Empirical evidence on corporate real estate ownership, firm investment and stock performance

(1)

Empirical Evidence on Corporate Real Estate

Ownership, Firm Investment and Stock Performance

Abstract This thesis examines the relationship between corporate real estate and corporate investment in a sample of firms from nine different European countries. Empirical tests have led to the conclusion that firms are more likely to invest less when they have a higher ratio of corporate real estate over total assets. Also, an increase in risk on the real estate markets is associated with a decrease in corporate investment. Literature regarding “sale and lease-back of corporate real estate” shows in general a negative effect of corporate real estate ownership on firm’s performance. In this thesis a negative relation between corporate real estate and firm’s stock performance is found as well. This suggests that there could be a relation between both negative effects of corporate real estate. Student: R. Dorré (11293063) Supervisor: Prof. Dr. M. Dröes Master Thesis: Corporate Finance and Real Estate Finance Date: 15-08-2017

(2)

2 | P a g e

Statement of originality

I, Robin Dorré, declare that this thesis, submitted as part of the requirements for the award of Master in Science at the school of Finance and Real Estate Finance at the University of Amsterdam, is my own work. I hereby declare that, my thesis is not interfering with anyone’s copyright or violates the intellectual property rights of other people or institutions. Work that is used as reference is fully acknowledged according to the needed reference practices. This work has not been submitted for a degree or diploma in any University yet.

Acknowledgements

In this small paragraph of acknowledgements, I would like to thank the people who supported me. In the first place I would like to thank my parents for always supporting me in doing what I think is important. Then I would like to thank my supervisor Dhr. Dr. Dröes for assisting me in writing this thesis and overcoming obstacles which came along when writing this thesis. Also, I would like to thank my fellow students Quirijn Otten, who helped me with some of the econometric parts of this thesis, and Lotte Warendorf for taking the time to check this thesis for grammar mistakes. I would like to thank the University of Amsterdam for facilitating a good working environment and a well-structured education in Finance. At last I would like to thank the ASRE (Amsterdam School of Real

(3)

3 | P a g e

1. Introduction

A large component within all capital markets is assigned to real estate. Of all the real estate in the world, a large share consists of corporate real estate (CRE). CRE consists of buildings, land and other properties, which can be held by a company itself or by an (institutional) investor who rents it to a company (Brounen and Eichholtz, 2005). CRE is seen as a major factor of production. Together with labor and capital it represents the three biggest factors of production. To maintain their daily business activities, companies need to expand and maintain their CRE portfolios. For these reasons it is reasonable to conclude that CRE is of great importance for corporations and directly affects the whole economy. A lot of research on the effects of CRE on companies has already been done and is widespread documented. Apart from the CRE ownership itself, the relation between CRE ownership and corporate performance has been researched extensively (Liow et al. 1999; Liow 2004; Chaney et al., 2012; Deng et al., 2017). Performance can be tested by various variables. The first papers focus on the relationship between CRE and firm performance, mainly examined operating income and subsequently on stock performance as well. Later, this was extended by financing parameters like debt financing and leverage. The most recent literature on the relationship of CRE and firm decisions focuses on the parameter investment (Deng et al., 2017). In their paper Deng et al. (2017) introduce a new independent variable, namely real estate risk and the effect on corporate investment. Here the standard deviation of the FAREIT index of US real estate stocks is used as volatility measure, which represents risk. In this thesis I will use this measure as well, but then for European stocks and countries. Conducting a research like this is difficult since there are in general no corporate real estate indices for individual European countries.

Deng et al, (2017) show a negative relationship for firms with real estate holdings, between real estate risk exposure and corporate investment. In his research they find that the same holds for long-term external financing. This implies that owning real estate and simultaneously bearing the risk on this real estate position, is commonly a bad characteristic of a firm. Although, researching the reason for these results that comply with the results of Deng et al. (2017), this is opposed as further research by them, will fall outside the scope of this research. It is very well interesting to conduct the research for the European market and evaluate the consequences. One of these consequences might be that stock returns of firms with high exposure to real estate will perform differently than their peers within the same industry. Brounen and Eichholtz (2005) tested this phenomenon. They found a significant effect for real estate holdings to stock returns. In this paper I will also research the CRE ownership effect on stockholder returns.

