Housing Prices in the Northern Netherlands: Considering
Earthquakes and Demographic Population Decline
Master thesis of: Nannette Evelien Stoffers Student number: S2189259
Supervisor: Professor J.P. Elhorst
Date: June 24th 2016
Abstract:
This master thesis investigates how housing prices in the Northern Netherlands have been influenced by earthquakes before 2013, whilst controlling for demographic population decline. It first presents, and then uses, theory on how to control for demographic population decline in a hedonic price model to control for demographic population decline in a sub-sample, the Province of Drenthe. Expanding the sample to the entire Northern Netherlands and adding expected earthquake magnitude to the model leads to the conclusion that there is no effect of earthquakes on housing prices before 2013. However, after estimating this model separately for different types of houses, a significant effect is found for apartments at the 10% level.
JEL Codes: J19, R21, R31
2 1. Introduction
Since the earthquake with magnitude 3.6 in Huizinge on the 16th of August 2012, earthquakes in the Northern Netherlands have become national news. On the 21st of January 2013 SodM (Staatstoezicht op de Mijnen, State Supervision of the Mines) received a letter from NAM (Nederlandse Aardolie Maatschappij BV, Dutch Petroleum Company) stating that though previous analyses predicted a maximum earthquake magnitude of 3.9 with a relatively small uncertainty, new analyses based on the gas field of Groningen only predict a 50% chance of an earthquake with a magnitude above 3.9 during the production period that would be ongoing for the next 50 years (NAM, 2013). A letter from SodM to the Minister of Economic Affairs, Henk Kamp, states that NAM and SodM agree that a reduction in gas extraction would reduce the number of earthquakes and reduces the chance at earthquakes with a magnitude above 3.9 reduces proportionally to the reduction in the number of earthquakes (State Supervision of the Mines - Ministry of Economic Affairs, 2013).
The uncertainty following the earthquake in Huizinge induced demand for extra investigations. As the earthquakes cause damage to the houses in the earthquake area one of the questions is whether the earthquakes impact housing prices. On the third of July 2013 Minister of Economic Affairs Henk Kamp sent a letter notifying the House of Representatives of the results of the investigation by Ortec Finance: so far the effect of earthquakes on housing prices seemed to be insignificant (Francke and Lee, 2013; Ministry of Economic Affairs, 2013). Insignificant in this context meaning that there is no effect on housing prices before 2013. Koster and Van Ommeren (2015) do find an effect before 2013. According to their paper housing prices have reduced by 1.7% from 2002-2007, and by 1.9% from 2008-2013. Koster (2016) in a follow up to Koster and Van Ommeren (2015), reduces the effect size of the earthquakes from 1.9% to 1.6%. However, these papers do not explicitly take into account demographic population decline, which also has a negative impact at housing prices. Therefore, their results of the effect of earthquakes on housing prices might be overestimated. Other studies, (Francke and Lee, 2014; CBS, 2015; Bosker et al., 2016) did not find a significant effect of earthquakes on house prices before 2013. Bosker et al. (2016) do find an effect for earthquakes, after the earthquake in Huizinge, of about 2%.
3
with magnitudes comparable to the earthquake in Huizinge: in 1997, in Roswinkel with a magnitude of 3.4; and in 2006 in Middelstum with a magnitude of 3.5 (NAM, 2008).
There is a plausible cause for this insignificance, namely that the perceived risk of earthquakes was lower before the earthquake in Huizinge (Bosker et al., 2016). They note that although there have been earthquakes before Huizinge, the earthquakes in Huizinge and all earthquakes thereafter have received much more media attention than the earthquakes before the earthquake in Huizinge, and this attention increased the perceived risk of earthquake damage. Additionally, the insecurity about the possible magnitudes of future earthquakes has increased since the report of NAM (2013) to SodM, this may also have increased the perceived risk of earthquake damage after 2012.
Since the literature is not conclusive on this matter, and the paper by Koster and Van Ommeren (2015) does not take into account demographic population decline, it will be very interesting to discover whether there might be an effect of earthquakes on housing prices before 2013 when demographic population decline is taken into account.
This thesis, concordant with Bosker et al. (2016), is hypothesizing that demographic population decline could have a negative impact on housing prices that needs to be controlled for when estimating the effect of earthquakes on housing prices. Without controlling for demographic population decline the effect of earthquakes on housing prices cannot be isolated and might be overestimated. The following is an example of how demographic population decline could affect prices:
The declining share of children in the neighbourhood of Oudemolen, Drenthe, has led the municipality of Tynaarlo to reject the construction of a new school building for the Meester Crone primary school in Oudemolen (Dagblad van het Noorden, 2011). The school was closed in 2013 (Voortman, 2013). A reduction in the level of services like this one can decrease the demand for housing in this area, and thereby have a negative impact on housing prices.
Therefore, variables related to demographic population decline are added to the estimation of housing prices, using the model by Koster and van Ommeren (2015) as a starting point.
4
noticeable effect on housing prices before 2013. The final aim is to discover whether the prices of different types of houses are affected differently by the earthquakes in the period 2008-2012. Regarding the first aim a theoretical framework is built using previous literature after which the housing prices in the Province of Drenthe are estimated. In the Province of Drenthe there are, in comparison to the Province of Groningen, relatively few earthquakes, but there is population decline. Based on the hedonic price equation by Koster and van Ommeren (2015) an equation is formulated including the variables for demographic population decline from the theoretical framework.
Secondly, the focus shifts to the entire Northern part of the Netherlands, consisting of Groningen, Friesland, and Drenthe, and the hedonic price equation previously estimated for Drenthe will now include an earthquake variable, thus differentiating between demographic population decline and earthquakes. The null hypothesis is that there is no effect of earthquakes on housing prices before 2013.
Finally, according to E.Hol from Invisor Omgevingsmanagement BV (personal communication, 2016) it might not be appropriate to assume homogeneity of variance for all housing types. After testing for homogeneity of variance, four separate equations are estimated. One equation for each type of housing: apartments, terraced, semidetached, and detached houses, with the null hypothesis that there is no effect of earthquakes for different types of houses.
