Vintage effect on house prices in the Netherlands

(1)

Vintage effect on house prices in the

Netherlands

Bachelor Thesis

Michiel van Ommen

10197559

BSc in Economics & Business University of Amsterdam

Thesis supervisor: Dorinth van Dijk

Abstract

This study finds significant evidence that there is a vintage effect in the Netherlands. It is also shown that this vintage effect differs across different levels of urbanization. For this thesis data from the NVM, The Dutch Brokerage Association, has been used to set up a hedonic pricing model to estimate the house price with the use of eleven characteristics. With the use of a dataset from the CBS, The Dutch Central Statistics Office, it is possible to make a distinction between different levels of urbanization. This is achieved by merging the municipality codes from the CBS dataset with the NVM dataset. It is shown that there is a stronger vintage for houses built in the period from 1931 to 1944, this is the case for all levels of urbanization.

(2)

Introduction

For this thesis a research will be done on the vintage effect on house prices in the Netherlands. A vintage effect is a price appreciation over time due to design and preferences, so the characteristics of a house and neighborhood with a certain construction year (Wilhelmsson, 2007).

In October 2015 the ECB published an article about the dynamics of house prices in the Netherlands. Besides these dynamics that the ECB published in their paper there are also other factors that can influence the house prices such as the vintage effect. In October 2015 the Swiss Bank UBS published the Bubble Index, which is designed to track the risk of housing bubbles in global financial centers. Amsterdam is on position 6 of this index and has a bigger chance on decreasing house prices in the future. According to Asabere and Huffman (1991) there is an increasing interest in historic districts, with their research they find a price premium of 131% for these districts.

As far as I am aware of, there is no research done on the vintage effect of the house prices in the Netherlands. The research question for this paper is:

To what extent will the house prices be influenced by the vintage effect on the houses in the Netherlands and is there a difference between different levels of urbanization?

First, it will be researched whether there is a vintage effect on house prices in the Netherlands. Secondly, it will be researched if this vintage effect, when it occurs, is of bigger or smaller interest in cities compared to the countryside, and vice versa. Furthermore in this paper there will be a comparison between the differences over time of the vintage effect. For this research a hedonic pricing model will be used. To measure the vintage effect several characteristics are included in the hedonic pricing model, where the time dummies (the dummy of the construction period) are the most important dummies to measure a vintage effect. The control characteristics are the type of a house, if the house is a monument, the municipality, the size of a house, the size of the garden, the maintenance and if the house has a parking space.

After making four regressions, one about the entire model and three other regressions to make a distinction between three different levels of urbanization it is shown that there is a vintage effect in the Netherlands. This vintage effect has

(3)

approximately the same effect across the three levels of urbanization. However according to the Chow test, these comparable results differ significantly from each other. In all levels of urbanization and in the total model houses built in the period from 1931 to 1944 have the strongest vintage effect. Furthermore it is shown that there is a vintage effect in the Netherlands on houses that are built before 1960. These houses are sold for more compared to other houses with the same

characteristics.

This paper is structured as follows, the first section consists of the literature review, which explains the vintage effect, outcomes of other research done on this subject and an explanation of a hedonic pricing model. Secondly the dataset from the NVM will be described. In the third section the methodology is given for this research and the hypothesis. In section four all the results of this research are shown and in the end the appendixes and references are given.

Literature review

Vintage effect

The definition ‘vintage effect’ varies across existing research. One of the earliest indications of this definition is by Hall (1971) who studied the automobile market and he defines the vintage effect as the reflection of goodness. Randolph (1988) had a different definition of the vintage effect. Randolph studied the housing market where he suggested that the vintage effect is the initial price increase due to quality of location and structural characteristics associated with homes built in a given year.

Randolph’s empirical results indicate that the assumption of stable unobserved quality is likely to be superior to ignoring vintage effects. However, his results also indicate that it may be reasonable to ignore the possibility of a vintage effect if the right measures of structural and neighborhood characteristics are included in the estimated regression. Additionally his results suggest that landlords do not perfectly maintain their rental housing property, Randolph recommends to investigate into the economic determinants of maintenance expenditures and physical depreciation of housing (Randolph, 1988).

(4)

Weis (1994) used for his research a large number of explanatory variables from the 1989 American Housing Survey to come up with locational and structural characteristics in 10 local markets. His results showed a positive age coefficient that is significant at the 5 percent level. This positive and significant age effect is called the vintage effect (Clapp & Giaccotto, 1997).

Holland (2008) says that it is well known that people prefer new built houses instead of older houses and therefore she says in her paper that age is a substitute for quality. However she suggests that the relative short supply of older houses may also mean that year produces a negative price effect. However this could also mean that older houses have a vintage effect, which attract a premium over the standard houses that are build in the recent years in California (Holland, 2008).

