DEGREE PROJECT
REAL ESTATE AND CONSTRUCTION MANAGEMENT BUILDING AND REAL ESTATE ECONOMICS
MASTER OF SCIENCE, 30 CREDITS, SECOND LEVEL STOCKHOLM, SWEDEN 2018
TECHNOLOGY
DEPARTMENT OF REAL ESTATE AND CONSTRACTION MANAGEMENT ROYAL INSTITUTE OF TECHNOLOGY
Property Developers’ Pricing Strategies and Time on Market
The effect of pricing strategies on time on market in newly built apartments projects in Stockholm
Tom Hagen
Rodriguez Meshe
Master of Science thesis
Title Property Developers’ Pricing Strategies and
Time on Market
Author(s) Tom Hagen
Rodriguez Meshe Department
Master Thesis number
Real Estate and Construction Management TRITA-ABE-MBT-18449
Supervisor Mats Wilhelmsson
Keywords Real Estate, Residential Real Estate, Pricing
Strategies, Time on Market, Overpricing
Abstract
This study aims to investigate the causal linkage between developers’ pricing strategies and time on market (TOM) in the primary market for residential properties, i.e. a market where the seller is a developer or construction company, since this market differs significantly from the secondary or succession market.
Regression models are estimated using a dataset from 11 500 newly built apartment units in the Stockholm, Sweden, sold between June 2010 and March 2018. To describe the investigated data and test the hypothesis if overpricing affects TOM a regression analysis was conducted.
The Ordinary Least Squares (OLS) technique was applied explaining the size of TOM as a function of changes in a set of characteristics and conditions (independent variables) in one single equation. To measure an objective Degree of Overpricing (DOP), expected price was obtained by using market data and a hedonic price model controlling property attributes and market conditions. DOP is measured as the normalized difference between selling price and expected price.
By using a constructed price model when studying the relationship between the price and multiple independent variables, the empirical results show that an increase in variables such as competitive supply, distance from city centre, monthly fee and selling time lead to a decrease in price while higher floor level lead to an increase in price. Looking into the degree of overpricing and its effect on TOM, a Time on Market model was applied. Generally, the empirical results demonstrate that higher DOP results in longer TOM. An increase in competitive supply and monthly fee result in shorter TOM. When it comes to size, smaller apartments seems to sell faster than larger apartments. Different modulations detect varying significances among the independent variables investigated. Overall, the models display a positive correlation between DOP and TOM.
The originality and value of this study lies in the analysis of data collected from several development projects. This study is one of the first study that empirically examine the price- TOM relationship in the Stockholm primary housing market.
Acknowledgement
Firstly, we would like to express our gratitude to our thesis supervisor Prof. Mats Wilhelmsson, KTH for guiding us through the entire thesis process and his honest and wise opinions. His help was very valuable and we truly appreciate the time he contributed to improve the quality of the results of our work.
Furthermore, we would like to thank Prof. Björn Berggren, KTH and Jon Lekander, Brunswick Real Estate for your data contribution as well as the examinator, Kerstin Annadotter for her patience and understanding. Furthermore, we would like to express our gratitude to people at Booli, Valueguard, Mäklarstatistik and other industry professionals who contributed to this thesis in one way or another.
Lastly, we are very grateful for the support from our families and friends during our studies.
Our peers for making the past two years enjoyable.
Best,
Rodd and Tom
Examensarbete
Titel Bostadsutvecklarnas Prissättningsstrategier
och annonstid
Författare Tom Hagen
Rodriguez Meshe Institution
Examensarbete Master nivå
Fastigheter och Byggande TRITA-ABE-MBT-18449
Handledare Mats Wilhelmsson
Nyckelord Fastigheter, Bostäder, Prissättningsstrategier,
Annonstid, Överprissättning
Sammanfattning
Denna studie syftar till att undersöka det kausala sambandet mellan bostadsutvecklarnas prissättningsstrategier och försäljningstiden (TOM) på primärmarknaden för bostäder, dvs.
marknaden där säljaren är en utvecklare eller ett byggföretag, eftersom denna marknad skiljer sig avsevärt från den sekundära marknaden eller successionsmarknaden.
Regressionsmodeller har beräknats med hjälp av ett dataset om 11 500 nyproducerade lägenheter i Stockholm, som såldes mellan juni 2010 och mars 2018. För att beskriva de undersökta uppgifterna och testa våra hypoteserna, regressionsanalys tillämpades – en statistisk teknik som förklarar storleken av en variabel, kallad beroende variabel, som en funktion av förändringar i en uppsättning andra variabler, som kallas oberoende variabler, genom kvantifiering av en enda ekvation. Metoden Ordinary Least Square (OLS) har använts för att förklara försäljningstidens storlek storleken som funktion av förändringar i en uppsättning egenskaper och villkor (oberoende variabler) i en enda ekvation. För att beräkna graden av överprissättning (DOP), förväntat pris har erhållits med hjälp av marknadsdata och en hedonisk prismodell som kontrollerar fastighetsattribut och marknadsförhållanden. DOP har beräknats som normaliserad skillnad mellan försäljningspris och förväntat pris.
Genom att använda en konstruerad prismodell när man studerar förhållandet mellan pris och multipelständiga variabler visar de empiriska resultaten att en ökning av variabler som konkurrenskraftig tillförsel, avstånd från centrum, månadsavgift och annonstid leder till en prisminskning medan högre våning nivå leder till en ökning av priset. När det gäller storlek verkar mindre lägenheter sälja snabbare än större lägenheter. Om man tittar på graden av överprissättning och dess effekt på TOM, tillämpas en TOM-modell. De empiriska resultaten visar att högre DOP resulterar i längre TOM. En ökning av befintliga utbudet och månadsavgiften resulterar i kortare TOM. Överlag, trots vissa motstridiga resultat, visar grundmodellen en positiv korrelation mellan DOP och TOM.
Originaliteten och värdet av denna studie ligger i analysen av data som samlats in från flera olika utvecklingsprojekt i Stockholmsområde. Denna studie är en av de första studier som empiriskt undersöker förhållande mellan prissättning och TOM på Stockholms primära bostadsmarknad.