5 2. Theoretical Framework
The theoretical framework consists of three parts: the definition of demographic population, its relation to housing prices, and how to control for demographic population decline.
2.1 Demographic population decline defined
Regarding the first aim of this thesis it is important to know how demographic population decline is defined. The first definition of demographic population decline that may come to mind is the reduction in the number of inhabitants of a certain region. However, this definition is too narrow for the purpose of this thesis. Van Dam, de Groot, and Verwest (2006) mention three ways in which demographic decline manifests itself:
- A reduction in population numbers;
- A reduction in population related to household composition; - A reduction in population related to demographics.
To clarify, a reduction in population numbers is not only a reduction in the number of inhabitants of a certain region, it could also be a reduction of the number of households. The number of inhabitants of a region is related to the number of households but not one to one. For example, the number of inhabitants may stay the same whilst the number of households increases because people divorce, increasing the number of single-person households. This is also a change in household size, bringing us to the next manifestation, household composition. Regarding household composition consider: household size, life phases, and income. An example of the influence of life phase is students. They move from their home region to a university city for a couple of years and after this phase they either move back, stay in the city of their university or move somewhere else. Then there are the changes in population demographics. Decreases in the share of the young, and changes in the percentage of immigrants to name a view examples.
2.2 Demographic population decline and its effect on housing prices
6
demographics are more important. Demographic population decline can influence both the supply and demand side of the housing market. For example, if there are less people migrating towards a certain area the demand for houses goes down, the demand for houses shifts inwards from D to D’ as in panel a of graph 1. If people migrate out of a certain area, supply of houses increases from S to S’ as in panel b. Although graph 1 illustrates both effects separately these effects are not mutually exclusive. Both of these tendencies lower housing prices, therefore, the direct relation between demographic population decline and housing prices is expected to be a negative one.
Graph 1
panel a panel b
Note: P denotes the price of houses, Q denotes the quantity of houses, S denotes supply of houses, and D denotes
demand for houses.
According to Van Dam et al. (2006) demographic population decline does not always have to be problematic. When the housing market is controlled by the suppliers of houses the housing market is under pressure. Demographic population decline can reduce this pressure, with some of the market power going from the suppliers’ to the consumers’ side of the market. Consumers might in this situation be enabled to better realise their preferences. However, in the Northern Netherlands, areas with demographic population decline usually do not face a housing market under pressure, consumers already have relative power compared to the suppliers of houses. Therefore, chances are that a reduced demand for houses would lead to an over-abundance of houses in the less attractive share of the housing stock (Van Dam, 2009).
7
relations, and life phases affects the number of households as well as the demand for and supply of houses, as do changes in demographics. These two types of demographic population decline could affect the prices of different types of houses differently. For example, a decrease in the average household size due to a decrease in fertility could increase the demand for apartments and decrease the demand for family homes.
In the longer term demographic population decline might not only be a burden, it could also be an opportunity (Gerrichhauzen and Dogterom, 2007; Van Dam et al., 2006; Haartsen and Venhorst, 2010). It might be a chance at improving the quality of the housing stock. Also, it is an opportunity to specialize in certain types of housing for which there is increasing demand, such as housing for the elderly. In addition, it provides opportunities to improve the quality of the neighbourhood surroundings, as has been done in neighbourhood Wold A in Lelystad (Gemeente Lelystad, 2002). In the long run this may increase the housing prices again. The long term effects of demographic population decline on housing prices are thus depending on (local) government policies on housing.
In conclusion, in the short term the overall effects of demographic population decline on housing prices are expected to be negative. However, there could be differences in the effects between different types of houses. Also, in the longer term, the effects of demographic population decline on housing prices depend on (local) government housing policies.
2.3 How to control for demographic population decline
There are numerous ways of controlling for demographic population decline. In this section an overview of methods of measuring demographic population decline directly and indirectly is presented.
2.3.1 Directly
8
The average number of persons per household includes information of demographic as well as household composition: fertility, life expectancy, the forming of relations, divorces, and life phase. Controlling for the average number of persons per household might thus partially control for demographic population decline.
Demographic population change also manifests itself in the share of children, the share of the elderly, and the share of immigrants. Even though in the future the passing of the baby boom generation in combination with a fertility rate below the replacement rate (Eurostat, 2015) will lead to population decline, the increase of the share of elderly people is, at first, demographic population growth. This might even have a positive effect on housing prices of certain types. A reduction in the share of children must be seen as demographic population decline, as it could for instance be caused by a reduced fertility rate or by migration of families with children. Finally, the share of immigrants could control for ethnic segregation in general and for what is called ‘white flight’: “the departure of whites from places (as urban neighbourhoods or schools) increasingly or predominantly populated by minorities” as defined by the Meriam Webster Dictionary (2016).
2.3.2 Indirectly
It is also possible to control for demographic population decline in a less direct way, namely by controlling for the factors that influence the surrounding environment of the house, neighbourhood or municipality. The relation between demographic population decline and the surrounding environment is mostly unidirectional, from the surrounding environment towards demographic population decline.
If the amount of services goes down, for some people the surrounding environment of the house will become less attractive because people will have to travel greater distances to services such as hospitals, primary schools or supermarkets. Therefore, the in-migration – migration towards a new neighbourhood – might decline.
9
changed, over the years consumers have become more mobile and less dependent on their local shops.
According to Van Otterdijk (2011) the discontinuation of an ER or hospital has a big emotional impact a rural area. In addition, The Netherlands Institute for Social Research (2007) notes that people complain about the discontinuation of shops, ATMs, libraries, leisure services, primary schools and health care services. This gives a clear indication of which services are of most importance to an area.
10 3. Models and Methods
In the previous chapter defined demographic population decline and explained the theory behind controlling for demographic population decline. This section first explains the hedonic price model and how demographic population decline is added to this model, after which earthquakes are included in the model.
