Determinants of the time on market for new home projects

(1)

Determinants of the time on

market for new home projects

Title: Determinants of the time on market for new home projects Student: ir. M.R. (Mark) Keuter

Student number 0628972

E-mail: mark.keuter@am.nl

(2)

Date: September 2018

School: Amsterdam Business School

University: University of Amsterdam

Programme: MBA Big Data and Business Analytics

Attestation

I understand the nature of plagiarism, and I am aware of the University’s policy on this. I certify that this thesis reports original work by me during my University project.

Mark Keuter

(3)

Samenvatting

Summary

Where the Dutch housing market is dominated by private buyers and sellers, ten percent of the yearly transactions are a result of new supply offered by real estate developers. This new supply differs from the resale market since multiple similar dwellings are put on the market at once and the transaction price is fixed instead of a result of negotiations between buyer and seller.

A real estate developer can often influence two factors while putting a project on the market: the project size and price of the dwellings. Since maximizing profit is usually their goal, developers are looking for the optimal mix to be able to carry out the project.

In previous researches adding supply is often viewed as the result of costs and revenues. When the costs of construction and land are low in relation to the market price of existing dwellings, companies will cease the opportunity to add new supply and make money. Studies have shown that markets should be considered at a regional level, since housing markets operate very locally. In addition, recent research has shown that prices can’t be considered without taking the transaction volume into account. The sum of price and volume defines the liquidity of the housing market. The time on market of dwellings is determined by the market pressure and the asking price.

All transactions in the period 2011 to 2017 by real estate developer AM are analyzed to determine which factors affect the time on market of a project instead of an individual dwelling. The target is to realize the sales threshold of 70 percent to be able to start construction. A regression analysis is performed based on the project’s characteristics in relation to the market conditions and macro-economic developments.

Next, the effect of an early success has been researched. When the time on market for the first sales in a project are a proxy for the time on market for the whole project, the early success could be used as an estimator in the regression. This is referred to as path dependency.

The largest effects on time on market are the changing market circumstances and the changing macro-economic variables. Specifically, a change in the mortgage interest rate hugely effects the time on market, although the sign is counterintuitive. The effect of early success is substantial and statistically very significant, which proves the supposed path dependence within the sales process. The effect of price setting and project size by the developer on the project’s time on market seems very small.

(4)

Samenvatting

Waar de Nederlandse huizenmarkt wordt gedomineerd door particuliere kopers en verkopers, bestaat 10 procent van de jaarlijkse transacties uit nieuw aanbod dat door projectontwikkelaars wordt aangeboden. Dit nieuwe aanbod verschilt van aanbod uit de bestaande voorraad, omdat in één keer meerdere vergelijkbare woningen op de markt worden gebracht en dat de transactieprijs vaststaat in plaats van het resultaat van onderhandelingen tussen koper en verkoper.

Een projectontwikkelaar heeft bij het daadwerkelijk vermarkten van een project vaak nog invloed op twee factoren: de omvang van de verkoopfase en de prijs van de woningen. Omdat winstmaximalisatie meestal het doel is, zoekt de ontwikkelaar naar de optimale mix om het project daadwerkelijk tot uitvoering te kunnen brengen.

In eerder onderzoek wordt het toevoegen van nieuw aanbod vaak als een resultante van kosten en opbrengsten gezien. Zijn de kosten van bouw en grond relatief laag ten opzichte van de marktprijs van bestaande woningen, dan zullen bedrijven een kans zien om nieuwe woningen aan te bieden en hiermee geld te verdienen. Hierbij is het wel van belang dat de lokale markt wordt beschouwd, aangezien huizenmarkten bijzonder regionaal opereren. Tevens blijkt uit recent onderzoek dat huizenprijzen niet los van het transactievolume kunnen worden gezien. Het totaal van prijs en volume definieert de liquiditeit van de huizenmarkt. De verkooptijd van huizen wordt bepaald door de marktdruk en de vraagprijs. Om te bepalen welke factoren invloed hebben op de verkooptijd van een nieuwbouwproject in plaats van een individueel huis, zijn de transacties in de periode van 2011 tot 2017 van projectontwikkelaar AM geanalyseerd. Het doel is om de zogeheten verkoopdrempel van 70 procent binnen een project te behalen, zodat de bouw gestart kan worden. Op basis van projectkenmerken in relatie tot de marktomstandigheden en macro-economische ontwikkelingen is een regressieanalyse uitgevoerd.

Daarnaast is gekeken wat het effect is van een vroegtijdig succes. Indien de verkoopsnelheid in het begin een goede indicatie geeft voor de uiteindelijke verkooptijd, kan deze als schatter worden gebruikt. Dit effect wordt padafhankelijkheid genoemd. De grootste effecten op de verkooptijd blijken de wijzigende verkooptijd van de concurrerende huizenmarkt en de wijzigingen in de macro-economische variabelen. Met name een wijziging in de hypotheekrente blijkt een groot effect te hebben, alhoewel dit een onverwachte richting heeft. Het effect van een vroegtijdig succes is ook substantieel en statistisch zeer significant, wat een bewijs levert voor de veronderstelde padafhankelijkheid. De invloed van prijsstelling en omvang van het project door de ontwikkelaar op de verkoopsnelheid lijkt zeer beperkt.

(5)

Acknowledgements

Two years ago, I was given the opportunity by the management at my company to broaden the mind and start an MBA study with a specialization on Big Data & Business Analytics. Although it was an intense period, it was very insightful, a fantastic learning experience and has brought me much. I would like to thank Mario, Sander and Elsbeth for their support and giving me this opportunity.

With regard to this thesis, I wish to thank Dr. Martijn Dröes and Prof. dr. Marc Salomon for their supervision and guidance during the research and writing of the report. Special thanks go to my colleagues Lex, Marco and Jeffrey for helping me gather the data from the different systems for the research.

I would like to thank all my fellow MBA students, my family and my friends for supporting me during this study. Finally, a special thanks to my girlfriend Sabine for her sacrifices, support and patience.

Mark Keuter

(6)

Table of contents

1 Introduction

In the Dutch housing market approximately ten percent of the yearly transactions of privately sold dwellings is new supply. To meet the demand in housing, new developments are initiated. This new supply will be added to the existing stock. Although research (Mayer and Somerville, 1996) suggests these developments are driven by hotness or coldness of the market, the development process usually takes several years because of the required permits by the authorities. The market circumstances might encourage or delay housing starts depending on the sales time and house prices. In the Dutch market, specifically the Randstad, the demand for dwellings is typically larger then supply, so there is hardly a risk of a surplus. Also, property developers and construction companies need to keep their organizations in operation, which requires continuous sales and construction to generate revenue.

