The relationship between European real estate returns and European stock returns in times of the financial crisis : a structural breaking point analysis

(1)

UNIVERSITY OF AMSTERDAM, FACULTY OF ECONOMICS AND BUSINESS

The relationship between European real

estate returns and European stock returns in

times of the financial crisis

A structural breaking point analysis

Boyd Ridder Student number: 6297668/10003659 Boydridder@hotmail.com Supervisor: dr. E. Giambona University of Amsterdam Master Thesis

Program: MSc Business Economics

Track: Dual track Finance & Real Estate Finance Date of submission: June 25h_{, 2016}

Abstract

This study aims to investigate what the exact relationship is between monthly EREIT return series and monthly stock return series. It does this for five selected European countries in the times before, during and after the recent financial crisis. With a fixed effects regression model it is investigated if the financial crisis can be marked as a structural breaking point for the relationship between the return series of both asset classes. This study finds both return series integrated with each other, for four of the five selected countries. Only for Belgium, these two return series are segmented from each other. In addition, the recent financial crisis cannot be marked as a structural breaking point for the investigated relationship, for none of the chosen countries. The findings of this study will help investors, with a European focus, in creating their optimal-risky portfolios, when combining these two asset classes.

(2)

2

Statement of Originality

This document is written by Student Boyd Ridder, who declares to take full responsibility for the contents of this document.

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

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

(3)

3

Acknowledgments

At the end of the process of writing my master thesis, I would like to show my sincere gratitude to all the people from the past six years that helped me complete my education at the University of Amsterdam, and writing my master thesis in the end. Without the help and support of all the teachers and fellow students I met in these past six years, I would have not succeeded. However, I would like to mention a couple of them in particular.

First, I would like to thank my thesis supervisor dr. Erasmo Giambona, for all his help and guidance during the months when I wrote my thesis. His critical views on my work have helped me a lot in improving my thesis. Also, his quick responses, encouragement and patience made it possible for me to present this study as it is now.

Furthermore, I would like to show my sincere appreciation to my parents, who have always supported me during my time at the University of Amsterdam. Their love,

encouragement and (financial) support during these times, were essential for me in completing my studies.

Finally, I would also like to thank my girlfriend, friends and the rest of the family for their never-ending believe in me during the time that I was studying at the University of Amsterdam.

(4)

4

1. Introduction

Investors and investment portfolio managers are always trying to maximize the return of their portfolios at given levels of risk. Most of the time they do this by constructing investment portfolios which combine several asset classes, which all have their own risk-return profiles. When two different asset classes are combined inside an investment portfolio, investors have to know what the exact relationship is between the total return series of both asset classes. Either these series are integrated with each other, or they are segmented from each other. The question, what the exact relationship is between the two asset classes, is of central importance when constructing the optimal risky portfolio.

When the return series of both asset classes are integrated with each other, this means both asset classes have the same risk-return profiles. This means the return series move in the same directions, and have explanatory power in predicting the total return of the other asset class. An important implication, therefore, is that both asset classes can be used as substitutes inside an investment portfolio, depending on the personal taste of the individual investor. On the other hand, when the return series are segmented from each other, this means both return series move independent from each other. When this is the case, this means both asset classes can be combined together to diversify away the

unsystematic risk inside an investment portfolio, which leads to the desired well-diversified portfolios by investors (Sharpe, 1992).

The relationship between the return series of Equity Real Estate Investment Trusts (further denoted as EREITs) and stocks will be studied in this research. Knowing this exact relationship will be of great value for investors when constructing their optimal portfolios. Can they be used as substitutes inside the investment portfolio? Or do the returns series move independent from each other, which means they can be combined together to diversify away the unsystematic risk inside the investment portfolios? This study will try to give an answer to both questions. What makes these questions interesting, are the special features of both asset classes. In comparison with other asset classes, stocks are highly liquid and transparent in nature. This means they can be traded with ease due to the high demand for stocks, and their share prices incorporates market information without any significant barriers. On the other hand, EREITs are an investment vehicle that makes investing in real estate accessible to any investor. Investors now do not have to reserve a

(6)

6

great part of their capital when they want to invest in the real estate sector. As house prices in the European Union are rising again after the recent financial crisis (Eurostat, 2016), this makes investing in real estate a valuable option for investors once again.

Stock prices and EREIT prices are both sensitive to changes in economic conditions. For example, both asset prices are strongly influenced by changes in national GDP, interest rates or inflation rates. Another example of a phenomenon, which had a huge impact on both asset prices, is the recent financial crisis. The roots of this financial crisis lie in the U.S. housing and mortgage market. The combination of a surplus of available funds from abroad together with a historical low interest rate led to a housing bubble, which began to burst in August 2007. At the same time, investors were still looking for investment alternatives which were regarded as safe and profitable. The generation and selling of financial derivatives like Mortgage Backed Securities (MBS) and Collateralized Debt Obligations (CDOs) stoked up this financial crisis even further. In 2007, interest rates started to rise again, which dampened the demand for real estate and mortgages around that time. This lowered the demand for real estate and led the house prices to fall dramatically. Also global financial institutions, which were heavily exposed to these financial innovations,

experienced significant losses. The crisis reached its peak in September 2008, when the (nearly) failure of some big market participants caused the financial markets to panic. As a consequence, market participants lost their confidence in the global financial system and share prices started to drop all over the world (Bullard, 2009). This negative spiral persisted until half-way 2009.

As described above, the financial crisis had its impact on both asset classes. EREIT returns were affected indirectly due to the drop in overall real estate values, while stock prices declined due to a loss of confidence in the global financial system. The index values and return series of both asset classes were severely affected by the impact of the financial crisis. However, what is most interesting for investors and investment portfolio managers is how the relationship between the return series of both asset classes is affected due to the financial crisis. Were they integrated or segmented before the financial crisis? And did this relationship change after the crisis? Can the financial crisis therefore be marked as a

structural breaking point in the relationship between the two return series? The answers on these questions yield investors and investment portfolio managers interesting insights when constructing their optimal risky portfolios. Both asset classes can be used as each other’s

(7)

7

substitute or can be combined together to obtain diversification benefits inside the investment portfolios. Investors, who make the right choice in the use of the two asset classes, are most likely the investors who obtain the highest profits and most optimal risky portfolios in the end.

Numerous studies have studied whether the real estate market and stock market are integrated, while the results have been diverse due to differences in sample periods, data, methodology and research focus. Almost all studies with respect to the

integration/segmentation question, or with a focus on the impact of the financial crisis, have a focus on the U.S. However, looking at the development stage of both markets, it can be said that Europe has the most-developed markets coming second after the U.S. A study about the relationship between the return series of European EREITs and European stocks in times of the financial crisis is not carried out yet. Therefore, it is the intention of this study to fill in this gap of research. The focus of this study is therefore to look how this

relationship was in the times before, during and after the financial crisis. Special attention is given to the question if the recent financial crisis can be marked as a structural breaking point for this relationship between the return series of both asset classes.