Förord
För det första vill vi uttrycka vår tacksamhet till vår handledare, prof. Mats Wilhelmsson, KTH för vägledning genom hela arbetsprocessen och hans ärliga och kloka synpunkter. Hans hjälp var mycket värdefull och vi uppskattar verkligen den tid han bidrog med för att förbättra kvaliteten på resultaten av vårt arbete.
Vidare vill vi tacka prof. Björn Berggren, KTH och Jon Lekander, Brunswick Real Estate för era bidrag samt vår examinator Kerstin Annadotter för tålamodet och förståelse. Vidare vill vi uttrycka vår tacksamhet till Booli, Valueguard, Mäklarstatistik och andra branschkunniga som bidragit till denna uppsats på ett eller annat sätt.
Slutligen är vi mycket tacksamma för stödet från våra familjer och vänner under hela vår studietid. Våra klasskamrater för dessa två angenäma år.
Vänligen, Rodd och Tom
Table of Contents
Abstract ... I Acknowledgement ... II Sammanfattning ... III Förord ... IV
1. INTRODUCTION ... 7
Background ... 7
Aim and research questions ... 8
Hypothesis ... 8
Delimitations ... 9
2. THEORETICAL FRAMEWORK ... 10
Swedish Housing Market Overview... 10
Real Estate Asset Pricing Theories ... 11
Price Discovery Mechanism ... 11
Search Theory, Listing Price and Overpricing ... 12
Time on Market (TOM) ... 14
3. METHOD ... 16
Approach... 16
Data Description and Collection ... 17
Variable Definitions ... 20
Credibility of the Study... 21
Shortcomings ... 22
4. RESULTS AND ANALYSIS ... 23
Price and Sales Data ... 23
Price Model ... 26
Time on Market Analysis ... 30
Days on market ... 30
Time on Market Model ... 31
Developer Specific Modelling... 39
5. CONCLUSIONS... 40
Further Studies ... 41
REFERENCES ... 42
APPENDICES ... 45
Appendix 1: Selling Performance ... 45
Appendix 2: Time on Market Distribution Group 1 ... 46
Appendix 3 : Time on Market Distribution Group 2 ... 46
Appendix 4: Time on Market Model for Developer 5 ... 47
Appendix 5 : Time on Market Model Group 10 ... 49
Appendix 6: Time on Market Group 20 ... 50
Appendix 7 : Distribution of DOP1 ... 52
Table of Figures
Figure 1: Sample Distribution over time --- 23
Figure 2: Price development and distribution over time --- 24
Figure 3: Valuation vs Sample Benchmark --- 25
Figure 4: Time on market distribution for the study sample --- 30
Figure 5: Weibull regression on selected developers with sale as event of failure --- 39
Figure 6: Time on market distribution for Developer5 (5 percent TOM=0) --- 45
Figure 7:Time on market distribution for group1 (10 percent TOM=0) --- 46
Figure 8: Time on market distribution (20 percent TOM=0) --- 46
Figure 9: Distribution of DOP1 (valuation after Svensk Mäklarstatistik) --- 52
Figure 10: Distribution of DOP2 (Estimated valuation by the price model) --- 53
Table of Tables Table 1: Overview of sample data ... 19
Table 2: Pricing model ... 28
Table 3: Correlation between DOP and Pricing ... 32
Table 4: TOM model adjusted sample ... 33
Table 5: TOM model adjusted sample with index ... 34
Table 6: TOM model sample vs adjusted 0 ... 35
Table 7: TOM model DOP1 vs DOP2 ... 35
Table 8: TOM models after observation period ... 36
Table 9: TOM model after location ... 37
Table 10: TOM model after competing supply ... 38
Table 11: TOM model for Developer5 ... 47
Table 12: TOM model Group10... 49
Table 13: TOM model Group20... 50
1. INTRODUCTION Background
During the past years Stockholm has been growing by around 40 000 inhabitants per year, as a consequence of this growth new housing construction has been on records levels and housing prices have reached unduplicated heights in the last two decades. Despite the increase in newly produced housing recorded during the past years, the demand has been estimated to remain high. The increase in housing prices in the Stockholm area, has made it very attractive for property developer in the area to start new development projects.
After years of high growth, both housing construction and housing prices has been slowing down. In 2017, the recorded decline of tenant-owned apartment prices in Stockholm was around 9,0 percent on annual basis, which can be compared to a decline around 6,5 percent in Sweden, according to Valueguard (2018). The number of sales-initiated development projects in the Stockholm area has decreased by 65 percent in the first quarter of 2018 compared to the same period in 2016, according to data from Svensk Nyproduktion (2018).
The inertia in the housing market means that the supply of unsold units is growing further.
Multiple developers have been forced to take actions as a result of the changed market situation in order to protect themselves by postponing or selling entire or parts of projects, changing the types of tenure, offering discounts or selling furnished apartment units etc. These actions have an impact on the selling prices as well as the time it takes to fully sale a development project. Recent markets analysis conducted by Svenska Dagbladet (2018) shows a large price difference between pre-signed apartments that individuals are selling through real estate brokers and the list price of corresponding apartments listed by developers.
For a developer, it is important to sell all units of a development project at a specific price and within the stipulated timeframe. Therefore, using an optimal pricing strategies can make or break developers since setting an optimal listing price can attract potential buyers and drive sales. Setting too high or too low prices than expected affect the marketability of a development project, which in turn could affect the total expected project returns.
Determining and optimal pricing has appeared to be difficult since price is affected by numerous factors. For developers it is crucial to sell their development projects within a specified timeframe and since, setting an optimal time on market (TOM) can improve the total realisable return of a development project. However, the developer generally faces a trade-off between the optimal pricing and optimal TOM.
In this study, we investigate whether or not overpricing affects the number of days a newly built apartment unit is on the market. Our main goal is to explain why TOM may or may not increase depending on the variation in the degree of overpricing (DOP) focusing on the Stockholm primary housing market.
Our contribution to existing research, is that this study solely focusing on transactions where the seller is a developer. Existing studies that investigate the relationship between pricing and time on market only use data collected from one project. The data utilised is collected from multiple development projects in Stockholm, an area which has not been covered to that extent before.