In the housing market the price setting of a new property on the market often is based on a hedonic approach (Rosen, 1974). Based on all the characteristics of the house, comparable dwellings can be used to determine the price. The real estate agent hired to sell the property can give a good estimation on the price level as well as the expected selling time based on experience and references. Finally, at the end of this process, the buyer and seller agree on a transaction price. This transaction will then be used as a reference for future supply.

The same approach cannot be used for new supply offered by real estate developers. Once a development is put up for sale, it acts as new supply on the market. But these new dwellings have a different playing field. The prices are fixed rather than a result of a supply and demand driven market. The characteristics of the dwellings are not specified in detail, since the first owner can choose different floorplans and in case of single family homes one often choose to enlarge the living space. The specific location often doesn’t have any transaction to compare with, which makes it relatively difficult to determine the market value. In such an environment other metrics, such as the time on market of new supply, play a more important role.

Since the development and especially construction of property is capital intensive, additional regulation is put in place to protect consumers of seller’s bankruptcy during construction. The seller must sell 70 percent of the project within six months from the start of the sale, otherwise the construction cannot start, and the signed contracts will be terminated.

This research tries to identify what determines the speed in which this first 70 percent of a project is sold and whether these are the characteristics of the project, the housing market circumstances at the specific location or macro-economic effects. This is important

(8)

Introduction

information for developers since they want to maximize profit, but also need organizational continuity and seek the optimal tradeoff for price setting and project volume. In sum, this thesis examines the following research question:

What are the main determinants for the time on market for new home developments: project’s characteristics, the competitive housing market or macro-economic effects?

To research the time on market of new housing developments, all the transactions in the period of 2011 to 2017 of AM (a large Dutch real estate developer) will be used. In total there are 173 unique projects. First, the time on the market for these projects will be determined based on the sales threshold of 70 percent, which determines whether and when a project will be constructed. Next, the project’s price levels and volume are compared to the actual levels of the housing market in the same period and municipality. These are the main independent variables and a set of control variables includes the percentage of single family homes within the project and the local housing market’s median time on market. To control for macro-economic effects, a measure of consumer trust, the interest rate and a measure of the housing market confidence are included in the regression. Since the market circumstances and the macro-economic variables change during the project, they are controlled for by taking the relative changes into account in the baseline regression.

The effects of price setting and project size, which can be influenced by the developer, on the time on market are small. The largest effects come from the local market’s time on market, the consumer trust and the change of interest rate and housing market confidence. These regressors are all statistically significant and have comparable values across different robustness tests. However, mixed evidence was found in the regression results. The coefficients on local market’s time on market and consumer trust have the expected sign, but the interest rate as well as the housing market confidence have a counterintuitive effect. An increase of the interest rate will decrease the project’s time on market and an increase of the housing market confidence leads to an increase of the time on market. Several researches have been conducted on path dependency regarding urban development and economic science. Path dependency means that the probability of a particular event to occur is affected by the events that have taken place in the past (Frenken and Boschma, 2007). The contribution of this research lies in providing insight in the effect of early success on the threshold sales time of a new home development.

To research the effect of this early success on the 70 percent threshold time on market, the early success is defined as the time on market for the 25 percent threshold, e.g. the time it takes to sell the first 25 percent of a project. This is used as a variable in the baseline regression.

(9)

Introduction

The effect of the early success is statistically significant (p=0.000), where a change of 10 percent leads to a change of 3.3 percent on the time of market of the 70 percent threshold. The other variables have similar effects as seen in the baseline regression, although the effect of change in housing market confidence bisects.

The overall explained variance measured with R-squared is around 0.6, which suggests there might be other unknown characteristics which can increase the reliability of the predictions.

The model can potentially be used to forecast the expected time of construction starts, since this requires additional liquidity by the developer to finance the unsold properties. For AM, being part of a public listed company, it can even be used to forecast number of dwellings sold, which is an important metric in the annual reporting.

The remainder of this thesis is organized as follows. Chapter 2 starts with an overview of the existing literature on transaction prices in the housing market at macro level. Next, the regional aspect is discussed where local markets are seen as independently operating markets. The relation between transaction price and transaction volume gives a different approach on the housing market. In the end the theories on selling time – or time on the market – and path dependency are discussed.

Chapter 3 describes the used dataset and the additional data that is collected. The dataset is then cleaned, complemented and analyzed to create a model with the sales threshold as the dependent variable of the time on the market. The additional market data and data on macro level are added to the dataset and the methodology is described. After the analyzing first results additional tests on robustness are executed and path dependency is analyzed.

Recommendations on further research are provided in Chapter 4 and conclusions and discussion on the research question can be found in Chapter 5.

(10)

Literature

2 Literature

2.1 House prices: supply and demand

There have been several researches on the relation of housing supply, housing price and housing cost. Most studies were focused on the housing price. DiPasquale (1999) published a review paper to analyze why there was so little research on the supply side of the housing market. In her overview she started with Muth (1960), who conducted one of the first empirical studies on house prices and housing production. He found no statistical evidence for a relation between price and production. These results were later confirmed by Follain (1979) with more recent data. Critics argued that since both Muth and Follain used national data, there might have been aggregation bias in these studies. Stover (1986) introduced a regional effect in his model by distinguishing sixty-one housing markets, but he couldn’t find proof for a causal relation between house price and production either. With an investors perspective, Poterba (1984) and Topel and Rosen (1988) argued that the decision of adding supply was given by the cost of construction compared to the market price. They both ignored the cost of land since the information on these costs was not available. They also failed to find a significant relation between construction costs and housing starts.

Although the cost of land was considered to be relevant in the cost-oriented approach, it took a decade before the next research was published. DiPasquale and Wheaton (1994) took the cost of land into account, but to their own surprise they couldn’t find a significant relationship between the total costs and housing starts. They did find a relation between sales time and construction costs.

Mayer and Somerville (2000) developed a model which accounted for the time it takes to go through a development process. They constructed a model where housing starts are a function of price and cost changes rather than levels.

2.2 A regional approach

Most of these researches are on a macro-level. But since housing markets are considered local markets, this could end up in an aggregation bias. Goodman (1998) differentiates several factors that specify this local market heterogeneity. On the demand side this could be because of difference in income levels and demographic and ethnic composition of the population. On the supply side, there are geography, land-use regulations and local market power. Although the estimates on the aggregated level are close, preferably lower level information is used on housing market modeling.