3.1 The hedonic price model
According to Rosen (1974) hedonic housing prices are implicit prices of housing attributes, which are caught in the observed price of the house and the observed attributes. Using the model by Koster and Van Ommeren (2015), whilst excluding the earthquake variable, as a starting point, the hedonic price model uses the transaction price, the observed price of the house, as dependent variable. The independent variables are the housing attributes, and neighbourhood attributes. This leads to the model in equation (1).
(1) log(𝑝𝑖𝑡) = 𝛼0+ 𝑋𝑖𝑡𝛽 + 𝑍𝑖𝑡𝛾 + 𝜂𝑖+ 𝜃𝑡+ 𝜖𝑖𝑡1
where i denotes a house, and t denotes the time in years. On the left hand side there is the
logarithm of the transaction price, log(𝑝𝑖𝑡). On the right hand side there are the constant, 𝛼0, housing attributes 𝑋𝑖𝑡, neighbourhood attributes 𝑍𝑖𝑡, neighbourhood and time fixed effects 𝜂𝑖
and 𝜃𝑡, and an error term denoted by 𝜖𝑖𝑡.
Regarding the dependent variable, the logarithm is used to solve the non-linearity of the price variable. In addition, instead of using the logarithm transaction price per square meter as Koster and Van Ommeren (2015) do, the actual transaction price is used. Koster (2016) explains in his report on Koster and Van Ommeren (2015) that there should not be a difference in the coefficients when taking the logarithm of the transaction price per square meter or the logarithm of the transaction price. However, there is a difference in the R2. Using the price per square meter leads to an increase higher R2 due to the lower variance. This thesis argues that the increase in R2 is artificial, more is explained from less information. Since in general people purchase houses as a whole, trying to explain the entire transaction price by means of housing and neighbourhood attributes, instead of the transaction price per square meter seems more intuitive.
11
Following Koster and Van Ommeren (2015) the right hand side of the equation contains the implicit prices of housing and neighbourhood attributes. The housing attributes are: the living space in square meters; the number of rooms; the type of house (apartment, terraced, semi-detached, or detached); whether or not there is a garage or garden; type of heating; when the house was constructed; and whether the house is listed as a monument. Neighbourhood attributes are: population density; the share of the young, the elderly and immigrants; the average size of the household; and the shares of industrial or commercial land use, infrastructure, open space, and water.
Equation (1) also contains fixed effects. The time fixed effects, included as year dummies will control for spatial-invariant price effects that are specific to a certain year. The time fixed effects control for example for the recession that affected the Netherlands during the period under investigation. Neighbourhood fixed effects will filter out the time-invariant effects on the housing price of specific neighbourhoods.
3.2 The hedonic price model and demographic population decline
Equation (1) already controls for some demographic population decline as characteristics as the average number of persons per household, the share of the young and the share of immigrants are included. Koster and Van Ommeren (2015) expect that these neighbourhood variables, plus the share of the elderly and the population density, would control for low population growth and other disadvantages of parts of the area under their investigation.
12
Within the dataset and the period of investigation there is data on the distance to hospitals, primary schools, and supermarkets on municipality level. These are among the most important services indicated in the theory section. The view of E.Hol (personal communication, 2016) of reducing the amount of variables by adding the distance variables together is understandable. However, that would mean giving all the distance variables the same weight which is a rather strong assumption. Reasons why these distance variables might have a difference in weight are for example that usually people visit the supermarket more often than the hospital and that the elderly might visit the hospital more often than a primary school. To control for demographic population decline they will be added separately to equation (1), resulting in equation (2).
(2) log(𝑝𝑖𝑡) = 𝛼0+ 𝑋𝑖𝑡𝛽 + 𝑍𝑖𝑡𝛾 + 𝐷𝑖𝑡𝜁 + 𝜂𝑖 + 𝜃𝑡+ 𝜖𝑖𝑡
the distance variables are denoted as 𝐷𝑖𝑡.
3.3 The hedonic price model, demographic population decline and earthquakes
After adding all the demographic population decline data the model is almost complete. The next step is to add the earthquake variable. The earthquake variable that will be used for this purpose is the expected earthquake magnitude variable as designed by Ph. D. student at the University of Groningen N. Durán (2016), which estimated the expected magnitude of earthquakes per house using data on earthquakes up to the date the house was sold. Adding expected earthquake magnitude to the model the equation takes the following form:
(3) log(𝑝𝑖𝑡) = 𝛼0+ 𝑋𝑖𝑡𝛽 + 𝑍𝑖𝑡𝛾 + 𝐷𝑖𝑡𝜁 + 𝜉𝐸𝑖𝑡+ 𝜂𝑖 + 𝜃𝑡+ 𝜖𝑖𝑡
where the expected magnitude of earthquakes is denoted as 𝐸𝑖𝑡.
3.4 Methods
The previous paragraphs explained the models required for answering the research question. The next step is to estimate these models. First, model (2) is estimated for the Province of Drenthe, where there is demographic population decline, but where there are relatively few earthquakes, in order to examine demographic population decline separately from earthquakes and see if the distance variables have the expected negative signs.
13
Northern Netherlands before 2013’. The alternative hypothesis is: ‘Earthquakes have a negative impact on housing prices in the Northern Netherlands before 2013.’
(H1.1) H0: 𝜉 = 0
(H1.2) Ha: 𝜉 < 0
What makes this investigation interesting besides controlling for demographic population decline is that it examines the period before 2013. It is known that there have been earthquakes of comparable magnitude to the earthquake in Huizinge before 2013. Most studies (Francke and Lee, 2013; Francke and Lee, 2014; CBS, 2015; Bosker et al., 2016) seem to indicate that earthquakes did not impact housing prices before 2013. Koster and Van Ommeren (2015) do find an effect of earthquakes on housing prices, but do not control for demographic population decline. It is interesting to observe if the significance of the effect of earthquakes on housing prices would uphold after adding controls for demographic population decline, or if they would disappear as expected.