Aim and research questions
The purpose of this study is to investigate the causal relationship between property developers’
pricing strategies and time on market (TOM) of their newly built apartment development projects within the Stockholm municipality.
While many studies have been previously conducted to study the relationship between pricing and time on market, the majority of them have been focusing on the secondary market. This study focuses on the causal relationship in the primary market. Specifically, new studies can be conducted in this topic in order to provide a better understanding of the pricing strategies’
impact on the time on market.
The general research question we have striven to answer is how developers’ various pricing strategies affect the time that it takes to sell an apartment unit based on the fact that the selling price is a function of multiple variables such as location, size, monthly fee, floor level etc.
In order to answer the main question, the following question will be answered:
• What is the state of the art on time on market?
• What factors do influence selling time (time on market) in the Swedish market? Is there a trend in selling time?
• What are the similarities and differences between the Swedish and other markets?
Hypothesis
To effectively evaluate the causal relationship between developers’ pricing strategies and time on markets, the following hypothesis will be used:
• High degree of overpricing leads to longer time on market (TOM)
• Time on market should be longer for projects located within longer distance from the city centre, since the housing supply tends to be large compared to the supply within the central Stockholm.
• Smaller apartments should sell faster that bigger apartments
Delimitations
This study is limited to cover only newly built tenant-owned apartment projects in the Stockholm municipality. The limitation is considered relevant since the Stockholm housing market differs from the rest of Sweden in terms of size and impact on the overall economy and financial stability. The study is limited to housing projects sold in between 2010 and 2018.
Variables examined in this study are limited to by their observability. Some variables have been excluded because of their poor quality of the data, some have been excluded due to minimal impact on property prices and time on market.
2. THEORETICAL FRAMEWORK
This section provides gives an overview of the Swedish housing market, an introduction into the theory of pricing strategies and time on market gathered from different sources and reports. The section intends to provide basic understanding of how the Swedish market functions, supporting theories and prior research conducted in the field examining the relationship between pricing and time on market.
Swedish Housing Market Overview
The amount of residential apartments as at the end of 2017 was 4,9 million, allocated to 43 percent in single-family dwelling buildings, 51 percent in one or two-dwelling buildings, 5 percent in special housing and 2 percent in other buildings (Statistic Sweden, 2018). The housing market in Sweden can be distributed into three tenure categories, the ownership right which is the largest type of tenure with around 39 percent, the rental right with 38 percent and the tenant-owner right with 23 percent (Boverket, 2018).
A report from Boverket (2015), the Swedish National Board of Housing, Building and Planning, describes two ways how housing development projects are structured in Sweden. In the first way, the landowner estimates the costs of building a land parcel and what to expect to be paid for the property once constructed or alternatively the property value for continued ownership.
A construction company provides building services and materials and the landowner’s role is to ensure that the property is developed in line with current detailed plan. The second is when a developer purchases land from a landowner, usually a municipality, to build and thereafter manage or sell to a tenant-owner association. Boverket (2007) and Persson, Lind et al. (2015) define a tenant-owner right as the right to use a residential apartment in a property building owned by a tenant-owner association. The right can be used as a collateral for loan and can be transferred to another owner though various market transactions such as sale, inheritance or gift.
The transactions of residential real estate occur through various methods. In many markets, the most used method is the asking price mechanism, also called the private treaty. Through this method the seller of a property seller advertises and an asking price and start a negotiation process with several potential buyers (Kehzer, 2015). Another selling method of non-distressed properties is through auction. According to Eklöf and Lunander (2003), auctions are usually associated with the sale of distressed properties located in less attractive areas. However, in markets such as Australia, New Zealand, Scotland, Ireland and Scandinavia auctions are the most common selling mechanism for residential properties. In Sweden, the sale of properties through auctions is the most popular selling mechanism regardless of the property cycle and housing type (Hungria-Gunnelin, 2018).
In the Swedish primary housing market, a large amount of newly produced apartment are sold using a fixed price method, usually before the construction is started or early in the production process. The fixed price is set based on the supply and demand in the market and developers can adjust prices of their housing projects depending on the price changes in the secondary market. Before the production starts, many developers prefer to have 60 to 70 percent of the project sold, and thus secure financing (Isacson, 2006).
Developers use different methods when prioritising potential buyers of newly built apartments within a specific development project. The advantages with these systems include ensuring the potential buyer the selling price, since there is no auction or negotiation. They also facilitate for developers to predict the total revenue from a project and get sense of the demand. Some developers use a “first come first serve”-system, other developers use different of queue systems, which the candidate with most credit days is offered the opportunity to choose an apartment in the project (Hemnet, 2016 and JM, 2018).
Real Estate Asset Pricing Theories
One of the most difficult, yet important, issues that a seller must decide is to price its products, to be able to sell within a reasonable time and at the highest possible price (Gordon and Winkler, 2016). A significant number of theoretical studies has been made on the pricing strategies when pricing residential real estate. As many other studies have shown that pricing strategies may be influenced by perceptions of quality of many consumer goods (Shapiro, 1968 and McConnell, 1968). Miller and Sklarz (1987) also studies whether similar perceptions exist for real estate asset, since such perception may have an impact on selling price. They concluded that there exists an optimal pricing strategy for real estate that a seller may maximise the net present value of the selling price by choosing an appropriate price. Gordon and Winkler (2016) examine the effect of a listing price change on selling price. By using a sample of 13 461 single- family home sales in which 4 308 had a listing price reduction during the listing period, they concluded that properties with high listing prices take longer to sell and reducing the price signals that the seller may be willing to further price reductions. Knight (2002) examines also the impact on time on market and the selling price and found that listing price change has a significant impact on market behaviour and increase the time on market and decrease the selling price.
Price Discovery Mechanism
There is a significant number of theoretical studies on price discovery mechanism. In classical asset pricing theory, markets are considered to be informationally efficient and prices are adjusted directly when new information is available. However, in real estate markets, the use of appraisal to follow market activities in order to measure performance has become more common due to the lack of high-quality price data. This process allows appraisers to estimate property values based on fundamental variables and market information (Geltner et al., 2003).