Based on this regional aspect of the housing market, one would expect a relation between the added stock and the regional price development. De Vries and Boelhouwer (2005) tried

(11)

Literature

to find this relation and used the market areas from the Dutch real estate agent society (NVM) as grouping, since they are assumed to operate as more or less independent housing markets. They modelled the change in price as a function of the change in housing stock. They have noticed an inverse relationship between price development and housing production in specific regions where large developments were initiated, indicating that an increase in supply triggers a fall in prices. Unfortunately, because of scarcity ot their data, they could not prove a significant relation.

Ooi and Le (2012) focused on the Singapore market, where - because of government regulation - much more data is available on housing starts and transactions. In Singapore new supply is put on the market with a presale mechanism, comparable with the Dutch sales technique. The new dwellings are being sold before completion, so the transactions will influence future stock. They developed a VAR-model, where change in supply influences the change of price and vice versa. They also added change of average income and change of interest rate as exogenous variables. Interest rates have an inverse impact on house prices, income rates are not correlated with house prices. They separated two markets: the market for new homes as the primary market and the market for existing dwellings as the secondary market with different effects. New supply in the primary market has a negative relation on the price, which is consistent with their ‘competitor’ hypothesis that there is more to choose from. They also found a positive relation on new supply and the price in the secondary market. This supports their ‘contagion’ hypothesis, where developers are viewed as price leaders, because of their information advantage and ability to predict the market.

Almost a decade later Van Dalen and De Vries (2013) build on the earlier hypothesis of De Vries and Boelhouwer, but had more data available for their research. They found a small, but significant relation between the actual build dwellings and the price development of the existing stock, but there is a lag of 2 quarters. Apparently, it takes some time to develop a new equilibrium on the housing market when additional supply is added.

Although the local markets affect the sales price, Clapp et al. (1995) differentiate them in three sets that influence housing prices: (1) purely local determinants; (2) causes that affect neighboring towns, but not the entire area; (3) factors that exert their influence on the entire area. This so-called ripple effect was later researched in Ireland (Stevenson, 2004) and Taiwan (Chien, 2010; Lee and Chien, 2011), where house prices in most neighboring markets converge after a shock in one. But there are exceptions. A regional global city like Taipei City appears to be a market on its own and does not ripple out.

(12)

Literature

2.3 House prices and transaction volumes

The relation between price and volume in the housing market was the subject of a research by Clayton et al. (2008). Analyzing the data of 114 metropolitan areas in the United States from 1990 to 2002, they found that, next to the effect of exogenous variables on both home prices and trading volume, decreases in price reduce trading volume, but increase in prices have no effect. This is however a Granger causality, so the effect occurs in the next time period. Trading volume also Granger causes home prices, but only in markets with inelastic supply.

Droës and Francke (2017) researched the European housing market and stated that transaction volume should be taken in account when researching housing price development. In a study of sixteen European housing markets in the period of 1999 to 2013 they discovered an autoregressive effect in their VAR-model. Change of price and change of volume in the previous period both have a positive effect on the price. Change of price in the previous period has a negative effect on the volume but change of volume in the previous period still has a positive effect.

2.4 Time on the market

Yavaş and Yang (1995) introduced the time on the market to measure the selling time in relation to the price setting. Their results suggest that for mid-price properties, a higher asking price can significantly increase time on the market. In contrast, the asking price has no significant effect on time on the market for low-price and high-price properties.

How time-to-transaction depends on the tightness of the market is researched by Novy-Marx (2009). When the tightness is low, there are more sellers than buyers. The expected time-to-transaction for the buyers is relatively low and the expected time-to-transaction for the sellers is relatively high. In a market with high tightness this is vice versa. When the market is in equilibrium the expected time-to-transaction would be the same for buyers and sellers.

Carillo and Pope (2012) compared the time on the market in a hot market (2003) and a cold market (2007). They did not only look at the mean or median but observed the distribution of time on the market. They concluded the change in shift of the distribution is complex and the changes of house characteristics could only explain a very small part of this shift.

In his research of the effect of price change on the time on the market for the Sydney housing market Khezr (2015) found evidence that overpricing increases time on the market. This intuitive relation was consistent with what is expected for the housing market but more significant than similar findings in the literature.

(13)

Literature

Many of the researches focus on either the price or the time on market as the dependent variable. Dubé and Legros (2016) developed a two-stage least squared approach, where the market circumstances are considered as exogeneous variables used in the first stage to estimate both price and time on market, which were then used in the second stage to estimate respectively the time on market and the price. They found that houses of better quality and having more positive amenities sell at higher prices and take less time to sell.

2.5 Path dependency

The theory of path dependency has its origins in social sciences and economic in particular but has been used in urban development as well. It suggests inertia within a sequence of events and an unwillingness to change unless another contingent event intervenes. Booth (2011) concludes therefore it is not just a vague way of saying past events may influence future ones, but that it puts greater emphasis on earlier than on later events.

The concept of path dependency resulted in a framework developed by Frenken and Boschma (2007) to provide ‘a micro-foundation of economic geography in terms of the interplay between industrial dynamics and urban growth’. It illustrates how the occurrence of an event changes the probability of a next event to occur. In their example the occurrence of innovation within a company generated strong path dependencies in spatial concentration of industries and specialization of cities, where gains on both firm size and city size are also expected.

(14)

Data and methodology

3 Data and Methodology

3.1 Data

To research the time on market for new developments, a transactional dataset is used. The dataset consists of all transactions of AM in the period from January 2011 until May 2018. The transactions to investors are not being used since the registered signing dates are no result of market behavior. Since the dataset is originated from the financial system, the definition of a project differs from the observed truth. Based on the registered dates within the dataset a new grouping of projects is made, which results in 205 projects with a total of 4,316 observed transactions.

The dataset is pooled cross-sectional since each project is its own entity with characteristics. The transaction can only occur once, so for every transaction there is only one observation.

The projects are located throughout the whole country, but the majority is in the Randstad, the megalopolis of the Netherlands. A visualization of the municipalities where these projects are located is shown in Figure 1.

(15)

Each project has its own starting date. The spread of the projects throughout the years is shown in Figure 2.

Figure 2: number of projects per year

In the Netherlands it is typical for new home developments the dwellings are sold from drawings. The whole project is put on the market at once and the fixed selling price is set. Consumers can show their interest and when there are multiple interests per lot, the lot gets assigned to one of them. For the other dwellings it is custom to apply first come, first serve.