Finally, E.Hol (personal communication, 2016) reasons that it might not be appropriate to assume homogeneity of variance for all housing types. In order to discover whether the prices of different types of houses are affected differently by the earthquakes in the period before 2013 Levene’s test for homogeneity of variance is conducted for the four types of houses that are considered: apartments, terraced houses, semi-detached houses, and detached houses. The null hypothesis of this test is that the variances are equal for all types of houses. The alternative hypothesis is that the variances differ between different types of houses.
(H2.1) H0: 𝜎12 = 𝜎22 = 𝜎32 = 𝜎42
(H2.2) Ha: 𝜎12 ≠ 𝜎22 ≠ 𝜎32 ≠ 𝜎42
where the subscripts 1, 2, 3, 4 denote apartments, terraced houses, semi-detached houses, and
detached houses respectively.
If the null hypothesis is rejected, model (3) will be estimated separately for each type of housing, with null hypothesis (H3.1) that there is no effect of earthquakes on housing prices of each type of housing, and alternative hypothesis (H3.2) that there is a negative effect of earthquakes on housing prices that differs per type of housing.
(H3.1) H0: 𝜉1, 𝜉2, 𝜉3 ,𝜉4 = 0
(H3.2) Ha: 𝜉1, 𝜉2, 𝜉3 ,𝜉4 < 0, 𝑎𝑛𝑑 𝜉1 ≠ 𝜉2 ≠ 𝜉3 ≠ 𝜉4
where the subscripts 1, 2, 3, 4 still denote apartments, terraced houses, semi-detached houses, and
14
15 4. Data
This thesis uses a panel dataset on transactions of houses from 2008 to 2012. The time-series 2008-2012 has been chosen since these are the years for which the distance variables are available, and since the effects of earthquakes before 2013 can be measured.
J.P. Elhorst, Professor at the University of Groningen, provided the data for this research. The data on housing attributes originates from the NVM (Nederlandse Vereniging van Makelaars). Data on the neighbourhood attributes including the distance variables originates from the CBS (Centraal Bureau Statistiek, Statistics Netherlands). The earthquake variable has been constructed by Ph. D. student at the University of Groningen N. Durán (2016).
After cleaning the data and removing outliers, there are 11,294 observations left for the estimation of housing prices for Drenthe, 31,816 observations for the three Northern Provinces (Groningen, Drenthe, and Friesland) together, and respectively 5,432, 11,362, 7,206, and 8,253 for apartments, terraced houses, semi-detached houses, and detached houses in the three Northern Provinces. Removing outliers entailed removing observations with a price below 25,000 or above 1,000,000, and removing observations of houses with a living space below 25m2 or above 250m2 as has been done by Koster and Van Ommeren (2015) and Koster (2016). According to them removing these outliers does not have an impact on the analysis.
4.1 Descriptive Statistics
16 Table 1: Descriptives
Mean St. dev. Min. Max.
Transaction price 186,835.5 87,445.94 25,000 1,000,000
Expected Earthquake Magnitude 0.099 0.233 0 2.765
Living space 114.124 34.395 25 250 Number of rooms 4.561 1.263 1 15 Apartment 0.173 0.378 0 1 Terraced 0.365 0.481 0 1 Semi-detached 0.215 0.411 0 1 Detached 0.247 0.431 0 1 Garage 0.396 0.489 0 1 Garden 0.679 0.467 0 1 Heating 0.907 0.291 0 1 Monument 0.005 0.070 0 1 Constructed before 1945 0.226 0.418 0 1 Constructed 1945-1959 0.076 0.26 0 1 Constructed 1960-1970 0.184 0.388 0 1 Constructed 1971-1980 0.214 0.410 0 1 Constructed 1981-1990 0.121 0.326 0 1 Constructed 1991-2000 0.111 0.314 0 1 Constructed after 2000 0.068 0.252 0 1
Share land use: water 0.034 0.084 0 1
Share land use: open space 0.097 0.127 0 1
Share land use: industrial/ commercial 0.108 0.149 0 1
Share land use: residential 0.491 0.270 0 1
Share land use: infrastructure 0.043 0.086 0 0.75
Population density 3,392.697 2,874.467 1 14,382
Percentage of children <15 16.382 4.952 1 38
Percentage of elderly >64 16.997 8.208 0 57
Average household size 2.170 0.371 1.2 3.6
Share of immigrants 11.181 7.166 0 46
Distance primary school 0.776 0.122 .5 1.2
Distance hospital 8.604 5.963 2.6 60.7
Distance supermarket 1.144 0.419 .4 2.5
Number of observations 31,816 The average transaction price of a house in the Northern Netherlands from 2008-2012 is €186,835.50. The mean of the expected earthquake magnitude is 0.099. This value is so low because there are many houses within the sample that are not located in an earthquake area. Besides the transaction price and the expected earthquake magnitude the variables can be divided into three categories: housing attributes, neighbourhood attributes, and distance variables, where the distance variables are observed at the municipality level. These descriptive statistics are similar to those of Koster and Van Ommeren (2015), with relatively many old buildings, and few buildings constructed after 2000. The division between different types of houses is a bit unequal however, including about twice as much terraced houses as apartments.
17
year was used. The variables can all be included in the analysis as the share of agriculture is not included. Unfortunately,actual data on the share of agriculture as it is not part of the dataset. Regarding population density the mean appears to be very high, but the standard deviation is also rather large. This can be explained as follows: the value for population density per km2 is taken on a neighbourhood level. However, as there are far more houses to be sold in dense neighbourhoods than in less dense neighbourhoods the mean of population density is skewed to the right. The standard deviation is rather high due to the big difference in population density between rural areas and cities. To take this into account the models use the natural logarithm of population density.
