Is there a Real Estate Portfolio Premium?: An Empirical Analysis of Portfolio Premiums

DEGREE PROJECT

IN REAL ESTATE AND CONSTRUCTION MANAGEMENT BUILDING AND REAL ESTATE ECONOMICS

MASTER OF SCIENCE, 30 CREDITS, SECOND LEVEL STOCKHOLM, SWEDEN 2021

Is there a Real Estate Portfolio Premium?

An Empirical Analysis of Portfolio Premiums

FRIDA CARLSSON MALIN STRÖMBERG

KTH ROYAL INSTITUTE OF TECHNOLOGY

DEPARTMENT OF REAL ESTATE AND CONSTRUCTION MANAGEMENT

MASTER OF SICENCE

Title Authors

Department

Master Thesis number Supervisor

Keywords

Is there a Real Estate Portfolio Premium? An empirical analysis of Portfolio Premiums

Frida Carlsson & Malin Strömberg

Real Estate and Construction Management TRITA-ABE-MTB-21395

Åke Gunnelin

Portfolio Premiums, Price Premiums, Transactions, Swedish Real Estate Market, Property Segments

ABSTRACT

This thesis aims to explore if the existence of portfolio price premiums can be verified and if they differ in time and over property segments. The purpose is to contribute with valuable insights within the field of portfolio premiums in the real estate industry. In order to explore this further a regression model was developed. The model includes six portfolio variables controlling for size in the aspect of transactional value and number of properties included in the portfolio. We further test if the premium varies over property segments and over time. The data was provided by Cushman & Wakefield and consists of 825 property transactions. The results show that a portfolio premium for small, medium and large portfolios with a transactional value over 500 million SEK, and a discount on small portfolios with a transactional value below 500 million SEK is present.

The premium was found to be 17.5% for small portfolios, 16.8% for medium sized and 26.3% for

large portfolios. While premiums were found for portfolios with a transactional value over 500

million SEK an 13.7% discount was found for small portfolios with a transaction value below 500

million SEK. Which indicates that investors are willing to pay a premium, but only for larger

portfolios. Furthermore, the only segment test with significant results were residential and

industrial of which residential indicated a discount on small and medium portfolios with a

transactional value over 500 million SEK and industrial indicated a discount on small portfolios

with a transactional value over 500 million SEK. The test of variation of a portfolio premium over

time gave mixed results and showed that investors payed a premium for medium and large

portfolios with a transactional value over 500 million SEK during 2010 - 2015.

ACKNOWLEDGEMENT

This master’s thesis has been written at the Royal Institute of Technology in Stockholm during spring 2021. The thesis constitutes as the final examination of the authors education in Civil Engineering within the Master of Real Estate and Construction Management and the master’s program Real Estate and Construction Management with Real Estate Economics specialization.

We would like to take this opportunity to thank our supervisor at the Division of Real Estate Economics and finance, Åke Gunnelin, for the guidance.

We would also like to express our gratitude towards the real estate consultancy firm Cushman &

Wakefield for the provided information and valuable inputs. Without this, the master thesis would not have been possible to realize.

Stockholm, June 2021

Frida Carlsson Malin Strömberg

EXAMENSARBETE

Titel Författare Institution

Examensarbete Master nivå Handledare

Nyckelord

Finns det en portföljpremie? En empirisk studie av portföljpremier

Frida Carlsson & Malin Strömberg Fastigheter och Byggande

TRITA-ABE-MTB-21395 Åke Gunnelin

Portföljpremier, Prispremier, Transaktioner, Svenska Fastighetsmarknaden, Fastighetssegment

SAMMANFATTNING

Detta masterarbete syftar till att undersöka fenomenet portföljpremier och bidra till utökad kunskap om premier och möjliga förklaringar till varför de uppkommer. Författarna har undersök om det går att kvantifiera den påstådda portföljpremien och om denna skiljer sig över fastighetsegmenten och tid. Syftet med arbete är att bidrag med värdefulla insikter och kunskap om portföljpremier inom fastighetsbranschen. För att kunna besvara frågeställningen utvecklade författarna en regressionsmodell. Modellen innehöll sex portföljvariabler som bland annat kontrollerade för storlek i förhållande till transaktionsvärde samt antal fastigheter inkluderade i portföljen. För att undersöka om premien varierade över fastighetssegment och med tid utfördes fem olika segmentstest och två års tester. Data som användes i regressionerna tillhandahölls av Cushman & Wakefield. Resultatet av studien visar att det finns en portföljpremie på små, medelstora och stora fastighetsportföljer med ett transaktionsvärde över 500 millioner kronor.

Premien noterades till 17,5% för små portföljer, 16,8% för medelstora portföljer och 26,3%

för stora portföljer. Medans en premie noterades för portföljer med ett transaktionsvärde över 500 millioner kronor kunde en rabatt om 13,7% hittas för små portföljer med ett transaktionsvärde under 500 millioner kronor. Segmenttesten som genomfördes gav blandade resultat. De test som gav signifikanta resultat var segmentstest för industri och bostäder. Resultatet av regressionen visade att det finns en rabatt för små och medelstora bostadsportföljer med ett transaktionsvärde överstigande 500 millioner kronor samt en rabatt för små industriportföljer med ett transaktionsvärde över 500 millioner kronor.

Utöver segmentstesten gjordes även två tester där författarna testade om premien

varierade över tid. Likaså här gav testerna blandade resultat. Det kan konstateras att en premie

återfinns för portföljer handlade under perioden 2010 - 2015.

FÖRORD

Detta examensarbete har skrivits vid Kungliga Tekniska Högskolan i Stockholm under våren 2021.

Avhandlingen utgör den avslutande examinationen för civilingenjörsprogrammet Samhällsbyggnad, inom mastern Fastigheter och Byggande samt för mastersprogrammet Fastigheter och Byggande. Vi vill ta tillfället i akt och tacka vår handledare Åke Gunnelin vid institutionen för Fastigheter och Byggande för hans vägledning och engagemang.

Vi vill också uttrycka vår tacksamhet gentemot fastighetsrådgivningsföretaget Cushman &

Wakefield för vägledning och information. Utan det hade inte examensarbetet varit möjligt att genomföra.