This research question is investigated via a fixed-effects regression model. Therefore, panel data is needed to carry out this research. Data on monthly EREIT returns and monthly stock returns is gathered from the SNL Property Database and DataStream. This data is gathered from stock exchanges of the following European countries: Belgium, the Netherlands, France, Germany and the United Kingdom. Besides data for these main variables of interest, also data for several control variables is gathered for each country separately via DataStream. Due to data availability, the sample period where this data is gathered for spans from March 2006 – March 2016. This sample period is divided into three sub periods, namely: Before the financial crisis (March 2006-July 2007), during the financial crisis (August 2007-June 2009) and after the financial crisis (July 2009-March 2016). These different time periods are then represented in the panel regression model as time dummy variables. To account for the differences between the different countries in the dataset, fixed effects are also included in the model. To test for the actual relationship between the two return series, and the effect of the financial crisis on this relationship, a t-test will be performed on the stock return variable and the stock return-financial crisis interaction term respectively. With these data, tests and methodology it should be possible to determine

(8)

8

what the exact relationship is between the two return series and if the financial crisis can be marked as a structural breaking point for this relationship between these two asset classes.

The remainder of this study is organized as follows. In section 2, the relevant

literature about this subject is discussed. Section 3 will explain the methodology that is used in this study. Next, section 4 describes the data that is used for this study and also provides some descriptive statistics of the dataset. Section 5 will present the empirical study itself and is followed by section 6, which provides the robustness checks for the results. Finally, section 7 will discuss these results and provides the conclusion of this study.

(9)

9

2. Literature review

Relevant articles about this subject will be discussed from different dimensions in this section. First, some relevant literature about Real Estate Investment Trusts (denoted further as REITs) and its special features are reviewed (2.1). Second, the literature about stocks in general and its index returns is discussed (2.2). Next, some studies which found enough evidence in favor of integration between the two markets will be reviewed (2.3). Further, the relevant research that supports the view that the two markets are segmented will be discussed (2.4). This section ends with a discussion of some relevant studies about structural breaking points in the relationship between the EREIT- and stock market (2.5).

2.1 Real Estate Investment Trusts

REITs can be described as financial intermediaries which serve as a link to transfer funds from saving accounts to the real estate sector (Schulkin, 1971). These funds are used to obtain ownership positions in real estate and to obtain their shares in different types of mortgage loans. People can now buy shares of these REITs to indirectly invest in the real estate sector. REITs were developed in 1960 and are characterized by their special feature which exempts them from paying Federal corporate income taxes when they meet certain requirements. To qualify for this tax exemption REITs should meet the following

requirements (Schulkin, 1971): i) The REIT must be a passive investor, while the active manager of the properties can own up to a maximum of 35% of the REIT stocks, ii) at the end of each quarter 75% of the REITs’ total asset value must consist of real estate, cash and government securities, iii) at least hundred persons should hold shares of the REIT while five or less persons cannot own more than fifty percent of the shares, iv) at least 75% of gross income should come from rents, mortgage interests or gains from real estate sales and v) at least 90% of ordinary income should be distributed to the shareholders of the REIT as dividend.

Despite all the requirements to qualify as a REIT, this financial intermediary can still be divided into three broad categories. The first category is the ‘Construction and

Development Loan REIT (C&D REITs),’ which are specialized mainly in financing several construction and development projects in real estate through the use of mortgages (Schulkin, 1971). The second category is the ‘Long-Term Investment REIT,’ which can be

(10)

10

divided into three subgroups namely: Equity REITs (further denoted as EREITs), which specialize in direct ownership of income properties (like office, retail and apartments), Mortgage REITs (further denoted as MREITs), which specialize in mortgages behind the real estate properties and lastly the Hybrid REITs which combine several features of the two other subgroups. This study will focus solely on the returns of investing in the before mentioned Equity REITs. The third main category named by Schulkin (1971) is the group with all the REITs which do not qualify for the first two main categories, and are labeled ‘Miscellaneous REITs.’

The different kinds of REITs show a long-run linear relationship in stock prices. In a study by He (1998) for example, EREITs and MREITs show the same response to changes in market fundamentals for the period 1972-1995. One of those fundamentals is the change in real estate returns. This means there is a causal relationship between the two REIT markets returns. The results of this study imply that changes in the stock price of one of the two markets can simultaneously cause changes in the stock price of the other. From a different point of view, this means that it is not possible for investors to obtain diversification benefits in the long-run from investing in these two different types of REITs. This finding is also confirmed in a later study by Lee and Chiang (2004), who show that both types of REITs can only be used as substitutes inside an investment portfolio. When both type of REITs can be treated as substitutes, this means the returns of one type can be forecasted by looking at the returns of the other, so the two types of REITs can be treated a single asset class when constructing the optimal-risky portfolio.

A study by Chen and Tzang (1988) looked at the return determinants of EREIT and MREIT returns. They found both types of REITs were sensitive to long-term interest rate movements for the period 1973-1979. However, for the period 1980-1985, the two REIT classes were influenced by both short-term as long-term interest rate fluctuations.

However, the sources of these interest rate movements were different for both REIT classes. EREITs are only sensitive to changes in expected inflation, while MREITs are affected by changes in both expected inflation as real interest rates.

Another study by Seiler et al. (2001) finds that combining investments in public and private real estate do not yield any diversification benefits in the portfolio of investors. This means that the inclusion of EREITs inside a pure real estate portfolio does not diversify away the unsystematic risk inside the portfolio, nor does it add any more liquidity or transaction

(11)

11

cost reductions to the portfolio. However, a study by Giliberto (1990) finds when financial market effects are removed; the two return series are strongly positively correlated. This strong positive correlation between the residuals of the two return series implies that there is an unidentified common factor between the return series of both types of assets.

Unfortunately, this common factor is yet to be found.

Finally, a study by Eichholtz (1996) finds that the combination of different

international REITs can help to diversify away the unsystematic risk inside a pure real estate portfolio. This ‘’international diversification benefit’’ is even bigger for pure real estate portfolios than it is for portfolios with stocks and bonds.