Since financing constraints and lending by means of collateral have other characteristics and restrictions in Europe, it will be relevant to literature to conduct this research for the European

(5)

5 | P a g e

market. Also it is found that companies that have their origin in a European country, have a higher ratio of CRE than firms based in the US, especially for non-real estate firms (Laposa and Charlton, 2001). Laposa and Charlton (2001) also find that European firms are leaning more towards owning CRE than their peers in the US. Owning real estate seems to be more inherited in the European corporate culture. The relationship between corporate investment and CRE ownership has not been tested yet for the European market. Subsequently I will test on the same sample the effect of CRE and corporate investment on stock performance of a firm. Therefore, the main research question for this thesis is: “What is the effect of corporate real estate ownership on firm investment and stock performance in the European market?” Because literature differs among their findings about the advantages and disadvantages regarding CRE ownership, often is suggested that there has to be done more research at the explaining factors for this over-/underperformance. To expose the effect of CRE ownership on firm investment, in an extensive way, this thesis will partly fill this void. Several hypotheses will be tested. The general objective for these hypotheses is to find the relationship between CRE and firm investment. In order to do this, I will use a corporate real estate ratio as proxy. Secondly, I will add a variable representing the nominal value of corporate real estate holdings. Additionally, I will add a proxy for real estate risk. The variable investment will be measured by taking capital expenditure and divide it by the accounting variable Property, Plant and Equipment (PPE). One of the objectives is to test this effect for different industries, like Brounen and Eichholtz (2005) did in their research on the relation between CRE and firm returns. They find that the stock returns of firms diverge among industries and within these industries they find a significant negative relation of stock outperformance and CRE ownership. This might be due to fewer investments by the effected firms. One hypothesis that can be abbreviated from this, is that corporate investment is negatively affected by CRE ownership. Therefore, I will parallel to the research of Brounen and Eichholtz (2005), do the same for CRE ownership on investments and I will also control for industries. The same results are expected for my empirical tests, which is: a negative relationship that is more significant in less intense real estate industries.

To test my hypotheses, I obtained panel data on 6709 firms in 9 European countries which resulted in 81.036 observations. The empirical tests are constructed in a way where I first come up with a measure for real estate risk. This is done by following the model of Deng et al. (2017). Here they measure the fluctuation within time-series variation of the corporate real estate markets which

(6)

6 | P a g e

accounts for the volatility of corporate real estate markets. This measure is needed to test the hypothesis where I expect corporate real estate risk to have a negative relation with corporate investment. Subsequently, I conduct research on the relationship between corporate real estate value and the proportion of CRE with corporate investment. Both variables are controlled for the real estate risk measure as described earlier. This is to test the hypothesis where I expect CRE to have a negative relationship with corporate investment. At last I will test the effect of CRE on stock performance, if any. Here I expect CRE to have an influence on the stock performance which can vary across industries (Brounen and Eichholtz, 2005). This is done by a two stage least square test, where in the first step the beta of a firm’s stock price is calculated and in the second stage this beta is estimated with real estate variables. Here I expect that a firm’s beta is significantly influenced by its corporate real estate positions. In general, the results support the hypotheses. I provide complementary evidence that CRE ownership negatively affects corporate investment. Also, I provide empirical evidence for a negative effect of real estate risk on corporate investment. In other words, the results provide evidence to assume that higher proportion of CRE ownership leads to a decrease in corporate investment. Subsequently, the results support the hypothesis that stock performance is affected by corporate real estate, however this effect partly diminishes when controlling for firm fixed effects. The next sections are structured as followed. In section 2 I will discuss relevant previous literature. In Section 3 I will introduce the data for my analysis. Here I will extensively discuss how the different datasets are obtained and how several key variables are created. In Section 4 I will explain the model that will be used to test the hypotheses. Section 5 presents the results from which I can conclude if the results provide evidence for the hypotheses. In Section 6 I will shortly comment on the findings and summarize the results in a compact general conclusion which links the hypotheses to the results. At last, in Section 7 the limitations and suggestions for further research are discussed.

2. Review of prior literature

Many definitions for the term “corporate real estate” are used in literature. The core of all these definitions can be summarized as that CRE describes all real estate that is used as industrial space, office space and retail space. This space is used for business activities and is part of the production process of corporations that are not mainly operating in the real estate market. The effects of CRE ownership have been researched extensively. One of the reasons why this is such a prevalent factor in empirical research is because estimates show that more than 25% of the assets of a company consist of CRE. Also, it is found that only human resources and loan expenses are of higher costs for firms than the maintenance costs of their CRE (Rodriquez and Sirmans, 1996). From here we can conclude that it is very important for companies to know how to optimally manage these costs and

(7)

7 | P a g e

investments concerning CRE. This is why a lot of research is conducted in this field of corporate real estate. For European countries, CRE seems to be a more common investment. However, the management of their property is not an occupation where these companies seem to invest a lot of time in (Nappi-Choulet, 2003).