Stockholm, juni 2021

Frida Carlsson Malin Strömberg

TABLE OF CONTENTS

1. Introduction 6

1.1 Background and Context 1

1.2 Aim and Research Question 2

1.3 Limitations 3

1.4 Outline 4

2. Literature Review 5

2.1 Portfolio Theory 5

2.1.1 Price Premiums within Real Estate 6

3. Theoretical Framework 9

3.1 Modern Portfolio Theory 9

3.2 Arbitrage Pricing Theory and The Law Of One Price 9

3.3 Value Theory 10

3.5 Efficient Market Hypothesis 10

3.6 Transaction Cost 10

4. Research Method 12

4.1 Data Collection 12

4.1.1 Possible Shortcomings with the Data 12

4.1.2 Variables in the Regression Analysis 13

4.2 Regression Model 16

4.2.1 Possible Shortcomings with the Model 17

4.2.2 Hedonic Pricing Model and Rregression Analysis 17

5. Results 20

5.1 Model 1 20

5.2 Separate Estimates of Portfolio Premium for the Different Segments 22 5.3 Separate Estimates of Portfolio Premium for the Different Time Periods 27

6. Discussion 30

6.1 Discussion in Rregards to Theoretical Framework 30

6.2 Possible Composition of Portfolios Premium 32

6.3 Discussion on Sustainability and Enviromental Social Governance 32

7. Conclusion 34

7.1 Model and Data Limitations 35

7.2 Suggestions for Further Research 35

References

Appendix

1 1. INTRODUCTION

1.1 BACKGROUND AND CONTEXT

The Swedish commercial real estate market has been booming during the past years. Price levels have risen as vacancies have decreased and rental income increased. There are multiple ways a shift in ownership of property can occur, for example either by title deed or through share transfer.

Most of the larger property transactions in Sweden are carried out through company transactions to gain tax benefits. Property investors are constantly looking to optimize their stock of real estate, meaning that they will divest and acquire properties that fit their investment profile and generate the highest level of return. Properties can be acquired as single units or through portfolios containing multiple units. The latter is frequently used by companies who want to combine assets regarding risk and type to present an investment opportunity.

Portfolio transactions is a popular mean when divesting a large number or real estate assets.

Looking back, portfolio transactions have accounted for a large portion of the total transaction volume on the Swedish real estate market. In 2019, transaction volumes on the Swedish real estate market reached new record high levels and in Q1 - Q3 2019 Savills reported the highest transaction volumes compared to the same period in 2016. The transaction volume during that period amounted to 152 billion SEK and two-thirds of the volume consisted of portfolio transactions (Börsvärlden, 2019). At the end of 2019, Fastighetsvärlden published an article listing the 20 largest transactions of the year and several of these were portfolio transactions.

Year 2020 was no exception in regard to portfolio transactions. Among the top 20 largest transactions, we find seven portfolio transactions (Fastighetsvärlden, 2020). However, year 2020 was a special year with the outbreak of Covid-19 putting the investment appetite on hold. In spite of a slow start during the first half of 2020 the activity on the transaction market picked up, finishing the fourth quarter of 2020 strong.

Figure 1 illustrates the volume of transactions made above 40 million SEK in 2010 through the

first quarter of 2021. The purple line illustrates the total share of portfolio transactions while the

orange line displays the total number of portfolio transactions. To sum up, in 2020 portfolio

transactions accounted for 37 percent of the total transaction volume compared to 57 percent in

2019. During the first quarter of 2021, 35 percent of the total transaction volume consists of

portfolio transactions. However, the number of portfolio transactions made in 2020 accounted for

23 percent of the total number, compared to 24 percent in 2019 and 21 percent in the first quarter

of 2021 (Cushman & Wakefield, 2021).

2

Figure 1 - Transaction volume and share of portfolio 2010-2020 (Cushman & Wakefield, 2021)

Figure 2 illustrates portfolio transactions in 2020 by segment. A clear majority (67 percent) of the transactions were carried out within the residential and industrial segment. The same trend can be observed in the first quarter of 2021 where 51 percent of the total transaction volume was allocated towards industrial properties.

Left: Figure 2 - Portfolio transactions 2020 by sector (Cushman & Wakefield, 2021) Right: Figure 3 - Portfolio transactions 2021 Q1 by sector (Cushman & Wakefield, 2021)

The method of acquiring portfolios of properties have been popular among quoted property companies, since 2010 these have accounted for 38 percent of the total portfolio transaction volume. Other purchasers of real estate portfolios are private property companies (24 percent), institutions (20 percent) and private property vehicles (13 percent) (Cushman & Wakefield, 2021).

1.2 AIM AND RESEARCH QUESTION

When talking to consultants and investors within the real estate industry there is no doubt that they

perceive that there is a portfolio premium. Property advisors state that real estate portfolios

containing several properties generate a higher price i.e. a portfolio premium and that the premium

varies depending on segment. The existence of the highly discussed real estate portfolio premium

has not been studied empirically for the Swedish market; therefore, a research gap is evident. The

3 aim is therefore to investigate if there is a portfolio premium and if the premium varies over property segments and time. Leading to our research question and sub questions:

Q1: Is there an evident real estate portfolio premium?

Q2: If a portfolio premium exists, does it differ over property segments?

Q3: Does the premium differ over time?

We believe it is of great importance to research this topic to sort out if a portfolio premium exists empirically. With the results from this paper the authors hope to contribute with further knowledge on portfolio premiums and their allegeable existence. Moreover, this thesis will add to the already extensive research about portfolio premiums in general but contribute with an important aspect when real estate is included. Lastly, the research presented in this paper will be of importance for several actors within the industry. The result presented can aid investors, consultants and valuers when valuing portfolios.

1.3 LIMITATIONS

The data collection and analysis has been limited to the Swedish real estate market and transactions conducted during 2010 through the first quarter of 2021. The segments included in the study are public (education and healthcare), industrial, office, retail and residential. Only transactions above 40 million SEK have been included in the population sample. Furthermore, the authors have only included deal types classified as investment, which is the purchase of commercial real estate with the purpose to receive an income. Transactions without sufficient information have been excluded.

To obtain accurate net operating income figures, henceforth abbreviated NOI, the authors have used appraisal statements no older than 6 months prior or past the transaction date. For transactions conducted in the first to third quarter, NOI-figures from the transaction year have been used. For transactions conducted in the fourth quarter, NOI-figures from the year after have been used.

Lastly, a limit of 500 million SEK in transactional value has been used to create a clear division between small, medium, and large portfolio. The division was decided upon in accordance with Cushman & Wakefield

.

1 Cushman & Wakefield is a leading global commercial real estate service firm. Cushman & Wakefield is one of the largest commercial real estate services firms with approximately 50,000 employees and 400 offices in 60 countries (Cushman & Wakefield, n.d.).