2.2 Stock returns in general

Besides investing in REITs, investors can also choose to invest in regular stocks as a part of their portfolio. Ang and Bekaert (2007) find in their study that these stock returns are predictable, at least in the short term. This predictability can be derived mainly from short term interest rates. Similar results were found by an earlier study by Fama and Schwert (1977). Another determinant of stock returns was found in a study by French et al. (1987). They found that the expected market risk premium is positively related to the volatility of stock returns. This means if the difference between the expected return of a stock and the risk free return goes up, the volatility of the stock returns will rise as well. Another study by Fama (1981) found both the inflation rate and real domestic activity (like GDP) as two other important determinants of stock market returns. The inflation rate and stock market returns are negatively related, while the real GDP rates are positively related to the stock market returns. Reasons for these relationships can be derived from the money demand theory and the quantity theory of money.

Further, a study by Bekaert et al. (2009) finds that there is an upward trend in return correlations for the European stock markets. This means that these markets tend to co-move together and thus have the same risk-return profiles. Also, they find that these correlations are bigger for large growth stocks than small stocks and these differences tend to grow over time. An important implication of their study is that investors cannot use stocks from different European stock exchanges together to diversify away the unsystematic risk in their portfolios. On the other hand, a study by Eun et al. (2008) also finds enough evidence for co-movement between different large-cap stocks, because of their same

(12)

12

reaction to shocks in general market fundamentals. However, they also find that holding a combination of large-cap and small-cap stocks can yield the desired diversification inside the investment portfolio. The reason for this is that small-cap stocks are more influenced by local factors than by global factors. Therefore, small-caps do not move in the same

directions as large-cap stocks, and thus the combination of the two types of stocks can be used to obtain diversification benefits inside the investment portfolios of small and individual investors. The constrained accessibility of these small-cap stocks is the reason that this diversification benefit is not applicable for large institutional investors.

Another possibility to diversify away this unsystematic risk in the portfolio is to combine different types of asset classes, which have different risk-return profiles (Sharpe, 1992). Therefore, this study will investigate what the exact relationship is between both the EREIT market and the stock market, and if they co-move together in Europe in the short run. The outcome of this study will determine if it is possible for investors to obtain

diversification benefits inside their investment portfolio when they combine the two asset classes.

2.3 Integration between the national stock markets and national REIT markets

There are several other studies that examine the relationship between the stock market and the REIT market. One of the first papers that analyzed this relationship was the research in the U.S. by Liu et al (1990). The results were twofold. First they found evidence for co-movement between the national stock indices and EREIT return indices (integration). They define integration as the situation when only systematic risk is priced for both stocks and real estate relative to the overall market index. However, they found evidence in favor of ‘’segmentation’’ between both markets, when appraisal based returns were used, which means that both markets have different expected returns and risk profiles. Following Liow and Yang (2005) stocks and real estate are substitutable in the situation of integration. However, when ‘’segmentation’’ arises, the two assets can be used together to diversify away the unsystematic risk in a portfolio. Another study by Gyourko and Keim (1992) in the U.S. finds that the EREIT market and the S&P 500 index are integrated in the short run. They found a significantly positive correlation between both markets, which means that both markets co-moved in the same directions and had the same risk-return profiles for this sample period (1962-1990). Another implication of the findings of their study is that the

(13)

13

returns on the S&P 500 index had significant explanatory power in predicting the returns of EREITS in the short run. These findings are in line with the results of another study by Ambrose et al. (1992). They found that the total return series of both markets displayed strong tendencies consistent with a random walk model. These results were enough to conclude that segmentation does not exist between the stock market and the different real estate markets. This means there was not enough evidence to conclude that the

combination of utility stocks with REITs (both equity- and mortgage REITs) can be used to diversify away the unsystematic risk inside an investment portfolio.

Another study by Chaudry et al. (1999) investigates the relationship between both markets in the long run. With the Johansen’s test they allowed for seasonal influences in the real estate return data, caused by the irregularity of real estate appraisals. The findings of this study were that these two markets were cointegrated in the long run for the period 1978-1996. This cointegration was even stronger when CPI was included in their regressions. This means inflation is an important underlying factor linking the financial-asset markets with the real-estate-asset markets in the long run. On the other side, a more recent study by Lin and Lin (2011) studied this relationship for six Asian economies for the time period 1995-2010. For four of these six economies (China, Hong Kong, Taiwan and Japan) these two markets showed (partial) integration for the given time period. In some of these countries the stock market returns were leading the real estate market returns, while in other countries it was the other way around. They conclude that these results imply that both markets show a variety of inter-relationships depending on different economic and political policy environments. These inter-relationships determine the exact nature of the

relationship between the real estate market and stock market in each specific country. Quan and Titman (1999) found two possible explanations for these co-movements between the two markets which lead to integration. First, the correlation tends to increase if both rents and corporate profits are affected by the same external business cycle shocks. Second, the correlation between the two increases when the expectations about future rents and corporate profits are the same, so move in the same directions. However, this explanatory power tends to be dependent on the length of the measurement interval. Over short time periods, changes in these expectations are likely to be large relative to the actual rents and corporate profits. This means that over short time intervals changes in stock returns are a better predictor of real estate returns than rental rates. However, as the time

(14)

14

period lengthens, stock returns lose their explanatory power relative to rental rates. This means it is likely to expect the correlation between the returns of both markets to decline as the time period interval lengthens. Integration between the returns of both markets will then be less likely with longer time periods.

2.4 Segmentation between the national stock markets and national REIT markets

Eichholtz (1996) took his research outside the U.S. He found that REIT investments showed low international correlation with stocks and bonds in the short run for the period 1985-1994, which is now in favor of market segmentation between both markets. As a possible explanation for this, he mentioned that real estate is more affected by local factors than stocks and bonds. Those are more affected by global factors. Quan and Titman (1999) studied his subject for the same time period, but used a fixed regression model to study this relationship for seventeen countries. However, the results were the same, that there was not enough evidence in support of integration between both markets.

Wilson et al. (1996) studied the relationship between the two markets for Australia, testing for cointegration based on the Arbitrage Pricing Theory (APT) paradigm. They found the two markets to be partially segmented for the period 1975-1993. With their

methodology they ignored the special features of real estate, and instead assumed the presence of arbitrageurs on both markets. An implication of this approach is that the activity of those arbitrageurs would reduce the market segmentation by constructing efficient portfolios. This arbitrage would make sure both asset classes have the same risk-return profiles, which is not in line with market segmentation. However, even with this approach the results of the study were that both markets were partially segmented during the sample period. However, Wilson et al. (1996) also mention to take these results with caution. The regressions may lack from omitted variable bias (inflation, interest rate term structure, etc.) and there may be structural breaks in the used time series.