When signing the sales contract, there are multiple precedent conditions. One of them is that the majority of the dwellings within the project are sold within six months (usually 70 percent). If this condition is not met within six months, the developer can extend the period with another six months. If the condition is still not met, the sales contract will be terminated. This condition is required by Stichting Waarborgfonds Koopwoningen, a foundation to protect customers against developers and constructors taking too much financial risks during the construction of a project which is largely financed by the signed customers. The foundation issues an insurance to guarantee the houses will be build.

When the condition is met within 6 months, the real estate developer will grant contract to start the construction. This means there are no further precedent conditions and the original contract is binding. This is also the moment the buyers of the property can go to the notary for the deed. At the notary the first invoice is paid and at this time the real estate developer generates revenue.

Because the dataset lacks a reliable date for the start sale of a project, the date of the first sale within a project is considered to be the starting date. For all other transactions the

10 7 23 32 47 29 25 0 5 10 15 20 25 30 35 40 45 50 2011 2012 2013 2014 2015 2016 2017

(16)

sales dates are calculated to determine time on the market in days. Based on this time on market the running percentage of the cumulative sales is calculated. This results in a cumulative sales curve per project. The cumulative sales curves of a sample of six projects are shown in Figure 3. All curves show a steep slope for the first half of the project, after which the slope flattens until all dwellings within the project are sold. The horizontal line represents the 70 percent threshold as described above. Different projects reach the 70 percent threshold at different points in time. It is exactly the variation in this ‘project time on market’ that is investigated in this thesis.

Figure 3: sales curve per project

3.2 Dependent variable

Since at least 70 percent of the project needs to be sold within a specified time before it actually will be build, the time on market per project to achieve this threshold is considered an important measurement. To calculate this time for each individual project, the first sale above the threshold is considered as determinant for this period. The red dot in Figure 4 represents the first sale above the threshold on time k. The sales time of this dwelling determines the project’s time on market for the threshold, PTOM70i,k-t for project i during

(17)

Figure 4: sample sales curve to determine PTOM

Because the effect of an early success is taken into account, the time on market for this is calculated in a similar way. A threshold of 25 percent is set and the green dot in Figure 3 represents the first sale above this threshold. The sales time of this dwelling on time m determines the project’s time on market for the early success, PTOM25i,m-t.

Figure 5: histogram of project time on market for threshold 70 percent

Projects with a total sales time below ten days or a median time on market of over a year are considered outliers and are dropped from the dataset. Since many of the projects in

(18)

2018 are not completely sold and the market data is incomplete, the projects from the year 2018 are dropped as well. This resulted in a dataset with 173 observations. A histogram of the main dependent variable (PTOM70i,k-t) is shown in Figure 5. Although the larger part of

these projects have the time on market within half a year, there are some projects that exceed this period. Further descriptive statistics of PTOM70i,k-t and the other variables used

in the analysis are reported in Table 1.

Table 1: Descriptive statistics (N = 173)

variable

mean

std

min

max

PTOM70

120.422

90.317

5

357 logPTOM70

_4.447

_0.916

_1.609

_5.878

logPTOM25

2.870

1.183

0

5.673 TR

_0.167

_0.231

_0.006

_1.280

PR

1.103

0.241

0.363

2.136 PSFH

_0.899

_0.286

₀

₁

logMTOM

129.497

58.493

24

304 ΔMVOL

_0.120

_0.259

_-0.321

_1.263

ΔMPM2

0.024

0.047 -0.126

0.163 ΔlogMTOM

_-0.133

_0.273

_-0.974

_0.439

CT

1.312

18.826 -41

26 IR

_0.033

_0.008

_0,023

_0,052

HMI

96.295

20.622

51

121 ΔCT

-0.692

3.090 -14

13 ΔIR

-0.054

0.055 -0.247

0.056 ΔHMI

0.060

0.115 -0.119

0.574

3.3 Independent variables

Because the dwellings differ in size, the price per square meter is calculated to make it a comparable variable. These variables are aggregated to the project where the median of the price per square meter is used as the project price per square meter (PPM2it).

The number of dwellings within the project determine the total size of the project, described as the project volume (PVOLit).

Within a project there may be different types of houses. To capture this difference, the percentage of single-family homes within the project is calculated as variable PSFHit. This

variable is somewhat like a dummy variable, since many of the projects consists of only single-family homes or multi-family homes.

When a project is put on an existing (resale) housing market, the characteristics of this market should be considered. The Dutch Association of Real Estate Brokers and Real

(19)

Estate Experts (NVM) offers regional market statistics based on their market share. The market share of NVM is over seventy percent in The Netherlands (Dröes and Koster, 2016). The regional market statistics are added to the original dataset. A set of municipality records is used to provide geographically relevant statistics. This panel data includes the median price per square meter (MPM2tl), median time on market (MTOMtl) and the

transaction volume (MVOLtl).

Because all the project’s observed variables should be considered relatively to the surrounding housing market, several ratios are calculated. The price ratio (PRit) is

calculated by dividing the median price per square meter of the project (PPM2it) by the last

known median price per square meter of the municipality at the start sale of the project (MPM2tl). The transaction ratio (TRit) is calculated by dividing the project size (PVOLit) by

the last known quarterly transaction volume of the municipality (MVOLtl). The idea is that

particularly deviations from market prices and volumes have an impact on the time of market of the project. When prices are relatively high, the time on market will increase. When the project size is relatively high to the market’s transaction volume, the time on market will increase as well.

The direct relation of the project specific variables with the project time on market is shown in the different scatterplots in Figure 6.

Figure 6: scatterplots of variables

Most projects have transaction rates between 0 and 0.1, meaning the project volume is not larger than 10 percent of the quarterly transaction volume in the local housing market.

(20)

There is no direct relation between the transaction rate and the project’s time on market. Many of the projects have a price rate between 1 and 1.25, indicating the prices of the dwellings are set above the market’s median value. Again, no direct relation between the price rate and the project’s time on market can be observed. This also holds for the percentage single-family homes, but the dummy-like characteristic of this variable is visible. There are only a few projects with a mix of single and multi-family homes.

The market’s last known values are used since these are observed when determining the final price and volume of the project at the start of the sale. The changes during the sales of the project are taken into account by the relative change during the sales period.

ΔlogMTOMk-t,l is calculated by the difference in logMTOMkl at the threshold time on market

(time k) and the initial logMTOMtl at t=0. ΔMPM2k-t,l is the relative change of the median

price per square meter at t=k to the initial price at t=0. In a similar way, ΔMVOLk-t,l is the

relative change in the market’s transaction volume.