2.1 Corporate real estate management

Research on the effects of CRE management originates from the late eighties and begin nineties and this was mostly done in the US. The first researches to write about CRE management where for example Veale (1989), Nourse et al. (1993) and Kimbler et al. (1993). Here, Nourse et al. (1993) find that managers of firms in the US were not aware of the importance of a good management strategy for the CRE of the firm where they were working for. Also, Nouse et al al. (1993) find that a clear strategy concerning CRE was ignored and that in general the management of CRE was not included in companies’ main strategies. Ambrose (1990), performed an empirical research on this same matter and found that because firms were not paying attention at their CRE, they could not implement the value of their CRE into their strategy. This value was recognized by corporate raiders and therefore firms with high values of real estate became more attractive as take-over target. After the 90’s, empirical research focused more on the relation between CRE ownership and the performance of the firm. Deng and Gyourko (2000) suggest in their research that ownership of CRE results in a discrepancy between the main business of a firm. This leads to situations where firms are not capable of taking on beneficial projects because they already invested their money in long-term CRE properties. From their research and results, they conclude that rather risky non-real estate firms with a relative large CRE portfolio should be supported to sell at least some of the CRE assets that they are holding. This because from their research, Deng and Gyourko (2000) find that there is a return penalty associated with holding large proportions of CRE. The main finding in this research of Deng and Gyourko (2000) is that they found firms which sold part of their real estate performed better than their peers who held on to all their real estate. Linneman (1998), also finds in his empirical research that holding large proportions of CRE has a negative effect on non-real estate firms. He concludes that it is not favorable for firms to invest high cost capital outside their core business and main competences. He suggests that firms in general should avoid investing in relatively low return buildings when they have a high cost of borrowing capital themselves. This can be interpreted as when a firm has a high risk business and therefore has a high cost of capital, they should not diversify their business with the ownership of real estate because it diminishes their risk pay off. In the late nineties the CRE market recovered. The main reason for this recovery was the increase in amount of real estate investment firms, which step in to the opportunity of bad CRE management by the firms itself (Nappi-Choulet et al., 2009). This development made it more

(8)

8 | P a g e

attractive for non-real estate firms to outsource the ownership of CRE. This increase of the CRE investment firms and thereby the increase in the European institutional real estate market also resulted in a higher interest in the issue of CRE ownership. Following DTZ, companies have rediscovered their CRE ownership holdings in this era. This led to a higher focus on CRE management by the firms that still own high proportions of real estate. DTZ estimates that around 70% of all the European firms still own their real estate. This is in high contrast with the ratio of US firms that own real estate, this percentage is estimated at a 30% owner-occupation rate. This is in line with the results of the research of Laposa (2001), as mentioned before.

2.2 Corporate real estate and investment

Deng et al. (2017) examined in their paper what effect real estate risk has on corporate investment and financing decisions of firms. They state in their paper that in general real estate risk is associated with a negative effect on investment and potential debt raising projects. The findings of this paper, which are explained before, are contradictory to previous research where real estate value was documented to have positive effects. Two of these researches about corporate real estate value on firm investment are done by Chaney et al. (2012) and Gan (2007). Chaney et al. (2012) argue that the value of CRE functions as collateral. This indicates that an increase in corporate real estate value results in an increase in collateral and thus less financing constraints for investments. Chaney et al. (2012) find that for every 1$ increase in corporate real estate value, thus collateral, a US corporation investment increases by 0.06$. Gan (2007) researched is his paper the consequences of the land and market burst of Japan in 1990. The effect of the burst was mainly noticeable across firms with larger land holdings before the economic set back. In his research Gan (2007) is profoundly interested in the consequences of the burst on investment behavior and allocation. He, as well as Chaney (2012) finds that for firms that operate in the manufacturing business, collateral is from higher importance than for firms operating in other industries. Following the results and conclusions from Gan (2007), corporate real estate functions as collateral channel. Therefore, when shifts in real estate markets happen, this directly influences investments and therefore on the economy as a whole.

2.3 Corporate real estate and firm performance

Where literature before the year 2000 was predominantly positive about CRE, later literature differs in its conclusion about the associated relationship between CRE ownership and corporate performance. Liow (2004) and Deng and Gyourko et al., (2000), claim that firms with a high level of CRE holdings have lower returns and higher risks that lead to a relative lower performance. On the other hand, papers discuss the benefits of owning real estate as potential collateral. As mentioned before, Chaney et al, (2012) find that for every 1 dollar of real estate holding, a firm invests 0.06

(9)