Chiang and Lee (2002) also found enough evidence in favor of segmentation

between both market returns for the period 1975-1997. For this, they used a style analysis of Sharpe. The result of this analysis is that the price behavior of REITs was unique during the sample period when compared to the price behavior of equity, fixed-income securities and un-securitized real estate. This price behavior cannot even be duplicated by using a combination of the other three asset classes. Therefore, they suggest REITs should be

(15)

15

treated and included as a single asset class when investors are doing their asset allocation in their portfolios. Next, they go even one step further by stating REITs should even be

included in an investment portfolio, when un-securitized real estate is already present in the portfolio. This should be possible, as the risk –return profile of REITs is unique in its nature and the return series are independent from the return series of any other asset classes. These results are also in line with the results of an earlier study by Liang and McIntosh (1998). They use a different subsample and time period than Chiang and Lee, but follow the same methodology. However, they found enough evidence in favor of market segmentation for these two markets for the time period 1984-1997. As the R-squared statistic declined dramatically during the last five years of their sample period, this indicates that REITs have become more unique in nature and therefore have their own risk-return profiles. Therefore, Liang and McIntosh also state that the return series of the REIT market are independent of the stock return series. When the two markets are independent of each other, this

ultimately means that both asset classes can be combined together to diversify away the unsystematic risk inside an investment portfolio (Liang and McIntosh, 1998).

2.5 Structural breaking points in the relationship between the two markets

Wilson and Okunev (1999) did research solely on the relationship between REIT returns and stock returns for the period 1971 -1993. They did this for the U.S., the U.K. and Australia. For their research they used an ARFIMA long memory model with monthly index data. They found both markets partially integrated in the long run for the three countries. From 1971 – 1987 both markets co-moved in the U.S., while they moved in different directions from 1988 until 1993. In the U.K. these relationships were exactly the opposite. From 1971 – 1987 both markets moved independently from each other, while from 1988 – 1993 both markets co-moved in the U.K. In Australia the results were less clear. As a possible explanation for this phenomenon they mentioned the stock market crash in 1987 as a ‘’structural breaking point.’’ This could indicate that these structural breaking points could lead to a change in the relationship between the national stock markets and the national REIT markets. For investors, investing in U.S. securities, around that time, this means that both asset classes could be used as substitutes from 1971 – 1987. After the structural breaking point, however, this relationship changed and the two asset classes could then be used together to diversify away the unsystematic risk in their portfolios.

(16)

16

A study by Glascock et al. (2000) studied this relationship for the period 1972-1996. They used both cointegration and vector autoregressive models to explore the long-run relationship between the two assets classes. They found that REIT-returns behaved more like stocks since the 1990s, while they behaved more like bonds in the time period before. Research by Basse (2009) shows that the recent financial crisis indeed led to a massive structural break in the relationship between stocks and REITS in the U.S. This study finds that after the recent financial crisis REITS became more risky relative to utility stocks, which was a change in the relationship between the two. Similar results were found by Clayton and MacKinnon (2001) who found that during the 1990s the relationship between REIT returns and large cap stocks declined through time, while REIT returns showed a more closely link with small caps from then on, especially when the REIT market was in a downturn. However, this relationship appeared to be more cyclical in nature.

As the above literature suggests, it still remains unclear if the national stock markets and REIT markets are integrated or segmented at this moment. It is more likely that this relationship changes between cycles, with such a structural breaking point, as the stock market crash in 1987, indicating as a sign for the transition of the relationship. Therefore, this study will look if the recent financial crisis could be marked as such a structural breaking point, and thus if the relationship between the two asset markets has changed.

(17)

17

3. Methodology

3.1 Research question & Hypotheses

This research is investigating if the recent financial crisis had a significant impact on the relationship between the national stock market returns and national EREIT market returns. Therefore, the focus of this study is on if this relationship between both returns series has changed during and after the financial crisis. Such a disruption in the relationship between the two return series, due to a dramatic (economic) event, is called a ‘’structural breaking point’’ in a study by Wilson and Okunev (1999). The focus of this study is therefore

translated and investigated with the following research question:

‘’Was the financial crisis a structural breaking point for the relationship between European stock index returns and European EREIT index returns?’’

To come up with an answer for this research question, two hypotheses will be tested in the empirical section of this study. To know if the relationship between the two return series has changed, it should be firstly determined what this relationship was exactly around the time before the financial crisis. A study by Glascock et al. (2000) found that REIT returns in general behaved much like bond returns in the years before 1990. However, after the legal reforms concerning the ownership structure of REITs, the returns series of REITs behaved more like stock returns series after 1990. This means the EREIT returns move in a similar way as the stock returns do and thus that both return series have some explanatory power in predicting the total return of the other series at a particular point in time. As the

correlation tables in section 4.3.2. show, the returns series of both markets are highly positively correlated for almost each of the chosen countries. This high positive correlation for both return series, together with the results of the study by Glascock et al. (2000) lead to the first hypothesis of this study concerning the relationship between the EREIT returns and stock returns in the times before the financial crisis:

Hypothesis 1: The European stock market returns are integrated with the European EREIT

(18)

18

The study by Wilson and Okunev (1999) identified a dramatic (economic) event as an event that can disrupt the relationship between the two return series. For example, in the time period 1971-1987 the total returns of EREITs were integrated with the total stock returns in the U.S. However, after the stock market crash in 1987, this integration dramatically

disappeared and it would take years for this integration to re-establish again. This means that in the following years stock returns lost their explanatory power in predicting the total returns of EREITs, and the two asset classes could not be used as substitutes anymore inside investment portfolios. Research by Basse (2009) already studied the effect of the recent financial crisis on the relationship between the return series of both asset classes in the U.S. He found that after the recent financial crisis REITS became more risky relative to utility stocks, which was a change in the relationship between the two. Again, the two asset classes could not serve as perfect substitutes of each other inside investment portfolios in the years after the crisis. Therefore the following hypothesis will be tested in this study:

Hypothesis 2: The recent financial crisis had a significant impact on the relationship between

the European stock market returns and the European EREIT market returns.

The results of the tests for the second hypothesis will be of great value for investors and investor portfolio managers when constructing their optimal risk-return portfolio. If the two returns series are not integrated anymore with each other after the recent financial crisis this means the returns series are segmented from each other. In this case both asset classes can be combined together inside an investment portfolio to diversify away the unsystematic risk.

3.2 Research methodology

For this study, the same methodology as Quan and Titman (1999) used in their research on the relationship between both markets, will be used. This means that this study will make use of panel data, and therefore makes use of a fixed effects regression model. In this model country fixed effects (λ) will be included to account for the differences between the

different countries in the dataset. The European countries where this relationship will be investigated for are: Belgium, the Netherlands, France, Germany and the United Kingdom.