The consumer trust is an indicator determined by Statistics Netherlands (CBS). The outcomes of a monthly survey reflect the view of Dutch consumers on their economic situation. Consumer trust is a good indicator for the willingness to buy luxury products but can also be used for large purchases. The monthly indicator of this time series is used as variable CTt. ΔCT is the relative change of this indicator during the sales process.

Since the majority of Dutch dwellings for sale are financed with a mortgage, the interest rate could be relevant for the housing market. Based on the historical data provided by the Dutch National Bank (DNB), the monthly median interest rate for mortgages with a fixed interest for five to ten years of this time series is used as variable IRt. ΔIRk-t is the relative

change.

The Dutch Homeowners' Association (VEH) reports a monthly housing market indicator of the consumer trust in the housing market. This indicator is a three-monthly moving average based on a representative survey in The Netherlands. It is used as variable HMIt. Finally,

the relative change is given by ΔHMIk-t.

The year a project has been put to sale is determined to take the time aspect into account. These will be used as time fixed effects. For location fixed effects, the province where the project is located is used.

When estimating the project threshold time on market, the project’s size and median price per square meter are expected to have a large effect. Since the relative size and price are used, the regressors for TRit and PRit are both expected to be positive. In a similar

reasoning the market’s median time on market is also expected to have a positive regressor.

(21)

The percentage of single family homes within the project is expected to have a negative regressor, since these projects are typically easier to split in different sales phases. An apartment block cannot be constructed in different phases and therefor needs to be sold in one phase as well.

Consumer trust is an indicator to show how the consumer experiences the economic situation. A positive indicator would reflect trust in the economy, which would lead to more buyers on the market and a lower time on market for objects.

A lower interest rate would stimulate consumers to buy new dwellings, so a positive regressor is expected to increase time on market with higher interest.

The housing market indicator reflects the specific consumer trust in the housing market. It is more narrowed than the generic consumer trust since it also accounts for the tightness of the housing market.

An overview of the expected signs of the coefficients on the discussed variables above can be found in Table 2.

Table 2: expected signs of coefficients on independent variables

variable

expected sign

logPTOM25

+

TR

₊

PR

+

PSFH

_-

logMTOM

+

ΔMVOL

_-

ΔMPM2

-

ΔlogMTOM

₊

CT

-

IR

₊

HMI

-

ΔCT

-

ΔIR

+

ΔHMI

-

3.4 Empirical model

The objective is to determine the impact of different variables on the project time on market for the threshold of 70 percent. This target variable is not expected to be normally distributed. In addition, the estimated relationships are most likely not linear. As such, a log-linear model (Stock & Watson, 2015) is used. Since the regional housing market is

(22)

expected to be of importance, the relation between time on market, transaction price and transaction volume is analyzed.

The hypothesis is that the project time on market can be influenced by the price setting and the volume of the project. To determine the impact of the project specific variables on the project time on market, the log of time on market for the seventy percent threshold is estimated with a regression on the relative price (median price per square meter), the relative volume (project size per quarterly market transaction volume), the percentage of single-family homes in the project and the median time on market in the same municipality. This results in the following baseline regression:

log(𝑃𝑇𝑂𝑀70𝑖,𝑘−𝑡) = 𝛽0+ 𝛽1𝑇𝑅𝑖𝑡+ 𝛽2𝑃𝑅𝑖𝑡+ 𝛽3𝑃𝑆𝐹𝐻𝑖𝑡+ 𝛽4𝑙𝑜𝑔𝑀𝑇𝑂𝑀𝑡𝑙 + 𝛽5𝐶𝑇𝑡+ 𝛽6𝐼𝑅𝑡+ 𝛽7𝐻𝑀𝐼𝑡+ 𝛽8𝛥𝑀𝑉𝑂𝐿𝑘−𝑡,𝑙 + 𝛽9𝛥𝑀𝑃𝑀2𝑘−𝑡,𝑙+ 𝛽10𝛥𝑙𝑜𝑔𝑀𝑇𝑂𝑀𝑘−𝑡,𝑙+ 𝛽11𝛥𝐶𝑇𝑘−𝑡 + 𝛽12𝛥𝐼𝑅𝑘−𝑡+ 𝛽13𝛥𝐻𝑀𝐼𝑘−𝑡+ 𝛼𝑙+ 𝜏𝑡+ 𝜀𝑖𝑡𝑘 (1)

where 𝑃𝑇𝑂𝑀𝑖,𝑘−𝑡 is the project time on market for the seventy percent threshold in days for project i on time t, 𝑇𝑅𝑖𝑡 is the transaction rate, 𝑃𝑅𝑖𝑡 is the price rate, 𝑃𝑆𝐹𝐻𝑖 is the percentage single-family homes and 𝑀𝑇𝑂𝑀𝑡𝑙 the market’s median time on market in location l, 𝐶𝑇𝑡 is the consumer trust index, 𝐼𝑅𝑡 is the monthly interest rate and 𝐻𝑀𝐼𝑡 is the housing market indicator. 𝛥𝑀𝑉𝑂𝐿𝑘−𝑡,𝑙 is the relative change of the market’s transaction volume during threshold time k-t, 𝛥𝑀𝑃𝑀2𝑘−𝑡,𝑙 is the relative change in market’s median square meter price during threshold time k-t, 𝛥𝑙𝑜𝑔𝑀𝑇𝑂𝑀𝑘−𝑡,𝑙 is the change in market’s time on market during threshold time t, 𝛥𝐶𝑇𝑘−𝑡 is the relative change in consumer trust during threshold time

k-t, 𝛥𝐼𝑅𝑘−𝑡 is the relative change of interest rate during threshold time k-t and 𝛥𝐻𝑀𝐼𝑘−𝑡 is the relative change in housing market indicator during threshold time k-t. To control for time and location, they are added as fixed effects as dummy variables.1

Since there are observations in more than one time period, but the individual projects are not repeatedly observed across the periods, pooled OLS is used as estimation method. Because of heteroskedasticity of the variables, the normal standard errors cannot be used (Stock & Watson, 2015). To address this heteroskedasticity, robust standard errors will be used.

1_{This baseline regression uses the last known exogenous variables at t=0, e.g. the reported statistics of the}

previous moth or quarter. A dynamic model using actuals at t=0 as regressors was researched as well, see Appendix 1. This approach did not improve the model, so (1) is viewed as the baseline relationship.

(23)

The early success could be a good estimator for the 70 percent threshold time on market as well. Either the projects dwellings meet the market’s demand or the (lack of) popularity of the project influences the behavior of other customers on their buying decision.

