Rationality check on the Dutch housing market : an empirical study

(1)

Rationality check on the Dutch housing

market - An empirical study

Master thesis

August 15, 2017

Mariann Lesk´

o

11386371

mariann.lesko@student.uva.nl

Supervisor:

dr. Andr´

as P´

eter Kiss

Faculty:

Economics and Business

Programme:

MSc Economics

(2)

Abstract

This thesis analyses the ground lease system on the Amsterdam housing market from a time-discounting point of view. The new ground lease system allows for perpetual leasehold, bought out in one lump sum. Hedonic price regressions are estimated on asking prices, determining the price effect of ground lease compared to full ownership, and the effect of years paid in advance of ground rent. With an own collected cross-sectional dataset, apartments on ground lease are 6-9 percent cheaper than full ownership. Among ground lease apartments, the paid off period has no price effect. This is explained partly by the low share of lump sum compared to price, but also resembles of myopic and amnesiac behaviour. This calls for a policy of providing further information to citizens on the effects of buying out the lease for long periods.

Statement of Originality

This document is written by Student Mariann Lesk´o who declares to take full

responsi-bility for the contents of this document.

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in cre-ating it.

The Faculty of Economics and Business is responsible solely for the supervision of com-pletion of the work, not for the contents.

Acknowledgements

I thank my supervisor dr. Andr´as P´eter Kiss for his help and suggestions for improvement

and time he dedicated for me. I am grateful for the programming book recommendations for Luuc van der Zee.

(3)

1 Introduction

In some cities within the Netherlands there is a special regulation on housing market which is recently under debate to change. The municipalities of these cities (including Amsterdam and Den Haag (VEH, 2017)) do not sell the land, but lease it to house owners

through the constitution of ground lease (erfpacht in Dutch (van Weeren, 2012)). This

explicit differentiation of land and house ownership means that the house owner pays an annual rent for the right of usage of the land. However, it is also possible to pay the amount in advance for a long period, eg. for 50 years in Amsterdam. As a result, when selling the house, it is common that the paid period is not expired, and the house has

some years free of paying this amount (canon). There is recent change in the system

allowing for infinite paid off period. I question this regulation from a time-discounting point of view.

Time-discounting is an important part of assumptions when we analyse and set policies on markets including time dimension. However, it is not rare for consumer preferences to exhibit some time-inconsistency which can lead to market imperfections, sometimes to the exploitation of consumers. One of the most common example of this bounded rationality is hyperbolic discounting, where consumers have a present bias in their decisions. Thus, price-setting firms can benefit from consumers’ non-optimal choices when offering goods where the time of purchasing and consuming is different (DellaVigna & Malmendier, 2004). The possibility of having limited capacity over calculating with time periods raises questions regarding long-term investment decisions of inexperienced consumers, such as the decision of buying or selling a house.

In this thesis, I empirically analyse the Amsterdam housing market with my own col-lected dataset and would like to contribute to the debate from a new perspective. I would like to answer the question whether people discount future payments rationally at their long-term investment of buying a house. I (implicitly) assume that if people do not dis-count future rationally, then this bias appears in offered real estate asking prices. I choose Amsterdam to have a roughly comparable market, where the above mentioned regulation still exists uniformly.

(5)

I have three expectations regarding ground lease and the paid off ground lease period. I expect that (1) ground lease has a negative price effect compared to full ownership of the land. This is motivated by the restrained rights of ground lease. Furthermore, I expect that (2) there is a positive effect of the ground rent-free period on house prices which is in line with rational behaviour. If future fees’ present values are taken into account, then the longer free period belongs to a house, the higher the value of the house is. Nevertheless, I also expect that (3) this positive effect stops growing over time. That is, after a number of years, people can not distinguish the effect of free periods of varying length. This would mean that their discounting is time-inconsistent, thus not completely rational.

Using the hedonic price regression method, my calculations show that ground lease has indeed a negative effect on residential real estate prices. However, the within ground lease estimations for hypotheses 2 and 3 do not show a significant positive price effect of any length of the paid off period, and thus hypothesis 3 cannot be tested. The results show that legal restrictions of not having full ownership rights have a 6 to 9 percent weight in prices. They suggest that usual price-controlling characteristics are more important and possibly inconsistent time-discounting is not an efficiency problem in this market. A second interpretation could call attention for possible amnesia of past information, but this is less plausible regarding the nature of the market.

The rest of the thesis is structured as the following. Section 2 gives an overview of the nature and importance of Amsterdam housing market, and explains ground lease regulations, including the current debate on the subject. Section 3 covers the methodology of hedonic price regression, and a brief literature review on inconsistent time-discounting and includes the details of the models for the hypotheses. Section 4 introduces the used data and its collection method. The next section shows the results, then gives details on the main determinants of apartment prices. The last section summarizes the thesis, draws policy recommendations based on the results and outlines directions for further research.

(6)

2 Amsterdam housing market

2.1 General overview

Amsterdam is one of the most important cities in the Netherlands, a hub in North-Holland as the country’s capital. The municipality, Amsterdam and surrounding ar-eas, has 853 180 inhabitants, with a growing number of individual households (Statistics Netherlands, 2016). However, the city has been experiencing scarcity in terms of accom-modation places in the last decades, resulting continuously increasing house prices (CBS StatLine, 2017).

Amsterdam is also a focal city to international investors, especially after the Brexit vote outcome in the United Kingdom. The dense agglomeration where there is still growth makes the Dutch housing market a relatively safe option to invest in (Phillips, Roberts, & Watson, 2017). Housing rents are expected to grow further due to the increased demand for offices all over the city - partly form the relocating companies from London, partly from the growing number of technology companies and other occupiers. This has a further effect on the residential properties on sale as well, thus, the real estate market is expected to stay strong over the upcoming years (Cushman & Wakefield, 2017).

The regulation of the housing market is highly developed. When one considers buying an apartment, there are several options for ownership situation. Even when the desired apartment is offered for sale, chances are that the new buyer cannot own entirely the whole property and the land or parcel it lies on. The simplest method is full ownership, when the buyer purchases both the house and the land or parcel it lies on. The second option is the topic of this paper: ground lease. Here, the buyer can only purchase the house, but not the parcel, and has to pay periodically a fee to the owner, the local municipality for the right of using the parcel, thus, the municipality leases the land for house owners. The final option that one encounters when considering buying a real estate property is to only buy a membership right from a co-operative residential association for using an apartment. In this case, the property is owned by the association, and the bought right entitles the owner to use the property and common areas such as stairways (DGB Adviesgroep, 2017), therefore the resulting rights are more restricted than full ownership and ground lease.

(7)

When the building is split into apartments that can be sold separately, a union of owners must be established. There ownership associations (Vereniging van Eigenaars, VvE) must be registered at the Chamber of Commerce (Kamer van Koophandel, KvK) if they are active. Since 2008, all ownership associations are required to have a separate funding for maintenance works (VROM, 2008), they often provide members with a maintenance plan for the upcoming years. The operation of these associations vary widely hence the lack of strict regulation for them. They may ask for a monthly contribution for their services and for the maintenance fund, organize a common insurance for the building and require a regular meeting (montly, yearly) for the members. The more information is given about the ownership situation, and the more active it is, the more certainty is provided when selling the property (or property user rights).

Owning a house is frequently not a goal among the Dutch, the share of renting the

place of residence is historically high. Although according to Su´arez and Engelberts

(2005) there was a shift towards owner occupied properties from rental properties in the number of building permits granted, the share of rental building permits does not decrease in Amsterdam based on the most recent data available (CBS StatLine, 2014). On the contrary, the government sets incentives towards owning a house, with the National Mort-gage Guarantee, and tax changes in favour of repayment of savings mortMort-gages (Vrieselaar, Bhageloe-Datadin, Groenewegen, Hoving, & Oevering, 2017).