4.2 Multicollinearity
Multicollinearity is the problem of insufficient information to identify the individual effects of multiple variables because they vary insufficiently. It may cause inaccurate estimation results. Multicollinearity can be resolved by increasing the sample size, or dropping variables (Verbeek, 2012). To test for multicollinearity the variance inflation factor is used to detect possible multicollinearity and the correlation matrix to observe the correlation between the variables. A variance inflation factor of 1 means that there is no correlation of a variable with other variables. A rule of thumb that is often mentioned is that a variance inflation factor of above 10 is too high. However, according to Verbeek (2012) as long as the explanatory variable of interest is uncorrelated with the variables that are showing signs of multicollinearity, the amount of correlation between these variables does not have any impact on the standard deviation of the variable of interest.
The variable of interest in this thesis is the expected magnitude of earthquakes. In table 2 it can be observed that the variance inflation factor (VIF) of the earthquake variable is close to 1 for the entire northern part of the Netherlands, both when looking at all types of houses together as well as when looking at the different types separately. This means that the expected earthquake magnitude is almost entirely uncorrelated to the other explanatory variables.
Table 2 Multicollinearity - Variance Inflation Factor
all houses apartments terraced semi-detached detached
VIF Expected Earthquake Magnitude 1.09 1.16 1.09 1.07 1.15 Number of observations 31,816 5,432 11,362 7,206 8,253
18
Possible multicollinearity that has been observed is multicollinearity within the neighbourhood control variables and within the distance control variables. Regarding the neighbourhood control variables within the sample of apartments the variance inflation factors of the Share land use: residential and of Average household size were just above 10: 10.77 and 10.40 respectively.
When looking into the correlation matrix the highest correlation for Share land use: residential was the correlation between residential and Share land use: industrial / commercial, namely -0.5102. This might be because the reference category Share land use: agriculture is particularly small within this sample, as most apartments are expected to be in urban areas, rather than in areas with a large share of agriculture.
Regarding Average household size, this variable has a strong correlation with the variable Percentage of children <15 in all samples. The lowest correlation between these two variables is 0.7681. This is the value within the sample of detached houses – for all other samples the correlation between these two variables is above 0.8. Therefore, the estimates of these variables might be inaccurate.
Finally, regarding the distance variables, for the sample of the Province of Drenthe the variable Distance supermarket presented a variance inflation factor of 11.21. The correlation matrix shows a high correlation between all three distance variables, causing inaccurate estimates. However, this high correlation is only present in this sub-sample. Therefore, it this does not appear to be a problem for the analyses of the Northern-Netherlands.
4.3 Homoskedasticity
For homoskedasticity table 3 shows that the null hypothesis of homoskedasticity is rejected in all samples:
Table 3 Heteroskedasticity
Breusch-Pagan/ Cook-Weisberg test
Drenthe all houses apartments terraced semi-detached detached 𝝌𝟐 334.95 594.52 12.60 350.21 371.98 534.24 𝑷 > 𝝌𝟐 0.0000 0.0000 0.0004 0.0000 0.0000 0.0000 Number of observations 11,294 31,816 5,432 11,362 7,206 8,253
19
municipality level, and all neighbourhood attributes are added at neighbourhood level, Koster and Van Ommeren (2015) will be followed by clustering at the neighbourhood level.
4.4 Endogeneity
20 5. Results
In this chapter the regression results are presented in order to investigate the hypotheses from chapter 3. Section 1 describes the estimates of regression model (2) for Drenthe, section 2 describes the estimates of model (3) for the three northern provinces, and section 3 presents Levene’s test and the estimates of model (3) separately for the various types of houses.
5.1 Drenthe and demographic population decline
Using the theory from chapter 2 and 3, model (2) is estimated for Drenthe to show how demographic population decline is controlled for. The results of this regression are presented in the column 'model (2)' of table 4.
Most important are results of the distance variables, which control for demographic population decline. Remember that in theory increasing distances to services has a negative impact on housing prices, as the area becomes less attractive for migrants. Without migrants moving in with a fertility rate below the replacement rate, there would be demographic population decline. In line with this theory the coefficients for the distance to the nearest: primary school, hospital, and supermarket are all negative. Thus an increase in distances to services indeed lowers housing prices. That the coefficients for the distance to a primary school and a hospital are statistically significantly different from zero shows that it is indeed important to control for demographic population decline as it has a negative influence on housing prices. Additionally, the coefficients for the three distance variables: primary school (-0.257), hospital (-0.010), supermarket (-0.013) differ from each other. The impact on housing prices of an increase in the distance to a primary school is thus different from an increase in the distance to a hospital. This shows that it is indeed important to add these variables separately than to the model then to add the sum of the distances.
Furthermore, as expected there is a really clear trend in the coefficients of the variables that control for the period in which the house is constructed. If the house is constructed between 1945 and 1980 it impacts the price of a house negatively, whereas construction after 1980 has a positive impact on the price of the house. The house being a monument also has a positive price impact. Other housing attributes also have the expected sign: an increase in living space, and an increase in the number of rooms would increase the price of a house, as do having modern heating, and having a garage.