Price discovery refers to the process by which market participants’ opinions of an asset value are put together in order to establish the asset’s price based on accessible information in the market (Barkham & Geltner, 1995; Geltner et al., 2003). The process of price discovery is developed based on the concept where two markets have similar characteristics regarding the asset value; most important price information that exists in one market is transferred to the second market. The price discovery process between private and public real estate markets, since they have similar characteristics of value has been examined by Giliberto (1993) and Moss
& Schneider (1996), but also Geltner et al. (2003) that review a large number of literature related to real estate asset pricing, flow of information between markets where real estate is traded and price discovery.
According to existing literature, there is strong relationship between private and public market, and the public markets have a stronger ability to integrate new information into price faster in comparison to the private market even though this integration process may occur in different ways. For example, the study by Moss and Schneider (2003) compared the transmission of information from public to private real estate markets in the US and the UK and concluded that the transmission is faster in the more complete in the UK than the US, due to greater homogeneity and portion of securitised properties in the UK.
Search Theory, Listing Price and Overpricing
The selling process of residential real estate is time-consuming for a property seller (developer) and requires planning and thus, choosing the optimal pricing and marketing strategy is very crucial. It also requires time to match a specific property and a specific potential buyer (Knight 2002). Before entering the selling process, the seller must decide whether to the objective is maximise transaction price or minimise the time on market (Miller 1978). Through the selling process reservation prices for both sellers and buyers can be observed and thereafter estimated the selling price. Since there exists contradictory objectives between sellers and buyers, the search theory is used to explain the relationship between the matching process in the market.
Search theory is used in several branches of economics and related disciplines, but it is known for its application to labour markets, where it is used to study unemployment and goods markets, where is can be used to formalize the role of money in exchange process (Moscarini
& Wright, 2010). In real estate, search theory is mostly used to explain the matching process between sellers’ and buyers’ behaviours in the selling process of residential real estate, but also the expected trade-off relationship between sellers’ pricing strategies and time on market (Knight 2002). In Quan (2002) the search theory is applied to explain this relationship, where two pricing discovery mechanisms are presented, the search market framework and auction setting.
According to the search market framework or search market, a pool of sellers is randomly matched with a pool of buyers during a certain period where the buyer has various reservation prices. The sellers’ search is driven by the eagerness to maximise the discounted present value of future sales profits based on the choice of a listing price that result in lower marginal costs of pursuing the search that will generate higher marginal profits. The buyers’ search among potential homes available in the market is motivated by the desire to maximise the utility (Knight, 2002). Both sellers and buyers have costs that that incur when entering the selling or buying process. The sellers’ trade-off is between selling quickly, since holding the property vacant or unsold for too long results into higher costs or waiting to find a buyer who is willing to pay a high price, assuming the wait might generate a higher selling price. The buyer faces a trade-off between paying lower price and pursuing the search for a cheaper property that maximise the utility, since the cost of continuing the search increases with the time spending on the search (Quan 2002).
As mentioned before, the main objective of a seller is to sell a property at the highest price possible price in the shortest possible time. List price plays a very significant role in the selling process. All things being equal, a higher list price may lead to more days spent waiting for a buyer who is ready to accept that list price (Kehzr, 2015). Nonetheless, a higher list price may signal high quality of the property or that the seller might be willing to decrease the price (McConnell, 1968). An important part during the selling process is for the seller to choose the optimal list price at which the property will be listed in the market. According to literature the list price set by the seller will have a direct impact on the selling price and the time on market, even though this list does not have to be the final selling price, because the seller may adjust the price based on the demand in the market (Knight, 2002).
Some literature notably examine the role of the listing price from a search theoretic point of view. According to Horowitz (1992) and Yavas and Yang (1995) list price indicate the seller’s reservation price and signals the prices that will be accepted with certainty if offered by the buyer. Therefore, it is very important the listing price is set carefully, since a high list price relative to value may reduce the number of prospective buyers, which may lead to an increase in time on market and thus, higher holding costs. A low listing price increases the probability of quick sale, as the result of a bigger pool of buyers, but at the same time higher risk for the selling price ending up lower than what could be achieved with longer time on market (Knight, 2002). The seller faces the risk that the property become stigmatised, when it has been on the market for too long, this has strong negative impact on the selling price. According to Taylor (1999), a property become stigmatised when it avoided by potential buyers for reasons that are unrelated to its physical conditions or features (Tomei, 1992).
Studies such Knight (2002) and Asabere et al., (1993) look into the impact of overpricing on selling price and thus the time on market. Asabere et al., (1993) investigate the relationship between price and optimal time on market. Their findings indicate that both overpricing and underpricing result in sub-optimal selling prices. This study concluded that the optimal time on market is lower than the actual time spent on the market and overpricing increases the optimal
time on market. Therefore, it is very important for seller to be able to set the right price when selling residential real estate, since revised list price have a strong negative impact on selling price. Knight (2002) finds that overpricing of a property not only take longer time to sell, they also end up being sold at a lower price mostly because they attract fewer potential buyers. In other words, overpricing is costly to the seller both in money, since the costs of holding increase and in time, due to longer time on market.
Time on Market (TOM)
While the main motivation of this study is to investigate whether pricing strategies affect the time on market, a large number of studies exists relating to pricing and time on market. The time on market or selling speed along with the selling price are the most important factors that the seller has to consider before entering the selling process of residential real estate. As mentioned before, it is crucial for the seller of a property to decide whether the objective is to sell the property quickly or the maximise the selling price.
In general, time on market (TOM) also called days on market (DOM), refers to the number of days a property is listed on the active market. Time on market is a very important factor when measuring liquidity in the real estate market (Zhu et al., 2016). A long TOM indicates that a property a been active on the market for a while, which in turn could signal the seller’s desperation and that there is something wrong the property or a slowdown in the market.
Short TOM, on the other hand, indicates high demand on the markets (Badino, 2018).