2.2 Ground lease, current developments

A ground lease is a legal right, halfway between ownership and rent, a property right connected with the land and not with the person occupying the land (City of Amsterdam, 2015). It can be transferred without the need of permission of the owner of the land. Ground lease was introduced in Amsterdam more than 120 years ago, in 1896. Since then in about every ten years, the system has been changed and developed to adjust to changes in political and financial environment (van Veen, 2005; Couzy & Koops, 2017). The City of Amsterdam owns about 80% percent of area within the municipality. Figure 1 shows that older (central and more expensive) areas of the city have lower fraction of ground lease, as scholars warn (Ploeger & De Wolff, 2014; Gautier & van Vuuren, 2017).

(8)

Figure 1: Area of the municipality of Amsterdam.

Red: issued in ground lease, yellow: full ownership. Source: amsterdam.nl/wonen-leefomgeving/erfpacht/

The bottom line of the concept of ground lease is to provide a steady income to the municipality by not selling, just leasing parcels to house owners. In a sense, it is a way of taxing land property by the municipality. Furthermore, the City claims that ground lease has an important role in practising land control policy (City of Amsterdam, 2015). They prevent land speculation and use the revenue to finance the Municipality’s annual budget.

Ground rent is the annual payment lessees pay to the municipality. It can be paid in half-yearly instalments, or in a lump sum for a longer period, till the end of the period

of lease. The General Conditions regulate the basics of the lease contract. However,

these Conditions have changed several times since 1986 (further General Conditions were created in 1915, 1934, 1937, 1955, 1956, 1966, 1985, 1994, 1998 and 2000 (City of Am-sterdam, 2015)). The maximum length of the period of lease is 75 years for contracts before 1966 and maximum 50 years for contracts after 1966, however, the lump sum can be paid for 100 years for contracts of General Conditions 1994. Recently, in July 2016 the municipality of Amsterdam introduced a new option for ground lease, the perpetual lease for newly built houses. With this option, ground lease contracts are evaluated only

(9)

once and ground rent does not change in every 50 or 75 years, except for being indexed with inflation. House owners can buy out perpetual lease for an infinite time period. This still differs from full ownership, as all lease contracts have to be re-evaluated when the purpose of the land usage changes.

Given that the value of land and building laying on it are treated separately, ground lease is an instrument introduced to help urban citizens finding homes. With the system they do not have to pay for the entire land, only the building on it and the smaller ground lease rent, which aims to be a more affordable housing policy. Furthermore, the ground lease rent is also interpreted as a tax on land use, which is preferred (George, 1879) compared to the taxation of more elastic goods. Moreover, as Ploeger and De Wolff (2014) argues, free and unlimited ownership does not exist, as there are land usage plans and other public law restrictions.

Moreover, some argue (Couzy & Koops, 2017) that increases in the value of a house is not a result of the house owner, but as a result of society. The government supplies infrastructure, so the government should gain the benefits from a rise in land prices, not the owner of the house. This is more of a moral or political economy argument for ground lease.

However, there is an ongoing social debate on the regulation, questioning the benefits and efficiency of the system. With the 50 or 75 years renewal of the lease contract, home owners face the risk of unexpectedly raising ground rents, especially those who have a fixed annual rent instead of a variable. Land values continuously grow in Amsterdam as the city becomes more important and local home owners bear the result (Couzy & Koops, 2017; de Lange, 2017). Additionally, mortgages supporting buying a house are harder to get for a house on land with ground lease due to the uncertainty of future payments. The City Council thus decided to introduce the perpetual leasehold, where home owners can decide to have a final evaluation of the ground and pay a fixed amount yearly (adjusted with inflation), or to buy out the ground rent for forever. A local association of ground lease holders did not agree with the original plan, claiming that the mandatory change would have been too expensive and the home owners were left out from the decision. They collected signatures for a referendum which was not held (de Lange, 2017). A long debate

(10)

was conducted in the City Council at the end of June 2017 on the details of changing the systems. The change from the old system is now voluntary, except for new ground lease contracts. However, the questions whether buying out the leasehold rights is consistent, and whether it has an effect on house prices remain untouched.

2.3 NVM, Funda.nl market share

NVM, the Dutch Association of Real Estate Brokers and Real Estate Valuers (NVM, 2015) is the largest association of real estate agents in the Netherlands. About 75 percent of Dutch houses are sold by agents who are members of the association (NVM, 2017a). The association differentiates itself from other participants of the market by maintaining a code of practice (NVM, 2017b) as a signal of quality. Furthermore, NVM stresses the importance of updated professional knowledge, and requires its members to participate in annual trainings. As a general rule of thumb, the extra costs due to agencies and administrative procedures is about 6-7 percent compared to the indicated asking price (DutchNews.nl, 2015), which is easily available for any consumer considering to buy a house. NVM operates Funda.nl, the biggest housing website in the Netherlands.

Regarding the market, the research focuses on real estate residential property mar-ket in Amsterdam. Looking at the individual sale advertisements gives the research a microeconomic approach. Although DiPasquale and Wheaton DiPasquale and Wheaton (1996) notes that at the micro level, residential and commercial property markets are closely related, the sale advertisement structure on Funda.nl covers mostly residential property characteristics. By not devoting any specific place for office or retail property characteristics makes the assumption feasible that the advertisers are inexperienced home owners. Additionally, Funda (NVM) maintains a specific website for providing service on

non-residential property, Funda in business. This site covers the markets for offices, and

agricultural lands, making the market distinction clear.

3 Methodology

Using the hedonic regression approach, I estimate real estate offered selling prices (val-ues) by observed characteristics, including the expiry date of the paid ground lease

(11)

pe-riod. In this section I give an overview of the hedonic regression approach, inconsistent discounting behaviour and specify my own models.

3.1 Hedonic regression approach

There are three traditional ways of estimating real estate prices, the cost-based ap-proach, the cash-flow approach and the hedonic approach. The cost-based method es-timates the costs of reconstructing a given building at current technology (replacement value), then takes depreciation into consideration. It is used for valuation of unique con-structions such as stadiums (Pagourtzi, Assimakopoulos, Hatzichristos, & French, 2003), where comparable buildings are not sold on the market frequently. The cash-flow approach takes the present value of all expected future cash flow from the property. Valuations for investment decisions when the owner is presumably will not be the occupier of the prop-erty use the cash-flow approach, such as transfers of ownership of hotels. The hedonic approach, or comparable method estimates real estate value by its observed characteris-tics. Thus, needs the information on the value of other comparable properties of the same market with varying characteristics, to conduct a regression analysis.

There are commonly used variables to describe the characteristics of residential real estate properties. The most important variables are the size (area) and inner structure (number of rooms), the number of bathrooms, garden, storage space. Real estate values strongly depend on location (DiPasquale & Wheaton, 1996), therefore variables describing the house’s or apartment’s location are also included. These can be postcode, neighbour-hood, or district level, describing access to public transportation, distance from city centre in time or the story an apartment is found. Separating value components of land and the residential structure laying on it and treating them as additive value components is a common practice, see for example Francke and van de Minne (2017), Diewert, de Haan, and Hendriks (2015) and Francke and Vos (2004). The choice of scale and details depend on market definition and availability of data.

I use asking price as the dependent variable of the hedonic regression. It is common to use asking price for such equations as it approximates market value well. Ball (1973) summarizes studies searching for attributes of housing the consumer derives utility from.

(12)

Studies are listed with dependant variables of both asking price and end of bargain selling price as market value. Song (1995) shows the strong connection between asking price and selling price, although bargaining outcome, the relative difference between the two, is higher for small houses. Limsombunchai (2004) also uses asking prices as dependent variable at comparing hedonic regression with other methods for predicting real estate value of houses in New Zealand. For the closely related rental house market, Moll (2012) shows a detailed list of hedonic regression literature using asking and agreed rents for offices.