4 1.4 OUTLINE

Chapter 1 – Introduction

In the very first section of this master thesis the reader is introduced to the topic through a brief introduction to the subject. The aim and purpose of the master thesis is also presented along with the research questions and the limitations.

Chapter 2 - Literature review

In the second chapter a summary of previous literature within the field of research is presented.

The literature presented touches upon portfolio premiums and financial assets, speculative investor behavior and portfolio premiums connected to property transactions.

Chapter 3 - Theoretical framework

Relevant theories for the thesis are presented in the third chapter. These will act as a foundation when explaining and discussing the results of the study. Theories presented are Modern Portfolio Theory, Arbitrage Pricing Theory, Efficient Market Hypothesis and theory regarding Value Additivity among others.

Chapter 4 - Research method

The aim of the section is to introduce the reader to what methods have been used to collect the needed data for the thesis. Furthermore, the regression model is presented along with the variables included in the analysis.

Chapter 5 - Results

In the fifth chapter the results of the regression model are presented. The presented results consist of the main model as well as regression output from each segment and time period test.

Chapter 6 - Discussion

A discussion is presented regarding the results of the different regressions. With the help of theories from chapter three the results are discussed, and research questions answered.

Chapter 7 - Conclusion

In the last chapter of the thesis, a summary of the results and the discussion following is presented.

Furthermore, the authors provide insight in what research areas should be explored henceforth.

5

2. LITERATURE REVIEW

In the following chapter previous research within the field is presented. This is primarily to introduce the reader to the subject and to give context. Moreover, the aim with this chapter is to prove that a research gap is present which justifies the research question. The chapter primarily touches upon portfolio theory and price premiums within real estate.

2.1 PORTFOLIO THEORY

The literature on portfolio premiums related to risk and diversification is extensive. However, there is a lack of literature on factors which drive and induce real estate portfolio premiums. Therefore, the literature review has been broadened to touch up on other research areas about premiums.

Friedman (1971) conducted a study presenting that the very same mathematical models used to select and evaluate stock portfolios also could be used on real estate portfolios. Friedman also showed that real estate portfolios are superior to stock portfolios regarding return. Portfolios containing real estate assets have been shown to have a negative return covariance with other assets because real estate is traded on local markets with varying economic conditions. Moreover, real estate assets are seen as a long-term investment and are therefore not as sensitive to market movements. John Burr Williams (1938) arrived at the conclusion that the value of a stock should equal the present value of its future dividend streams. Simaan (1987) later studied asset pricing and portfolio selection. Simaan arrived at the conclusion that investors are willing to pay an optimization premium. Meaning that investors are willing to pay more for a diversified portfolio which maximizes the expected utility. Harry Markowitz (1991) conducted several studies on portfolio theory and evolved Williams theories. Markowitz states that dividend streams are to be seen as uncertain and therefore the value should depend on the expected future dividend streams.

Related to Simaans research on portfolio optimization and selection of financial assets are Elton

et al. (1976; 1978) research on criteria’s for optimal portfolio selection. The study from 1976

investigates portfolio optimization without mathematical formulas whilst the research from 1978

used mathematical correlations to find the efficient frontier. A portfolio placed on the efficient

frontier can be seen as efficient, meaning it offers the highest expected return for a defined level

of risk. Later, Jianping & Lee (1994) used mathematical expressions to study the expected returns

on five financial asset portfolios. The authors used a multi-factor model and arrived at the

conclusion that when real estate assets are placed in portfolios, a higher diversification can be

achieved. Incorporating real estate assets with other financial assets leads to less exposure to risk

and a real estate factor premium. Another researcher, Stephen Morris (1996) conducted a study

based on J. Micheal Harrison and David M. Kreps previous research on speculative investor

behavior. Morris's research departure from a scenario where investors have heterogeneous beliefs

about the asset's fundamental values initially, but their beliefs change over time. Morris arrived at

the conclusion that speculative premiums never disappear and depend on expectations of the future

divestments. José Scheinkman and Wie Xiong (2003) developed a model which studied trading

volumes and bubbles that result from speculative behaviors. Their research was based on

heterogeneous beliefs and overconfidence among investors. The authors investigated financial

6 assets and arrived at the conclusion that when a short-sale opportunity is present an investor can sell the asset to another part with an optimistic future belief. This opportunity is valued by the investor and they are therefore willing to pay a price that exceeds their valuation. Mahamoud Qadan and David Y. Aharon (2019) studied the theoretical framework within financial decision- making to investigate the relationship between investor sentiment and stock size premium. The authors found that investors overvalue small stocks when they are less risk averse and that the size premium correlates with investor sentiment.

2.1.1 PRICE PREMIUMS WITHIN REAL ESTATE

Anchér, Grewin and Jacobson (2006) investigated the motives for price premiums on real estate portfolios and the factors that affect the price premium. A qualitative study with interviews were conducted where actors from different parts of the real estate industry shared their thoughts regarding motives for portfolio premium. According to the respondents the most important factors affecting the price premium in portfolio transactions are that investors are willing to pay more to reduce the risk exposure. Another factor motivating premiums is the reduction of transaction costs with portfolio transactions instead of separate deals. The thesis also states that actors like property vehicles and foreign investors have increased the demand and therefore also prices. The interviewees state that the price premium on real estate portfolios is between 5 - 20% at that time (Anchér et al, 2006).

Miles and McCue (1982) analyzed Real Estate Investment trusts (abbreviated REIT) using regression analysis. They found evidence that diversification by property type produces higher risk-adjusted cash yields than diversification by location. In a later study, Miles and McCue (1984) strengthened their conclusion and showed that diversification with property types is more effective since their return did not correlate as much as the portfolios differentiated by region did.

Diversification by property type was more effective to collect a lower risk premium. In 1986, Hartzell et al. identified five ways to diversify a real estate portfolio to minimize the risk premium.

The five ways are region, property type, property size, lease maturity and metropolitan statistical

area (core area). In a later study Hartzell et al. (1987) continued their work and stated that

diversification by region is preferable but the sectioning between regions should be done after

economic fundamentals. Areas with the same economic state are rated equal and diversification is

formed by owning properties in areas with different economic states. That diversification after

economic factors is a superior strategy was confirmed by Malizi and Simons (1991) and Mueller

(1993). Eichholtz et al. (1995) also studied the diversification of real estate portfolios and the

generic approach is to use different property types and geographical regions. According to the

study, in the US diversification over regions is more effective for retail properties but not for

offices since they had similar performance across regions. The opposite result was found in the

UK, retail properties were regionally uniform and for office diversification by both property type

and regions was favored.