To examine the scope of the effect of path dependency three additional specifications are estimated. First, logPTOM25i,m-t is added to the baseline regression to see whether it has

an effect on top of the project characteristics. Then, logPTOM25i,m-t is assumed to capture

all relevant information at the start of the project and the project characteristics are omitted. Instead a separate equation estimates whether the effect of project characteristics goes through the success at the start of the project. These equations are stated below.

As mentioned, the time on market for the 25 percent threshold is added to the baseline regression. log(𝑃𝑇𝑂𝑀70𝑖,𝑘−𝑡) = 𝛽0+ 𝛽1𝑇𝑅𝑖𝑡+ 𝛽2𝑃𝑅𝑖𝑡+ 𝛽3𝑃𝑆𝐹𝐻𝑖𝑡+ 𝛽4𝑙𝑜𝑔𝑀𝑇𝑂𝑀𝑡𝑙 + 𝛽5𝑙𝑜𝑔𝑃𝑇𝑂𝑀25𝑖,𝑚−𝑡+ 𝛽6𝐶𝑇𝑡+ 𝛽7𝐼𝑅𝑡+ 𝛽8𝐻𝑀𝐼𝑡 + 𝛽9𝛥𝑀𝑉𝑂𝐿𝑘−𝑡,𝑙+ 𝛽10𝛥𝑀𝑃𝑀2𝑘−𝑡,𝑙 + 𝛽11𝛥𝑙𝑜𝑔𝑀𝑇𝑂𝑀𝑘−𝑡,𝑙+ 𝛽12𝛥𝐶𝑇𝑘−𝑡+ 𝛽13𝛥𝐼𝑅𝑘−𝑡 + 𝛽14𝛥𝐻𝑀𝐼𝑘−𝑡+ 𝛼𝑙+ 𝜏𝑡+ 𝜀𝑖𝑡𝑘 (2)

where 𝑙𝑜𝑔𝑃𝑇𝑂𝑀25𝑖,𝑚−𝑡 is the log of the project time on market for the twenty-five percent threshold in days for project i during time m-t.

To determine the effect of an early start on the seventy percent threshold, the known variables at the start of the project are omitted from the regression and the project’s time on market is regressed on the early success and the changes of the exogenous variables during the sales process.

log(𝑃𝑇𝑂𝑀70𝑖,𝑘−𝑡)

= 𝛽0+ 𝛽1𝑙𝑜𝑔𝑃𝑇𝑂𝑀25𝑖,𝑚−𝑡+ 𝛽2𝛥𝑀𝑉𝑂𝐿𝑘−𝑡,𝑙 + 𝛽3𝛥𝑀𝑃𝑀2𝑘−𝑡,𝑙+ 𝛽4𝛥𝑙𝑜𝑔𝑀𝑇𝑂𝑀𝑘−𝑡,𝑙+ 𝛽5𝛥𝐶𝑇𝑘−𝑡 + 𝛽6𝛥𝐼𝑅𝑘−𝑡+ 𝛽7𝛥𝐻𝑀𝐼𝑘−𝑡+ 𝛼𝑙+ 𝜏𝑡+ 𝜀𝑖𝑡𝑘

(3)

Then, the logPTOM25i,m-t is estimated on the project characteristics and the market

circumstances at the start of the project. The changes of exogenous variables are not taken into account, since these are measured on a monthly or quarterly base and therefore expected to have no effect.

(24)

Data and methodology log(𝑃𝑇𝑂𝑀25𝑖,𝑚−𝑡) = 𝛽0+ 𝛽1𝑇𝑅𝑖𝑡+ 𝛽2𝑃𝑅𝑖𝑡+ 𝛽3𝑃𝑆𝐹𝐻𝑖+ 𝛽4𝑙𝑜𝑔𝑀𝑇𝑂𝑀𝑡𝑙 + 𝛽5𝐶𝑇𝑡+ 𝛽6𝐼𝑅𝑡+ 𝛽7𝐻𝑀𝐼𝑡+ 𝛼𝑙+ 𝜏𝑡+ 𝜀𝑖𝑡𝑚 (4) 3.5 Estimates

The results from the models discussed in Chapter 3.4 are shown in Table 3. The regression on the baseline, see equation (1), show variable results. The regressor on transaction rate has a positive sign, as expected, but is not statistically significant. Doubling the project size, and therefore doubling the transaction rate, would increase the project time on market by more than seven percent.

The regressor on price rate has a counterintuitive sign (increasing the dwelling’s prices would decrease the time on market). It is statistically significant on a 95% level. Increasing the median price with 10 percent relatively to the median market price would decrease the time on market with 6.4%. This might be indicative of price and time being simultaneously determined.

Most of the projects consist of only single-family homes or multifamily homes, so PSFHit is

almost a dummy variable. The project time on market for the seventy percent threshold is 40% lower than for multifamily homes and is statistically significant at the 90% level. The market’s median time on market has a dampened effect on the project time on market. When the market’s time on market is 10% higher, the project’s time on market 4% lower. This effect is statistically significant.

Changes of housing market circumstances during the sales process also have an effect on the project’s time on market. An increase of 10% in transactions decrease the project’s time on market of 2%. An increase of the median market price per square meter by 10% increases the project’s time on market by 10%.

The change of the market’s time on market has an unexpected negative effect. When the market’s time on market increases with 10%, the project’s time on market decreases with 5%. A possible explanation could be the dwellings sold in the secondary market are priced with another reference, making the project’s dwelling more (or less) attractive than the alternative.

From the macro-economic variables, the consumer trust has the most significant effect. The sign is as expected, and the effect is large, where an increase of consumer trust by 1 point leads to a decrease of the sales time of 4%.

(25)

The coefficient of the interest rate has a negative sign, but it is not significant at 10% significance level. The effect is quite large, where a decrease of 0.1% in the interest rate increases the project’s time on market by 12,5%. Although one would expect a decrease in interest rate to have a decreasing effect on sales time, the opposite could be true as well. Since the interest have been decreasing for the past years, an increase of the interest rate could signal the lowest point has been reached.

The housing market indicator has a small effect with a counterintuitive sign but lacks significance. An increase of 1 point leads to 0.3% increase of the project’s time on market. It was expected to be a better predictor, since it is an indicator that is focused on the housing market.

A change in consumer trust has a small, but insignificant effect. The sign of the coefficient is as expected.

A decrease of the interest rate by 5% has a large effect and significant on the project’s time on market. It will increase by more than 40%. It has the same sign as the coefficient on the interest rate at the start of the project, so the same reasoning can be applied.