There exists several studies that investigate the relationship between time on market and various variables including pricing strategies, search theory, brokerage firms’ marketing efforts or macroeconomic parameters. Kalra and Chan (1994) examined the effect of macroeconomic parameters on time on market. Using the time on market as a dependent variable and independent variables such as mortgage rate, employment level and price concession, the found that high price concessions and high total employment lead to shorter time on market but high mortgage rate increase the time on market. On the other hand, in a market with increasing unemployment rates, it is reasonable to expect longer time on market and a similar correlation can be observed when looking into property prices (Hui and Yu, 2012). Another macroeconomic factor that has strong impact on property prices, and thus impact the time on market is inflation. According to Leung et al., (2002), all other things equal, high inflation level results in low real property prices, which raises the incentive to buy a property today, since future prices will be high. Generally, there is strong correlation between the time on market and the conditions in the housing and financial market. According to literature the effect of single parameters seems to be difficult to distinguish since the correlations seem to be strongly dependent on the data used and when the study was conducted.
There is a number of studies that have connected the seller’s pricing strategies and time on market. The general conclusion indicates that the more a property is overpriced, the longer is the time on market. This conclusion is also supported by Miller (1978), Ong and Koh (2000) and
Anglin and Wiebe (2003). More closely, Yavas and Yang (1995) examine the relationship between time on market and listing price. The study provides a way to determine the optimal listing price and investigates the impact of listing price on time on market. The findings indicate that the relationship between list price and time on market is unclear but some empirical evidence show positive correlation between time on market and listing price for mid-priced properties but non significance for higher or lower listing prices. In addition, the study also found that there are some factors that are mispriced and have an impact on time on market.
Björklund et al. (2006) investigate the whether or not the offer price affects the transaction price and the time on market. The findings indicate that a setting a high offer price or an offer price equal to expected will generate a high transaction price and the opposite result when setting an offer price below the expected price. The study concluded that setting a high offer price is the best pricing strategy since a high offer price lead to a longer time on market compared to a low offer price, but result in a high transaction price. These findings are contradictory compared to the study of Beracha and Seiler (2014) that examine three pricing strategies such as round pricing (thousands digits is 0 or 5), “just below” round pricing (thousands digit is 9 or 4) and precise pricing ( all digits) and found that the properties listed using the “just below” strategy are associated with largest discount negotiated relative to the asking price and concluded that the “just below” strategy is the most effective strategy for the seller in terms of greater yield relative to value. Knight (2002) also found that high listing price result in longer time on market and low selling price and Anglin et al. (2003) that analysed the relationship between listing price, transaction price and time on market, found that increase in listing price lead to increase time on market especially for house with low variance in list price.
There are few studies that examine the relationship between time on market and the search theory and the findings and conclusions differ. Some studies concluded that longer time on market increase the likelihood the receive a better offer, while other conclusions state that longer time on market lead to the stigmatisation effect, which in turn leads to lower selling price (Hui and Yu, 2012). Haurin (1988) utilise the search theory to conclude that high variance of the distribution of offers for a property, leads to longer expected time on market. The study also concludes that the greater the atypicality of a property, the greater the variance of offers but also that atypicality of a property lead to longer time on market, i.e. atypical properties are difficult to sell.
3.
METHODS
ApproachThis study is targeted at factors affecting the selling time of condominiums in the Stockholm housing market. The selling time of a project has a big impact on its profitability. Property developments projects require high capital investments. Parts of these initial cost can be offset through early disposal (pre-sale) of apartment units hence it is costly to hold unsold apartments as they do not generate any value. For property developers and financing institution it is of interest to determine any attributes explaining selling times of apartments units and to quantify their impact. Other research suggests that there are several apartment characteristics explaining differences in selling times.
Studies on time on market in other housing markets serves as impetus to analyse the Swedish market, hence the approach of this study is derived from other research papers in this field.
The first part of the analysis comprises a short overview of existing literature and research focusing on the relationship between pricing and time on market (TOM) to give an overview about the topic and common theory answering the first question.
The second part of the analysis creates a quantitative estimate of the theoretical economic relationships assumed from the literature review. In an empirical study a dataset from Booli.se, which is the largest open database for final property prices in Sweden. The analysed dataset contains information about 11 500 sold housing units from the Greater Stockholm area.
To describe the investigated data and to test the hypotheses of research question number two a regression analysis is applied – a statistical technique explaining the size of one variable, called dependent variable, as a function of changes in a set of other variables, called independent variables, through the quantification of a single equation. The most common used method for regression analysis is Ordinary Least Squares (OLS). This regression estimation technique calculates the respective estimates so as to minimize the sum of the squared residuals, which makes the method mathematically very straightforward and powerful in its significance (Studenmund 2010).1
Other research stresses the simultaneity problem when investigating time on market in relation to selling prices. Due to the reciprocal impact it is difficult to determine the causality between the two variables. To dodge solving this controversial question, two different models are used in this study. The inherent simultaneity problem provides the conclusion that factors describing prices can describe selling time as well. Hence, this study starts with a price model and follows with time on market modelling. Similar approaches with two stage models have been carried out in prior research from Yavas & Yang (1995), Li (2004) and Gustavsson & Vahtola (2014).
1 Given a set of specific assumptions and criteria (for more detailed explanation see the appendix).
The first model serves the purpose to best possible estimate selling prices using price as the dependent variable. The model estimates the movement of selling prices for condominiums by comparing sales of 11 500 units and the specific selected characteristics (independent variables) of these units. The result reveals the significance and magnitude of price-impact for these characteristics.
The second model uses the output from the first model to test hypotheses on market time using time on market as the dependent variable. An index from Mäklarstatistik is used to compute a degree of overpricing for each observation relative to market levels. Like in the first model, the second model’s result give evidence on the significance and magnitude of the impact of a set of characteristics. In addition, survival models (which ones) are presented to graphically illustrate the impact of selected characteristics.
The conclusion of the study brings together the results from the literature study and empirical study. To explain the identified similarities and differences economic theory is amended by inductive reasoning. Due to the data limitation and the lack of comprehensive research on the Stockholm housing market, it seems reasonable to formulate novel content rather than solely confirming established theories and theses.