(19)

19

In this model the national monthly EREIT index returns (RE) are used as the dependent variable, while the other variables are used as explanatory and control variables. Other variables that are included in the regression equation to avoid omitted variable bias are: monthly stock returns (STOCK), monthly percentage change in inflation (INF), monthly percentage change in GDP (GDP), the three-month interest rate (INT) and the 10-year government bond rate (LTB). Furthermore, time dummy variables will be used to indicate the three different time periods of this study: Before, during and after the recent financial crisis. From these three time periods, two dummy variables will be included in the used regression. Dummy variables will only be included for the time periods during the financial crisis and after the recent financial crisis. This is done to be able to avoid the dummy variable trap, which is a common problem with regression techniques when equations are unsolvable because multiple variables are perfectly collinear with each other (Stock & Watson, 2009). This problem is avoided by leaving one of the three time dummy variables out of the regression. Interaction effects between the time dummies and the stock return variable are then also included in the regression to see if the financial crisis had a significant effect on the relationship between the European stock market returns and European EREIT market returns. This is done for the whole sample, as well as for each of the chosen

countries separately. Therefore the following OLS-panel regression equation will be used:

REit = β0 + β1STOCKit + β2INFit +β3GDPit + β4INTit + β5LTBit + β6FC+ β7(STOCK*FC)+

β8PFC+ β9(STOCK*PFC)+λ+εit

With:

 REit = monthly EREIT return at time t for entity i  STOCKit = monthly stock return at time t for entity i

 INFit = monthly percentage change in average CPI at time t for entity i  GDPit = monthly percentage change in average GDP at time t for entity i  INTit = three-month interbank rate at time t for entity i

 LTBit = 10-year government bond rate at time t for entity i

 FC = time dummy variable indicating the time period during the financial crisis (August 2007 – June 2009)

(20)

20

 PFC = time dummy variable indicating the time period after the financial crisis (July 2009 – March 2016)

 λ = country fixed effects

 εit = error term

First, a t-test will serve as the instrument to test for the relationship between the monthly stock returns and EREIT returns in the time period before the financial crisis. To allow for this t-test a normal distribution of the dataset is assumed. As can be seen from the

correlation table in section 4.3.2 both return series are positively correlated when a holding period of one year is assumed (which is nearly equal to the time period before the financial crisis). This positive correlation together with the results of an earlier study by Basse (2009) led to the first hypothesis of integration between the returns series of both asset classes. In this case, both return series follow the same pattern which means that if the value of one of the series goes up, the other follows, and vice versa. Therefore, it is expected that the coefficient on the monthly stock return is significant, which can be tested by a two-sided t-test. The first testable hypothesis will then be:

H0: β1 = 0 H1: β1 ≠0

If the null hypothesis is rejected and the alternative hypothesis is accepted, this means monthly stock returns had a significant effect on monthly EREIT returns in the times before the financial crisis. If this is the case, this means there is enough evidence in favour of the first hypothesis that both return series were integrated around that time.

The second hypothesis tests if the recent financial crisis had a significant impact on the relationship between the monthly stock returns and monthly EREIT returns, and thus can be marked as a structural breaking point. This means both return series were not integrated anymore in the times of the financial crisis. A two-sided t-test will now be performed on the coefficient of the interaction term between the monthly stock returns and the financial crisis dummy to see if the financial crisis had its influence on this relationship. This leads to the second testable hypothesis:

(21)

21

H0: β7 = 0 H1: β7 ≠0

If the null hypothesis is rejected in favor of the alternative hypothesis, this means the second hypothesis is confirmed that the recent financial crisis had a significant impact on the relationship between the European stock market returns and the European EREIT market returns. In this case, this means monthly stock returns do not have explanatory power in predicting monthly EREIT returns in the period during and after the financial crisis. Without this explanatory power, this means both return series are independent (and segmented) and can be used together inside an investment portfolio to diversify away the unsystematic risk. These tests will be performed for the European countries as one group, as well as for each of the five chosen countries separately.

(22)

22

4. Data and descriptive statistics

4.1 Sample selection

In this study the short-term relationship between the returns of the national REIT market indices and the national stock market indices is investigated for five European countries, namely: Belgium, France, Germany, the U.K. and the Netherlands. The reason to choose for these economies differs per country. The reason to choose for the U.K. is straightforward. With a total of 33 listed REITs, this market accounts for 6.95% of the Global REIT Index according to the EPRA (2015). This means data for the EREITs in this country is mostly interesting as this is the largest and most developed REIT market in Europe. This also means data for this market is widely available. The same reasoning, although on a smaller scale, applies for the Netherlands with a total of 5 listed REITs, accounting for 3.05% of the Global REIT Index. The U.K. and the Netherlands are therefore the two EREIT markets with the most activity in Europe, which means that the findings of this study will have the most impact for these countries as well. On the other hand, France and Germany are two of the most developed countries in Europe in terms of their financial systems. Therefore, the exact relationship between the REIT- and stock market is considered as important for investors and investment portfolio managers as they try to optimize their portfolios with respect to return and risk when combining the two asset classes. The Belgian REIT market is chosen because of its close economic link with the four before mentioned countries.

4.2 Data description

Due to data availability, monthly data for REIT index returns is collected for this study. This data is collected from the SNL Property Database, and includes the total returns of the EREITs which have their headquarters in the five chosen countries for the period March 2006 until March 2016. This means the dataset includes only data on EREITS, which are real estate companies which earn at least 75% of their revenues through investing in and owning their own properties. However, for Germany, this data is only available since March 2007. Therefore, for this country the time period of study is from March 2007 until March 2016. For the other countries the time period of study starts from March 2006.

(23)

23

Next, data of the total returns of the national monthly stock indices are gathered for the main stock index of each country. Therefore, the dataset will include data for the following stock indices: BEL20 (Belgium), CAC40 (France), DAX30 (Germany), AEX (Netherlands) and the FTSE100 (U.K.). Again, this data is collected for the time period March 2006 until March 2016. This data is coming from the database DataStream. From this database index values were collected for each national stock index. To derive the total return of each index for each month, the following formula is used: [(Index valuet – Index valuet-1)/Index valuet-1]. In

this formula Index valuet stands for the value of the particular index at time t, while Index

valuet-1 stands for the index value at time t-1. With this formula the percentage change in

index values is calculated between each subsequent month. This percentage change can be seen as the total return from investing in each particular index each month separately.

Data for the different control variables is also collected via DataStream. This means data is collected for national monthly inflation rates, interest rates and long-term bond rates for each country separately. As a first control variable the monthly inflation rate is chosen as EREITs are sensitive to changes in monthly inflation (Cheng & Tzang, 1988). The monthly inflation rate is calculated by taking the average Consumer Price Index (denoted further as CPI) for each month in each country separately, and then looking at the monthly percentage change in this average CPI for each month.