7 In 1996 REIT’s were studied by Capozza and Lee. They found that retail portfolios trade at a premium, however industrial properties are traded at their net asset value. The authors found that securitization increases the value for retail properties whilst the opposite is evident for industrial properties.

In addition to research on portfolio premiums in general Pivo and Fisher (2011) examined the effect of walkability on investment returns and property values. Walkability refers to the degree to which an area located in proximity of the property encourages walking for functional or recreational purposes. They found that a greater walkability affected office, retail and residential property values positively. However, industrial properties were not affected by walkability hence, no premium could be derived to this segment. The article also states that walkability does not have any significant effect on historical investment returns and that walkable properties have as much or less potential to generate returns as less walkable properties. But this statement assumes that the properties are priced correctly.

Lekander (2016) states that real estate can fulfill four different roles; an asset class reducing portfolio variance, an inflation-hedging asset class, a liability matching asset class and one generating high returns. Ang (2012) studied what role real estate plays in the asset allocation puzzle and states that different kind of investors seek different assets. An asset investor believes that contracts with shorter terms are more attractive because it better disturbs the asset risks of the bond portfolio. For an investor seeking returns, the focus is on wrongly priced assets or markets. For inflation-hedging investors, long contracts with strong arrangements is wanted.

Jadevicius recently (2019) did a study on real estate portfolios and the benefits of global diversification for core property funds. According to the study, an optimal portfolio has equal parts invested in Europe, USA and Asia Pacific because the currency differences lowers the risk and gives the highest return.

Another factor that influences portfolio premiums according to Bertrand M. Roehner is speculative behaviour. In 2001, Bertrand M. Roehner wrote about speculation and the effects it has on financial and real estate markets. The study investigated the mathematical mechanism of specialism and how it affects markets. The author could conclude that speculative behavior contributes to price peaks where real estate has a positive amplitude that is correlated with the price. Shortly after, Gregory W. Brown and Micheal T. Cliff (2005) conducted a study on asset valuation and investor sentiment. Where investor sentiment refers to the overall mood among investors shows towards a certain financial asset or market. The survey displayed a positive relationship between investor sentiment and market returns indicating that these findings have the ability to explain assets deviation from their intrinsic value. Later, Jim Clayton, David C. Ling and Andy Naranjo (2008) investigated the role of investor sentiment and fundamentals in commercial real estate markets.

The result of the study showed that behavioral paradigms open for irrational behaviors and limits

in the arbitrage opportunity. By using a behavioral paradigm, the authors could conclude that

8 trading volumes, investor sentiment and capital flows affect the determination of asset prices.

Schiller (2014) states that market prices of speculative assets reflect the taste and the technology available as well as the future possibilities. The prices also reflect the social psychology and sociology today as well as possible changes in these.

Kungliga Vetenskapsakademin (2013) reports on Eugene Fama, Lars Peter Hansen and Robert Shiller’s research on financial markets and asset prices. In the article, which is a summary of their research that was awarded with the 2013 Nobel prize in economic science, behavioral finance is discussed and the implications this has on asset prices. Asset values are often calculated in a rational way by looking at the future dividend and discounting these using a constant discounting factor (per period). However, Shiller noticed that the fluctuations in price given these circumstances did not match rational behaviors with investors. This led to questions of why the discounting factor would change and why the fluctuations in the discount factor would explain the changes in price. One possible explanation is to abandon the reasoning that investors are rational and see to behavioral finance where unrealistic expectations are present and high prices can be explained by an optimistic view of future cash flows. This might be one of the reasons for asset prices to deviate from their intrinsic value.

Price premiums in sale- and leaseback transactions has been studied by Sirman and Slade (2010).

Hedonic price estimation was used, and a conclusion was found that sale- and leaseback

transactions had a higher price than other market transactions. When compared to other

transactions a price premium of 13.86% was evident. However, when adjustments had been made

and income had been accounted for, the price premium disappeared. They therefore stated that the

market is efficient.

9

3. THEORETICAL FRAMEWORK

In the following section the authors will present relevant theory that will be used later in the thesis to discuss the results found.

3.1 MODERN PORTFOLIO THEORY

The very first study on portfolio theory was presented by Harry Markowitz in an article, which was published in the Financial Journal in 1952. Markowitz proved that investors could create portfolios by combining assets to generate the highest expected return at a given risk level. He showed that one could maximize portfolio returns by diversification, i.e. investing in different assets to minimize risk. The level of portfolio risk was measured through standard deviations.

Markowitz stated that if assets have a correlation that is equal to one, the level of risk in the portfolio equals the weighted mean of the assets. If the correlation for the assets was < 1 the diversification is larger, and the level of risk is reduced. The findings by Markowitz suggest that an investor should diversify their portfolio so that the covariances between securities are minimized to yield the highest return. When calculating the level of risk of the portfolio the assets covariances need to be calculated in regard to systematic and unsystematic risk (Markowitz, 1952).

3.2 ARBITRAGE PRICING THEORY AND THE LAW OF ONE PRICE

To determine asset prices one can use the arbitrage pricing theory. The theory is based on the law of one price which states that two identical securities cannot sell at different prices (Elton, 2014).

In a situation, where two identical securities are priced differently there is an arbitrage opportunity, i.e. arbitrage appears when one can make a profit without taking any risk. Law of one price assumes a frictionless market where there are no transaction or transportation costs, or legal restriction and the currency exchange rates are the same (Berk & DeMarzo, 2017).

3.2.1 VALUE ADDITIVITY

Closely connected to the arbitrage pricing theory is the theorem of value additivity. The theorem assumes that no arbitrage opportunities exist and that the price of an asset with a linear return must be given by the same prices as of the other assets (Varian, 1987).

Price (C) = Price (A + B) = Price (A) + Price (B)

Value additivity is defined by the law of one price. The equation above describes the theory. If there are two securities A and B, and a third security C with the same cash flow as A and B combined, security C must have the same value as A and B. Hence the same price according to the law of one price. Value additivity states that the price of asset C must equal the price of asset A and B - the portfolio. Value additivity and the law of one price state that the net present value of a portfolio of securities must equal the net present value sum of all independent assets (Berk &

DeMarzo, 2017). Value Additivity assumes asset prices which already reflects the value of the

portfolio, also called equilibrium prices. However, if the value of a portfolio exceeds the value of

10 the assets included an arbitrage opportunity exist. This opportunity could be exploited by investors who could buy the assets separately and sell them in a portfolio. The same goes for assets with values which is below the combined value. The portfolio could then be unbundled, and investors could make a pure profit (Varian, 1987)

3.3 VALUE THEORY

An investment opportunity consists of multiple cash flows that occur over a longer period of time.