3.2 Inconsistent time discounting

The act of buying a house is a term investment decision, which requires long-term time discounting. Time-consistent choice is modelled with exponential discounting. Economic literature shows examples of agents with time-inconsistent choices which lead to inefficiencies. This results in a distortion of prices, under- or overconsumption of goods and services.

The most frequent models assume hyperbolic or quasi-hyperbolic discount functions instead of the exponential form (Strotz, 1955; Phelps & Pollak, 1968). With these prefer-ences, agents have decreasing discount factor over time or they discount all future periods with an extra discount factor. This means that agents have a higher discount rate between present and the next period than any other subsequent periods in the future. DellaVigna and Malmendier (2004) shows that rational firms with fitting contract design can exploit the naive consumer who has biased estimate of her own short-run discount factor, while the sophisticated consumer can use these contract types as a commitment device. Exam-ples list businesses of investment goods (costs are present, but benefits are in the future), including sport club monthly subscription with overestimated future activity, and leisure goods (benefits are realized before costs) such as credit card usage of high interest rate, or gambling.

Most of the models aim to explain present bias, namely that people are impatient

towards future consumption, but this impatience varies by the time of decision. Al-though hyperbolic discounting have been criticized (Rubinstein, 2003), more general

(13)

func-tional forms have been developed to explain also increasing and decreasing impatience (Bleichrodt, Rohde, & Wakker, 2009; Gerber & Rohde, 2010).

Strongly connected to patience, the processing of available information can also vary by time. Myopia is defined as ’discounting information from anticipated future events, with the discount rising progressively as the event becomes less imminent’ in Pryce, Chen, and Galster (2011), while Gabaix and Laibson (2017) argues that perfectly patient agents can also exhibit inconsistent choices of preference reversals1_{, if their ability to predict future}

is limited. Amnesia is a choice process in which past events play no role (Dow, 1984), giving scope for inconsistent decisions.

Pryce et al. (2011) uses the concepts of myopia and amnesia to describe housing mar-kets’ respond to increasing flood damage caused by global warming. By overweighting information from the present, they argue, past information and [flood-related] losses fade

away, and market amnesia might appear even at the presence of individual amnesia.

3.3 Models

Following the above described hedonic regression method, I run two types of regression

models. In the first, I show that apartments on ground lease have lower price than

apartments with full ownership situation.

pi = β0+ βGLDGL+ βControlsi+ ui (1)

Here β0 shows the coefficient of a home with full ownership, and βGL indicates the price

difference due to the different ownership situation of ground lease.

Then, regarding properties only on ground lease contract, the main regression model is:

pi = β0+ βDEP 1DEP 1i + ... + βDEP kDEP ki+ βControlsi+ ui (2)

where pi is the (logarithm of) property offered selling prices, and DEP 1 to DEP k are the

10 years long expiry period (EP) dummies. Here β0 means the coefficient of an expiry

1_{With the example of Benhabib, Bisin, and Schotter (2010), it occurs, when a subject prefers $10 now} rather than $12 in a day, but he/she prefers $12 in a year plus a day rather than $10 in a year.

(14)

period less than a year. Controls include the usual variables used in real estate value estimating: (log) size, number of rooms, age (year of construction), level of energy level, as suggested in van Dijk and Francke (2015). I also include location variables as location is one of the most important factors when evaluating a house (DiPasquale & Wheaton, 1996, Chapter 4). This is possible, as addresses are also indicated on the advertisements,

at least on postcode level. More details on Controls are given in Section 5, where the

main determinants of house prices are explained.

3.4 Expectation for outcomes

For Equation 1 negative coefficient is expected for βGL. This is interpreted as a discount

for the limited ownership rights and verifies hypothesis 1.

As explained in the Introduction, I expect positive coefficients for all DEP j (j = 1...k)

in Equation 2, and it would be rational to be an increasing effect, higher coefficient as

j grows. More valuable estate for longer payment-free periods, keeping everything else

constant, based on hypothesis 2. But I expect that after a given break point, the increase in coefficients stops and until then, the difference between them decreases. For example, people can distinguish 0 versus 10 years free of payment, but they cannot make the difference between 40 and 50 years free of payment at the time of thinking of buying a house, supporting hypothesis 3. Therefore I expect that the prices offered to buyers shows this pattern.

4 Data

4.1 Method of acquiring it

The largest housing website in the Netherlands is Funda.nl which has at least 60% stable market share of all housing websites (van Dijk & Francke, 2015). The advertised houses and apartments are mostly brokered by the Dutch Association of Real Estate Brokers and Real Estate Valuers (NVM) (Su´arez & Engelberts, 2005), which is the industry association for brokers, housing agents. The advertisements on the website include the asking price, the size, address, the mentioned expiry date (if applicable), and other characteristics of the apartment or house on sale.

(15)

Using theBeautifulSoup package of Python language2, I parsed through the html codes of the search results for Amsterdam and constructed my cross-sectional database of apart-ments (and houses) on sale on the 7th May 2017. Then the string variables were translated

from Dutch to English3_{, and new numeric variables were created containing the}

informa-tion available. Based on the dates, I constructed a Time till expiry date variable and

broke it into dummy variables of 10 (and 5) years long periods.

There are advertisements where the variables of interest or the most important control variables are not present. This is not a mistake of the parsing algorithm, as after manually checking several of them, it turned out that the advertiser or estate agent had not filled the given places. See for example Figure A.1, A.2 and A.3 in Appendix A.1.3 where the ownership situation is missing from the advertisement, and therefore could not be taken into account in the final analysis.

In some cases of control variables, usually it is not indicated if the item is missing. Therefore I assume that if there is no information available on an item, then it is not present at the given property. For example, if there is no data on the existence of a storage room, neither on its size, location, the insulation and other variables describing it, then I assume that the property has no separate storage room. I use the same assumption for gardens, insulation and parking facilities.

As for the energy label, the website provides an estimate for apartments that have no energy label. This is based on the characteristics of the property and lies on the

judgement of the housing agent. In the dataset, these preliminary energy labels are

treated as given energy labels. Ranging from A (most energy efficient) to G (least energy efficient characteristics), the energy label suggests information about expected heating costs. An efficient label (A-C) is associated with warmer and more comfortable living conditions.

2_{My Python knowledge heavily depends on Severance and Andrion (2016) and on Sweigart (2015).} 3_{Fundla.nl provides English platform as well, but the Dutch version was more reliable in terms of} filling of the advertisements.

(16)

4.2 Descriptive statistics

In this section I provide some descriptive statistics of my database.

Based on the advertisements on Funda.nl on 7th May 2017, there were 2148 different apartments on sale that I constructed my database from. After dropping the observations where ownership situation was not clearly defined, a total of 1846 observations stayed in the dataset. However, in a small number of observations the ownership situation is not indicated on the advertisements, it is provided by request. Moreover, for 12 observations, only a membership- or usage right can be purchased: this entitles the owner to live in the property and use the common areas. Finally, out of these for 3 apartments on sale, the price was only available on request, thus, had to be deleted. See Table 1 for distribution of ownership situation and summary statistics of price. Apartments with full ownership are on average more expensive than apartments with a ground lease contract. Furthermore, Usage right and hiding the ownership information for the first glance both have a lower mean price.

Table 1: Summary statistics of asking price, by ownership situation

Ownership situation mean sd N

Full ownership 712620.9 1300167 676 Ground lease 407359.5 320812.8 1137 Usage right 295791.7 316979.6 12 On request 289889.4 77807.8 18 Total 517453.6 840163.8 1843 Source: apartments.dta

In the model of Equation 2, I compareotherwise similar apartment prices whose date of

ground lease expiry differs. Therefore I include the apartments on sale with a ground lease ownership situation. Even in the cases for indicated ground lease ownership structure, in 48 observations it was impossible to determine whether it was a yearly amount or a paid off period until a given date. Therefore I decided to leave these observations out.

