝑛𝐻𝑆!+ 0.2990.672 ∗ 𝜋!+ 3.8990.099 ∗ 𝑔!− 6.5710.00 ∗ 𝑈!− 2.1520.001 ∗ 𝑟!0.01230.858 ∗ 𝑆!− 0.00390 0.290 ∗ 𝐶!+ 𝜀 (5)
Second, Model 2 is discussed. When looking at Table 1, you see that the p-value for the F-test is also equal to zero. Again, with the chosen significance level of α= 0.05 this means that the p-value is smaller than the significance level (𝛼 > 𝑝 → 0.05 > 0.00). Hence, we can reject the null-hypothesis. Moreover, the R-squared and the adjusted R-squared are close to 1, this means that the conducted regression is a good fit between the dependent and independent variables. The assumption about the error-term 𝜀~ 𝑖. 𝑖. 𝑑. 𝑁(0, 𝜎! is checked by
conducting a residual plot. To fulfill the assumption, the residual plot needs to be identically and independently distributed around zero. Figure 4 represents the residual plot and fulfills the assumption; the residuals are randomly scattered around zero. The second multiple regression model is:
ln 𝑃!= (0.005) +−12.02 2.6710.00 ∗ ln 𝑃𝑜𝑝!− 0.006 ∗ 𝑙𝑛𝑊0.127 !− 0.1710.00 ∗ 𝑙𝑛𝑖!+ 0.2640.068 ∗ 𝑙𝑛𝑌!
− 1.0020.236 ∗ 𝑙𝑛𝐻𝑆!+ 0.2860.685 ∗ 𝜋!+ 0.089 ∗ 𝑔3.863 !− 6.4740.00 ∗ 𝑈!− 2.2120.00 ∗ 𝑟! + 𝜀
Figure 4: residual plot
Third, Model 3 is discussed. When looking at Table 1 you see that the p-value for the F-test is equal to zero. Again with the chosen significance level of 5%, we can reject the null-hypothesis. The R-squared and the adjusted R-squared are both close to 1, so they indicate a good fit. However, they are much smaller than Model 1 and 2. This implies that Model 1 and 2 better explain the dependent variable than Model 3. The third regression model is:
𝑙𝑛𝑃!= (0.00) +3.318 (0.00) ∗ 𝑙𝑛𝑌0.826 !− 0.2260.00 ∗ 𝑙𝑛𝑖!+ 0.6460.625 ∗ 𝜋!+ 𝜀 (7)
5.1.2 Hypothesis 2
The obtained F-test is used to test unrestricted models against the restricted models. First, we take Model 1 as the unrestricted model and we take Model 2 as the restricted model. The chosen significance level is 0.05. When looking at Table 2, you see that the obtained p-value of the F-statistic is higher than the significance level. As a result, we do not reject the null-hypothesis. This means that we cannot reject that the added parameters, the parameters from the stock market and from consumer trust, are equal to zero. Hence, we prefer Model 2 because the added parameters in Model 1 are likely to have no effect on the dependent variable.
The second test is to compare Model 2 against Model 3. The unrestricted model is Model 2 and the restricted model is Model 3. The chosen significance level is 0.05. When looking at Table 3 you see that the obtained p-value from the F-statistic is smaller than the significance level. As a result, the null hypothesis can be rejected. This means we can assume
that one of the extra parameters in Model 2 is not equal to zero. Hence, we prefer Model 2 because the added variables are likely to influence the dependent variable.
F(6,170) 77.00 Prob.>F 0
Table 2: Comparing Model 1 and Model 2
F(2,168) 0.60 Prob.>F 0.55
Table 3: Comparing Model 2 and Model 3
5.1.3 Hypothesis 3
We only consider Model 2 from this moment on because both hypotheses mentioned above prefer Model 2 to the other models (see Figure 5). The t-test is used to test the contribution of a single independent variable on the dependent variable. The p-values from every parameter can be found in the equation in the parentheses. These p-values will be tested against the significance level of 5%. For the population, wealth, mortgage interest rates, unemployment, rent and the constant the null-hypothesis is rejected against a significance level of 1%. The p-value from income and economic growth is higher than the significance level of 0.05. However, if we increase the significance level to 0.10, the null-hypothesis can be rejected. In Model 2 only housing stock and inflation can possibly have no influence on the house prices. This does not mean that we remove them from the regression, because they can have an important influence on the house prices together with another variable. Mulder (2006) also stated that demographics have an influence on housing because the opportunities for population change when the supply of housing changes. Inflation can also not be removed, because it is the fundament of the long-term relationship between house prices as mentioned in Boelhouwer et al. (2018). The inflation makes sure that the variables are not nominal but real. The general increase in the level of price is controlled by inflation. Also, inflation fluctuates extremely in the short term. Prices do not react immediately to short-term inflation fluctuations. They react slower to inflation so the long-term inflation is taken into account.