9 | P a g e

dollar extra. This is in line with the findings of Gan (2007), who found that a shock in collateral influences firm’s investment for the market of Japan. Also, Ambrose (1990) explored the effects of CRE ownership. He found that firms did not consequently report on the performance and development of its CRE portfolios. This seemed to indicate that firms were not managing their CRE portfolios actively. In this period of the late eighties, corporate raiders came into the market. They noticed the inefficiency regarding the management and exploitation of CRE. From this perspective the corporate raiders were the only actors interested in this gap of inefficiency. They reacted on the fact that there were hidden values which could be exploited. This development resulted in a consolidation wave, because all firms with inefficiently managed CRE portfolios became a takeover candidate. However, this research was done with data until 2000, this clearly connects to the results and suggestions of later research. This literature states that the corporate real estate ratio is decreasing over time. They suggest that a possible explanation could be that before the year 2000 there were few options to lease corporate real estate (Brounen and Eichholtz, (2005); Deng and Gyourko, (2000)). This last finding is in contrast with the findings of Redman and Tanner (1989), who found that almost half of the firms were leasing their CRE and this paper is from well before the Brounen and Eichholtz (2005) paper. Though, from the 2000’s on, literature finds a high increase of commercial real estate equity funds. This could be an indicator that firms apparently rather lease than own their real estate and leasing or sale and lease-back become more popular. Or better said, the benefits became clearer. Deng and Gyourko (2000) suggest several reasons which could explain this trend. The main finding is that companies in the past neglected the influence of CRE on corporate performance and decision-making.

In the paper of Brounen and Eichholtz (2005) they examined the relationship between CRE ownership and stock performance for firms of which their primary business is not real estate related. The data they used was obtained by collecting balance sheet information. From there they conducted CRE ownership ratios for firms in nine different countries. One of the main findings of their paper is that CRE ownership ratios differ across industries. Also they found that over time, within their sample, the general CRE ownership ratios faced a significant decrease. Namely, for the year 1992 they found an average ratio of 0.34 and in 2000 this same ratio had an average of 0.29. This ratio is the nominal value of PPE over total assets. Following the suggestions of the authors, this could be due to the increasing leasing alternatives. Furthermore, they find different results among industries between the relationship of CRE ownership and corporate stock performance. For this last effect they suggest that this can be explained by the different strategies within industries. For example, an industry intense sector exploits more real estate than a service based sector and those

(10)

10 | P a g e

companies should adjust their strategies to it. They finally conclude that more research should be done about the effect of different industries on corporate performance.

Also Liow (2004) conducted empirical research on the relation between CRE ownership and firms’ stock performance. However, Liow (2004) did this for the Asian and more particular Singapore market. In his research he finds that in general CRE ownership is associated with a negative effect on stock performance. The tests he performed provide evidence for a negative relation between CRE ownership and firm returns, risk, systematic risk and abnormal return performance. This negative effect of CRE ownership on performance remained constant for all non-real estate firms in seven industries and different portfolios. The main conclusion of this research is that CRE ownership is not associated with advantages for non-real estate firms in terms of risk-return reward. Following Liow (2004), it is reasonable to assume that these non-real estate companies own CRE for another reason. Liow (2004) suggest that these reasons could include institutional, financial and cultural factors. Therefore, a cross-country analysis is opposed. This raises a discussion whet ether firms rather lease than own their real estate. About this matter there is extensive literature. Although this is outside the scope of this paper, it is part of the whole picture in which CRE influences corporate investment and performance.

2.4 Corporate real estate ownership or - lease dilemma

As discussed in previous paragraphs, owning CRE can lead to negative effects on investment, performance and other factors. When this is true, firms should benefit from selling their CRE. Although this seems quite easy, this is not always the case. For instance, the real estate market is in essence an illiquid market. When a company wants to sell his CRE, but there are no potential buyers for the right price, the firm faces a risk of selling for a lower price and puts itself at risk on losing value. In a situation like this, the possibility of sale-and-lease-back can be of help. As mentioned at the end of the previous paragraph, this raises a question whether to own or lease real estate property. Therefore, this dilemma is widely discussed and should be included in the literature review concerning this thesis.

As stated in paragraph 2.2, Redman and Tanner (1989) find in their sample within their research that there are more companies that lease than own their CRE in general, although this is an opposite finding to other literature (Laposa and Charlton, 2001). The proportion of leasing in the CRE market, makes it an interesting part of corporate real estate to gain knowledge about. In the same research, Redman and Tanner (1989) try to find reasons for the disposal or leasing of firms’ CRE. Another part of the leasing sector within CRE is sale-and-lease-back (SLB). Slovin et al. (1990) explored the effect of SLB on stock performance. By conducting an event study they find that SLB events create significant abnormal returns. Also, they find that this does not hold for other types of external

(11)