Data Description and Collection
The dataset serving as the foundation for the empirical study is provided by the Swedish data company Booli. Booli.se is one of Sweden's largest platforms for residential housing listing housing up for sale, closing prices, value indicators and others. The transaction information is collected directly from the Swedish real estate agencies as well as property developers (Booli, 2018). The analysed sample comprises 11 500 observations of newly built apartment units within the greater Stockholm area. The objects have been listed respectively sold between June 2010 and March 2018. The information provided for each observation is: development project, floor number, area, selling price, size, monthly fee, apartment type, number of rooms, developer, date of listing, date of sale, apartment number.
To increase the explanatory power of the models developed and obtain more significant results secondary and additional data was gathered by the authors to amend the dataset. This information was added based on common modelling for estimating housing prices and time on market as well as the availability to obtain it.
Distance to City Centre, to weigh the different apartments based on their geographic location a distance component was added. The variable shows the distance from the respective area to the city centre (Stockholm central station) to be travelled by car according to google maps. See more detailed description in the appendix.
District, to characterize the different areas on a high-level detail the authors created three general districts: Central Stockholm (comprising Essingen, Kungsholmen, Östermalm and Vasastan-Norrmalm), Western Stockholm (comprising Bromma-Västerled, Essingen, Solna- Sundbyberg, Sollentuna and Spånga-Kista) and Southern Stockholm (comprising Södermalm, Hägersten-Liljeholmen, Farsta-Väntor and Nacka).
Index (2018 = 100), since the observation period is over a time span of eight years the single prices need to be adjusted to be comparable. The index chosen to adjust the sample size is taken from the information platform Valueguard.se which provides estimated indices for different housing types in Sweden. The selling prices from the sample were corrected with the index for flats in the Stockholm area so that May 2018 represents a value of 100 and May 2011 a value of 157,8.
Outliers and measurement errors in the sample set can cause serious problems in the later modelling. To mitigate such issues observations deviating too much in value have been dropped from the sample size. The result is a more homogenous and comparable data set which allows more accurate explanations. Variables this filter was applied to are size, floor, price, room, selling time, apartment type. (See the appendix for more detailed info regarding dropped information.)
The following Table 1 presents the variables used in the final models. For other variables see the appendix.
Table 1: Overview of Sample Data
Variable Variable
Type Mean Std. Dev. Min. Max.
Selling Price [SEK] Continuous 3 438 412 1 693 199 1 008 800 17 900 000
Adjusted Price [SEK] Continuous 4 483 716 2 175 253 1 306 101 26 600 000
Log. Adj. Price Continuous 15,212 0,4512703 14,083 17,10
Price per sq. m. [SEK/m2] Continuous 47 039 17 003 14 467 200 000
Time on Market [days] Continuous 181,0 199,4 0,0 730,0
Degree of Overpricing [%]
(Mäklarstatistik valuation) Continuous -0,61% 2,24% -10,45% 9,54%
Degree of Overpricing [%]
(Sample valuation) Continuous -0,04% 1,22% -8,60% 4,47%
Floor Continuous 3,717766 3,01737 0 20,00
Size [m2] Continuous 73,36775 23,28511 20 177,00
Size Group 1 (up to 35m2) Dummy 0,0240529 0,1532212 0 1
Size Group 2 (35m2 to 50m2) Dummy 0,1631052 0,3694807 0 1
Size Group 3 (50m2 to 65m2) Dummy 0,2189532 0,4135582 0 1
Size Group 4 (65m2 to 80m2) Dummy 0,2048106 0,403584 0 1
Size Group 5 (bigger than 80m2) Dummy 0,3890781 0,4875663 0 1
Number of rooms Continuous 2,791421 1,017314 1 5
Competitive Supply [units] Continuous 210,1347 170,3753 1 617
Driving distance to city centre [km] Continuous 7,34919 3,575722 0 17,40
Commuting time to city centre [min] Continuous 20,8561 7,478632 0 45
Monthly Fee to Tenant association
[SEK] Continuous 4026,845 1176,495 0 8726
Central District Dummy 0,2136884 0,4099305 0 1
Southern District Dummy 0,4280995 0,4948289 0 1
Western District Dummy 0,3581088 0,4794691 0 1
Trend (month number over total
observation period) Continuous 29,71828 16,89884 1 77
Trend2 (squared month number) Continuous 1168,718 1093,801 1 5929
Lowprice Quarter Dummy 0,3333333 0,4714289 0 1
Highprice Quarter Dummy 0,3333333 0,4714289 0 1
Variable Definitions
Most variables state the physical characteristics of the apartments. Macroeconomic variables describe the condition at selling date. Following are some short specifications of the variables used in the models:
- Competition, to see if supply has a significant impact on the pricing level we created a variable with competitive supply of similar size within the same area. The competing period is set as a timeframe of one year with six months before and after the selling date. The supply variable is only considering units from the sample meaning new produced units. In this study it is assumed that these units are best comparable, and buyers have a clear preference to either buy new produced units or condominiums in private hand.
- Degree of Overpricing, the earlier discussed simultaneity problem prevents including rice as an independent variable in the time on market model. Instead a measurement of the deviation from the expected selling price is used. In theory, an apartment unit with an observed selling price significantly higher than the market value (or expected price) should take longer to sell. Not many buyers are willing to pay a high premium in comparison to apartment units of the same value.
- Developer, to derive statements on different projects respectively developers dummy variables for the ten biggest developers of the sample were created. The study does not disclose neither the development projects nor the individual developing companies included in the data set.
- Month, the different month show the effect of seasonal social respectively buyer behaviour. The variable takes the form of a dummy equalling one when matching the respective month.
- Monthly Fee, condominium buyers in Sweden pay a fixed monthly fee to their tenant association. This fee covers for maintenance, operation and investments concerning common areas but also the amortization of loans should the association have any.
The monthly fee is mainly depending on two factors, size of apartment and outstanding total debt of the association.
- Selling Price, the apartments selling price is stated in thousands of Swedish Krona.
Depending on the model, prices are adjusted with indices to enable comparability and transformed into logarithmic form to fit mathematical dependence.
- Price Tertial, to detect trends within different price ranges the observations are grouped in three price categorizes subject to the distribution of the selling prices in the sample.
- Size Group, to detect trends depending on apartment size the observations are grouped in seven size categorizes. Since there is already a size variable size group takes the form of a dummy variable equalling one when matching the respective group.