(17)

The distribution of asking price is far from normal or even log-normal. Both the Shapiro–Wilk and Shapiro–Francia normality tests reject the hypotheses that the price (in euros), the specific price (euros per square meter), or the logarithms of these would be normally distributed. Figures 2 and 3 show their distributions.

Figure 2: Distribution of asking price

Figure 3: Distribution of asking price per square meter

The age of the apartments are not always present precisely in the advertisements. However, the construction period is given in most cases where the year of construction

(18)

is missing. Therefore I use Construction period and not age of the building as a control variable for the price. Table 2 shows frequencies of each construction period by ownership type. Ground lease contracts are always more frequent except for apartments built before 1906.

Table 2: Construction period

Ownership situation

Construction period Full ownership Ground lease Usage right On request Total

Before 1906 352 76 2 3 433 1906-1930 187 214 0 2 403 1931-1944 33 110 0 1 144 1945-1959 17 34 0 3 54 1960-1970 5 124 9 4 142 1971-1980 5 43 0 0 48 1981-1990 10 109 0 0 119 1991-2000 20 111 1 1 133 2001-2010 28 212 0 4 244 2011 and after 18 105 0 0 123 Total 675 1,138 12 18 1,843 Source: apartments.dta

Summary statistics of the other non-categorical control variables are in Table 3. Area, the number of rooms and bathrooms, and vertical location follow a skewed distribution. The rest of the variables in the table are dummy variables created based on the raw data. A considerable percentage of apartments have garden, parking facility, and insulation. A relatively small percentage are new property or have porch or construction quality marks, or accessible for the elderly. However, the reduced variation does not imply lack of effect, as it is explained in Section 5.

(19)

Table 3: Summary statistics of controls

Mean Min Max SD N

Area 92.36 15 1047 56.05 1843

Number of rooms 3.26 1 15 1.29 1843

Number of bathrooms and toilets 1.62 0 6 .91 1843

Story (located at) 2.46 0 23 2.39 1843

New property .02 0 1 .12 1843

Has storage .53 0 1 .49 1843

Has insulation .77 0 1 .42 1843

Has construction quality marks .07 0 1 .24 1843

Has porch .03 0 1 .16 1843

Accessible for elderly/disabled .04 0 1 .20 1843

Has garden .98 0 1 .13 1843

Has parking facility .73 0 1 .44 1843

Source: apartments.dta

5 Results

5.1 Model 1

For the dependent variable, the estimations use level and logarithm of offered selling price. This way, the coefficients of characteristics can be interpreted in absolute and relative terms: in euros and in percentages. However, estimations on the logarithm of asking price are more reliable, taking the distribution of asking price into account.

Apartments with ground lease ownership situation are on average cheaper than apart-ments with full ownership. Table 4 contains the coefficients of different ownership types compared to full ownership situation, the dependent variable is logarithm of asking price. Even after controlling for other characteristics, apartments on ground lease are 5 to 9 per-cent cheaper compared to full ownership: (eβGroundlease − 1) ∗ 100 gives -9.26, -5.53, -8.98

and -5.58 percent significant differences respectively. Columns 1 and 3 are controlled for 3-number postcode levels (pc3), while columns 2 and 4 have 4-number postcodes (pc4)

(20)

as location variables. This explains the difference between the estimated coefficients: the central older districts have traditionally fewer ground lease contracts, however, the cen-tral areas tend to be more expensive in cities (DiPasquale & Wheaton, 1996). The more detailed pc4 variable captures price changes due to location, nevertheless the discount for not owning the land remains significant.

Table 4: Estimation Results

(1) (2) (3) (4)

All All Not PA Not PA

Ground lease -0.0972∗∗∗ -0.0569∗∗∗ -0.0941∗∗∗ -0.0574∗∗ (0.0115) (0.0116) (0.0167) (0.0199) Usage right -0.1275∗ -0.0491 -0.1217∗ -0.0553 (0.0569) (0.0450) (0.0580) (0.0457) On request -0.0570 -0.0343 -0.0440 -0.0023 (0.0525) (0.0414) (0.0524) (0.0356) R2 _0.920 _0.942 _0.933 _0.946 Adjusted R2 _0.917 _0.938 _0.930 _0.941 RMSE 0.178 0.154 0.176 0.162 N 1842 1842 995 995 Postcode PC3 PC4 PC3 PC4

Standard errors are in parentheses. Controls are used but coefficients are not reported. ∗ _{p <}_0.05,∗∗ _{p <}_0.01,∗∗∗ _{p <}_0.001

These results are in line with the general perception that freedom of full ownership in-creases value (Ploeger & De Wolff, 2014). Gautier and van Vuuren (2017) shows a 10 per-cent price difference for Amsterdam ground lease, however, they only use neighbourhood-level location variables which is more similar to pc3. Columns 3 and 4 restrict the sample for ground lease advertisements which are not paid in advance, hence their time till expiry date is 0. The estimations are robust to the sample restriction. Tyvimaa, Gibler, and Zahirovic-Herbert (2015) find a 5 percent discount for ground lease dwellings against free-hold land in Helsinki. The sign of the coefficients for Usage right and Unknown ownership situation are negative as expected, but not significant possibly due to their small sample

(21)

size.

5.2 Model 2

The main hypothesis of Equation 2 is that years paid in advance of the ground lease contract has positive effect on house prices. Table 5 shows the coefficients of dummies of the years paid in advance put into ten years long bins. The base group is the apartments on sale with a ground lease contract, but without having a period free of payments.

Table 5: Price effect of expiry preiods of years paid in advance

(1) (2) (3) (4) LnP LnP P P Dep10 2=0 (base) 0 0 0 0 Dep10 2=1 0.027 -0.032 1715.3 -2430.0 (0.039) (0.032) (25289.5) (25275.4) Dep10 2=2 0.022 -0.043 49351.1∗ 12686.8 (0.030) (0.028) (23934.5) (20950.1) Dep10 2=3 -0.011 -0.025 28596.4 13171.0 (0.030) (0.025) (24946.1) (19688.8) Dep10 2=4 -0.020 0.0013 6763.3 23934.1∗ (0.016) (0.014) (12813.0) (11585.1) Dep10 2=5 -0.027 0.0086 -5922.0 18244.1 (0.017) (0.015) (12812.5) (11987.0) Dep10 2=6 0.027 -0.036 -57269.5 -87169.4∗ (0.041) (0.038) (42383.3) (35039.8) Dependent var lnp lnp p p Postcode PC3 PC4 PC3 PC4 R2 _0.905 _0.941 _0.834 _0.886 Adjusted R2 _0.900 _0.934 _0.826 _0.873 RMSE 0.170 0.137 142368.662 121404.037 N 1090 1090 1090 1090

(22)

Table 6: Price effect of years paid in advance

(1) (2) (3) (4)

TTE ys TTE ys TTE ys TTE ys

b/se b/se b/se b/se

timetillend ys -0.00047 0.000043 -402.1 62.9 (0.00031) (0.00028) (275.3) (245.8) Dependent var lnp lnp p p Postcode PC3 PC4 PC3 PC4 R2 _0.904 _0.940 _0.832 _0.884 Adjusted R2 _0.900 _0.934 _0.825 _0.872 RMSE 0.169 0.137 142599.240 121935.788 N 1090 1090 1090 1090

Table 6 shows the effect of years paid in advance on the logarithm and level of asking prices. None of the coefficients are significant. This can be interpreted as the indication that among ground lease properties, traditional variables have more effect on prices than the length of payment-free period. The lack of significance might be due to the small sample size, compared to the number of controls used.