Figure 5: Display of Model 2 against the residual and the fitted model
5.2 Analyses
In the first and third hypothesis is Model 2 preferred over the other models. When combining these hypotheses, we can assume that Model 2 is the best regression for explaining the dependent variable, price. When analyzing this regression we can see the influence of each independent variable, while keeping the other variables constant.
5.2.1 Significant positive effect
When population increases 1%, the house prices increases 2.671%. This is in line with the expectation. Because if there are more people in the city there is more demand for housing and the price will increase. When income increases 1%, the price of houses increases 0.264%. Again, this seems correct because when you have more income you will have more money to buy a house and this will put pressure on prices. Lastly, when economic growth increases with 1 relative to the prior month, the house prices will increase by 386.3%. This seems much, however economic growth relative to the prior month has a mean of 0.002 (see Table 4). So, for a more realistic analysis we would say; when economic growth increases with 0.001 relative to the prior month then the house prices will increase by 0.3863%.
5.2.2 Significant negative effects
When the variable deposits increases by 1%, the house prices will decrease 0.127%. This does seem logical because you expect that when people have more assets these assets are not invested somewhere else. If the amount of assets in the variable deposits is low and the overall wealth of people stays the same, the rest must be invested somewhere else for example houses. This would cause an upward pressure on house prices. When the unemployment level increases with 1, the house prices will decrease 647.4%. Again this seems much, but the mean is 0.0688 (see Table 4). A more realistic analysis would be; when unemployment increases with 0.01 then the house prices will decrease by 6.474%. The negative relationship seems logical because when there are more people unemployed this implies a negative pressure on prices. Furthermore, the parameter of unemployment implies a larger effect compared to other variables in this regression. When the mortgage interest rate increases 1%, the house prices will decrease 0.171%. This seems realistic because if the mortgage interest rate is higher, monthly interest expenses will be higher and you can borrow a smaller amount, keeping the rest of the variables constant. When the rent level relative to the prior month increases with 0.001, the house prices will decrease 0.2212% keeping other variables constant. Again, the step is smaller because it gives a more realistic analysis (see relative rent Table 4). However, the negative relationship does not seem logical. The ING Economics Department (2017) expected a positive relation between rent and house prices. This difference can be caused by omitted variables, by calculating the wrong trend line or can be explained by the adjustment relationship mentioned in Section 2.2. Gallin (2008) stated that the response of the price effect is faster than the rent effect. So, when prices are relatively high compared to rents, the growth of real rents is higher than normal and the growth rate of real house-prices will diminish compared to the prior period.
5.2.3 Insignificant effects
When the housing stock increases by 1%, the house prices will decrease 1.002%. The negative effect seems logical because when the housing stock increases, there are more houses available and this increases the supply and therefore reduces the upward pressure on the prices. Lastly, when the inflation level increases with 0.001, the house prices will increase by 0.0286%. Again a smaller step is taken because the average inflation is around 0.00127. These results imply a relatively small effect of inflation on prices. The p-value for the t-test is
high and can imply that inflation has no effect on house prices. Because inflation is a control variable for the general increase in prices, the variable itself will not explain much.
(1) (2) (3) (4) (5)
VARIABLES N Mean SD Min Max
Price 180 283,066 45,992 209,766 417,652 Population 180 780,393 37,865 736,627 856,536 Variable deposits 180 236,836 46,683 154,743 297,624 Housing stock 180 399,537 17,735 377,153 432,712 Income 180 30,333 3,352 25,000 36,800 Consumer trust 180 -0.0765 0.942 -8 3 Stock 180 0.00659 0.0537 -0.188 0.132 Unemployment 180 0.0688 0.0128 0.0440 0.0910 Mortgage rates 180 0.0460 0.00816 0.0281 0.0562 Economic growth 180 0.00209 0.00186 -0.00340 0.00400 Inflation 180 0.00127 0.00483 -0.0110 0.0120 Rent 180 16.21 3.837 9.593 22.34 Relative rent 180 0.00475 0.00605 -0.0124 0.0308
Table 4: Summary table variables
6. Conclusion
The main question in this thesis is which factors have influenced the increasing house prices in Amsterdam over the past fifteen years. After discussing relevant literature, the independent variables included in the best-fitted regression were population, wealth, housing stock, income, unemployment, mortgage interest rates, economic growth and relative rent changes controlled by inflation. Consumer trust and the return on the stock market did not contribute to a better regression as well as to the long-term regression proposed by Boelhouwer et al. (2018). The results imply that population, income, inflation and economic growth have a positive effect on the house prices whereas wealth, housing stock, unemployment, mortgage interest rates and relative rent changes have a negative effect.