A change in the housing market indicator has a large and unexpected sign as well. An increase of 5% would also increase the time on market by more than 15%.

Overall there are several variables that have significant value as predictor on all models. The largest effects are given by the changing market circumstances during the sales period of the project, locally with the changing market’s time on market (ΔlogMTOMk-t,l) and on

macro-economic level on changing interest rate (ΔIRk-t) and housing market indicator

(26)

Table 3: regression results on log project time on market treshold

(1)

logPTOM70

(2)

logPTOM70

(3)

logPTOM70

(4)

logPTOM25

TR

0.037024

(0.2844364)

0.0514549

(0.2508198)

-0.372513

(0.374197)

PR

-0.6418338 **

(0.2613614)

-0.4569552 **

(0.2276525)

-0.5487263

(0.458991)

PSFH

-0.3943325 *

(0.223477)

-0.4791539 **

(0.1952545)

0.1162301

(0.301753)

logMTOM

0.3905042 **

(0.1732956)

0.4398496 ***

(0.1447251)

0.02291

(0.2740713)

logPTOM25

0.3134101 ***

(0.0604752)

0.3272136 ***

(0.0621072)

ΔMVOL

-0.2040502

(0.2292453)

-0.0456483

(0.1907305)

0.1463785

(0.1630791)

ΔMPM2

1.075147

(1.52109)

0.8374602

(1.1152)

0.7838082

(1.208093)

ΔlogMTOM

-0.5128917 **

(0.2448103)

-0.3538927 *

(0.2024799)

-0.5397706 ***

(0.2067261)

CT

-0.0395812 ***

(0.0128996)

-0.0203975 **

(0.0093697)

-0.0515695 ***

(0.0173373)

IR

-125.7584

(87.29597)

-94.46404

(74.41016)

-83.32733

(87.72879)

HMI

0.002998

(0.032697)

-0.0072045

(0.026824)

-0.0294605

(0.0284962)

ΔCT

-0.0048028

(0.0140042)

-0.0129702

(0.0120268)

-0.0182191

(0.0128069)

ΔIR

-8.635329 ***

(3.237411)

-6.82665 **

(2.889766)

-6.318412 ***

(1.95372)

ΔHMI

3.193444 **

(1.552742)

1.529348

(1.192139)

1.551238 **

(0.6879192)

Location fixed effects Yes

Yes

Time fixed effects

Yes

R-squared

0.5591

0.6611

0.6187

0.2315

RMSE

0.66694

0.58677

0.60764

1.1145

Observations

173

173 Note: robust standard errors are reported in the parentheses below the estimated

coefficients. Individual coefficients are statistically significant at the *10%, **5% or

***1% significance level. Full report can be found in Appendix 2.

Table 2 reported the expected signs for the coefficients of all variables. In Table 4 the same variables and expectation can be found and the test on the regression results are added. Although half of the variables have a different sign as expected, for two out of three with the largest effects (ΔlogMTOMk-t,l and ΔIRk-t) an alternative reasoning is provided.

(27)

Table 4: resulting signs of coefficient on variables

variable

expected sign test on results

logPTOM25

+



2

TR

+



PR

+



PSFH

-



logMTOM

+



ΔMVOL

-



ΔMPM2

-



ΔlogMTOM

+



CT

-



IR

+



HMI

-



ΔCT

-



ΔIR

+



ΔHMI

-



2_{The independent variable logPTOM25 is added in equation (2) and (3). The results are discussed in Chapter}

(28)

3.6 Robustness of baseline results

Until 2014 the Dutch housing market was struggling because of the financial crisis and the recession that followed. From 2015 most, regional markets were moving in an upwards direction. Based on the historical trend of the consumer trust as shown in Figure 7, two periods can be identified, with one part predominantly negative and the other positive. The dataset is split in two periods: 2011-2014 (5) and 2015-2017 (6). The regression results are reported in Table 5.

Figure 7: Dutch Consumer Trust (source CBS)

-50 -40 -30 -20 -10 0 10 20 30 jan -11 me i-11 se p -11 jan -12 me i-12 se p -12 jan -13 me i-13 se p -13 jan -14 me i-14 se p -14 jan -15 me i-15 se p -15 jan -16 me i-16 se p -16 jan -17 me i-17 se p -17

Consumer Trust

(29)

Table 5: robustness of baseline model – two time periods (2011-2014, 2015-2017)

(5)

logPTOM70

(6)

logPTOM70

TR

0.1765191

(0.3034428)

0.3381441

(0.533367)

PR

-0.1380207

(0.4085177)

-0.6807442 *

(0.3759703)

PSFH

-0.3715726 *

(0.2160368)

-0.3302105

(0.374481)

logMTOM

-0.1151142

(0.2906822)

0.3974231 *

(0.2286648)

ΔMVOL

0.3144959

(0.247225)

-0.7945165 *

(0.45099)

ΔMPM2

-0.9680321

(2.115332)

1.806041

(2.179508)

ΔlogMTOM

-0.4512193

(0.395965)

-0.6152779 **

(0.3085586)

CT

-0.0316288 **

(0.0128791)

-0.0579518

(0.0355786)

IR

-3.127266

(75.47092)

-205.3098

(174.644)

HMI

0.0216083

(0.0273067)

0.0029059

(0.0668527)

ΔCT

0.004122

(0.0182918)

-0.0082372

(0.0269326)

ΔIR

-9.959757 ***

(1.524077)

-7.553561

(5.657757)

ΔHMI

2.534104 *

(1.44119)

4.377953

(2.7208)

Location fixed effects

Yes

Time fixed effects

Yes

R-squared

0.7820

0.4719

RMSE

0.49535

0.76567

Observations

72

101 Note: robust standard errors are reported in the parentheses below the estimated

coefficients. Individual coefficients are statistically significant at the *10%, **5% or

***1% significance level. Full report can be found in Appendix 2.

The results for period 2011-2014 report coefficients as preliminary expected. An increase in transaction rate has a larger effect than on the baseline regression, although it is not statistically significant. The effect of the interest rate is much smaller, but also lacks significance. The other coefficients are comparable with the baseline regression. The overall explained variance measured by R-squared increased to 0.78, which is quite

(30)

substantial. But because there are only 72 observations during this period, there is a possibility of overfitting.

For the period 2015-2017, the results are very similar as the total dataset. The loss of significance on the changing macro-economic variables is notable. The effect of the interest rate at the start of the sale doubles, where in the previous period it was much smaller. The overall explained variance decreased to 0.47. Overall, the baseline regression passes the robustness tests for the period during and after the Dutch housing crisis.