In their paper, Rehm, Filippova and Stone (2006) discuss the relationship between a house’s vintage or period of construction and the property value. In their study they use a database from New Zealand. The main source of data for their study is the 1996 official database of all detached or semi-detached residential real estate transactions in New Zealand in which they studied the four largest markets, namely Auckland City, Wellington City, Christchurch City and Dunedin City. For each city, three hedonic regressions were conducted. The models differ only in

specification of house vintage. One model used the most common specification, using a linear, continuous age variable input as year of construction (Rehm, Filippova, & Stone, 2006).

They found a different vintage effect across the four markets; this is given in Figure 1 on the next page. They set the 1990s vintage as the benchmark for their study in New Zealand. The outcome shows that there are premiums and discounts associated with each vintage. For Christchurch and Dunedin there is a sharp discount from 1980 until 1940 and this discount remains steady for Dunedin, but it slightly increased for Christchurch. However, Auckland and Wellington share a similar pattern in terms of vintage effects, with discounts from the 1980s until 1960s and premiums from 1930s through 1880s. Thus the results of a vintage effect are not entirely uniform across markets.

(5)

Figure 1: Vintage effect in four different markets in New Zealand

Furthermore they conclude that one of the three models provides greater explanatory power for vintage effects, namely the linear model which measures vintage in terms of year of construction.

They also find that these premiums exist in the two markets with highest per capita income, so the vintage effect do vary considerably in socioeconomics terms. They offer a hypothesis that the price premiums or discounts associated with a given vintage may relate to the gain or loss of the psychic income, the intangible value that is derived from products (Rehm, Filippova, & Stone, 2006).

In their paper, Coulson and McMillen (2008) suggest that some houses built in a given year age more elegant than others. In their paper they suggest that the depreciation rate of a house may depend on the vintage of each house. According to Coulson and McMillen it is hard to measure a vintage effect within the housing

market due to the high correlation between age and vintage of a house.

Asabere and Huffman (1991) find in their paper that historic districts can produce a positive external benefit. This study has taken place in the city of

Philadelphia and they found that, other influential factors held constant, the effect of historic districting on the value of federally certified parcels are positive. Furthermore they find that historic districting may have differential impacts across different land zones. The price premium they find for residential parcels located within historic districts is 131%. However the price premium for nonresidential parcels within a historic district is insignificant.

(6)

On the other side, Rubin finds within the framework of the implicit pricing model that newer houses generally request a rent premium over older houses. Evidence in his paper suggests that consumers in many markets may have a pure taste for age. However he finds several near to zero coefficients for the effect for age on the house prices (Rubin, 1993).

The outcome of the paper of Coulson and McMillen (2008) suggests that the regression coefficients do not provide accurate measures on separating the effect of age and cohort on house prices; this is due to the extremely high correlation between age and vintage.

Furthermore they find perfect multicollineairity when time of sale is introduced into the model. However when they use a nonparametric estimator and the use of a large sample, the estimator is feasible and generates interesting results. In both cases, nonparametric and within the hedonic pricing model, prices are indicated to decline in the first few years after the construction of the house. Within the

nonparametric model the outcomes impose a discount on houses build in the 1940s relative to homes of close cohorts. The nonparametric results show a U-shape age effects, which means that very old houses trade at a significant premium (Coulson & McMillen, 2008).

Wilhelmsson researches the depreciation rates of house prices depending on the level of maintenance of the property and the location of the property. His

empirical analysis is based on cross-sectional data that included 968 transactions for single-family houses in 2000 of Stockholm.

Wilhelmsson finds that the depreciation rates for maintained properties are significantly lower than for properties that are not maintained. However he does not find a variation in depreciation rates for houses that vary in space in the same market (Wilhelmsson, 2008).

Hedonic Pricing Model

Rehm, Fillipova and Stone (2006) used three different hedonic models to conduct the vintage effects in the four largest markets of New Zealand. The most common model, which is a linear model, uses a linear, continuous age variable input as year of construction. The complete model they come up with is:

(7)

𝑃 𝑿 = 𝑃(𝑳, 𝑺, 𝑵, 𝑨, 𝑬, 𝑪, 𝑽), where P(X) is the house price function and this function maps the following characteristics: L is the characteristic of land (site area), S is the structural area (floor area), N is the neighborhood characteristic (per capita income), A is the accessibility (the distance to the central business district), E is the

environmental externality (water view and the quality of surrounding improvements), C is the physical condition (condition of interior and exterior) and V is the vintage (year of construction).

The hedonic model used in the paper of Asabere and Huffman (1991) is different compared to the one of Rehm, Fillipova and Stone (2006). Asabere and Huffman are using a hedonic model that includes the dummy variable Historic District that has the value 1 if the location is within a historic district and 0 is the location is outside a historic district. Also in this model they use the log price transformation for estimating the price effect. Furthermore they include four different zoning types such as residential, commercial, industrial and others. They also include variables for location such as Neighborhood, Time of sale, and Lot Size (Asabere & Huffman, 1991).