The monthly interest rate is taken as the second control variable, as it controls for omitted variable bias (Quan & Titman, 1999). For this variable the three-month interbank rate was chosen, which is the interest rate banks charge when lending out funds to each other for a period of three months. Data for this variable is nearly the same for all countries, except the U.K., as all the countries which joined the Euro zone have to charge the three-month Euribor-rate. Data for the long-term government bond rates (the third control

variable) is collected for the bonds with a maturity of ten years, for each country separately. The last control variable in the regression is the percentage change in GDP value for each country. This data, however, can only be obtained on a quarterly basis. Therefore, the linear interpolation technique is used for the time period Q1-Q4 for each year separately. With this technique the quarterly data is transformed into monthly data, without any missing values. Linear polynomials are used to construct new data points in the range of the known data points. Therefore, according to Chow and Lin (1971), the dataset still remains discrete and even without any outliers. This technique is also used if there are other missing

(24)

24

values in any of the other control variables. Data on each of these variables is collected to control for omitted variable bias in the regression.

4.3.1 Descriptive statistics for the variables of interest

Descriptive statistics of the total monthly return series for each of the five chosen countries are provided in Table 1. The table reports the number of observations, mean, standard deviation, minimum- and maximum value for the total returns of both EREIT and stock indices. For Belgium, the Netherlands, France and the U.K. there is a total of 121

observations, which equals the monthly time period of interest of this study (March 2006 – March 2016). For Germany, there are 108 observations, and therefore the period of interest is April 2007 – March 2016.

Furthermore, during the sample period, the highest mean EREIT return was found in France, with a mean monthly return of 0.36%. On the other side, the mean monthly return for the U.K. was -0.04%. So on average investors incurred losses when investing in U.K. EREITS, while they made the highest profit by investing in French EREITs during the sample period. For the stock market the highest profit on average was made in German stocks (0.45%), while the biggest losses were incurred with Belgian stocks (-0.81%). With respect to the volatility of both return series, it can be derived from the table that Germany had the most volatile EREIT market with a standard deviation of 10.8%. This can also be seen from the minimum and maximum value for the total EREIT returns with a minimum and

maximum of -49.3% and 49.9% respectively. Finally, for the stock market, Belgium was the most volatile with a standard deviation of 10.3%.

However, the results of this table might be sensitive to outliers, so inferences from this table should be made with certain caution. Therefore, the same descriptive statistics are provided again in appendix A, but are now winsorized with a tail of 2.5% on both the upper side as the bottom side. This is done to control for the effect of potential outliers. This winsorized dataset is also used later on for the empirical study of this research. For some descriptive statistics for the control variables, see appendix B.

(25)

25 Table 1: EREIT returns and stock returns (2006-2016)

Country #Observations Mean Std. Dev. Minimum Maximum

EREIT Belgium 121 0.0012 0.047 (0.220) 0.115 The Netherlands 121 0.0007 0.075 (0.260) 0.239 France 121 0.0036 0.069 (0.268) 0.213 Germany 108 0.0021 0.108 (0.493) 0.499 United Kingdom 121 (0.0004) 0.083 (0.340) 0.436 STOCK Belgium 121 (0.0081) 0.103 (1) 0.108 The Netherlands 121 0.0040 0.056 (0.216) 0.125 France 121 0.0029 0.051 (0.141) 0.126 Germany 108 0.0045 0.059 (0.208) 0.143 United Kingdom 121 0.0042 0.044 (0.144) 0.098

* All values in this table are rounded up to three decimals, except for the mean (four decimals) ** All values should be multiplied with 100 to obtain the actual percentages

Source: SNL Property Database & DataStream

The same EREIT returns and stock returns are plotted over the time period, in Graph 1. In this graph both return series are displayed for each country separately over the sample period. The time variable on the horizontal axis is in monthly units, while the vertical axis displays the total return in base points. As can be seen from this graph EREIT returns behaved more volatile in both directions in the first part of the sample period. This is the case for every country in the sample. Around month 40 (which is nearly equal to the start of the period after the financial crisis), it looks like this volatility is dampened. However,

conclusions about the relationship between the return series of both asset classes cannot be taken by looking at these graphs only. These conclusions can only be taken after doing some empirical tests, which will be done and discussed in section 5 of this study.

(26)

26

Graph 1: Monthly EREIT returns and stock returns over the sample period (2006-2016) -.2 -.1 0 .1 .1 0 _-.1 _-.2 0 50 1 0 0 1 5 0 0 50 1 0 0 1 5 0 0 50 1 0 0 1 5 0 B e lg iu m Th e Ne th e rl a n d s Fr a n c e G e rma n y Un it e d K in g d o m E R E IT R e tu rn s S to ck R e tu rn s T im e

(27)

27

4.3.2 Correlations for different holding periods

In table 2 below, the correlations between the monthly EREIT returns and stock returns are displayed for each country separately. As this study follows the same methodology as the study by Quan and Titman (1999), it is investigated if these correlations tend to decrease as the chosen holding period lengthens. Quan and Titman (1999) argue that these correlations decrease with longer holding periods, as the effect of changes in (expectations of) external business cycles decrease with the chosen measurement interval. Over short time periods, changes in these cycles are likely to be large relative to the actual rents of real estate and profits of corporations. Therefore, stock returns have significantly better explanatory power in predicting EREIT returns than rental rates. However, as the time period lengthens, stock returns lose their explanatory power relative to rental rates, because changes in

(expectations of) business cycles are then relatively small when compared to actual rents and corporate profits. In other words, if these correlations are significant, this means the returns of both the EREIT market and stock market move together over time. Thus, both markets are then integrated and both asset classes can be used together inside an investment portfolio as substitutes.

As can be seen from Table 2, the correlation between the EREIT returns and stock returns is positively significant for four of the five chosen European countries. These correlations are all positive for every chosen holding period. This means that at first sight the two markets are integrated, irrespective of the holding period, and thus that the returns of both markets move together in the same direction. Only for Belgium this correlation is near zero, which assumes both markets are independent of one another. For holding periods of one year and ten years this correlation even becomes negative. Therefore, by looking at these correlations only, at this point in time it can be assumed that both return series are dependent of each other. Only for the Belgian markets, one can consider segmentation between the returns of both markets by looking at Table 2.

(28)

28

Table 2: Correlations between EREIT returns and stock returns for different holding periods

Holding period Belgium The

Netherlands

France Germany United Kingdom 1 Year (0.1123) 0.5255 0.5071 0.3734 0.3147 3 Years 0.1258 0.6059 0.5630 0.3243 0.4951 5 Years 0.0696 0.6032 0.5017 0.3422 0.3874 7 Years 0.1103 0.5849 0.5126 0.3414 0.3729 10 Years (0.0646) 0.5720 0.4788 NA 0.3654

* The 10-year holding period for the national EREIT market and the national stock market in Germany is not applicable, as for this country the data interval only goes from March 2007 until March 2016.