To value cash flows that occur at different points in time, the income streams needs to be discounted. To get the present value of a stream of cash flows the present value of the cash flow is calculated using the following formula (Berk & DeMarzo, 2017):

𝑃𝑃𝑃𝑃 = C (1 + 𝑟𝑟)

PV = Present Value

C = Cash flow r = Interest rate

n = Periods to discount

The method of discounting future cash flows with a risk adjusted discount rate is the most common method used when evaluating an investment opportunity. In addition to this, there are multiple other value theories. Two of these are Intrinsic and Subjective value theory, where the intrinsic value refers the underlaying value of an asset (Kenton, 2021) and the subjective value implies that the value of an asset is determined by how useful the asset is to the individual and that the value cannot be measured properly (Kagan, 2021).

3.5 EFFICIENT MARKET HYPOTHESIS

Relating to the arbitrage pricing theory is the efficient market hypothesis. According to the efficient market hypothesis capital markets are efficient when the price fully and correctly reflects all information available on the market. When the price is set in efficient markets all available information is used and is incorporated in the price. Solely new information can alter the price (Berk & DeMarzo, 2017). If there is a portfolio premium and investors are willing to pay a portfolio premium one could say that the market is not efficient, and the efficient market hypothesis does not apply (Eatwell, 1994). Efficient market hypothesis states that there are no arbitrage opportunities and that competition between investors will eliminate all positive net present value prospects (Berk & DeMarzo, 2017).

3.6 TRANSACTION COST

The size of the transaction costs is very relevant to the real estate industry from all perspectives.

The real estate markets have high transaction costs compared to other markets like the stock

market. The real estate market has high transaction costs because of its relative intransparency and

11 the involvement of many agencies (Nozeman, 2010). Transaction costs have been defined several times by different authors. Coase (1937) was the first to mention the concept. The definition he used for transaction costs was “The costs of using the price mechanism in the market”. The price mechanism is the costs of using a consultant or subcontractor instead of doing or making it yourself. Nordensky and Bulatovic (2005) have studied the transaction costs in commercial real estate deals in Sweden and according to them transaction costs can be defined as the costs that arise when purchasing something instead of producing it. These costs can be divided into the cost of finding a purchaser or vendor, to find the right price and to complete the deal. According to Geltner (2007) transaction costs can be divided into property transfer tax, agent fees, legal fees and due diligence. Due diligence can be divided into six major categories: legal, financial, physical, building services, environmental and regulatory.

Nordensky and Bulatovic (2005) also states that in the Swedish market, the stamp duty diminishes

when properties are traded through company transactions lowering the transaction costs. If the

properties before sale are placed in a portfolio, the transaction cost will be even lower since only

one process is conducted instead of several.

12

4. RESEARCH METHOD

In the following section the method used to collect and analyze required data arguments to why the chosen method is presented. In addition, the regression model and variables included in the model will be presented.

The study will be conducted with a quantitative research approach due to the nature of the research question. When following a quantitative process, a structured methodology is important. Mainly for the purpose to make the study replicable and reliable. The sample must be large enough so the result can be generalized. The validity of the process needs to be considered to make sure that the measures taken are appropriate. Quantitative research is associated with a deductive approach, using data to test theories (Saunders, Lewis & Thornhill, 2016). Quantitative research uses numerical statistical analysis to prove or reject the hypothesis (Williams, 2007). The collection of data requires an understanding of the relationship between the variables used (Soiferman, 2010).

4.1 DATA COLLECTION

All data used in the thesis will be secondary compiled data, where the authors have sorted and selected data that is to be used. The dataset consists of transactional data from the Swedish real estate market between 2010 through the first quarter of 2021. The data has been provided by Cushman & Wakefield, and the majority of the data has also been collected by the company. Some transactional data has been collected by the authors, which has been done through searching for articles about the specific transaction online as well as in Datscha

. The net operating income figures (abbreviated NOI) which have been used in the regression have been retrieved from the Cushman & Wakefield valuation database. The data set provided by Cushman & Wakefield has been prepared by the authors in accordance with the limitations listed in section 1.3. This was done in order to only include relevant data in the regression analysis.

4.1.1 POSSIBLE SHORTCOMINGS WITH THE DATA

Data samples used in quantitative research should be collected with the random sampling method, according to Gaganpreet Sharma (2017). This is a method where each data point in the population has an equally large chance of being selected as a sample. This means that the population has an equal probability to be chosen. The population is all properties in Sweden, both sold and unsold and our study only represents a part of the population.

A shortcoming with secondary data is that it might have been collected for another purpose. The quality of the data must be checked, and the sources evaluated (Saunders, Lewis & Thornhill, 2016). In addition, the data set that will be used contains data that has been interpreted by the part that has collected the data. Meaning that interpretations made by this person may vary from

2Datscha is a leading software as a service company with three offices in Sweden, Finland and UK. (Datscha, n.d.)

13 interpretations the authors would have made. However, this cannot be controlled for and can therefore be a possible shortcoming.

Selection bias is yet another possible shortcoming of the data. Selection bias occurs when the sample used does not accurately mirror the target population, i.e. one has failed to ensure randomization of a population sample. This can cause some groups or individuals in the population to be less likely to be included in the sample. When this error is present the statistical analysis can be distorted, affecting the statistical outcome and test (Hill et al., 2018).

4.1.2 VARIABLES IN THE REGRESSION ANALYSIS

Below all variables in the regression analysis are listed including information about their definition and the collection process.

PRICE

Price is the dependent variable in the regression model and is expressed as SEK per square meter.

The transaction price of each transaction has been gathered by Cushman & Wakefield.

TRANSACTION YEAR

The dataset covers transactional data from 2010 through the first quarter 2021. To account for changes in the overall macroeconomic state of the Swedish economy and the state of the Swedish property market, time dummies have been included. The data is expressed in years where 2021 only includes transactions made in the first quarter.

PORTFOLIO

The variable illustrates the portfolio size regarding both value and square meter. The category has been divided into:

1. Small portfolios which include five or less properties. Where yet another parting has been made; small portfolios with a transactional value under 500 million (Small_Portfolio_under500) and small portfolios with a transactional value over 500 million (Small_Portfolio_over500).