The hedonic price model used in the paper of Wilhemsson (2008) also uses a log price transformation in his hedonic price model. Furthermore he uses the variable Maintenance as an interaction term to allow houses in different conditions to have different depreciation rates. Also in his paper he includes the variable Distance that measures the distance from the central business district.

However there is heteroscedasticity induced by the dwelling age in hedonic house price equations. This will be the case for repeat sales equations. This is due the fact that the property must be older to be held longer. Another fact is that buyers and sellers frequently improve their home at the time of sale. The changes to a house will be less extensive when the time between buying and selling is shorter (Goodman & Thibodeau, 1996).

Also Randolph uses a hedonic price model to estimate the house prices, he includes several characteristics that indicate house prices. The variables he uses are distance to work, crime rate, median income, basement, parking space and the number of persons per household (Randolph, 1988).

Randolph estimates the vintage effect on house prices in his paper from 1986. In his model he includes indicators such as neighborhood and building type. Building type should matter according Randolph since the economics and technology of

(8)

maintenance behavior are more likely to diversify across building types. Also neighborhood is important, because the value of maintenance is likely to vary with neighborhood quality. In order to measure the neighborhood characteristics

Randolph uses the variables Median Income and Number of People. Furthermore to measure the vintage effect Randolph modeled the following equation: ℎ_!" = 𝑎 ∙ 𝑣_! + 𝜃_!, where ℎ_!" is unit j’s unmeasured quality at the time it is built. ‘a’ is the vintage effect, 𝑣! is the year of construction and 𝜃! is a random disturbance (Randolph,

1988).

Data

For this research the data that is used is made available via the NVM, which is The Dutch Brokerage Association. With this data it is possible to make a hedonic price model because of the different characteristics and home sales information from transaction in the Netherlands. Furthermore the timeframe is from 1992 until October 2014, within this period a regression will be done to measure if there is a vintage effect on house prices in the Netherlands.

The raw dataset contains 2,370,226 observations of housing transactions in the Netherlands. After taking out the errors the pure dataset contains 1,636,973 observations. The following Table 1 give all the information about the variables that are included in this model.

Table 1: All information about the variables from the total regression

Variable Description Mean Type

Log (Price) Log transformation of transaction price

12.1416 Numerical

Sales Sale transactions per

period 78,215 Numerical

Log (Parcel) Log transformation of the size of the parcel in m2

5.3634 Numerical Log (Size) Log transformation of

the size of the house in cubic meters

5.9226 Numerical Log (Garden) Log transformation of

the size of the garden in m2

(9)

Rooms Indicates the number of rooms in the house

4.77 Numerical

Time Date of transaction n/a Date

Monument Indicates if the house is a monument (0 or 1)

n/a Dummy

Neighborhood Indicates where the

house is located n/a Dummy

Building period Indicates when the

house was built n/a Categorical

Type of building Indicates what type of building the house is

n/a Categorical

Urbanization level

Indicates the level of urbanization (1-5)

n/a Categorical

Maintenance inside

Indicates the level of maintenance of the inside (1-9)

7.00 Categorical

Maintenance outside

Indicates the level of maintenance of the outside (1-9)

7.02 Categorical

Tables 8-13, given in the appendixes, give all the information about the possible dummy variables. The characteristics building period consists of nine variables that indicate in which period the house has been built, this information is given in Table 8. Furthermore, Table 9 indicates the amount of transactions per year. The dummy variable for municipality is merged with the dataset of the CBS to

calculate the level of urbanization that is shown is Table 10. The dataset of the CBS contains of five levels of urbanization. In this research there will be three levels of urbanizations used, namely a high, medium and a low level of urbanization. These three levels consist of the combination of the two highest levels of urbanization, a medium level and the combination of the two lowest levels of urbanization. The dummy variable monument indicates when the house is monument, the amount of observations are given in Table 11. The characteristic for the type of house consists of 16 dummy variables that indicate what type of building the house is and this is shown in Table 12. In Table 13 the possible dummy variables for the characteristic parking are given, this variable consists of 6 dummy variables.

This dataset contains 458 municipalities, which corresponds to the municipal division of 2006 from the CBS. Big cities have the following codes; Amsterdam has code 0363, Rotterdam has code 0599, Den Haag has code 0518 and Groningen has

(10)

code 0014 (CBS, 2006). With the use of this information it is possible to make a distinction between a city and the countryside, the urbanization level that can have level 1 to 5. Where level 1 is the highest level of urbanization and level 5 is the lowest level of urbanization.

Methodology

The model

When old houses are sold for more compared to other houses from another period with the same characteristics, there is a vintage effect. Also with the use of the municipality codes it is possible to investigate if in different levels of urbanization the vintage effect is stronger. Thus it is first researched if there is a vintage effect in older building periods compared to other building periods and there will be made a

distinction between three levels of urbanization and it is researched whether a house is sold for a higher price due to the urbanization level where the house is in.