(29)

29

5. Results

5.1 The relationship before the financial crisis

To start with, a simple and a multiple regression are applied to look for the relationship between the stock market movements and the EREIT market movements in the months before the financial crisis. As already discussed in the methodology section of this study, it is expected that changes in stock market movements have a significant effect on changes in the EREIT market movements.

First, a simple regression is carried out to look for the direct effect of stock market movements on EREIT market movements. This leads to a regression of EREIT return values on a constant, stock return values and an error term, as can be seen in Table 3 below.

Table 3: Simple fixed effects regression model of EREIT returns on stock returns for the period March 2006 – July 2007

Variable Coefficient Std. Error t-value P-value 95% Confidence Interval

α -0.0014 0.0023 -0.61 0.544 -0.006 0.003

Stock 0.4599 0.0481 9.56* 0.000 0.366 0.554

* R-squared: 0.1340

** Number of observations: 592 *** Number of groups: 5

**** Country fixed effects are included in the regression to account for differences between the different countries in the dataset over time

From table 3 it can be seen that the coefficient on the variable on stock returns proves significantly positive at both the 5% as the 1% level. As both the dependent as the independent variables are in percentages, these results can be interpreted as follows: ‘A

one percentage change in monthly stock returns is followed by a 0.4599% change in monthly EREIT returns.’ This result implies that both asset classes could serve as each

other’s substitutes inside an investment portfolio in the times before the financial crisis. However, data on stock returns during that time period can only explain 13.4% of the EREIT movements, as can be seen by looking at the R-squared statistic. One can conclude that this regression is suffering from omitted variable bias, and therefore the results of this

regression should be taken with certain caution. To control for this omitted variable bias, data on the following control variables is included in the next regression: Monthly change in

(30)

30

inflation, monthly change in GDP, monthly interest rates and long-term government bond rates. Time dummy variables are included in this regression, to make the distinction between the three different time periods. The results of this regression can be found in Table 4.

Table 4: Multiple fixed effects regression model, for European countries during the period March 2006-March 2016

Α 0.0054 0.0141 0.38 0.704 -0.223 0.033 Stock 0.5025 0.2062 2.44* 0.015 0.097 0.908 Inflation 0.0402 0.135 0.3 0.766 -0.225 0.305 GDP 4.3772 1.2982 3.37* 0.001 1.827 6.927 Interest rate -0.0072 0.0056 -1.28 0.201 -0.018 0.004 LT-Bond rate -0.003 0.0127 -0.24 0.812 -0.028 0.022 FC -0.003 0.0097 -0.31 0.757 -0.022 0.016 PFC -0.01 0.011 -0.95 0.344 -0.031 0.011 FC*Stock -0.0162 0.2222 -0.07 0.942 -0.453 0.421 PFC*Stock -0.1281 0.2177 -0.59 0.557 -0.556 0.299 * R-squared: 0.1945 ** Number of observations: 546 *** Number of groups: 5

**** Country fixed effects are included in the regression to account for differences between the different countries in the dataset over time

When the control variables are added in this multiple regression model, the coefficient on monthly stock returns remains significant at the 5% significance level. The first test of this study is tested with a two sided t-test and should test if the coefficient on stock returns is significant. In both the simple as the multiple regression the coefficient stays well above the critical value of 1.96 (Stock &Watson, 2009) at a significance level of 5%. Therefore, the null hypothesis is rejected in favor of the alternative hypothesis for the first part of the overall research question. This means the European EREIT returns were integrated with the European stock returns in the times before the recent financial crisis. As both variables are quoted in percentages, the coefficient on monthly stock returns can be interpreted as follows: ‘In the times before the financial crisis, a one percentage change in stock returns

leads to a 0.5025% change in EREIT returns.’ Therefore, enough evidence was found to

conclude that the movements in the returns of both markets were partly integrated in the times before the financial crisis. This means the two asset classes could be used as

(31)

31

substitutes inside an investment portfolio around that time. However, they cannot be marked as perfect substitutes as a one percentage change in stock returns is followed by a less than one percentage change in EREIT returns. However, by looking at the stock return series pattern, one could predict the future EREIT return series pattern and adapt his/her investment strategy to gain the most profit around that time. The finding that both returns series are strongly correlated, and thus integrated, are in line with the findings of the earlier study in the U.S. by Basse (2009). Except for the variable on monthly stock returns, only the variable on GDP growth rate proves to be significant in explaining monthly EREIT returns. This is in line with the findings of the study by Fama (1981), where the GDP growth rate proves to be an important determinant of monthly stock returns. As both asset classes moved in the same directions around that time, it can be assumed that the GDP growth rate can also be marked as an important determinant for monthly EREIT returns. In this sample, this means that a one percentage change in the monthly GDP growth rate is followed by a roughly 4.37% change in monthly EREIT returns.

5.2 The effect of the financial crisis on the relationship between EREIT returns and stock returns

To see if the financial crisis had a significant impact on the relationship between monthly stock returns and monthly EREIT returns, the same fixed effects regression model as

presented in Table 4 is used. For this part of the research question the variable of interest is the interaction term between the monthly stock return variable and the time dummy standing for the period of the financial crisis. If this interaction term proves to be

significantly different from zero (positive or negative), this implicates that the financial crisis had a significant impact on the relationship between monthly stock returns and monthly EREIT returns. A two-sided t-test is used to test for this significance. This yields a critical value of 1.96, which the t-statistic should pass to reject the original null hypothesis (Stock &Watson, 2009). Looking at Table 4, it can be seen that this interaction term has a t-value of -0.07, which is equal to a P-value of 0.942. This means the interaction term is not

significantly different from zero for both the 5% and the 1% significance level. Therefore, the null hypothesis cannot be rejected in favor of the alternative hypothesis, which means that the coefficient of the interaction term of interest is not significantly different from zero. This result implies that there is not enough evidence that the recent financial crisis had a

(32)

32

significant impact on the relationship between monthly stock returns and monthly EREIT returns. In other words, the recent financial crisis did not change the relationship between monthly EREIT returns and monthly stock returns, so this economic event cannot be marked as a ‘structural breaking point’ for this relationship. Therefore, it can be said that the return series of both asset classes remain integrated during and after the financial crisis. This means they can only serve as each other’s substitutes inside an investment portfolio. However, these results were generated with a sample group which consists out of all the five chosen European countries. To see if these results are different for the individual countries, the same model will be estimated again, but now for each country separately. The results will be presented and discussed in the next section.