- Time on Market, to measure how long an apartment was on the market the difference between selling date and the listing date is calculated. Time on Market is stated in number of days.
- Trend, to include and represent a time trend during the observation period two consecutive integer variables were created. The first one equals the year in the observation period starting at 1, the second equalling the squared value of the first variable modelling a linear and exponential trend.
Credibility of the Study
To be regarded as credible, an empirical study needs to provide reliability through approved models and good input data as well as validity by its fundamental approach.
The models used are conventional methods within econometrics. They are simple to build with powerful conclusions making them easy to repeat in similar contexts, time periods and other markets. Booli.se is a professional data provider with good reputation in the market.
However, there can be quality differences between the different original data sources since the data collected by Booli is provided by different real estate agents often the developers themselves. We worked in close collaboration with Booli and discussed potential outliers and errors. According to the knowledge of the authors, the data has not been manipulated in any way.
In comparison, validity can be a concern in studies of this type. The internal validity is dependent on the actual mathematical models used. They must be designed robust enough to produce scientific significant content. The external validity comprises the economic theory supporting the approach. In hypothesis testing, researchers tend to narrow down their data sets to specific samples thereby obtaining more robust and significant conclusions. In the real estate research, it is common practice to investigate either one market in one specific sector of the real estate market over a short period or have a higher detail level approach with more generalisations.
This is study represents a trade-off between these two approaches. On one hand the exclusion of some data is necessary, as information should produce conclusive results. On the other hand, findings of this study should be general applicable to some extent. Hence, some simplifying assumptions must be made when working with cross-geographical data.
By not using the complete sample population and applying careful data due diligence the authors tried to minimise the negative impact of generalisation (Studenmund 2010).
Shortcomings
Although additional data was gathered, more information on the observed properties would have allowed to build better models producing more significant results. The following list represents information not available or accessible the authors deem to be important for investigations like the one carried out: Total floors, geographic coordinates, number of bathrooms, competitive supply including private market, balconies, finishes, view, finance and mortgage details.
There is no data on listing prices included. The selling prices of the apartments are simultaneously considered to be the listing prices since the properties were directly sold via the developers’ platforms. Hence, a comparison of listing and selling price is not part of this study.
When looking at new production one must consider differences compared to the private market. The latter is quite homogenous apartments entering and leaving the market on a continuous basis. New supply does not enter the market in a constant stream but in specific batches. Depending on the distribution curve, this creates a rather swaged distribution wave when analysing observations. Regarding price indices and models, this means that single or several projects can dominate the price formation during specific periods. Hence, we can regard such results as biased.
Since the study only considers newly produced housing units, the results are to some extent biased towards certain geographic areas and financing options. Development projects in metropolitan areas are often restricted to certain locations due to available space and building permits whereas the private market can be considered as omnipresent. New tenant associations usually start with a significant amount of debt passing on this financial burden to their members. House buyers of new housing are therefore much more exposed to credit risks than buyers who enter associations with lower debt. This must be considered when observing final selling prices as well as the representative buyer group.
Based on the reasoning above, this study assumes that house buyers have a clear preference to buy either new produced units or condominiums from the private market. Housing units of other types are therefore not included in the data, the same applies to competitive supply. However, the indices used to modify the data are based on the open condominium market including private sales. To further increase the general robustness and significance of this study, observations from the private market could have been included in the samples as well. Allowances for such data exceeded the scope of this study and was hence omitted.
4. RESULTS AND ANALYSIS Price and Sales Data
In the next chapters, the analysis and results of the study are presented. First, a model for selling price is introduced and factors explaining selling prices are discussed. The second model aims to explain time on market. Variables affecting selling time are investigated followed by a discussion on differences and similarities between the two equations. The analysis is concluded with an extract on other factors affecting pricing strategies.
In the primary market, new supply does enter the market in distinct batches. This often creates a rather steep distribution wave when illustrating the observations. Regarding models for price indices this means that single or several projects dominate the price formation during specific periods given a high concentration of sales. Figure 1 shows the distribution of the whole sample size.
Figure 1: Sample Distribution over time
The distribution of observed sales is concentrated in period from May 2011 to April 2017. The graph illustrates how the sales distribution is following the listing curve with a small lag. Several four-month lags can be identified, e.g. Jun11-Sep11, Mar14-Jun14, Sep14-Dec14, Sep15- Dec15, May16-Aug16. Interestingly the significant amount of sales in August 2013 does not seem to have an equally significant preceding listing-peak. An explanation could be the gradual built-up of constant supply from June 2012 to June 2013.
A time lag between the actual selling of apartment units and the respective reporting to a platform such as Booli could explain the decreasing number of observed sales from the end of 2015. An economic reasoning could be that the developers simply did not produce and/or sell as much units from 2016 on compared to the years before. This would indicate either a saturation of the market or a delay in construction. Given this conclusion is true it is interesting if there a corresponding impact on selling time can be observed.
Figure 2, shows the average price development of the complete sample as well as the movements of the Valueguard Index used to adjust selling prices in the sample. The index is based on sales data from real estate agencies from the Greater Stockholm area. The market shows a strong upward trend during the observation period for all three series. The fluctuating curves and steep outliers illustrate the price implication of single projects compared to the two indices generated from homogenous markets. For Stockholm Centre projects with relative high prices can be observed being sold in September 2014 and late summer 2015. There is a general price spike for 2016 which is due to the low number of observations during that period. This spike should not be seen as representative for the central Stockholm but only as a description of the sample (confer Valueguard Index). In general, one can observe how the low liquidity in 2016 causes strong fluctuation for all series. The series Stockholm Lån comprises the complete sample. Its movement follow the index with periodic signs of over and underpricing. Compared to the Central area, the western and southern parts of Stockholm provide a flatter and smother price development. Housing in the Western seems cheaper than in the Southern.
Figure 2: Price development and distribution over time
Figure 3: Valuation vs Sample Benchmark
Due to the simultaneity problem between price and selling time, we introduce the concept of mispricing meaning a divergence between the expected market price and the actual selling price. Within the models the variable Degree of Overpricing is included to illustrate the time implications of pricing. The economic reasoning behind this variable is that an apartment having an observed selling price which is significantly higher than the price the market would suggest for that unit, will take longer to sell. In this case the unit would be considered as overpriced and vice versa, an underpriced apartment should sell significantly faster.