11 | P a g e

leasing. The firms who apply a SLB mechanism benefit from a reduction in future taxes. A similar research has been conducted by Grönlund, Louko and Vaihekoski (2008). They state that European firms increased their real estate disposals during the last decennia. In this study they find evidence that supports their hypothesis. In this hypothesis they state that the positive effect on stock performance of a firm that announces a sale and leaseback, is a consequence of the revealing hidden value of the assets of the company. In a comparable research of Rutherford (1990) the same effects are documented. He also finds that disposal of CRE is associated with positive abnormal returns for the selling firm. Subsequently, he finds even more significant results than the ones found by Slovin et al. (1990). Complementary to these studies, there is a study of Owers and Rogers (1986), who studied the spinoffs of CRE in general. Also in this paper, spinoffs of CRE are associated with positive abnormal returns. From the previous described literature, we can abbreviate the conclusion that a disposal, spinoff or SLB transaction is in general associated with a positive reaction on the stock market. This is completely in line with the findings of a decline in average CRE holdings by non-real estate firms (Brounen and Eichholtz, 2005; Linneman, 1998). Adams and Clarke (1996) study the effect of SLB for the UK stock markets. In their research they find results which do not comply with the ones of Slovin et al. (1990) and Rutherford (1990) for the US market. They find that the share price of the lessee is decreasing, when a firm chooses to make use of a SLB construction. The reason for this negative reaction which is suggested by the authors of this paper is that when a firm announces a sell-off or disposal, this is associated with financial distress and thus an action to generate cash. A reason why the US and UK market react differently is not opposed. Another research that is conducted regarding SLB transactions, is the paper of Fisher (2004). In his research the results also showed a positive abnormal return after a SLB transaction. However, in his research he controls for the length of the leases. For the empirical study he divides the lease terms into two groups. One group has a lease term of 15 years or shorter and the other group has a lease term of more than 15 years. The results show a significant positive effect for the short term leases and significant negative effect for the long term leases.

Generally, literature states that SLB transactions positively affect firm performance or their stock price. This again, would be in line with the literature which states that there is a decline in CRE ownership over the past decennia. Before SLB transactions became popular, these constructions were mostly used by distressed firms. Therefore, it still has an image where stockowners might suspect a bad financial situation for a firm who decides to dispose their CRE. This is also why SLB still might be undervalued by the market. This is in line with the suggestions of Adams and Clarke (1996). The possibilities concerning leasing and owning CRE and the implications that come from these decisions, show once again the importance of CRE and the management and strategy around this.

(12)

12 | P a g e

Drawing conclusions from the prior literature, there is predominantly more literature, especially in the last two decades which provide evidence against CRE ownership holdings. Most literature find that CRE negatively influences stock performance. Also the SLB literature shows that there is a positive relation between the disposal of real estate and stock return. However, from Chaney et al. (2012) we can see an increase in the amount of money invested by a firm, which should in general be viewed as positive. It is worth mentioning that almost all literature is focused on the US markets and that other outcomes for European markets are expected because of the different financial constraints and the different levels of CRE ownership (Laposa and Charlton, 2001).

3. Data

To collect the data for this thesis I use three different databases. I start with the database COMPUSTAT; from which I collect accounting data on European firms. This data is obtained via the Wharton Research Database Service (WRDS). The second database is DataStream from which I collect monthly closing prices and returns of the share price of the firms which were selected in COMPUSTAT. DataStream is an add-in on excel which provides the requested data by entering a list of firm codes (ISIN, CUSIP, Ticker, SEDOL etc.) and manually selecting the variables that are required. At last, I collect data on each real estate index per country which is needed for estimating the risk factor of CRE. This data is not available in the databases that are commonly used like Wharton Research Data Services, Bloomberg, Reuters or Thompson one. Therefore, I use the country unique dataset of the European Public Real Estate Association (EPRA). They have composed indices of European real estate firms that are securitized. Since there have only been REIT’s to test for CRE risk, the EPRA has come up with another benchmark for the real estate sector. The FTSE NAREIT is a common used index for the US real estate markets, but for European countries there is not such an index. The EPRA FTSE NAREIT indices are constructed of European institutional real estate firms. To be included in the EPRA FTSE NAREIT index, a company must derive a specified percentage of their revenue from activities in the real estate business. Official REIT’s in the US have to oblige to similar rules. Essentially, for companies to be included in the EPRA index, they have to oblige to four elemental ground rules (source: ground rules EPRA1_{). The result of this composition is a country} specific real estate index which is similar to the FTSE NAREIT index2. This index is used for empirical

1_{“The Free float market capitalization has to be at least 50 million, traded volume should be a minimum of 25}

million over three months annualized period, firms have to derive at least 75% EBITDA from relevant real estate related activities and must produce a set of annual accounts in English” (www.epra.com).

2_{The FTSE NAREIT ALL REITs index is an index that includes all real estate funds that are tax qualified. The}

index is capitalization-weighted and all REITs that fit the criteria and are listed on NYSE, the NASDAQ or the ASE are included in the composition of the index. However, these are only US firms and the index is the benchmark for the US real estate (www.reit.com).