For this research a hedonic price model will be used to estimate the price effect. This model is a statistical model, which compares the transactions with the included variables. All variables, as discussed in the previous section, will be part of the hedonic price model.

As discussed in the literature review, Randolph includes the type of building into his hedonic pricing model (Randolph, 1988). Furthermore Rehm et al. (2006) includes the characteristics maintenance into their hedonic pricing model. Besides this they include the characteristic floor that is comparable with the size of the house in cubic meters in this research. Further they include the building period, the vintage of a house, as a characteristic (Rehm, Filippova, & Stone, 2006). These variables are the control variables that will also be included in this hedonic pricing model for the houses in the Netherlands.

For this research a log transformation of the price is done to make it easier to see the percentage effect of the variables on the price of a house. With a log

transformation the marginal changes in the explanatory variables are interpreted in terms of percentage change in the dependent variable.

(11)

𝐿𝑜𝑔 𝑃𝑟𝑖𝑐𝑒_! = 𝛽_! + 𝛽_!!∗ 𝑇𝐼𝑀𝐸 _!+ 𝛽_!!∗ NEIGHB _!+ 𝛽_!!∗ 𝐵𝑈𝐼𝐿𝐷_!+ 𝛽_! ∗ 𝑃𝑁𝑀 + 𝛽_! ∗ 𝐿𝑜𝑔 𝑆𝑖𝑧𝑒 + 𝛽_!!∗ 𝑇𝑌𝑃𝐸_! + 𝛽_!∗ 𝐿𝑜𝑔 𝐺𝐴𝑅𝐷𝐸𝑁 + 𝛽_!∗ 𝑅𝑂𝑂𝑀𝑆 + 𝛽_!! ∗ 𝑀𝑎𝑖𝑛𝑡𝑒𝑛𝑎𝑛𝑐𝑒 𝐼𝑁𝑆𝐼𝐷𝐸 !+ 𝛽!"!∗ 𝑃𝐴𝑅𝐾𝐼𝑁𝐺!+ 𝜀!

With this hedonic price model all the available characteristics are included to investigate the effect on house prices. With the use of an OLS regression it is

possible to investigate the effect of the year of construction on the price, the vintage effect. A distinction of the vintage effect between the urbanization levels is done via the municipality data, which is based on the merging of the two data sets from the CBS and NVM.

Hypothesis

𝐻_!: 𝑉𝑖𝑛𝑡𝑎𝑔𝑒 𝑒𝑓𝑓𝑒𝑐𝑡 = 0 𝐻_!: 𝑉𝑖𝑛𝑡𝑎𝑔𝑒 𝑒𝑓𝑓𝑒𝑐𝑡 ≠ 0

Based on the papers of Rehm et al. (2006) and Hall (1971) it would be expected that there is a positive vintage effect in the Netherlands. Furthermore there is an

expectation that this effect will be stronger in cities compared to the countryside, because there will be a higher density of older buildings and a higher supply and demand side on the housing market.

Results

In Table 2 the results from the total regression are given, the omitted variables are the base variables. With this base variable you compare the betas from the

regression with the base variable. The regression has a high 𝑅!_{of 0,88, which}

indicates that 88% of the variation is explained by the variables within the model. In Table 2 the regression indicates that almost all the variables are significant at the 1% significance level. However the house type multiple level apartments is insignificant. Also in Table 2 there is indicated that the time dummies and the zip code dummies, which indicate the municipality, are significant in the partial F-test.

(12)

Table 2: Total regression without a distinction of urbanization levels

Variable Beta t-statistic

Built before 1905 Built 1906-1930 Built 1931-1944 Built 1945-1959 Built 1960-1970 Built 1971-1980 Built 1981-1990 Built 1991-2000 Built after 2001 Monument Log Size HT Simple HT Single-family HT Canal house HT Mansion HT Living Farm HT Bungalow HT Villa HT Manor HT Estate

HT Ground floor app. HT Top floor app. HT Multiple level app. HT app. w/porch HT app. w/gallery HT Nursing home

HT Top and ground floor app. Log Garden

Number of rooms

IS very poor maintenance

IS very poor to poor maintenance IS poor maintenance

IS poor to average maintenance IS average maintenance

IS average to good maintenance IS good maintenance

IS good to excellent maintenance IS excellent maintenance

No parking Parking Carport Garage

Garage and carport Garage for more cars

0.061*** 0.039*** 0.070*** 0.012*** -0.018*** (omitted) 0.054*** 0.120*** 0.141*** 0.092*** 0.578*** -0.063*** (omitted) 0.190*** 0.110*** 0.079*** 0.192*** 0.249*** 0.257*** 0.195*** -0.117*** -0.025*** 0.000 0.083*** 0.129*** -0.325*** -0.075*** 0.082*** 0.010*** -0.299*** -0.283*** -0.188*** -0.177*** -0.119*** -0.089*** (omitted) 0.058*** 0.062*** (omitted) 0.049*** 0.069*** 0.118*** 0.140*** 0.138*** 64.36 56.46 95.68 16.40 -32.10 (omitted) 98.03 197.04 147.98 43.82 737.22 -74.77 (omitted) 55.97 201.35 30.36 198.47 253.20 109.93 5.27 -65.16 -6.37 0.02 19.44 19.22 -6.60 -10.87 352.18 72.83 -99.06 -56.24 -148.73 -74.72 -207.34 -105.65 (omitted) 56.72 130.26 (omitted) 70.23 91.24 308.87 127.93 151.58