An additional interpretation of the lack of significance of all time-related variables is the presence of myopic behaviour of buyers, or amnesiac behaviour of sellers. If buyers do not count with future payments of ground lease, then the benefits and present value of already paid rents do not play a role in their decision of buying a house. Consequently, sellers cannot ask a compensation for their previous investments. Alternatively, if sellers exhibit amnesiac behaviour, or treat paid off periods as sunk costs, then it follows that they do not ask for an extra compensation other than due to house characteristics.

In my database, for 291 observations the precise ground rent was available. These were the advertisements with ground lease ownership situation, but without any years of

(23)

payment in advance, their expiry period is 0 years. The base group for the regressions estimating Equation 2 was constructed from them. For these advertisements a hypothet-ical lump sum can be calculated for given time periods. This lump sum is the present value of the ground rents being bought out, therefore is based on the following elements: the index-linked land value, the ground rent rate and the premium factor, P F (City of Amsterdam, 2015). The first two serves to calculate the current ground rent, based on the current market value of the land. This is then multiplied with the premium factor which is the sum of a geometric sequence based on the quarterly published discount factor

(purchase rate, translated from afkooppercentage) and the number of years the buyout is

calculated: P F = 1 − 1 (1+pr)n 1 −_1+pr1 = 1 + pr pr 1 − 1 (1 + pr)n ! (3) where pr is the purchase rate and n is the number of years of the lump sum paying for. For the purchase rate 3.28% for buying out in the second quarter of 2017 (Amsterdam.nl, 2017), the premium factor is 14.9751, 19.5299, 22.8283, 25.2169 for 20, 30, 40 and 50 years respectively.

Table 7: Hypothetical lump sum as a share of asking price

Variable mean p25 p50 p75 min max sd

20 years lump sum share .032 .007 .022 .041 .0007 .2752 .037

Source: apartments.dta

Assuming that the ground rent would be similar if they were bought out immediately, the hypothetical lump sum payment (or premium) can be calculated. Table 7 shows the descriptive statistics of this calculated lump sum as a share of asking price. The lack of significance of the above analysis might be also from the fact, that the calculated premiums are relatively small compared to the asking price. Even the 75th percentile for

(24)

paying off 50 years is less than 7 percent, without taking location effects into control. It can be concluded that house prices are determined by other factors.

Unfortunately, the hypothetical yearly ground rent could not be calculated for the rest of the advertisements. It is because the advertisements lacked the separate land value used for the lease contract. Even with the suggestions of the Municipality for Land Pricing Policy, it could not be estimated because the land value also depends on the value of the building laying on it apart from size and location. This historical information cannot be obtained without full access of each building’s selling history. Therefore an analysis of comparing hypothetical rents with current asking prices is not feasible for this paper.

5.3 Determinants of house prices

In this section I give more details on control variables and their effect on asking price. The regression outputs contributing to this section are in Appendix A.3. After fitting several functional forms and control variables, the estimations in Table A.2 on page 34 are chosen to introduce the main determinants of apartment prices. All of these models estimate the logarithm of asking price, as these models had always better explanatory power than regressions on the price or the price per square meter or the logarithm of the latter. Consequently, the estimate interpretations of the rest of the chapter are based on

the formula effect = (exp(β) − 1) · 100 percent, and coefficients are regarded significant

from 5 percent significance level or stronger.

The table contains the control variables of eight estimates. Columns 1-2 are the control variables of columns 1-2 in Table 4, estimating model 1. These take full ownership as a reference category and estimate the effect of ground lease situation. Columns 3-4 contain the control variables for Table 6, where the number of paid years in advance is the main explanatory variable. The last 4 columns cover the logarithmic estimates from Table 5, and the similar models with 5-year-long-period bins instead of 10.

The models presented are different in the level of location variable. 3- and 4-digit postcode levels are presented. The rationale behind using more detailed location variables is that location is correlated with other characteristics (such as age of a building), and

(25)

to the similarities of neighbouring areas, multicollinearity issues arise, and the variance inflation factors of several 4-digit postcodes are indeed high. This affects the standard errors of the other estimates in the column and reduces significance even at existing effects. Pc3-level postcodes approximate city districts (neighbourhoods) well, with a bit of more details: 10 categories instead of 8 city districts. For pc3 level, the central district, the oldest part of the city is chosen as reference category, postcodes starting with 101.

Compared to the city centre the neighbourhood Zuidoost (with postcodes 110x) is the

cheapest, where similar apartments can be bought for half the price. The second cheapest

area is Nieuw-West (with postcodes 106x), and the third is Noord (with postcodes 103x

and 102x). Postcodes of Zuid are not different significantly from the city centre. This

observation about prices of neighbourhoods is in line with rental prices (RentSlam, 2017). Pc4 coefficients are not reported, but they suggest the same price structure across the city. However, when pc4-level variables are present, then other variables which correlate with location (area, construction period) become less important predictors both in absolute terms and in significance.

Following location, area and structure variables are presented first. Area is in logarithm of units of living area (square meter), and thus the coefficient can be interpreted as elasticity when the dependent variable is also in logarithm. It is slightly below 1 in all estimates, but not significantly different from it. This means that one percent increase in area increases asking price with almost or close to one percent (the elasticity of price with regard to area is close to one). When we control for area, the number of rooms becomes a less important factor, with a negative coefficient but not significantly different from

zero4_{. These affects should be interpreted as marginal effects, and therefore the variable}

Number of rooms refers to structure and not additional space. This suggests that living

area is a more important determinant than the disposition of the apartment. However, the number of bathrooms or toilets has a significantly positive effect in most cases, with one unit increase associated with 1.6-2.3 percent increase in the asking price, keeping everything else constant.

The age of the buildings is also an important factor. The oldest properties are chosen for reference category, built before 1906. These are the most expensive properties hence they

(26)

have the most prestigious interior. Houses built between 1960 and 1980 are the cheapest, approximately 20-25 percent cheaper than the base group, controlled for location and ownership category. Construction period has an effect on the years paid in advance as young apartments’ ground lease cannot have been paid long periods in advance, however, taking their interaction terms into the regression does not change the outcome. With the more precise location control it can be shown that houses built after 2010 have 8-10 percent higher asking price compared to the oldest ones.

The energy label of apartments is based on label A, resulting negative coefficients for

all the other labels. However, significant jumps are at label B and E. The label not

applicable also has a negative effect for the estimates only on the ground lease sample.

This latter effect can be interpreted as a discount for the lack of information. Energy labels include insulation and other quality effects that are in connection with heating and

sustainability which explains the lack of significance of the dummy variables Insulation

andConstruction quality marks and Type of roof. However, based on statistics of Adjusted R-squared and Akaike information criteria (Wooldridge, 2013), it is worth including these

variables despite the lack of significance. The same applies to the type of apartment, and

the residential layer the apartment is located at: Story, existence of storage space, porch,

garden, or parking lot, whether it is accessible to the elderly or disabled and whether it is

part of an ownership association,VvE. However, the size of the garden, the different type

of construction quality marks, the volume of the apartment, the garden location specifics (eg. by forest, by busy road, by park, by water, etc.), the information on the type of garden, the orientation of the garden, the number of stories in the building are deselected during the model specification process.

One more important variable affects apartment asking prices, namely whether it is

a resale or a new property (newbuiltD). New properties tend to be more expensive by

approximately 30 percent. It is important to note that these properties are all built after 2010 and therefore their comparison category is the last construction period.

(27)

6 Conclusion

This thesis introduced and analysed the Amsterdam housing market from a time dis-counting point of view. Based on the results of estimated models on the collected database, it seems that apartment prices depend mostly on regular factors and indication of incon-sistent time discounting does not appear in the way the research question addressed the issue. It is rather the discount that arises from the reduced ownership rights that appears in prices, and the length of the paid off period does not seem to account for a significant share of the asking price. This contradicts with hypothesis 2 and makes hypothesis 3 not testable.