(13)

13 | P a g e

research on the US market in prior literature (Deng, 2017). The European compositions of the EPRA’s FTSE NAREIT indices list the daily returns of only nine countries. These countries are: The United Kingdom, Belgium, the Netherlands, Switzerland, France, Sweden, Germany, Italy and Spain. These will be the countries where my research is aimed at. Databases of EPRA are only available from 1990 onwards. Indices of these countries composed by ERPA are backwards composed on historical data from 2000 onwards. Following Deng et al. (2017), I deploy the FTSE NAREIT composites to measure CRE property returns. Funke et al., (2010) state that there are reasons for both including REITs and excluding REITs from empirical tests. The main reason for excluding REITs, following Funke et al., (2010), is that by excluding REITs from the market portfolio real estate risk is isolated from the market risk. This is used in most asset pricing tests. However, by including REITs returns, there is less chance of information asymmetry. Since CRE risk is an important factor of my empirical study, in where I need to isolate the market risk from real estate risk, I choose to exclude the REITs and isolated them in the EPRA’s FTSE NAREIT indices. Therefore, REIT returns measure real estate market performance in an accurate way, this because the measure of REIT returns is not influenced by other investments, but sorely on real estate gains.

Concerning the accounting data from COMPUSTAT it was unfortunately, by law, not obligatory anymore to report Property, Plant and Equipment (PPE) in the form of the separate variables Building, Land and improvement, Machinery and Construction. Therefore, from 1995 onwards, firms just report PPE. This results in some kind of a measurement issue, so from here on it is not possible to precisely test what the amount of real estate (building) is on a firm’s balance sheet. Following prior literature of Brounen and Eichholtz (2005), I use the PPE as parameter for measuring the value of CRE. Hereby, Brounen and Eichholtz assume that the share of building within PPE was not fluctuating significantly to make it a bad measure. Since I want this research to be as extensive as possible, the full length of the period 1995-2016 is chosen. Also, I do not want to exclude firms for which I do not have complete data because this would create a kind of survivor bias. Therefore, I will use firms which are not listed anymore but have historical data available, as well.

3.1 Accounting data

For the first selection in COMPUSTAT I will select all firms for the nine mentioned European countries without missing total assets (COMPUSTAT No. 6). Then I will exclude all financial firms and REIT’s by SIC. REIT’s are excluded because I focus on the effect of real estate ownership for non-real estate firms. Financial firms are excluded because their business in terms of credit supply for mortgages etc. has too much conflict and correlation with CRE. Also firms who fail to have consecutive data of every year or have missing data on total assets are excluded. Last requirement

(14)

for firms is to have at least three years of data. This provides 6.709 firms and a total of 81.036 firm-14 | P a g e

year observations over the period 1995-2016. I assume that real estate assets are located in the country where them headquarter is located. This assumption has to be done because firms do not list their real estate assets/PPE per region or exact location. This assumption is common in prior literature (Brounen and Eichholtz, 20015; Chaney et al., 2012).

Table 1. Distribution of firms and observations across countries for the full sample. The UK accounts for the

largest share of the sample and Belgium for the smallest.

Country Firms Observations %

Belgium 163 2.164 2,67% Germany 1.017 13.149 16,23% Great-Britain 2.643 29.500 36,40% France 1.027 12.905 15,93% Italy 383 4.741 5,85% Netherlands 266 3.245 4,00% Sweden 728 8.543 10,54% Spain 204 2.655 3,28% Swiss 278 4.134 5,10% Total 6709 81.036 100% Next, I will collect data on the value of real estate assets of each firm which meets the mentioned criteria. For this variable I will use the book value of the COMPUSTAT variable PPE which is a composition of the value of building, land and improvements and construction in progress of a firm. Furthermore, the real estate ratio is used, which will give an interpretation of the proportion of real estate holdings. This real estate ratio is constructed by dividing the book value of PPE by total assets and will be seen as the CRE weight of a firm, from now on CRER. As stated before, this is the same measure as is used by Brounen and Eichholtz (2005), as well as by other papers. !"#" = &&# '()*+ ,--.)-

(15)

15 | P a g e

As specified in the methodology section, I will first test the CRER on investment. Next I will add the nominal value of CRE. This is represented by the book values of the variable PPE. I expect the CRER to be a more reliable measure. I do this because of the simple reason that in the CRER variable I control for size. Graph 1. Average percentage of proportion corporate real estate ownership across the full sample. Graph 1 shows the average CRER for all countries and industries within the full sample. Completely in line with prior literature, it is clearly visible in the graph that the CRER is decreasing over time. It is interesting to note that from 2005 onwards the decline stagnates. The last paper to report on the decline in CRE holdings is dated from 2004, by Brounen and Eichholtz. As mentioned before, the general suggestion in literature is that this effect is caused by the increasing popularity of leasing CRE. The dependent variable Investment is measured by using Capital Expenditure (COMPUSTAT No. 128) over the PPE (COMPUSTAT No. 8) lagged by one year.