(13)

𝑅! !"# RMSE Observations 0.8818 0 .14717 1,098,318 Partial F Time dummy (monthly)

Zip code dummy

3784.40*** 629.00***

HT = House type, app. = apartment, IS = inside, OS = outside, *** = significant at 1%

Table 2 indicates that there is a house price vintage effect in the Netherlands. As can been seen in Table 2 the dummy variables for time, the building period, are for all, except for houses built between 1960 and 1970, positive. This indicates that houses built in all the building periods except for 1960 until 1970 have a positive effect on the house prices compared to houses built in 1971 till 1980.

The vintage effect can been seen by the fact that houses built before 1960 have a positive effect compared to houses built in the period 1971 to 1980, which means that these houses are sold for more compared to houses built in the period from 1971 to 1980 with the same characteristics. The strongest vintage effect is for houses built in the period from 1931 to 1944.

Also Table 2 indicates that a villa and a manor have the highest impact on the transaction price, an increase of 25% compared to a single-family house. Also

maintenance have an effect as expected, from good maintenance onwards it has a positive effect. Further having a parking space, garden and more rooms positively affect the transaction price.

In short, there is a vintage effect in the Netherlands and this effect is the strongest for houses built in the period from 1931 to 1944. These houses are sold for more, which is the vintage effect.

The following tables below, Table 3 to Table 5, show the difference between the regressions done between the different levels of urbanization. In these

regressions there has been made a rearrangement between the five levels of urbanization as distinguished by the CBS. The first group, the highest level of urbanization consists of the two highest levels of urbanization (level 1 and 2). This highest level of urbanization and the results are given in Table 3. Thereafter the medium level of urbanization will be shown in a table, this medium level consists of the urbanization level 3. The medium level of urbanization and the outcomes from the regression are given in Table 4. And finally the lowest level of urbanization is

(14)

regressed, this group consist if the two lowest levels of urbanizations (level 4 and 5). This low level of urbanization and the outcomes from the regression are given in Table 5 below.

Table 3: Regression with a high urbanization level

Number of rooms

IS good to excellent maintenance IS excellent maintenance No parking Parking Carport 0.043*** 0.021*** 0.048*** -0.004*** -0.037*** (omitted) 0.041*** 0.104*** 0.126*** 0.087*** 0.598*** -0.066*** (omitted) 0.190*** 0.114*** 0.132*** 0.200*** 0.257*** 0.295*** -0.012 -0.116*** -0.032*** 0.003 0.057*** 0.117*** 0.007 -0.070*** 0.081*** 0.012*** -0.295*** -0.284*** -0.191*** -0.177*** -0.120*** -0.092*** (omitted) 0.059*** 0.065*** (omitted) 0.046*** 0.061*** 29.79 18.68 42.11 -2.79 -34.78 (omitted) 43.77 97.16 80.73 31.88 521.83 -54.55 (omitted) 47.83 150.19 17.13 116.71 169.13 50.26 -0.18 -61.41 -7.79 0.57 11.51 13.56 0.05 -10.19 223.11 60.32 -68.48 -43.17 -109.76 -57.77 -148.72 -79.04 (omitted) 45.53 97.26 (omitted) 46.30 50.37

(15)

Garage

0.108*** 0.150*** 0.124*** 178.15 62.61 78.62 𝑅! !"# RMSE Observations 0.8929 0.14355 504,403 Partial F Time dummy (monthly)

Zip code dummy

2060.03*** 913.39***

In the output for a high level of urbanization, as shown in Table 3, almost all the variables are significant at the 1% significance level. Only three house types, estate, multiple level apartment and nursing home are insignificant. Also in Table 3 it is indicated that the dummy variables for time and the zip code are significant, which is tested with the partial F-test.

As is shown in Table 3 above houses built before 1971 first decrease in price but before 1960 it starts to increase compared to houses built in the period from 1971 to 1980. These houses that are built before 1960 are sold for more, compared to houses built between 1971 and 1980, which is the vintage effect. The vintage effect is the strongest for houses built in the period between 1931 and 1944 in areas with a high level of urbanization.

Also in an area with a high level of urbanization a villa and a manor have a higher premium on house prices compared to single-family houses. But a manor has a higher premium than a villa, 25.7% and 29.5% respectively. Also maintenance have an effect as expected, from good maintenance onwards it has a positive effect on the transaction price. Further a parking space, garden and more rooms are as expected, they have a positive effect on the transaction price.