6.1 Policy recommendation

Until now, ground lease system received criticism because of its complicated layout and the negative effects of uncertainty arising from the regular re-evaluations of 50 or 75 years. Thus the price decrease associated with ground lease ownership situation can be a result of uncertainty-avoiding behaviour of home owners, and not just the restriction regarding the purpose of usage of the land. Furthermore, uncertainty causes harder mortgage approvals, even with the present low mortgage rates. Although houses are cheaper with ground lease instead of full ownership and therefore mobility of citizens increases, uncertainty might diminish this effect. Furthermore, the average tenant does not change the purpose of a parcel while living in a home, thus they do not suffer disutility from the restriction. For the municipality, ground rents and premiums are considered as stable revenue with less distortion than other sorts of taxes, thus its rational goal is to keep the regulation. As a result, the initiative of the new perpetual lease may be more generally approved welcome from this point of view. It provides stable payment structures and thus certainty both for home owners and the municipality.

Based on the results, more information should be given to individual home sellers and buyers about the advantages and disadvantages of paying off ground lease in advance. Even though that the present value of paid off ground lease is relatively small to the overall value of a house or apartment, it should appear in the price determining decision process for properties on ground lease. The new ground lease system allows for paying off the ground rent for an infinite period of time, and information is given to current home

(28)

owners whether it is worth switching from their old contracts. The municipality launched a website (City of Amsterdam, 2017) which calculates the costs of possible change for perpetual ground lease, the hypothetical yearly rent or the ground premium paying for infinite periods, given enough information on current contract, last land evaluation, the applied general conditions and other characteristics of the real estate. The website gives an initial advice whether the home owner should change to the new system or stay with the old contract.

However, the online tool compares the present value of payments only for the case where the property stays with the current owner forever. There is no indication what would be the best for the owners if they plan to sell their properties after a period of time. The possibility of not counting with previously paid ground rents at determining the asking price can cause inefficiency at the decision of paying a premium. Sellers do not get compensated for their previous long-term investment of freeing the ground lease owner from paying the ground rent for forever.

Even though the perpetual ground lease system helps the uncertainty most complaints were about, more effort should be dedicated to educate home owners on this aspect of their decision. From then on, asking prices might take previously paid amounts into account and not just the ownership situation itself. This would help more efficient decisions in a market where sellers and buyers are both inexperienced citizens.

Additionally, further research is needed to determine whether this pattern is present in bigger sample and other dates. There might be a small price effect of the length of paid off period which was too small to be significant in a sample of roughly 1000 observations. If there is, hypothesis 3 should be tested, namely checking whether longer paid off periods have always greater positive effect on asking price, even when the periods compared are far in the future.

6.2 Limitations, further research

The thesis has its limitations and thus further directions of research are described here. As pointed out before, an analysis of bigger sample size is needed to determine the effect of years paid in advance more reliably. Moreover, an analysis of final selling prices is

(29)

needed to confirm the results. Gautier and van Vuuren (2017) shows a small but positive effect of years paid in advance on selling prices, however, their flexible specification also varies strongly and mostly has negative coefficients (compared to the not paid in advance ground lease base group) which awaits explanation. Furthermore, it is not clear form their working paper whether they report standard errors or p-values under coefficients which also calls for further discussion on the topic.

It should be stressed again that most houses on Funda.nl are sold through a housing agency that is a member of NVM. The profit maximising behaviour of real estate agencies should be a further reason to account for the whole value of a house at determining the asking price. On the contrary, future buyers might be the ones with lower reservation prices, since they might lack the ability of consistent time discounting. As a result, agents use heuristics and experience at giving advice on determining asking prices. Alternatively, buyers might have a stronger bargaining power which diminishes the value increase of paid off period. The latter is not plausible though, as the Amsterdam housing market supply is considerably scarce. However, buyers and sellers do not differ strongly in this market for homes, and further information on the present value of already paid off periods should solve the inconsistency. Housing agencies might be a platform for the implementation of such awareness campaigns.

(30)

A

Appendices

A.1 Documentation for data cleaning

A.1.1 Process

The process for building the database took four steps. First, a Python script went through the search results at Funda.nl to get the links for the advertised properties on sale. Then, a second script went through the advertisements one by one to save the html codes of them. Subsequently the third script went through the saved html codes to get the data out of them in string format to a data frame, saved in .csv extension, separately for apartments and houses which were then saved in different sheets of an Excel workbook. I admit there must be nicer ways to do it, but this solution was the most intuitive to deal with character coding issues. Finally, the raw data from the Excel sheet was imported into Stata where the variables got English names and labels instead of the originally Dutch ones. Then, new numeric and categorical variables were constructed based on the string variables, to use them more efficiently in the later regressions.

I declare that I do not use the obtained data for any commercial purposes and I only use it for the sake of this research. Given that now Funda.nl tries to prevent automated web-scraping of browsers driven by bots, for further research and extension of the database their explicit permission and approval is needed. Until the finalizing of the thesis they refused to reply for further data requests.

A.1.2 Codes

Here I give access to view the Python codes produced to create the database. 1. Collecting links of advertisements

(a) https://drive.google.com/open?id=0B6qe5GRBbhxcSW81OFVaZktlNHM/view 2. Saving the html codes from the links

(a) https://drive.google.com/open?id=0B6qe5GRBbhxcQ0lLcWJvaUN2SmM/view 3. Creating the database for Stata

(31)

A.1.3 Example for missing data in an advertisement

Figure A.1: Saved screen shot for an advertisement without ownership situation

Source: Funda.nl, http://www.funda.nl/koop/amsterdam/appartement-48312140-lijnbaansgracht-14-1/

(32)

Figure A.2: Saved screen shot for an advertisement without ownership situa-tion Source: Funda.nl, http://www.funda.nl/koop/amsterdam/appartement-48312140-lijnbaansgracht-14-1/

(33)

Figure A.3: Saved screen shot for an advertisement without ownership situa-tion Source: Funda.nl, http://www.funda.nl/koop/amsterdam/appartement-48312140-lijnbaansgracht-14-1/

(34)

A.2 Amsterdam population

The municipality of Amsterdam has 853180 inhabitants which number is increasing since 2006. Table A.1 shows the population of Netherlands and the region where Am-sterdam belongs to (Great-AmAm-sterdam). Furthermore it includes a neighbouring town without ground lease regulation, Amstelveen, and Rotterdam as a second business hub in the country.

Table A.1: Population development; Region per month - Population at the end of the pe-riod (Bevolkingsontwikkeling; regio per maand - Bevolking aan het einde van de pepe-riode)

Period Netherlands Great-Amsterdam (CR) Amstelveen Amsterdam Rotterdam

2002 16192572 1184987 78095 736562 599651 2003 16258032 1194634 78866 739104 598923 2004 16305526 1204536 79036 742783 596407 2005 16334210 1209471 78774 743070 588697 2006 16357992 1213535 78945 742884 584058 2007 16405399 1222305 78980 747093 582951 2008 16485787 1235514 79768 755605 587134 2009 16574989 1251327 80695 767457 593049 2010 16655799 1267128 81796 779808 610386 2011 16730348 1280170 83363 790110 616260 2012 16779575 1291735 84379 799278 616294 2013 16829289 1305307 85015 810767 618357 2014 16900726 1320295 87162 821752 623652 2015 16979120 1329572 88602 833624 629606 2016 * 17089690 1348503 89298 848861 638221 2017 January* 17092727 1349536 89350 849799 638714 2017 May* 17121333 1353980 89585 853180 639596 *: Preliminary figure

(35)

A.3 Control coefficients

Table A.2: Control variable coefficients

(1) (2) (3) (4) (5) (6) (7) (8)