21 Table 4
OLS regression results: model (2) for Drenthe and model (3) for the Northern Netherlands
Dependent variable: ln transaction price
Model (2) Model (3)
Expected earthquake magnitude 0.004
(0.02792) Ln Living space 0.594*** 0.638*** (0.02216) (0.01774) Number of rooms 0.014*** 0.015*** (0.00289) (0.00186) Terraced -0.016 0.040*** (0.54874) (0.01447) Semi-detached 0.087*** 0.135*** (0.2723) (0.01619 Detached 0.335*** 0.345*** (0.03059) (0.01714) Garage 0.075*** 0.084*** (0.00666) (0.00473) Garden -0.028*** -0.025*** (0.00680) (0.00591) Heating 0.095*** 0.105*** (0.01116) (0.00715) Monument 0.227** 0.086*** (0.09904) (0.02834) Constructed 1945-1959 -0.095*** -0.036*** (0.01392) (0.01327) Constructed 1960-1970 -0.095*** -0.048*** (0.01279) (0.00944) Constructed 1971-1980 -0.057*** -0.025** (0.01647) (0.00988) Constructed 1981-1990 0.009 0.050*** (0.01546) (0.00982) Constructed 1991-2000 0.105*** 0.149*** (0.01708) (0.01144) Constructed after 2000 0.207*** 0.245*** (0.02220) (0.01331)
Share land use: water -0.023 0.209*
(0.44923) (0.11135)
Share land use: open space -0.237 0.018
(0.15378) (0.08053)
Share land use: industrial/ commercial -0.117 0.083
(0.13887) (0.07465)
Share land use: residential -0.134 0.059
(0.14262) (0.06479)
Share land use: infrastructure -0.138 0.046
(0.14401) (0.08094) Ln Population density 0.004 -0.022* (0.01494) (0.01243) Percentage of children <15 0.003 0.000 (0.31497) (0.00224) Percentage of elderly >64 0.005*** 0.001 (0.00119) (0.00163)
Average household size -0.015 0.012
(0.03786) (0.02417)
22
(0.00183) (0.00143)
Distance primary school -0.257** -0.276***
(0.10579) (0.06679) Distance hospital -0.010*** -0.011*** (0.00367) (0.00334) Distance supermarket -0.013 -0.106*** (0.04391) (0.02597) Constant 9.375*** 9.296*** (0.22940) (0.15655)
Construction dummy 2(6) Yes Yes
Time fixed effects (4) Yes Yes
Neighbourhood fixed effects (229) Yes Yes
R2 0.79585 0.78861
Number of observations 11284 31816
Note: The cluster robust standard errors are in the parenthesis; *, **, *** denote a 10%, 5%, and 1% significance level respectively.
5.2 The Northern Netherlands earthquakes and population decline
By expanding the sample to the entire Northern Netherlands and inclusion of the expected earthquake magnitude the effect of earthquakes on housing prices before 2013 can be examined. In table 4 these results are found under ‘model (3)’.
Note that the results are very much similar to those in table 4 model (2), all significant variables have the same sign in both models. In addition, the R2 is still approximately 0.8. Both of these findings speak in favour of the model.
The most important result of this estimation is that after controlling for demographic population decline the coefficient of the expected earthquake magnitude is not significantly different from zero. Meaning that the null hypothesis (H1.1) of there being no effect of earthquakes on housing prices in the Northern Netherlands cannot be rejected in favour of the alternative hypothesis. This result supports the conclusion from the studies by Francke and Lee (2013, 2014), CBS (2015), and Bosker et al. (2016) that there is no significant effect of earthquakes on housing prices before 2013 in the Northern Netherlands.
5.3 The Northern Netherlands and types of houses
It is important to know whether earthquakes affect different types of houses in a different manner. Using Levene’s test for homogeneity of which the results are presented in table 5, it is
23
shown that there is indeed heterogeneity of variance for the different types of housing. The null hypothesis (H2.1) of homogeneity of variance is rejected.
Table 5 – Levene’s test for homogeneity of variance
Apartment Terraced Semi-detached Detached
Levene’s test statistic 167.314 2559.676 794.103 890.756 𝑷 > 𝑭 0.00000 0 0.00000 0.00000 Number of observations 31816 31816 31816 31816
24 Table 6 – OLS regression results of model (3) for separately for each type of house
Dependent variable: ln transaction price
Apartment Terraced Semi-detached
Detached
Expected Earthquake Magnitude -0.223* 0.035 0.045 0.016
(0.12383) (0.05284) (0.05583) (0.05800) Ln Living space 0.710*** 0.541*** 0.515*** 0.681*** (0.03157) (0.03251) (0.02325) (0.02272) Number of rooms -0.004 0.019*** 0.023*** 0.014*** (0.00510) (0.00301) (0.00271) (0.00310) Garage 0.100*** 0.117*** 0.058*** 0.068*** (0.01806) (0.00882) (0.00631) (0.00841) Garden 0.074*** 0.003 -0.014* -0.065*** (0.00909) (0.00597) (0.00831) (0.00668) Heating 0.065*** 0.102*** 0.130*** 0.116*** (0.01061) (0.01092) (0.01242) (0.01320) Monument 0.022 0.095 0.203*** 0.108** (0.03060) (0.07202) (0.07080) (0.05033)
Share land use: water 0.294 -0.105 0.236 0.151
(0.31132) (0.18985) (0.17061) (0.29205)
Share land use: open space -0.136 0.058 0.055 -0.135
(0.20568) (0.09336) (0.07400) (0.10052)
Share land use: industrial/ commercial -0.132 0.095 -0.024 -0.057
(0.17045) (0.08890) (0.07365) (0.08940)
Share land use: residential -0.155 0.060 -0.041 -0.051
(0.15725) (0.08856) (0.06347) (0.08243)
Share land use: infrastructure -0.651*** 0.051 -0.028 0.294**
(0.24774) (0.11430) (0.12398) (0.13591) Ln Population density 0.038 0.022 -0.002 -0.024** (0.09176) (0.01922) (0.01264) (0.01197) Percentage of children <15 0.015*** -0.007*** -0.001 0.007 (0.00569) (0.00172) (0.00261) (0.00411) Percentage of elderly >64 -0.001 -0.002** -0.001 0.005*** (0.00490) (0.00104) (0.00128) (0.00193)
Average household size -0.023 -0.005 0.004 0.118***
(0.04457) (0.02127) (0.02681) (0.04445)
Share of immigrants -0.002 -0.003* -0.005*** -0.000
(0.00389) (0.00188) (0.00176) (0.00230)
Distance primary school -0.308 -0.181** -0.209*** -0.126
(0.19810) (0.08175) (0.07983) (0.14244) Distance hospital -0.010 -0.011*** -0.007** -0.002 (0.01148) (0.00337) (0.00329) (0.00444) Distance supermarket -0.166 -0.123*** -0.077* -0.024 (0.12075) (0.03170) (0.03974) (0.05523) Constant 8.30*** 9.524*** 9.843*** 8.849*** (0.86492) (0.22091) (0.18485) (0.25551)
Construction dummy (6) Yes Yes Yes Yes
Time fixed effects (4) Yes Yes Yes Yes
Neighbourhood fixed effects (194), (435), (552), (703)
Yes Yes Yes Yes
R2 0.81746 0.77171 0.82140 0.70653
Number of observations 5432 11362 7206 8253
25
What stands out is that different neighbourhood attributes affect the prices of various types of houses differently. For example, a high percentage of children under the age of 15 has a slightly positive effect on the price of apartments, but a slightly negative impact on the price of terraced houses. A high percentage of elderly has a slightly negative impact on the price of terraced houses but a slightly positive impact on the price of detached houses.