In short there is a vintage effect in the Netherlands in areas where there is a high level of urbanization and this vintage effect is the strongest for houses built in the period from 1931 to 1944.

(16)

Table 4: Regression with a medium urbanization level

Number of rooms

0.068*** 0.046*** 0.087*** 0.016*** -0.012*** (omitted) 0.053*** 0.128*** 0.158*** 0.077*** 0.599*** -0.060*** (omitted) 0.140*** 0.099*** 0.080*** 0.199*** 0.237*** 0.274*** 0.252*** 0.048*** 0.065*** 0.009 0.195*** 0.175*** -0.485*** -0.058 0.080*** 0.008*** -0.256*** -0.253*** -0.167*** -0.163*** -0.111*** -0.084*** (omitted) 0.054*** 0.060*** (omitted) 0.041*** 0.068*** 0.122*** 0.145*** 0.142*** 32.86 35.08 63.35 12.10 -11.73 (omitted) 52.56 114.64 92.74 16.09 379.53 -34.91 (omitted) 16.64 94.98 12.51 107.90 140.85 58.72 3.95 7.63 5.28 0.70 23.03 16.04 -8.36 -1.29 179.47 30.06 -43.41 -22.31 -66.43 -31.17 -98.12 -48.00 (omitted) 25.86 66.77 (omitted) 33.18 50.27 174.70 69.53 81.96

(17)

𝑅! !"# RMSE Observations 0.8883 0.14144 281,111 Partial F Time dummies (monthly)

Zip code dummies

983.36*** 972.31***

In Table 4 the output of the regression of a medium level of urbanization is given. Also in this table almost all the variables are significant at a significance level of 1%. There are two house types that are insignificant; these two are the multiple level apartments and the top and ground floor apartments. Furthermore it is given that the dummy variables for time and zip codes are significant within the partial F-test. Also in this regression the 𝑅!_{is high, 0.89.}

In Table 4 it is shown that there is a vintage effect for houses that are located in in an area with a medium urbanization level. Houses built before 1960 will receive a premium compared to houses built from 1960 to 1980. This is the vintage effect and this effect is the strongest for houses built in the period from 1931 to 1944.

Also in Table 4, houses in an area with a medium level of urbanization, it is shown that that house types villa, manor and estate will receive a higher premium on the transaction price compared to single-family houses. Also the variable

maintenance is as expected and will have a positive effect on the transaction price from good onwards. Further a parking space, garden and more rooms are as expected, they have a positive effect on the transaction price.

In short there is a vintage effect in the Netherlands in areas with a medium level of urbanization and also in these areas houses built in the period from 1931 to 1944 have the strongest vintage effect.

Table 5: Regression with a low urbanization level

Built before 1905 Built 1906-1930 Built 1931-1944 Built 1945-1959 Built 1960-1970 Built 1971-1980 Built 1981-1990 0.067*** 0.046*** 0.076*** 0.015*** -0.014*** (omitted) 0.062*** 39.18 35.35 47.96 11.83 -14.68 (omitted) 66.52

(18)

Built 1991-2000 Built after 2001 Monument Log Size HT Simple HT Single-family HT Canal house HT Mansion HT Living Farm HT Bungalow HT Villa HT Manor HT Estate

Number of rooms

0.125*** 0.142*** 0.103*** 0.535*** -0.060*** (omitted) 0.185*** 0.100*** 0.083*** 0.181*** 0.251*** 0.247*** 0.315*** 0.057*** 0.108*** 0.041** 0.122*** 0.135*** -0.027 -0.020 0.085*** 0.007*** -0.348*** -0.309*** -0.201*** -0.193*** -0.123*** -0.089*** (omitted) 0.055*** 0.060*** (omitted) 0.064*** 0.082*** 0.124*** 0.140*** 0.152*** 122.37 79.43 23.52 368.61 -38.04 (omitted) 19.47 87.14 25.52 117.55 126.34 78.51 4.96 5.76 6.57 2.41 7.14 5.05 -0.25 -0.46 204.71 27.95 -58.43 -29.73 -78.71 -37.84 -109.30 -52.74 (omitted) 23.68 60.04 (omitted) 40.56 57.39 180.06 88.95 102.07 𝑅! ! ! RMSE Observations 0.8593 0.15493 312,804 Partial F Time dummies (monthly)

Zip code dummies

852.41*** 315.48***

In Table 5 the output is given from the regression within the area with a low level of urbanization. Almost all the variables are significant at a 1% significance

(19)

level, however the multiple level apartments variable is significant at 5%. The house types nursing home and top and ground floor apartment are insignificant. However the dummy variables for time and zip codes are significant at a 1% significance level for the partial F-test and also the 𝑅!_{is high in this regression, with a value of 0.85.}

In areas with a low urbanization level there also exists a vintage effect on house prices. Houses built before 1960 will receive a premium compared to houses built in the period from 1971 to 1980. This vintage effect is the strongest for houses built in the period from 1931 to 1944.