Ownership Ownership Years Years Dep5 Dep5 Dep10 Dep10

Log or area 0.9893∗∗∗ 0.9452∗∗∗ 0.9932∗∗∗ 0.9398∗∗∗ 0.9859∗∗∗ 0.9416∗∗∗ 0.9852∗∗∗ 0.9425∗∗∗ (0.0239) (0.0214) (0.0341) (0.0259) (0.0353) (0.0261) (0.0354) (0.0265) Number of rooms -0.0031 0.0058 -0.0226∗ -0.0103 -0.0211 -0.0100 -0.0211 -0.0112 (0.0093) (0.0085) (0.0109) (0.0080) (0.0108) (0.0079) (0.0109) (0.0081) nbathtoilet 0.0230∗∗∗ 0.0166∗∗ 0.0288∗∗ 0.0119 0.0277∗∗ 0.0099 0.0288∗∗ 0.0125 (0.0064) (0.0054) (0.0088) (0.0069) (0.0091) (0.0071) (0.0090) (0.0070) Before 1906 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (.) (.) (.) (.) (.) (.) (.) (.) 1906-1930 -0.0391∗∗ -0.0044 -0.0656∗∗ 0.0031 -0.0655∗∗ 0.0059 -0.0655∗∗ 0.0035 (0.0135) (0.0130) (0.0248) (0.0222) (0.0245) (0.0224) (0.0246) (0.0226) 1931-1944 -0.0890∗∗∗ 0.0091 -0.1245∗∗∗ 0.0148 -0.1235∗∗∗ 0.0174 -0.1238∗∗∗ 0.0170 (0.0174) (0.0189) (0.0257) (0.0270) (0.0258) (0.0273) (0.0257) (0.0272) 1945-1959 -0.0712∗ -0.0407 -0.1446∗∗∗ -0.0893∗ -0.1445∗∗∗ -0.0909∗ -0.1458∗∗∗ -0.0892∗ (0.0298) (0.0352) (0.0387) (0.0431) (0.0388) (0.0434) (0.0388) (0.0436) 1960-1970 -0.2429∗∗∗ -0.2214∗∗∗ -0.2808∗∗∗ -0.2343∗∗∗ -0.2982∗∗∗ -0.2235∗∗∗ -0.2956∗∗∗ -0.2222∗∗∗ (0.0290) (0.0298) (0.0342) (0.0320) (0.0357) (0.0325) (0.0355) (0.0325) 34

(36)

1971-1980 -0.2633∗∗∗ _-0.1845∗∗∗ _-0.3108∗∗∗ _-0.1804∗∗∗ _-0.3326∗∗∗ _-0.1866∗∗∗ _-0.3158∗∗∗ _-0.1752∗∗∗ (0.0323) (0.0343) (0.0390) (0.0401) (0.0406) (0.0399) (0.0390) (0.0409) 1981-1990 -0.1359∗∗∗ _-0.0898∗∗∗ _-0.1690∗∗∗ _-0.0717∗∗ _-0.1918∗∗∗ _-0.0399 _-0.1842∗∗∗ _-0.0420 (0.0233) (0.0211) (0.0303) (0.0258) (0.0380) (0.0308) (0.0355) (0.0303) 1991-2000 -0.1259∗∗∗ _-0.0515∗ _-0.1315∗∗∗ _-0.0092 _-0.0913∗ _0.0487 _-0.1287∗∗∗ _0.0059 (0.0241) (0.0240) (0.0326) (0.0273) (0.0414) (0.0336) (0.0363) (0.0307) 2001-2010 -0.1240∗∗∗ _-0.0211 _-0.1594∗∗∗ _0.0007 _-0.1474∗∗∗ _0.0089 _-0.1530∗∗∗ _-0.0008 (0.0223) (0.0204) (0.0318) (0.0266) (0.0323) (0.0278) (0.0324) (0.0278) 2011 and after -0.0270 0.1004∗∗∗ _-0.0658∗ _0.0887∗∗ _-0.0529 _0.0905∗∗ _-0.0597 _0.0830∗∗ (0.0227) (0.0239) (0.0312) (0.0283) (0.0317) (0.0294) (0.0313) (0.0292) A 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (.) (.) (.) (.) (.) (.) (.) (.) B -0.0571∗ _-0.0472∗ _-0.0822∗∗ _-0.0655∗∗ _-0.0794∗∗ _-0.0568∗∗ _-0.0841∗∗ _-0.0636∗∗ (0.0222) (0.0194) (0.0252) (0.0216) (0.0256) (0.0213) (0.0256) (0.0217) C -0.0261 -0.0307 -0.0446 -0.0409 -0.0380 -0.0350 -0.0436 -0.0388 (0.0232) (0.0200) (0.0274) (0.0225) (0.0272) (0.0220) (0.0275) (0.0226) D -0.0354 -0.0176 -0.0496 -0.0016 -0.0434 0.0016 -0.0470 -0.0023 (0.0241) (0.0204) (0.0286) (0.0236) (0.0282) (0.0230) (0.0287) (0.0236) E -0.0608∗ -0.0517∗ -0.0828∗∗ -0.0393 -0.0738∗∗ -0.0387 -0.0765∗∗ -0.0421 (0.0245) (0.0220) (0.0284) (0.0240) (0.0283) (0.0233) (0.0286) (0.0240) 35

(37)

F -0.0634 -0.0714∗ _-0.0903∗ _-0.0718∗ _-0.0872∗ _-0.0550 _-0.0976∗ _-0.0665∗ (0.0334) (0.0286) (0.0395) (0.0324) (0.0396) (0.0321) (0.0394) (0.0325) G -0.0506∗ _-0.0449∗ _-0.0748∗ _-0.0607∗ _-0.0699∗ _-0.0535 _-0.0744∗ _-0.0598∗ (0.0243) (0.0212) (0.0306) (0.0277) (0.0305) (0.0283) (0.0307) (0.0285) not applicable -0.0313 -0.0205 -0.0762∗∗ _-0.0382∗ _-0.0711∗∗ _-0.0346 _-0.0745∗∗ _-0.0381∗ (0.0216) (0.0178) (0.0250) (0.0192) (0.0248) (0.0187) (0.0251) (0.0193) apartment 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (.) (.) (.) (.) (.) (.) (.) (.) bel-etage 0.0759 0.0831 0.1366 0.1043∗ _0.1427 _0.1093 _0.1385 _0.1039∗ (0.0471) (0.0519) (0.0749) (0.0473) (0.0818) (0.0623) (0.0755) (0.0473) corridor 0.0683 0.0906 0.0639 0.1012 0.0747 0.1044 0.0744 0.1003 (0.0504) (0.0778) (0.0673) (0.0991) (0.0777) (0.0986) (0.0711) (0.0945) double upstairs -0.0323 -0.0454∗ _-0.0113 _-0.0202 _-0.0066 _-0.0192 _-0.0080 _-0.0216 (0.0198) (0.0183) (0.0244) (0.0220) (0.0248) (0.0219) (0.0248) (0.0221) basement -0.1990∗ _-0.1977 (0.0798) (0.1155) Story 0.0059 0.0052∗ _0.0055 _0.0051 _0.0053 _0.0054 _0.0052 _0.0054 (0.0032) (0.0026) (0.0037) (0.0028) (0.0038) (0.0028) (0.0038) (0.0028) Has storage -0.0094 0.0006 -0.0438∗∗ -0.0273∗ -0.0400∗∗ -0.0239∗ -0.0428∗∗ -0.0278∗ (0.0103) (0.0093) (0.0137) (0.0112) (0.0139) (0.0114) (0.0138) (0.0113) 36

(38)