Regarding the distance variables, they do not seem to have a significant effect on the price of apartments. The reason for this result could be that apartments are often build in urban areas where the distances to services are rather small, therefore this might not influence the housing decision that much. Detached houses are often located in rural areas, and people who want to live in a detached house may find it more important to have space than to live close to services, thus providing a possible explanation for why distance to services does not have a significant impact on the price of a detached house.
The most important result however, is that the expected earthquake magnitude is significantly below zero for apartments at the 10% level after controlling for demographic population decline. This means that the null hypothesis (H3.1) can be partially rejected in favour of the alternative hypothesis (H3.2). The effect of earthquakes on different types of houses is thus not the same and there is already a slight negative effect of earthquakes on apartments before 2013. This means that if the expected earthquake magnitude increases with 1%, the logarithm of the price decreases with approximately 0.223% which means that the price of a house decreases with approximately 1.25%. Regarding the insignificance of the expected earthquake magnitude for the other types of houses, though previous studies (Francke and Lee, 2013, 2014; CBS, 2015; Bosker et al., 2016) did not look at the different types of houses separately, it is in line with these studies not to expect a significant effect from earthquakes on housing prices before 2013.
26 6. Discussion
6.1 Summary
The Northern Netherlands are facing earthquakes as well as demographic population decline, both of which need to be taken into account when estimating housing prices. Demographic population decline is expected to lower housing prices by increasing the supply and decreasing the demand for housing. Based on the results of previous studies (Francke and Lee, 2013, 2014; CBS, 2015; Bosker et al., 2016) earthquakes are not expected to have an impact on housing prices in the investigated time period, 2008-2012.
First, variables to control for demographic population decline need to be found and tested. Controlling for demographic population decline can be done directly using variables that directly influence the components of population change. And indirectly, controlling for influences on the surrounding environment of the house. In their model Koster and Van Ommeren (2015) use neighbourhood attributes and might already capture a part of demographic population decline. Furthermore, the thesis theorizes that the distance to the services: primary school, hospital, and supermarket, control for demographic population decline indirectly due to their relation with the attractiveness of the surrounding environment of a house. After developing the theory housing prices are estimated using a hedonic price estimation. The results from model (2) for the Province of Drenthe are in line with the theory on how the distance variables control for demographic population decline, a higher distance to services has a negative price impact.
In order to analyse the impact of earthquakes on housing prices in the Northern Netherlands before 2013 the expected earthquake magnitude is added to the model. The results of this model are similar to those of the model without earthquakes, which is clearly in support of the model. As expected from the literature, the earthquake variable is insignificant in the time period from 2008-2012 once controlling for demographic population decline. The null hypothesis (H1.1) cannot be rejected.
27
Which can possibly be explained by the fact apartments are often part of higher buildings, whereas the other housing types are mostly constructed on the ground.
6.2 Limitations
This thesis contributed to the existing literature by investigating the effects of earthquakes on housing prices before 2013, while controlling for demographic population decline. And by estimating the effect of earthquakes on housing prices separately for the different types of houses after finding heterogeneity of variance. Yet, this thesis has its limitations. The limitations of this thesis can broadly be divided into limitations of the data, and limitations to the analyses.
The first data limitation is the number of observations. Although the original dataset comprises 160903 observations, in the biggest sample used in this thesis there are only 31816 observations left. The original sample ranged from 2003 to 2014, but for the analyses all observations after 2012 dropped out, and as the distance variables were only available from 2008 onward all observations before 2008 also dropped out.
A related data limitation is that the lot size is not taken into account, because if doing so even more observations would drop out. Observations would drop out because taking the logarithm of the lot size to account for non-linearity would lead all houses with a lot size of zero – those that are not build on the ground – to have no value, and therefore drop out of the sample. Fortunately, not adding lot size might not be a big issue as Koster (2016) found no significant difference in adding lot size and not adding lot size to the hedonic price equation.
A final data limitation worth mentioning is that it has not been possible to construct a variable for the difference in the number of households for municipalities from one year to the next. The municipality borders are changing over time, when calculating the change in the number of households in a municipality from one year to the next, these changes should be controlled for. It has, due to time constrained, unfortunately not been possible to construct such a variable whilst controlling for the changes in borders.
Then there are the limitations to the analyses. The first limitation, also noted by Koster (2016), is the possibility that there are housing attributes or other characteristics that are of importance for home owners which are not observed or not observed in the correct way.
28
Lastly, one might note the possibility of earthquakes affecting demographic population decline. It is not unlikely that people might want to migrate from an area with earthquakes towards an area without earthquakes. Though aware of its possible existence, measuring or controlling for this effect is outside the scope of this thesis. Fortunately, the period under investigation is before the NAM (2013) came with the news that it is impossible on the basis of the little available data on the Groninger Veld to come to a trustworthy prediction of the maximum earthquake magnitude that could be expected in the future. Therefore, the impact of earthquakes on demographic population decline might still be small.
6.3 Further research
In this thesis the period 2008-2012 has been under investigation. An important topic for further research might be how earthquakes affect the different types of houses differently, in more recent times, after the earthquake in Huizinge.
Also, recommended for further research is delving deeper into the why of the negative impact of earthquakes on the housing prices of apartments. Results from this thesis indicated that, significant at a 10% level, the housing prices of apartments were negatively affected by earthquakes, whereas the housing prices of the other types of houses were not significantly affected by earthquakes. The question arises if the housing prices of apartments might be more vulnerable to the influence of earthquakes then other types of houses. It would be interesting to see whether this result upholds after 2012.