/01.-)2.0) =

!*3.4

&&#

₅₆₇

In all regressions I will use control variables for the potential heterogeneity among firms from the sample. These control variables are Leverage, Log Cash, Profit and the interest rate3. The latter is

composed by taking the monthly quoted annual interest rate and dividing it by 12 to obtain the monthly rate. Subsequently, I will calculate the rolling average of the last 12 months to come up with the average interest rate. The other control variables cash, Leverage and profit are common 3_{Source: World Databank, long-term interest rate for European region.} 0 0,05 0,1 0,15 0,2 0,25 0,3 0,35 0,4 1995 1997 1999 2001 2003 2005 2007 2009 2011 2013 2015

CRER

(16)

16 | P a g e

variables in empirically testing the relation between accounting variables. These control variables control for size, solvency and the cash position of firms within the sample. Cash is the sum of cash and short-term investments, leverage is the ratio of liabilities over total assets and profit is the ratio of operating income before depreciation over the total assets of a firm.

Table 2. The descriptive statistics of the key variables which are included in the regressions. The sample

consists of 6709 firms with a maximum total of 80.819 firm*year observations for the key variables. All variables are winsorized at 1 percent on both sides.

Variables Number Mean Median SD. Min Max

Investment % 68.349 47.76 21.25 85.8 0.01 499.11 CRER 80.819 23.09 16.28 22.3 0 85.61 Log(PPE) 78.863 2.63 3.08 3.26 -3.86 10.28 Leverage % 80.662 20.09 16.52 18.88 0 77.43 Log(Cash) 78.369 2.63 2.58 2.76 -3.73 9.04 Profit % 80.599 3.94 8.97 21.5 -87.22 34.27 Interest rate (%) 23 2.80 2.33 2.61 -0.1 11,44 Rem Volatility 207 5.03 4.5 2.21 1.58 17.89 Beta 64.374 0.77 0.69 0.66 -0.58 2.71

Because CRE has relatively high maintenance costs compared to other corporate assets, having a larger amount of corporate real estate will result in higher maintenance and construction costs. The financial resources spent on these adjustment costs will reduce the amount of cash available for investments. Therefore, the firm that has a higher value of real estate, will have higher adjustment and maintenance costs and less cash available for investments according to Gan (2007). Therefore, the cash resources of a firm is a good or even required control variable for performing tests of CRE on investment.

Table 3 shows the average CRER and investment per industry. As can be seen, there is a lot of differentiation across industries. When conducting tests on this data, this differentiation among industries is vital to control for in the methodology. The ratio of corporate real estate is highest for retail stores, construction and mining. This is the same result as reported by Brounen and Eichholtz (2005).

(17)

17 | P a g e

Table 3. Distribution of average CRE ownership ratio and investment. The average of the whole sample will differ from the average of the percentages below since some industries are represented by more firms than others. Brounen and Eichholtz (2005) found ratio’s which are similar to the results presented in this Table.

Industry CRER Investment

1 Food 33,7% 22,5% 2 Mining and Mineral 55,6% 79,7% 3 Oil and Petroleum Products 34,0% 71,5% 4 Textiles, Apparel & Footwear _21,4% _36,9% 5 Consumer Durables _24,4% _32,8% 6 Chemicals _30,8% _33,4% 7 Drugs, Soap, Perfumes, Tobacco _15,1% _45,6% 8 Construction and Construction Materials 45,7% 27,9% 9 Steel works 34,8% 19,3% 10 Fabricated Products 26,2% 27,5% 11 Machinery and Business Equipment 18,0% 42,9% 12 Automobiles _26,2% _28,6% 13 Transportation _38,2% _29,9% 14 Utilities _31,2% _18,9% 15 Retail Stores _35,3% _34,6% 16 Banks, Insurance Companies, and other Financials - - 17 Others 17,0% 61,3%

3.2 Stock and index data

After collecting the accounting data and computing the required measures, I will collect data on stock prices, real estate indices and market indices. As explained before, the real estate market index data is retrieved from the EPRA’s databases. This is data on the monthly returns for the selected countries within the index. This return data however is computed as daily return on a base value, which for most countries is 1000 at the beginning of the year 2000. From these return prices I will first computed the percentage monthly change. Additionally, I will collect stock and market data for the required countries and firms via DataStream. All the afore mentioned data is collected in one

(18)

18 | P a g e

excel file which is afterwards merged with the accounting data in Stata. This merge is done by matching the variables’ fiscal year and company ISIN.