Insulation -0.0264∗ _-0.0079 _-0.0154 _0.0111 _-0.0182 _0.0096 _-0.0163 _0.0108 (0.0107) (0.0096) (0.0137) (0.0120) (0.0136) (0.0121) (0.0137) (0.0121) cqualitymarksD -0.0434∗ _-0.0309∗ _-0.0457∗ _-0.0419∗∗ _-0.0431∗ _-0.0426∗∗ _-0.0440∗ _-0.0430∗∗ (0.0173) (0.0152) (0.0196) (0.0159) (0.0195) (0.0159) (0.0197) (0.0160) porchD -0.0133 -0.0191 -0.0210 -0.0190 -0.0258 -0.0213 -0.0230 -0.0173 (0.0220) (0.0219) (0.0296) (0.0302) (0.0295) (0.0295) (0.0295) (0.0304) newbuiltD 0.4808∗∗∗ _0.4378∗∗∗ _0.2780∗∗ _0.1626 _0.2579∗∗ _0.1593 _0.2772∗∗ _0.1609 (0.0960) (0.1070) (0.0892) (0.0912) (0.0930) (0.0902) (0.0893) (0.0919) Unknown 0.0025 -0.0004 -0.0175 -0.0113 -0.0187 -0.0088 -0.0163 -0.0115 (0.0140) (0.0125) (0.0201) (0.0173) (0.0203) (0.0172) (0.0201) (0.0173) Flat -0.0138 -0.0122 -0.0145 -0.0090 -0.0126 -0.0048 -0.0117 -0.0093 (0.0136) (0.0124) (0.0194) (0.0167) (0.0197) (0.0165) (0.0195) (0.0166) Combination roof 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 (.) (.) (.) (.) (.) (.) (.) (.) Gable roof 0.0223 0.0244 0.0271 0.0267 0.0268 0.0301 0.0286 0.0277 (0.0191) (0.0185) (0.0257) (0.0256) (0.0259) (0.0255) (0.0260) (0.0255) Other -0.0112 -0.0362 -0.0240 -0.0317 -0.0197 -0.0266 -0.0216 -0.0298 (0.0307) (0.0232) (0.0711) (0.0457) (0.0728) (0.0442) (0.0700) (0.0458) accessD 0.0059 0.0247 -0.0175 0.0017 -0.0117 0.0000 -0.0136 -0.0014 (0.0210) (0.0202) (0.0228) (0.0212) (0.0232) (0.0207) (0.0231) (0.0212) 37

(39)

garden -0.0052 -0.0193 -0.0132 -0.0210 -0.0084 -0.0177 -0.0110 -0.0211 (0.0269) (0.0230) (0.0396) (0.0307) (0.0391) (0.0315) (0.0387) (0.0303) parking 0.0056 0.0016 0.0068 -0.0026 0.0053 -0.0014 0.0051 -0.0019 (0.0101) (0.0092) (0.0139) (0.0117) (0.0139) (0.0119) (0.0139) (0.0118) 101 0.0000 0.0000 0.0000 0.0000 (.) (.) (.) (.) 102 -0.4405∗∗∗ _-0.4101∗∗∗ _-0.3949∗∗∗ _-0.4042∗∗∗ (0.0364) (0.0414) (0.0416) (0.0409) 103 -0.4984∗∗∗ _-0.4292∗∗∗ _-0.4101∗∗∗ _-0.4239∗∗∗ (0.0317) (0.0369) (0.0380) (0.0370) 105 -0.1178∗∗∗ _-0.0674∗∗ _-0.0680∗∗ _-0.0716∗∗ (0.0132) (0.0218) (0.0220) (0.0223) 106 -0.5656∗∗∗ _-0.5048∗∗∗ _-0.4994∗∗∗ _-0.5030∗∗∗ (0.0191) (0.0220) (0.0224) (0.0222) 107 0.0306 0.0996∗∗∗ _0.1046∗∗∗ _0.1006∗∗∗ (0.0157) (0.0250) (0.0257) (0.0257) 108 -0.2233∗∗∗ _-0.1551∗∗∗ _-0.1522∗∗∗ _-0.1579∗∗∗ (0.0280) (0.0315) (0.0323) (0.0320) 109 -0.1666∗∗∗ -0.0938∗∗∗ -0.0941∗∗∗ -0.0947∗∗∗ (0.0166) (0.0201) (0.0206) (0.0206) 38

(40)

110 -0.7936∗∗∗ _-0.7343∗∗∗ _-0.7359∗∗∗ _-0.7408∗∗∗ (0.0257) (0.0284) (0.0302) (0.0297) VvE -0.0051 -0.0111 0.0308 0.0139 0.0278 0.0151 0.0286 0.0155 (0.0160) (0.0149) (0.0241) (0.0217) (0.0237) (0.0212) (0.0241) (0.0219) Constant 8.3844∗∗∗ _8.5362∗∗∗ _8.5393∗∗∗ _8.7776∗∗∗ _8.5788∗∗∗ _8.7639∗∗∗ _8.5690∗∗∗ _8.7750∗∗∗ (0.1315) (0.1246) (0.1835) (0.1502) (0.1898) (0.1494) (0.1892) (0.1525) Dependent var lnp lnp lnp lnp lnp lnp lnp lnp Postcode PC3 PC4 PC3 PC4 PC3 PC4 PC3 PC4 R2 _0.920 _0.942 _0.904 _0.940 _0.906 _0.942 _0.905 _0.941 Adjusted R2 0.917 0.938 0.900 0.934 0.901 0.935 0.900 0.934 RMSE 0.178 0.154 0.169 0.137 0.169 0.136 0.170 0.137 N 1842 1842 1090 1090 1090 1090 1090 1090

(41)

A.4 House prices

The main analysis is conducted on apartment prices, because this category of residential real estate had most observations. However, the advertisements also included houses on sale in Amsterdam. In this section, I give the main statistics and estimates on house prices as well. The conclusion does not differ from apartments.

Summary statistics on house prices are in Table A.3. Full ownership is again more

valuable than ground lease. The category of Usage right does not exists for individual

houses.

Table A.3: Summary statistics of asking price, by ownership situation

Ownership situation mean sd N

Full ownership 1438630 1360333 108

Ground lease 553779 553516.5 361

On request 548546.9 358082.9 9

Total 753604.9 886422.4 478

Source: houses.dta

Table A.4 describes the construction periods. Ground lease is most frequent except for the oldest houses category.

Table A.4: Construction period by ownership situation

Full ownership Ground lease On request Total

Before 1906 57 6 0 63 1906-1930 15 22 0 37 1931-1944 3 16 0 19 1945-1959 4 32 0 36 1960-1970 6 35 1 42 1971-1980 4 8 1 13 1981-1990 4 49 3 56 1991-2000 6 79 1 86

(42)

2001-2010 5 85 1 91

2011 and after 4 29 2 35

Total 108 361 9 478

Source: houses.dta

Table A.5 shows the estimates for ownership situation. Due to the small sample size,

most of the coefficients are not significant for the sub-samplenot paid in advance ground

lease observations. The main negative effect of houses on ground lease remains approxi-mately 9 percent.

Table A.5: Estimation Results

(1) (2) (3) (4)

All All Not PA Not PA

Ground lease -0.0894∗ -0.0968∗ -0.0233 -0.0525 (0.0449) (0.0475) (0.0691) (0.1306) On request -0.1368∗ -0.1573∗ -0.1178 -0.3222 (0.0576) (0.0703) (0.0881) (0.2166) R2 0.939 0.959 0.946 0.977 Adjusted R2 _0.931 _0.946 _0.919 _0.941 RMSE 0.180 0.159 0.226 0.193 N 341 341 120 120 Postcode PC3 PC4 PC3 PC4

Table A.6 shows the coefficients of the years paid in advance (time till end), similar to Table 6. The effect on logarithm of asking house prices with the more detailed location variable is almost the same as reported by Gautier and van Vuuren (2017). However, this .12 percent per an additional year is not significant at any conventional level.

Table A.7 contains the coefficients of ten year bins for the expiry date of paid off peri-ods. Both this, and Table A.6 estimates are controlled for the conventional explanatory

Rationality check on the Dutch housing market : an empirical study