The house type Estate receives the highest premium on the transaction price compared to a single-family house. If the house type is an estate it will receive a 31.5% higher price compared to a single-family house. Also the variable

maintenance is as expected, from good maintenance onwards it will have a positive effect on the transaction. Further the other variables are as expected, a parking space, garden and more rooms have a positive effect on the transaction price.

In short, there exist a vintage effect in areas with a low urbanization level in the Netherlands. In these areas houses built in the period from 1931 to 1944 have the strongest vintage effect.

In Table 6 below a summary is given from the vintage effect of the total regression and from the three different levels of urbanization in the Netherlands.

Table 6: Summary vintage effect in the Netherlands

Variable Overall effect High urbanization level beta Medium urbanization level beta Low urbanization level beta Built before 1905 Built 1906-1930 Built 1931-1944 Built 1945-1959 Built 1960-1970 Built 1971-1980 Built 1981-1990 Built 1991-2000 Built after 2001 0.061 0.039 0.070 0.012 -0.018 (omitted) 0.054 0.120 0.141 0.043 0.021 0.048 -0.004 -0.037 (omitted) 0.041 0.104 0.126 0.068 0.046 0.087 0.016 -0.012 (omitted) 0.053 0.128 0.158 0.067 0.046 0.076 0.015 -0.014 (omitted) 0.062 0.125 0.142

(20)

Figure 2: Vintage effect in the Netherlands with distinction of urbanization

As is shown in Figure 2 the vintage effect in the different levels of urbanization are more or less the same. The black line indicates the overall effect of the vintage effect in the Netherlands. This line has a U-shape. This U-shape effect is also indicated in the paper of Coulson & McMillen (2008), where their results show a U-shape age effect, which means that very old houses trade at a significant premium.

Because the vintage effect in all the four regressions are more or less the same level and in all the four regression the vintage effect is the strongest in the period from 1931 to 1944 a cross regression F-test is performed. With this F-test it is possible to make a distinction if the outcomes between the three levels of

urbanization are different from each other or not. 𝐹(!,!!!!!!!!)=

[𝑆𝑆𝐸_!− (𝑆𝑆𝐸_!+ 𝑆𝑆𝐸_!)]/𝑘 (𝑆𝑆𝐸_!+ 𝑆𝑆𝐸_!)/(𝑁_!+ 𝑁_!− 2𝑘) 𝑆𝑆𝐸_!: Sum of squared error term for pooled model 𝑆𝑆𝐸_!: Sum of squared error term for group 1 𝑆𝑆𝐸!: Sum of squared error term for group 2

k: # of estimated parameters (including constant) N: # of observations per group

-0,05 0 0,05 0,1 0,15 0,2 Be fo re 1 90 5 19 06 -1 93 0 19 31 -1 94 4 19 45 -1 95 9 19 60 -1 97 0 19 71 -1 98 0 19 81 -1 99 0 19 91 -2 00 0 Af te r 2 00 1 Overall effect High urbanization level Medium urbanization level Low urbanization level

(21)

𝑆𝑆𝐸_!"#!: 10,367.136 𝑁_!"#!: 504,403 𝑆𝑆𝐸_!"#$%&: 5,604.887 𝑁_!"#$%&: 281,111 𝑆𝑆𝐸!"#: 7,452.858 𝑁!"#: 312,804 𝑆𝑆𝐸_!"#$%: 23,700.998 𝑁_!"#$%: 1,098,318 k: 12 𝐹_!"#!/!"#$%& =644.081 0.02 = 32,204.06 𝐹_!"#$%&/!"# = 886.938 0.02 = 44,346.90 𝐹_!"#/!"#! =490.084 0.02 = 24,504.20

All of the F-values are outside the critical region (F>1.752), so the values between the different levels of urbanization are significantly different. There is a difference between the three different levels of urbanization.

Conclusion

This thesis has shown that there is a vintage effect on house prices in the Netherlands, however this vintage effect differ across the period when the house was built. On the overall effect in the Netherlands there is a vintage effect on house prices where the strongest vintage effect is measured in the construction period from 1931 to 1944.

However there can be made a distinction between the levels of urbanization, in this thesis three different levels of urbanization are used. From a high level of urbanization, followed by a medium level, and then the lowest level of urbanization in the Netherlands. Between these three levels of urbanization there exist different levels of a vintage effect, which is proved by the Chow test.

First, areas with the highest level of urbanization experience a preference for new built houses. However there exist a vintage effect in house prices in these areas. Houses built before 1971 first decrease in price but when a house is built before 1960 it starts to increase compared to houses built in the period from 1971 to 1980. These houses that are built before 1960 are sold for more, compared to houses built

(22)

between 1971 and 1980. The vintage effect is the strongest for houses built in the period between 1931 and 1944.