3.7 Early success

Table 1, columns (2) - (4), shows the estimates of equation (2) – (4). The main conclusion is that there is considerable path dependency in the project’s time on market on top of the project characteristics (column 2). This effect stands alone and is not determined by the project characteristics at the start (column 3 and 4).

In regression (2) the variable for early success was added. The time on market for the threshold of 25% determines the success of the early stage of the sales process.

This early success variable has a significant effect on the time on market for the 70% threshold. An increase of 10% results in an increase of 3%. The p-value is the lowest of all variables (0.000).

Most of the other variables have comparable results with the baseline regression (1). The signs and sizes of the coefficients are merely the same, as well as the significance levels. The effect of the change in volume during the sales period decreases a lot and is four times as small. The coefficient on change in housing market indicator bisects and loses its statistical significance. The overall explained variance measured by R-squared increases to 0.66.

These results indicate evidence for possible path dependency. To further research this hypothesis, another regression (3) is run on this early success variable omitting all variables that are known at the start of the sale.

Most of the coefficients are similar to those from regression (2). The early start

logPTOM25i,m-t again has the lowest p-value (0.000). Change of the market’s time on

market has a 1% significance level as well. The change of the market’s transaction volume changed sign but is not statistically significant. The R-squared is 0.62, which is higher than the baseline regression.

The last regression (4) has the early success logPTOM25i,m-t as dependent variable and

(31)

The only significant variable is the consumer trust. In a period where it is 10 points higher, the time on market for the 25% threshold is 50% lower. The housing market indicator has a similar effect: when it is 10 points higher, the time on market is 30% lower. The explained variance measure by R-squared is 0.23.

The effect of an early success on the time on market for the 70 percent threshold is considered significant and substantial, but the largest effect is given by the changes in macro-economic variables. The early success itself is harder to estimate, where consumer trust has a small, but statistically significant effect.

(32)

Limitations and further research

4 Limitations and further research

This researched was done with the available dataset of transactions by one real estate developer. It would be recommended to run the analysis on a larger dataset with transactions of multiple developers. Although there is a dataset of transactions by SWK, this dataset lacks the characteristics of the dwellings, like the living surface. This makes the transactions hard to compare.

Since 2016 Statistics Netherland started to collect transactional data including qualitative characteristics to publish a price index for new residential supply. The dataset is not publicly available, but the contributors to this dataset, being the real estate developers, could share this information following the example of the real estate agents. A larger dataset could improve the results.

AM could also improve her own dataset by registering more details about the sales progress of her projects. Since no official start date of sales is registered, the first transaction of a project is used as a best guess. Splitting up a project during the sales is not registered either, since the accounting software makes no use of this information. Registering a publishing date per house would immediately increase the information level on the sales process.

This research used a linear regression model with the ordinary least squares method to estimate the with several endogenous en exogenous variables. Another way of modelling the expected sales time of a project is by duration analysis, for example a proportional hazards model (Kluger and Miller, 1990). A hazard function gives the probability of an event to occur at time t, given it has not occurred until time t. The outcomes of such a model would indicate the odds relative to the market circumstances given the characteristics of the project.

This research used location fixed effects on province and local market circumstances to account for location heterogeneity. This heterogeneity could further be specified by using local demographics, employment opportunities and market tightness. Although these effects should also be reflected by the local market equilibrium, they could be dampened by other effects which are less relevant for the sales of new home projects.

With a broader dataset and substantially more observations, creating a forecasting model would be an interesting approach. Several studies have been conducted on forecasting housing prices, starting with Case and Shiller (1990). They found a positive correlation on the previous housing price development at short horizons and a negative correlation at longer horizons. This could be true for time on market as well. Developing a model in sample and testing it out of sample could demonstrate the reliability of such a model. When

(33)

Limitations and further research

successful, machine learning techniques could be applied to optimize the forecasting model with all additional transactions throughout time.

The determinants of the 70 percent threshold are largely identified, but the explained variance on the early success is very limited. Because of the path dependency, identifying the determinants for this early success could reveal additional instruments for the developer to influence the project’s time on market. These might not be related to project’s characteristics, but marketing measures.

This research focused on the characteristics of the projects, but the underlying dwellings are all sold by individual transactions. Although the objects within a project can seem interchangeable on paper, the consumer might have preferences which are not captured by generalized characteristics. A recent trend is auctioning of new home projects (Trouw, 2018; Volkskrant 2018), where instead of assigning a lot to a customer, the highest bidder on a specific lot gets to buy the dwelling. Although this would undermine the consumers trust in pricing expertise by the developer because of information asymmetry (Ooi and Le, 2012), it can also reveal the true market value of the dwelling.

(34)

Conclusion and discussion

5 Conclusions and discussion

5.1 Conclusions

The goal of this research is to determine how the target threshold of a new residential development can be modelled in time given its price setting and size and the market circumstances it is in. The time on market is regressed on the relative price to the local market and the relative project size to the local market’s transaction volume. Next to that, the market’s time on market and exogenous variables interest rate and consumer trust are considered.

To analyze this correlation, all transactions between 2011 and 2017 by real estate developer AM are used. The number of days to sell the first 70 percent of a project is used as the target variable since this is used as a sales threshold to start the construction of a projects. The relative price and the relative volume are calculated as price rate and transaction rate and exogenous variables are added to the dataset.

The expected effect of the price rate and the transaction rate on the project’s time on market are high. Because these are the only variables a developer can influence, the project size and the price setting get a lot of attention during the sales preparation. The changing market circumstances are expected to affect the sales time as well, but the changes will only be relevant for the dwellings that need to be sold.

Based on the analysis in this research, the relative price of the project has an unexpected effect. Increasing the price of the dwellings of the project would shorten the time on market. But this effect is not statistically significant.

There are exogenous variables that do have a significant effect on the project’s time on market. These regressors are on the market’s time on market, the change of this time on market during the sales period, the consumer trust and the change of interest rate and change of housing market indicator during the sales period. The largest effects are given by the last two, which suggests that although a developer believes its own decisions on price and size are important, the influence of the market is much more relevant for the sales results in the end.

Overall the explained variance given by the R-squared values is 0.56. This improved to 0.78 when looking at a part of the dataset in the period from 2011 to 2014, which is during housing crisis, but the number of observations taken into account was only 72.

When taking the early success as variable into account, evidence for path dependency was found. Although this early success could not be explained by the project’s characteristics, it proved to be a very significant variable for estimating the time on market for the 70% threshold. The other significant regressors were change of market’s time on market,