2. Medium portfolios which include six to 20 properties. Where the same parting as above has been made; medium portfolios with a transactional value under 500 million (Medium_Portfolio_under500) and medium portfolios with a transactional value over 500 million (Medium_Portfolio_over500).

3. Large portfolios which include more than 20 properties. Where Large_Portfolio_under500

includes property transactions under 500 million and Large_Portfolio_over500 includes

portfolio transactions over 500 million.

14 Portfolio_No includes all transactions that are not a portfolio i.e. single unit transaction.

LOCATION

Location is a categorical variable and has therefore been converted into a dummy variable. The property transactions in the dataset are conducted all over Sweden and are therefore categorized into three locations: City, Regional and Rural. Table 1 illustrates what city locations that are included in each category. The purpose of the location variables is to group areas after attractiveness, to capture the willingness to pay for properties located in different areas. A map of all city, regional and rural locations can be found in Appendix 1.

Table 1 - Geographical areas included in the location variable, City, Regional and Rural

INVESTOR NATIONALITY

Account for if the investors are international or Swedish. Dummy variable, investor with Swedish nationality has been given the value 1 and international investors the number 0.

NET OPERATING INCOME

The NOI is the rental income minus the operating costs. This is a numerical variable and is expressed as the annual NOI per square meter. The NOI data is from Cushman & Wakefield valuations of the properties. The valuations have been conducted by Cushman & Wakefield within a time span of 6 months prior or after the transaction date. For the transactions conducted in Q1 - Q3, have been assigned the NOI from the transaction year. For transactions conducted in Q4 the NOI for the next coming year is used.

SIZE

The total lettable area of the property/properties included in the transaction, expressed in sqm.

OMITTED VARIABLES

Omitted variables have been used to solve the issue with dummy traps. The following variables have been omitted: Year_2010, Location_Rural, Portfolio_No. The omitted variables are included in the intercept and therefore act as our reference point.

SECTOR

To answer the second research question, if the portfolio premium varies over property segments

the model has been tested on different datasets where the authors have created a new dataset for

15 each segment. The same regression model is used but only transactions from the same property segment are included instead of all transactions like the first model.

Explanation of each variable is listed in Table 2 and a summarization with descriptive statistics is presented in Table 3.

Table 2- Description of variables included in the dataset

In Table 3 below the summary statistics is presented. There are in total 825 observations of each

variable. Of all time-dummies in the table, year 2021 has the lowest mean, which is natural since

only data from the first quarter of 2021 is used. Transaction intense years with high means are

2017 and 2019. Regarding the location of the transactions, location city has the highest mean

meaning that most of the transactions have been conducted in city locations. Most of the portfolios

have been small and had a transaction value under 500 million SEK.

16

Table 3 - Summary statistics of all variables in the regression analysis

4.2 REGRESSION MODEL

The research question investigates whether there is a portfolio premium or not. Since the research question is of quantitative kind the authors have chosen to perform a regression analysis. The aim is to investigate and quantify what independent variables that influence a higher price. Multiple linear regression is a commonly used technique for investigating the assumed causal relationships between a dependent variable and a set of independent variables which influences the dependent variable (Nayebi, 2020). A hedonic regression analysis will be used which is based on the hedonic pricing model. To examine statistical relationships different software can be used. The authors have chosen to conduct a statistical analysis with the help of the statistical software R.

Table 2 displays the variables used in the regression. Price is the dependent variable and is expressed in SEK per square meter and will be a logarithmic value in the regression.

The dataset which the analysis is based on consists of both non-numerical, which has been

translated into numerical data, and numerical data. To perform the regression analysis, dummy

variables have been created (see table 2). The dataset consists of 825 observations divided into 219

17 portfolio transactions and 606 non-portfolio transactions. The portfolios have been categorized into six size groups explained in section 4.1.2.

4.2.1 POSSIBLE SHORTCOMINGS WITH THE MODEL

Multicollinearity is a phenomenon that can arise when two or more variables are highly correlated with one another. Multicollinearity can be labeled as perfect or near multicollinearity. Perfect multicollinearity is present when there is an exact relationship between two or more variables. In contrast, near multicollinearity is when there is a non-negligible but not perfect relationship between two or more variables. If near multicollinearity would be present but ignored this could result in high R

values and high standard error with individual coefficients. With this present, the regression will look good, but the individual variables will not be significant. This arises when explanatory variables are very closely related because it is hard to observe individual contributions of each variable to the overall fit of the regression model. In addition, the regression can become sensitive to small changes. Removing or adding explanatory variables may cause large changes in the coefficient values or the significance of the other variables. Near multicollinearity can cause a wide confidence interval for the parameter, causing significance tests to give inappropriate conclusions. There are several ways of dealing with multicollinearity, one can either ignore it, drop one of the collinear variables, remodel the highly correlated variables into a ratio or collect more data (Brooks, 2014).

4.2.2 HEDONIC PRICING MODEL AND REGRESSION ANALYSIS

To examine statistical relationships which the quantitative approach enables, software packages are often used (Saunders, Lewis & Thornhill, 2016). In this study, R has been chosen as the statistical software. A hedonic regression model will be used to create a regression model of desired fit, given the nature of the research question. The hedonic price model was first innovated by Andrew Court where the term hedonic was used to describe the weighting given the importance of the components. Goodman (1998) and Rosen (1974) describe hedonic as “the implicit price of attributes that are revealed to economic agents from observed prices of differentiated products and the specific amounts of characteristics associated with them” (Rosen, 1974;34). Rosen also proved that hedonic theory could be used for market valuations of different products with respect to their utility-bearing functions. Based on this research by Rosen, hedonic price functions can be specified for the property market as this can be seen as a differentiated product market.

As previously mentioned, the hedonic regression model can be used to recognize heterogeneous goods that are described with attributes or characteristics. When applying this to commercial properties or housing these characteristics can be made up of locational or structural attributes.

Because you cannot sell characteristics there is no market for these, therefore the price of these are

not observed independently. However, the characteristics will contribute to the price of the

property (Haan & Diewert, 2013). The dependent variable will be the price of the good and the

independent variable will have different characteristics that might affect the price in property

transactions. The hedonic regression model is a multiple regression model which can be used to

18 understand relationships between several variables. Multiple regression analysis is well suited when there are several possible explanations to the relationship between the explanatory variables (Rubinfeld, 2000).