5.3 The relationship per individual country

To see if the financial crisis can be marked as a structural breaking point for one of the individual countries, both a simple regression model as a time series regression model are estimated for each country separately. Looking at the correlation table (Table 2), it can be expected that these results can be different for Belgium as an individual country. Table 2 shows the two return series have a correlation which fluctuates around zero for every holding period. When this is the case, segmentation between the return series of both Belgian asset classes can be expected. This statement is confirmed by the results in Table 5 and 6. As can be seen, the variable on stock returns is not statistically different from zero (with t-values of 0.45 and 0.85 respectively). This means monthly Belgian stock returns did not have a significant effect on monthly Belgian EREIT returns in the period before the financial crisis. Therefore, there is enough evidence to conclude that both return series are independent of each other, and both asset classes can be combined inside an investment portfolio to diversify away the unsystematic risk.

(33)

33

Table 5: Simple regression of EREIT returns on stock returns for Belgium for the period March 2006 – July 2007

Variable Coefficient Std. Error t-value P-value 95% Confidence Interval

α 0.0018 0.0038 0.49 0.628 -0.006 0.009

Stock 0.0402 0.0901 0.45 0.656 -0.138 0.218

* R-squared: 0.0017

** Number of observations: 121

Next, a time series regression model is used to see if the financial crisis had a significant impact on the relationship between the two asset classes in Belgium. As can be seen from Table 6, the interaction term between monthly stock returns and the financial crisis dummy proves to be not significantly different from zero. This means the two return series

remained segmented from each other during and after the financial crisis. The financial crisis, therefore, cannot be marked as a ‘structural breaking point’ for the Belgian return series. However, it can also be seen from Table 6 that none of the chosen variables is significant in this model. This can be due to the high correlation between several control variables which leads to multicollinearity (Stock & Watson, 2009). This problem can lead to biased results in the estimated model.

Table 6: Time series regression model, for Belgium during the period March 2006-March 2016

α 0.1929 0.0225 0.86 0.393 -0.025 0.064 Stock 0.2424 0.2861 0.85 0.399 -0.325 0.809 Inflation 1.4788 1.6319 0.91 0.367 -1.755 4.713 GDP 3.621 3.0896 1.17 0.244 -2.501 9.743 Interest rate -0.003 0.0069 -0.44 0.664 -0.168 0.108 LT-Bond rate -0.0048 0.0049 -0.99 0.325 -0.015 0.005 FC 0.001 0.016 0.06 0.949 -0.031 0.033 PFC -0.0062 0.0206 -0.3 0.765 -0.047 0.035 FC*Stock -0.35 0.3302 -1.06 0.292 -1.004 0.304 PFC*Stock -0.2545 0.3091 -0.82 0.412 -0.867 0.36 * R-squared: 0.0616 ** Number of observations: 121

(34)

34

The next country which is tested individually for a structural breaking point is the U.K. This country differs from the other countries, as it is the only country which is not part of the original Euro zone. As can be seen from Table 7, the two return series are originally

integrated with each other, as the coefficient of the variable on stock returns is statistically different from zero. However, although the U.K. is not part of the Euro zone, the results of the time series regression model (Table 8) are in line with the tests for the structural breaking point of the Netherlands (Appendix C), France (Appendix D) and Germany (Appendix E). These tests show that for none of these countries the financial crisis can be marked as a structural breaking point in the relationship between EREIT returns and stock returns. However, also for this regression none of the chosen variables proves to be significant. Therefore, these results should be interpreted with certain caution, as this model most likely suffers from a form of multicollinearity.

Table 7: Simple regression of EREIT returns on stock returns for the U.K. for the period March 2006 – July 2007

Variable Coefficient Std. Error t-value P-value 95% Confidence Interval

α -0.0049 0.0049 -1.01 0.317 -0.146 0.005

Stock 0.4933 0.1191 4.14 0.000 0.258 0.729

* R-squared: 0.126

** Number of observations: 121

Table 8: Time series regression model, for the U.K. during the period March 2006-March 2016

α 0.0335 0.0457 0.73 0.464 -0.057 0.124 Stock 0.2163 0.5299 0.41 0.684 -0.834 1.267 Inflation 0.8387 1.4053 0.60 0.552 -1.946 3.623 GDP 2.2683 3.0886 0.73 0.464 -3.852 8.389 Interest rate -0.0003 0.0061 -0.06 0.956 -0.012 0.012 LT-Bond rate -0.0073 0.0085 -0.86 0.390 -0.024 0.009 FC -0.0297 0.0199 -1.49 0.138 -0.069 0.01 PFC -0.0112 0.0311 -0.36 0.719 -0.073 0.05 FC*Stock 0.4241 0.5668 0.75 0.456 -0.699 1.547 PFC*Stock 0.1170 0.5513 0.21 0.832 -0.976 1.209 * R-squared: 0.2167 ** Number of observations: 121

(35)

35

6. Robustness tests

Comparing these results with the earlier study by Basse (2009), yields the insight that the financial crisis was a structural breaking point for the relationship between EREIT returns and stock returns in the U.S., while this was not the case for Europe. A study by Basse (2009) finds that the relationship between the two return series changed in that investing in REITs became more risky compared to investing in utility stocks. This was shown by a dramatic increase in the market return beta when only the time period of the actual financial crisis was considered. To see if these results also apply for Europe, the same methodology as in the study by Basse (2009) will be employed in this robustness section. If the outcome of these tests are in line with the earlier findings by Basse (2009) it can be said that there is an actual change in the relationship between the two return series, although they remain integrated with each other.

First, it should be determined if both time series of EREIT returns and stock returns are cointegrated with each other. This can be done by looking at the existence of unit roots in both time series. Now, Europe is investigated as a whole instead of per country, which changes the dataset into time series data. This means that for both return series the arithmetic mean is computed by adding the returns per country and dividing them by five. These regressions make use of the natural logarithms of both the dependent and the independent variable. As natural logarithms can only be taken from positive numbers, now the absolute index values are used in the regression, instead of both total return series. To determine the existence of a unit root, a Phillips-Perron-test (Phillips and Perron, 1988) should be employed as this test is robust in the presence of heteroscedasticity and serial correlation. For determining this stationarity of both variables, the following regression equation will be used:

ΔPt = α + βPt-1 +∑ Yi ΔPt-I + λt + εt (1)

Where P stands for the natural logarithm of both variables, t stands for the time trend and α, β, Yi, λ and ε are parameters that need to be estimated by the model. When β is not statistically different from zero, a unit root exists and the data is not stationary.Also, the first difference of P (named R) is tested for stationarity, by the following regression:

The relationship between European real estate returns and European stock returns in times of the financial crisis : a structural breaking point analysis