Secondly, in areas with a medium level of urbanization there also exist a vintage effect on house prices. Houses built before 1960 will receive a premium compared to houses built from 1960 to 1980. In areas with a medium level of

urbanization the vintage effect is strongest for houses built in the period from 1931 to 1944.

At last, in areas with a low urbanization level there also exists a vintage effect on house prices. Houses built before 1960 will receive a premium compared to houses built in the period from 1971 to 1980. Also in these areas the vintage effect is strongest for houses built in the period from 1931 to 1944.

Recommendations

For future study about the vintage effect in the Netherlands I would recommend to investigate to what extent the differences of a vintage effect between the urbanization levels are influenced by the income per capita of the respective area. Also Rehm et al. (2006) give a possible explanation for the differences in the vintage effect across cities they investigated. The reason for the differences between a premium or a discount is due to the fact that there are differences between the income per capita. Because of that a comparable research about the vintage effect could be more precise with the introduction of the income per capita into its model.

A possible drawback is the indication of a monument; according to the dataset there are 9,234 monuments within the transactions. But there are over 77 thousand transactions from houses that are built in the period from 1500 to 1905.

(23)

Appendixes

Table 8: Building periods with the number of observations

Code Building period Number of observations

1 1500-1905 77,921 2 1906-1930 190,925 3 1931-1944 138,393 4 1945-1959 112,865 5 1960-1970 230,319 6 1971-1980 323,015 7 1981-1990 278,061 8 1991-2000 233,093 9 >2001 52,381

Table 9: Amount of transactions each year

Year Number of transactions 1992 33,426 1993 36,866 1994 36,568 1995 50,380 1996 57,813 1997 66,904 1998 80,661 1999 81,330 2000 82,078 2001 94,309 2002 94,437 2003 93,341 2004 93,443 2005 104,317 2006 106,781 2007 100,636 2008 81,790 2009 60,822 2010 62,647 2011 56,684 2012 59,204 2013 56,331 2014 46,205

(24)

Table 10: Levels of urbanization

Urbanization level Explanation Number of observations

1 High level 194,956

2 Medium to high level 529,398

3 Medium level 407,732

4 Medium to low level 345,593

5 Low level 159,294

Table 11: Number of observations with dummy variable Monument

Monument Number of observations Yes (=1) 9,234

No (=0) 1,627,739

Table 12: Information about the types of building

Type of building Number of observations Type code

Simple house 75,386 2 Single-family house 1,134,819 5 Canal house 3,509 6 Mansion 195,037 7 Living Farm 11,547 8 Bungalow 63,099 9 Villa 65,370 10 Manor 14,638 11 Estate 110 12

Ground floor app. 12,075 21

Top floor app. 20,531 22

Multiple level app. 5,290 23

App. w/porch 21,039 24

App. w/gallery 13,362 25

Nursing home 359 26

Top and ground floor app. 802 27

Table 13: Information about dummy variable Parking

Parking Explanation Number of observations

0 No parking 910,000

2 Parking 71,837

3 Carport 64,804

4 Garage 500,000

6 Garage and carport 32,901

(25)

References

Asabere, P. K., & Huffman, F. E. (1991). Historic Districts and Land Values. The Journal of Real estate Research , 1 (6), 1-7.

CBS. (2006, Januari 1). CBS. Opgeroepen op Januari 6, 2016, van Centraal Bureau voor de Statistiek.

Clapp, J. M., & Giaccotto, C. (1997). Residential Hedonic Models: A Rational Expectations Approach to Age Effects. Journal of Urban Economics (44), 415-437. Coulson, E. N., & McMillen, D. P. (2008). Estimating time, age and vintage effects in housing prices. Journal of Housing Economics , 2 (17), 138-151.

Goodman, A. C., & Thibodeau, T. G. (1996). Dwelling Age Heteroskedasticity in Repeat Sales House Price Equations. Real Estate Economics , 1 (26), 151-171. Hall, R. E. (1971). The measurement of quality change from vintage price data. Price indexes and quality change , 240-271.

Holland, S. C. (2008). Hedonic Modeling of the Tucson Housing Market: The Effect of Educational Submarkets on House Prices. In S. C. Holland, Hedonic Modeling of the Tucson Housing Market: The Effect of Educational Submarkets on House Prices (pp. 145-229). Michigan: ProQuest LLC.

Randolph, W. C. (1988). Estimation of Housing Depreciation: Short-Term Quality Change and Long-Term Vintage Effects. Journal of Urban Economics (3), 162-178. Rehm, M., Filippova, O., & Stone, J. (2006). The influence of vintage on house value. Pacific Rim Property Research Journal , 3 (12), 232-253.

Rubin, G. M. (1993). Is Housing Age a Commodity? Hedonic Price Estimates of Unit Age. Journal of Housing Research , 1 (4), 165-184.

Wilhelmsson, M. (2008). House price depreciation rates and level of maintenance. Journal of Housing Economics , 1 (17), 88-101.