The hedonic regression model is typically linear and looks like the equation (1) shown below Y

= β

+ β

X

+ β

X

+ … + β

X

+ Ɛ (1) The variable Y

is the dependent variable. Β

represents the intercept and when all independent variables are zero the intercept will be stated. The variables x

, x

, x

…. x

will be set by t explanatory variables. The β -values will quantify the effect of each explanatory variable. It is important to consider the variables included in the regression model. Lastly Ɛ represents the error term which capture all influence on Y

this is not captured by the independent variables (Brooks, 2014). If variables that can be expected to affect the price are excluded, we may suffer from omitted variable bias (Rubinfeld, 2000).

There are several variations of the multiple regression model. One of these is the log-log model, which is often used in economic modeling. The log relationship is used to reduce and regularize data that is skewed. The logarithmic transformation is often used when Y

is a variable of monetary value such as: price, salary or income, the variable often measures the size. Variables of log-log characteristics often have positive values and are positively skewed to the right with a long tail.

This indicates that a small portion of the population has large values. The transformation of values to ln(x) makes them less extreme, hence values that have been transformed with ln(x) are much closer to a normal distribution for variables of this type (Hill et al. 2018). The log-log model can look like equation (2) displayed below:

ln(Y

) = β

+ β

X

+ β

X

+ … + β

ln(X

) + Ɛ (2) In the study one regression model was developed which was tested with eight different input sheets. Model 1 investigates whether portfolios are sold at a premium or not. The premium is measured by the coefficients for the portfolio dummies (Portfolio_Small_under500, Portfolio_Small_over500, Portfolio_Medium_under500, Portfolio_Medium_over500, Portfolio_Large_under500 and Portfolio_Large_over500). The different datasets used with model 1 examines whether the eligible premium varies over property segments and time.

Model 1:

Ln(Price) = β

+ Σ β

Year_xxxx + β

Portfolio_Small_under500 + β

Portfolio_Small_over500 +

β

Portfolio_Medium_under500 + β

Portfolio_Medium_over500 + β

Portfolio_Large_under500 +

β

Portfolio_Large_over500 + β

Location_City + β

Location_Regional + β

Investor_Nationality

+ β

ln(NOI_SEK_per_sqm) + β

ln(Size_area) + Ɛ (3)

19 Portfolio_No and Location_Rural is also included in equation 3, however these are omitted and are therefore included in the intercept.

SEGMENT AND TIME PERIOD TESTING

Model 1 has been used to investigate if there is a portfolio premium present based on the full data set. The model has also been used to for the segment and time period tests separately. When one segment test has been conducted, the observations in the regression is solely from that segment.

Same for the time period tests, only observations from the years in the timespan has been included.

These tests have been conducted separately and not with dummy variables in a joint model with

all segments included. This since it was evident that different segments react differently to the

variables included in the model. Because the dummy variables impact the segment coefficients

differently, a larger fraction of interaction variables would have needed to be included resulting in

few common variables left. Thus, the separate equations generate similar results as the model

containing all segments controlled for with dummy variables.

20

5. RESULTS

In the following section the authors will present the results. The results of the regressions will be discussed further in section six.

Table 4 - Number of observations and share of portfolio

Table 4 illustrates the number of observations and fraction of portfolios included in each dataset.

The purpose of this research study is to investigate if a portfolio premium within real estate is evident and if it differs with property segments and time. This has been done with the help of a hedonic price model.

H

: β = 0 No Portfolio Premium exists H

: β ≠ 0 A Portfolio Premium exists

To test whether to reject the null hypothesis or not the beta coefficients for the variables Portfolio_Small_under500, Portfolio_Small_over500, Portfolio_Medium_under500, Portfolio_Medium_over500, Portfolio_Large_under500 and Portfolio_Large_over500 will be estimated.

5.1 MODEL 1

In section 4.2.2 equation 3 is the model used for the regression.

The results from the first model are presented below. The dataset contained 825 observations and

219 portfolios. When analyzing the output sheet, the authors have chosen to adapt a significance

level of 0.05.

21

Table 5: Results from Model 1 test

The adjusted R

determines the R

value for the regression, adjusting for the number of independent variables. The R

determines the overall goodness of fit for the regression model. The adjusted R

value for the model is 0.6837.

The main variables of interest in this research are the portfolio premiums. From the results displayed in Table 5, we can see that the coefficient for all portfolios with a transaction value over 500 million SEK is positive and significant on a 5 percent level. The coefficient for small portfolios under 500 million SEK is negative and significant on a 5 percent level while the coefficients for large and medium sized portfolios with a transaction value under 500 million SEK are insignificant.

Small portfolios with a transaction value under 500 million SEK are traded at a discount of 13.7

percent but small portfolios with a transaction value over 500 million SEK are traded at a premium

of 17.5 percent. Medium and large portfolios with a transaction value over 500 million SEK also

have an evident portfolio premium of 16.5 percent and 26.3 percent respectively. Indicating that

22 investors pay extra but only for portfolios with a transaction value over 500 million SEK. The premium also increases with the number of properties included in the portfolio. Furthermore, the variables Location City and Regional are statistically significant with a positive coefficient.

Investors pay more for properties in large and regional cities than in rural locations.

5.2 SEPARATE ESTIMATES OF PORTFOLIO PREMIUM FOR THE DIFFERENT SEGMENTS

To investigate if the portfolio premium differs over property segments, several tests was made on each specific segment. The dataset for each segment test only contained data for the segment in question. The dataset for office contained 232 observations and out of these, 46 were portfolios.

In the retail segment there was 202 observations and out of these, 52 were portfolios. The residential dataset contained 126 observations and out of these, 43 were portfolios while the industrial segment contained 188 observations and out of these, 58 were portfolios. Lastly, the segment with education and healthcare contained 77 observations and out of these, 20 were portfolios. As mentioned above, a significance level of 0.05 has been adapted for all tests.

Table 6: Results from Segment test 1 - Office

23 The adjusted R

determines the R

value for the regression, adjusting for the number of independent variables. The R

determines the overall goodness of fit for the regression model. The adjusted R

value for the model is 0.7487.

Table 7: Results from Segment test 2 – Retail

The adjusted R

determines the R

value for the regression, adjusting for the number of

independent variables. The R

determines the overall goodness of fit for the regression model. The

adjusted R

value for the model is 0.7416.

24

Table 8: Results from Segment test 3 - Residential

The adjusted R

determines the R

value for the regression, adjusting for the number of

independent variables. The R

determines the overall goodness of fit for the regression model. The

adjusted R

value for the model is 0.5807.