The construction of a variable for the change in the number of households on a neighbourhood or municipality level, and investigating its impact on housing prices would be interesting for further research. In addition, it is very important to further investigate the interaction between earthquakes and demographic population decline. And attempt to find a way to truly isolate these effects if possible.
Furthermore, it is interesting to delve deeper into the impact of media attention on risk perception of earthquakes and the effect thereof on housing prices. For example by constructing a variable measuring media attention towards the earthquakes. Since it is theorized that the impact of earthquakes on housing prices might not be significant before 2013 due to lower risk perception Bosker et al. (2016).
29 Reference
Bosker, M., Garretsen, H., Marlet, G., Ponds, R., Poort, J., Dooren, R. van & Woerkens, C. Van (2016). Met angst en beven. Verklaringen vaoor de dalende huizenprijzen in het Groningse aardbevingsgebied. Atlas voor Gemeenten.
Cameron, A. C., & Trivedi, P. K. (2009). Microeconometrics using Stata (Vol. 5). College Station, TX: Stata press.
CBS (2015). Woningmarktontwikkelingen rondom het Groninger veld. 1e kwartaal 1995 tot en met 2e kwartaal 2015. Den Haag.
Dagblad van het Noorden (2011, July 15). Oudemolen krijgt geen nieuw schoolgebouw. Retrieved from http://www.dvhn.nl/archief/Oudemolen-krijgt-geen-nieuw-schoolgebouw-20690851.html (24-06-2016).
Deng, G., Gan, L., & Hernandez, M.A. (2015). Do natural disasters cause an excessive fear of heights? Evidence from the Wenchuan earthquake. Journal of Urban Economics, 90, 79-89. Durán, N. (2016, February 24). Expected Earthquake Magnitude Variable (eq_mag).
Eurostat (2015). People in the EU – statistics on demographic changes. Retrieved from
http://ec.europa.eu/eurostat/statistics-explained/index.php/People_in_the_EU_%E2%80%93_statistics_on_demographic_changes (24-06-2016).
Francke, M.F. & Kee, J.M. (2013). De waarde ontwikkeling op de woningmarkt in aardbevingsgevoelige gebieden rond het Groningerveld. Ortec Finance.
Francke, M.F. & Kee, J.M. (2014). De invloed van fysieke schade op verkopen van woningen rond het Groningerveld. Ortec Finance.
Gemeente Lelystad (2002). Het is Wold voor en na. Unieke herstructurering in Lelystad. Lelystad: Gemeente Lelystad/Centrada.
Gerrichhauzen, L., & Dogterom, J. (2007). Krimp als kans. Dordrecht: Gerrichauzen en partners.
30
Koster, H.R.A. (2016). Gaswinning, aardbevingen en huizenprijzen. Vrije Universiteit Amsterdam, afdeling Ruimetlijke Economie.
Koster, H.R.A. & Ommeren, J. van (2015). A shaky business: Natural gas extraction, earthquakes and house prices. European Economic Review 80, 120-193.
Lammerts, R. & Dogan, G. (2004). Warkelijk Waor: Onderzoek naar economische en sociale participatie en het voorzieningen gebruik door bewoners van kleine dorpen in
Midden-Drenthe. Utrecht.
Meriam Webter Dictionary (2016). White flight. Retrieved from http://www.merriam-webster.com/dictionary/white%20flight (24-06-2016).
Ministry of Economic Affairs (2013). Brief aan de voorzitter van de Tweede Kamer:
Voortgang onderzoeken inzake gaswinning Groningen. The Hague: DGETM- EM/13118286. NAM (2008). Lichte aardbevingen - als gevolg van het winnen van aardgas. Brochure lichte aardbevingen. Assen. Retrieved from
http://www.nam.nl/content/dam/shell/static/nam/downloads/pdf/brochure-lichteaardbevingen.pdf (24-06-2016).
NAM (2013). Brief aan Staatstoezicht op de Mijnen: Actualisatie seismologische inzichten groningenveld. Assen: EP201301221641.
Rogers, L., & Borsella, P. (2016). Growth or decline: understanding how populations change. United States Census Bureau. Retrieved from
http://census.gov/content/dam/Census/newsroom/releases/2016/cb16-cn43_blog.pdf (24-06-2016).
ROB/Rfv (2008). Bevolkingsdaling. Gevolgen voor bestuur en financiën.
Rosen, S. (1974). Hedonic prices and implicit markets: product differentiation in pure competition. Journal of political economy, 82(1), 34-55.
The Netherlands Institute for Social Research (2007). Het beste van twee werelden: plattelanders over hun leven op het platteland. The Hague.
31 Van Dam, F. (2009). Ruimtelijke Gevolgen. In: N. Van Nimwegen & L. Heering, eds.,
Bevolkingsvraagstukken in Nederland anno 2009. Van Groei naar Krimp. Een Demografische Omslag in Beeld. NIDI rapport nr. 80, pp. 87–112. KNAW Press: Amsterdam.
Van Dam, F. V., de Groot, C., & Verwest, F. (2006). Krimp en ruimte. Bevolkingsafname, ruimtelijke gevolgen en beleid. Rotterdam/Den Haag: NAi uitgevers/Ruimtelijk Planbureau. Van Otterdijk, Y. (2011). Demografische krimp en ziekenhuiszorg. Raad voor de
Volksgezondheid. The Hague.
Verbeek, M. (2012). A guide to modern econometrics. Chichester: John Wiley & Sons.
Voortman, M. (2013, April 28). Meester Croneschool in Oudemolen gaat dicht. RTV Drenthe. Retrieved from http://www.rtvdrenthe.nl/nieuws/73727/Meester-Croneschool-in-Oudemolen-gaat-dicht (24-06-2016).
State Supervision of the Mines - Ministry of Economic Affairs (2013). Brief aan de Tweede Kamer der Staten Generaal: Gaswinning Groningen-veld. The Hague:
DGETM-EM/13010946.