To avoid bias, different data were used from the sample to create a valuation benchmark. The observations are compared to square meter prices from Mäklarstatistik based on their specific districts. Figure 3 shows the sample prices with the respective valuations over time. A divergence between the valuation and sample for western area can be observed. The series MSt Western only comprises values from the district Bromma-Västerled. Since the sample includes apartments from other Western areas as well, the series does not fit the sample ideally. Notice that, within the statistic modelling different valuations were applied to get the optimal fit for each area2. Although showing strong signs of periodic mispricing the sample distribution follows the suggested valuation of the market for all series.
The observation period covers a time span of eight years. The selling prices from the sample were corrected with the index for flats in the Stockholm area so that May 2018 represents a value of 100 and May 2011 a value of 157,8. The next figure shows the price dynamics with prices adjusted by the index. The unadjusted movements are illustrated by dotted lines.
Figure 4. Index-adjusted Prices
Compared unadjusted and adjusted prices, the adjusted series show smaller movements suggesting that the price index captures some of the underlying macroeconomic price drivers as well. The index adjusted series show a correction of the upward price trend. On the contrary, it suggests that from a relative point of view (the timeless market) prices for newly produced units have been slightly decreasing compared to existing stock. Economically this can be explained by customer preferences and the limited space available in Stockholm. The prime locations for housing in cities are mostly built-up already. Development projects often are situated outside the City centre except for some few redevelopments. Naturally, more central units are valued higher than others (Fanning, 2014). A customer preference for old architecture towards newer units also explains why the latter achieve lower market prices.
Price Model
To describe the price data and eventually test the hypotheses of the second research question a hedonic regression analyses was performed. A hedonic regression model is a statistical technique explaining the size of one variable – dependent variable – as a function of changes in a selection of other set variables – independent variables – through the quantification of a single equation in the following form:
𝑌𝑖 = 𝛽𝑜+ 𝛽1+ 𝜀𝑖
Where 𝑌𝑖 is the dependent variable transaction price and 1 a vector of coefficients associated with the external explanatory variables, X. The interpretations of these regression coefficients are as estimates of the implicit prices of the respective attributes, e.g. size of apartment unit.
0 represents a constant term describing the part of Yi that cannot be expressed by 𝑋𝑖. 𝑖 is a
stochastic term that is added to the regression equation to introduce additional random variations in Yi which cannot be explained by the included coefficients 𝑋𝑖. The error term corrects for omitted data, measurement errors, incorrect functional forms and/or random occurrences. It is assumed to have a normal distribution and constant variance.
The idea behind a regression analysis is to take a purely theoretical equation like:
𝑌𝑖 = 𝛽𝑜+ 𝛽1+ 𝜀𝑖
and create an estimated equation like:
𝑌̂𝑖 = 𝛽̂ + 𝛽0 ̂ 𝑋1 𝑖
By using a real dataset, each “hat” indicates a sample estimate of the true population value from the dataset. The most common used method for regression analysis is Ordinary Least Squares (OLS). This regression estimation technique calculates the respective estimates in a way to minimize the sum of the squared residuals, which makes the method mathematically very straightforward and powerful in its significance. OLS provides the best linear unbiased estimator of the regression coefficients (minimum variance, linear, unbiased estimator) if the Gauss–Markov Theorem is met. For this the model needs to comply with the following assumptions: The regression model is linear with an additive error term having a mean of zero, being uncorrelated with the explanatory variables or other observations of the error term and having a constant variance (ideally it is normally distributed as well). Explanatory variables must not be perfect linear functions of one another. When these conditions are not met OLS still can provide reliable estimates, however other estimation techniques are to be preferred.
The two most important properties of a good estimator are unbiasedness and a minimum variance. For unbiasedness the expected value of the estimated coefficient must be equal to the true value of the coefficient. For a minimum variance the estimating distribution must provide the smallest variance of all unbiased estimators. The final regression model shows strong signs of Heteroskedasticity and Serial Correlation stating that different observations of the error term are correlated and that their variance is not constant. This does not affect the unbiasedness of our estimates since the resulting deviation is expected to be positive and negative to the same extent. However, it complicates hypothesis testing since it makes the estimated coefficients are less reliable significant. The models in this study use therefore robust estimators who compute the coefficients using heteroskedasticity-corrected standard errors.
Since heteroskedasticity and serial correlation causes problems with the variances but not the coefficients themselves it makes sense to improve the estimation of the variances in a way that doesn’t alter the estimates of the slope coefficients.
1Xi represents a combination of physical apartments characteristics which are believed to describe as whole the pricing of condominiums. These attributes are selected based on experience, other models and available data. The choice of the functional form is based on the underlying economic theory. The functional form of a price regressions suggests the dependent
variable price stated in logarithmic form implying a percentage interpretation. The coefficients of the selected characteristics (independent variables) express the approximate percentage change in the dependent variable following from a one-unit increase or decrease in the independent variables. It is important to stretch that a variable coefficient only explains the change in Y correctly when all the other variables are held constant (Studenmund 2010).
Final model form:
𝐿𝑛 (𝑖𝑛𝑑𝑒𝑥 𝑎𝑑𝑗𝑢𝑠𝑡𝑒𝑑 𝑝𝑟𝑖𝑐𝑒) = 𝛽̂0+ 𝛽̂1𝑋𝑖
The price model has a strong explanatory power which can be seen in the R-square value of almost 86 percent. The t-values indicate significance on the one percent level for all included independent variables. The combination of a high R-square and t-values can be regarded as evidence that the selected variables effectively explain changes in the log-transformed housing prices.
Table 2 shows the final variables significantly explaining selling prices. The variables and their respective impacts on pricing are discussed below. Note that the list of independent variables in the table only include those variables who are one side significant and on the other side deemed to be important for the time on market regression as well3.
Table 2: Pricing model
3 Omitting these variables did not affect the overall fit of the model nor add additional heteroskedasticity.