25

Table 9: Results from Segment test 4 – Industrial

The adjusted R

determines the R

value for the regression, adjusting for the number of

independent variables. The R

determines the overall goodness of fit for the regression model. The

adjusted R

value for the model is 0.6052.

26

Table 10: Results from Segment test 5 - Education & Healthcare

The adjusted R

determines the R

value for the regression, adjusting for the number of independent variables. The R

determines the overall goodness of fit for the regression model. The adjusted R

value for the model is 0.2215.

As previously mentioned, the variables of interest are the portfolio coefficients. Table 8 displays the results from the residential segment test. The output table shows that the coefficients for small and medium portfolios with a transaction value under 500 million SEK is negative. The coefficient for these variables is significant on a 5 percent level while coefficients for portfolio variables with a transaction value over 500 million SEK, as well as, large portfolios under 500 million SEK is insignificant. Small portfolios with a transaction value under 500 million SEK are traded at a discount of 30.8 percent and medium portfolios with a transaction value under 500 million SEK are traded at a discount of 44.9 percent. Furthermore, investors pay more for properties in city locations than in rural locations.

From the results displayed in Table 9 we can see that the coefficient for small portfolios, within

the industrial segment, with a transaction value over 500 million SEK is negative and significant

on a 5 percent level while the coefficients for all other portfolio variables are insignificant. Small

27 portfolios with a transactional value over 500 million SEK are, according to the model, traded at a discount of 41.9 percent. Furthermore, Swedish investors pay less than international investors for industrial properties.

In table 6, 7 and 10 we can also observe that all portfolio coefficients are insignificant on a 5 percent level for the office, retail and education & healthcare segment.

5.3 SEPARATE ESTIMATES OF PORTFOLIO PREMIUM FOR DIFFERENT TIME PERIODS A test over different time periods was conducted, and the dataset was divided into two groups to see if the portfolio premium differs over time. The reason for the chosen distribution is because there is a growth in price per sqm from 2016 to 2020 and after that a small decrease caused by Covid-19. The growth in price per sqm between 2010 and 2015 is more volatile. The dataset for the time period of 2010 to 2015 contained 409 observations and out of these, 105 were portfolios.

The dataset for the time period of 2016 to the first quarter of 2021 contained 416 observations and out of these, 114 were portfolios. As mentioned above, a significance level of 0.05 has been adapted.

Table 11: Results from time period test 1 - 2010 – 2015

28 The adjusted R

determines the R

value for the regression, adjusting for the number of independent variables. The R

determines the overall goodness of fit for the regression model. The adjusted R

value for the model is 0.6994.

Table 12: Results from time period test 2 - 2016 - 2021

The adjusted R

determines the R

value for the regression, adjusting for the number of independent variables. The R

determines the overall goodness of fit for the regression model. The adjusted R

value for the model is 0.6555.

From the results displayed in Table 11, we can see that within the time period of 2010 to 2015 the

coefficient for all portfolios with a transaction value over 500 million SEK is positive and

significant on a 5 percent level. The coefficient for small portfolios under 500 million SEK is

negative and significant on a 5 percent level while the coefficients for large and medium sized

portfolios with a transaction value under 500 million SEK is insignificant. From the results

displayed in Table 12 we can see that within the time period of 2016 to 2021 Q1 the coefficient

for all portfolios is insignificant on a 5 percent level.

29

Within the time period 2010 to 2015, small portfolios with a transaction value under 500 million

SEK are traded at a discount of 23.8 percent but small portfolios with a transaction value over 500

million SEK are traded at a premium of 28.0 percent. Medium and large portfolios over 500 million

SEK also have an evident portfolio premium of 27.7 percent and 35.9 percent respectively. That

is, investors within the time period of 2010 to 2015 pay extra, but only for portfolios with a

transaction value over 500 million SEK. The premium also increases with the number of properties

included in the portfolio. Furthermore, the variables Location City and regional are statistically

significant with a positive coefficient. Investors pay more for properties in the big cities and

regional cities than in rural locations.

30

6. DISCUSSION

In the following chapter the authors will discuss the results presented in chapter five. The results will be discussed using the theorical framework previously presented. Furthermore, the authors will present their own thought on the results.

6.1 DISCUSSION IN REGRADS TO THEORETICAL FRAMEWORK

In chapter five the results of the regressions are presented. Model 1 showed that a portfolio premium was evident for small, medium and large portfolios with a transactional value above 500 million SEK. Small portfolios with a transactional value over 500 million SEK are traded at a premium of 17.5 percent while medium and large portfolios are traded at a premium of 16.5 and 26.3 percent respectively. There is an evident portfolio discount on small portfolios under 500 million of 13.7 percent. Conclusions to as if there is a portfolio discount on medium and large portfolios with a transactional value under 500 million SEK cannot be drawn because of insignificance. The outcome shows that portfolio premiums are evident, and we can therefor answer our primary research question. To sum up, investors do pay a premium but only for larger portfolios and the premium increases with the number of properties in the portfolio. The findings of an evident real estate portfolio premium evoke the question to why investors are willing to pay more to acquire a large set of properties and volume at once.

The second research topic, questions whether the portfolio premium differs over property segments. To elaborate on this, different segment tests were conducted which are presented in section 5.2. The portfolio variables in the segment tests were mostly insignificant. However, the residential segment test showed discounts for small and medium portfolios with a transaction value under 500 million SEK as well as small portfolios over 500 million SEK in the industrial segment.

Because of the issue with insignificance, we cannot draw any conclusions to if the portfolio premium differs over property segments. A possible explanation for these results might be because there are too few observations in each segment which make the data insufficient.

The third research question investigates whether the portfolio premium differs over time. To elaborate on this, two time period tests were conducted, presented in section 5.3. In the years of 2010 to 2015 there is an evident portfolio discount on small portfolios with a transactional value under 500 million. In addition, an evident portfolio premium on all portfolios with a transactional value over 500 million was found. Between 2016 and 2021 Q1 all portfolio variables are insignificant. In the time period 2010 to 2015 we can reject the null hypothesis and state that there is a portfolio premium. However, in the time period 2016 to 2021 Q1 we can not reject the null hypothesis. A conclusion can therefore be drawn that the portfolio premium differs with time.

The presence of an evident portfolio premium raises the question of what motives there are to pay

a premium and if the premium is not justified given the asset included in the portfolio. One of the

reasons for investors, in particular property vehicles, to pay a premium for a portfolio of real

estate assets could be the requirement to invest a certain amount of raised capital within a certain

period.