7. Discussion
This research equation is mainly based on the procedure of Boelhouwer et al. (2017, 2018). If these regressions are not correct then the fundament of my regression is not adequate. Other possible limitations of the research are missing data and choosing the wrong dataset for the
variable. The variables rent and income of Amsterdam were missing some data in the period from 2003 until 2017 that is compensated by taking a trend line. This conducted data does not represent the actual amounts. Also, for some independent variables, a different dataset would be more accurate. For example, the AEX is used as an indication for return on investment in the stock market; however, nowadays it is just as easy to invest your money around the world instead of only in Amsterdam. The AEX might not be an accurate indicator that investors consider when investing their money. Another drawback of this thesis is that the time period taken into account is only fifteen years. When taking more years into consideration, more data can be used and a more accurate regression can be made. However, the missing data makes it more difficult to carry this out.
Furthermore, some expectations about independent variables were not in line with the results of this research. Increasing rent, for example, seems to have a negative effect on house prices. Already one possible explanation is mentioned; the used trend line is not the actual relative rent changes. Another explanation can be due to omitted variable bias. This means that independent variables that are correlated with rent are not included in the model. Also, a possible explanation could be that the period from 2008 until 2013, the financial crisis, caused a strange period. The financial crisis had an enormous impact on the economic environment and cannot be accounted for by only nine independent variables.
For further research, I would recommend including more variables and consider taking one regression with dummies to account for the different time periods or four separate formulas to compare with each other for example, before the crisis, during the crisis, after the crisis and the total period.
8. Bibliography
Ayuso, J., & Restoy, F. (2006). House prices and rents: an equilibrium asset pricing approach. Journal of empirical finance. 13(3), 371-388
Boelhouwer, P., De Haan, J., & Tu, Q. (2018). House prices and long-term equilibrium in the regulated market of the Netherlands. Housing Studies, 33(3), 408-432
Boelhouwer, P., De Haan, J., & Tu, Q. (2017). The mismatch between conventional house prices modeling and regulated markets: insights from the Netherlands. Journal of Housing and the Built Environment, 32(3), 599-619
Braspenning, J. (2017). Mortgage credit in the Netherlands. Journal of European
Consumer and Market Law (EuCML), volume (4), 180-184
Candas, E., Bagdatli Kalkan, S., & Yomralioglu, T. (2015). Determining the factors
affecting housing prices. From the Wisdom of the Ages to the Challenges of the Modern
World Sofia, Bulgaria, 17-21 May 2015
Gallin, J. (2008). The long-run relationship between House prices and rents. Real
Estate Economics. 36(4), 635-658
Horn, K., & Merante, M. (2017). Is home sharing driving up rents? Evidence from Airbnb in Boston. Journal of Housing Economics. 38(1) 14-24
ING Economics Department. (2017). Pricing on the Amsterdam housing market:
Foundation for significant price increase eroding. Retrieved from https://www.ing.nl/media/pdf_EBZ_Amsterdam-real-estate-prices_tcm162-123899.pdf
International Monetary Fund. (2018). Global Financial Stability Report: A Bumpy
Road Ahead. Washington, DC, April.
Kakes, J., & Van den End, J.W. (2004). Do stock prices affect house prices? Evidence for the Netherlands. Applied Economics Letters. 11(12), 741-744
Keller, G. (2008). Managerial Statistics (8). Mason, Ohio/USA: South-Western Cengage learning
Kraniotis, L. (2017, September 29). De explosieve groei van Airbnb in Amsterdam is voorbij. NOS. Retrieved from https://nos.nl/artikel/2195345-de-explosieve-groei-van-airbnb-in-amsterdam-is-voorbij.html
Mulder, C. (2006). Population and housing: a two-sided relationship. Demographic
Research. 15(1), 401-412
Shaheen, Salman (2017). The Netherlands and UK sit at the heart of the global tax haven industry. International Tax Review: London.
Suvillan, C. (14-11-2017). Amsterdam returns Dutch house prices to 2008 highs. The Financial Times. Consulted on June 13th, 2018. Retrieved from
https://www.ft.com/content/79dca9b4-c3b7-11e7-b30e-a7c1c7c13aab
Van Ooijen, R., & Van Rooij, M. (2016). Mortgage risks, debt literacy and financial advice. Journal of Banking and Finance. 72(1), 201-217
Unknown. (14-02-2018). Amsterdam hardst stijgende huizenprijzen van Europa. Het Parool. Consulted on April 6th, 2018. Retrieved from https://www.parool.nl/amsterdam/-amsterdam-hardst-stijgende-huizenprijzen-van-europa~a4569531/