In regression (1) of the methodology section, I construct a measure for the real estate market risk. With this data I am able to analyze the real estate risk over time. To come up with a visualization of both the volatility of the real estate market and the stock market, I will also compute the yearly standard deviation of the monthly returns of the stock market (market volatility) and I will include both volatility measures in one graph. This real estate market risk data is constructed the same way as it is done by Deng et al., (2017). Graph 2. Fluctuation of the average real estate market volatility and the stock market index volatility. As can been seen from the graph there was in 2008 for both markets a spike in volatility. Also the 2002 tech bubble is visible for the stock market, however apparently this did not influence the real estate markets in Europe. This differs from the data of Deng et al. (2017) where they constructed the same volatility measures for the real estate market and the stock market of the US. In their graph, the real estate market also had a peak in volatility during the tech bubble. This implies that both markets, the European and the US market, cannot be seen as identical. Also as expected, the real estate market is less volatile than the stock market which is indicated by the difference between the solid (stock market volatility) and dashed line (real estate market volatility). The dashed line permanently acts below the solid line.

For the estimation of the real estate market volatility I first obtained the real estate FTSE NAREIT returns from EPRA for each country as described earlier in this section. Subsequently, I subtracted 0,00 0,05 0,10 0,15 0,20 0,25 0,30 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 Market vola REM vola

(19)

19 | P a g e

the rolling average for the past 12 months of the monthly annual 10-year German bond rate4 (risk-free rate). The latter has also been done for the market return. This to get the excess return for the real estate market as well as for the stock market return. Subsequently, I will perform regression (1) from the methodology section. A measure of time-series risk. In this regression I follow the methodology of Deng et al., (2017) where I first calculate the standard deviation of the monthly residuals of the FTSE NAREIT real estate index return data. This is done for each individual year and constructed separately for each country. When all the data is obtained, it represents the real estate market volatility and this variable representing the data is named “REMkt vola” (Real Estate Market Volatility). This defines the real estate market risk. This variable is merged in the Stata file with the firm data. This data is merged by country name and fiscal year, to match each firm to the correct market risk per country and year. At last I calculate the beta of each firm of the sample. The data that is required for calculating the beta of each firm are the firms’ stock returns, index returns and the risk free rate. The firms individual stock price data is retrieved via DataStream. To obtain this data I computed a list of all ISIN codes from the Stata file that is containing the accounting data. The country specific market index returns are retrieved via DataStream as well.

4. Methodology

In this section I will explain the regressions which I perform on the data described in section 3. The results of these regressions will be analyzed in section 5 and will form the extensive answer to the main question. Namely, what the effect of CRE on firm investment is for European countries. Below I clustered the hypotheses which will be tested by the regressions that are explained subsequently in this section. Hypothesis I: Firm investment is higher when the value of CRE ownership is higher Hypothesis II: Firm investment is lower when the proportion of CRE ownership is higher Hypothesis III: Firm investment decreases when real estate risk increases Hypothesis IV: The effect of corporate real estate ownership on investment differs among industries Hypothesis V: Firm characteristics affect the relationship between CRE ownership and firm investment

Hypothesis VI: Stock or return performance of firms is affected by the level of CRE ownership

(20)

20 | P a g e

For the empirical tests of the above mentioned hypotheses, I will first come up with a measure of the risk for owning real estate. Therefore, I need to measure the fluctuation in the time-series variation of the real estate markets. According to Deng et al. (2017), I firstly relate the returns of each country’s real estate index (EPRA index) returns to the country’s market returns and subtract the risk-free rate for both:

"89,5− "< = =>+ @ABCD58E+ =7 "_FG,5− "< + H5 (1)

Where "89 is the return on the country specific EPRA index, "FG is the return on the country specific

market index. The two variables are measured in excess of "<, which is the 10-year German bond

and is defined as the risk-free rate. For this regression I used the data discussed in paragraph 3.2 of the data section. The real estate returns are here paired to the market returns of each country. The regression is a derivative of the Fama and French CAPM-model. Here the =7 is an estimate of the

influence of the market returns on real estate market returns. The monthly residuals in this regression are calculated at the hand of =7 and the standard deviation of these residuals represent the volatility of the real estate market. This will be a variable on its own, representing the corporate real estate market risk. This variable will be used in equation (4). Next, I will test the hypotheses for investment. Here /01 (investment is the CAPEX of a firm divided by PPE lagged by one year, !"#" is the proportion of CRE measured by PPE relative to total assets and the control variables are the ones described in section 3. As mentioned before, it is likely that the amount of real estate value is connected to the amount of maintenance costs. Since maintenance costs account for a high proportion of firm expenses, it is more reasonable to take the ratio of capital expenses over total assets. Here I test if the proportion of real estate has a significant influence on the dependent variable investment. I expect this relation to be negative. /01_I5 = =_>+ =₇ !"#"_I5+ =_J !(0)K(+-_I5+ L_I5 (2)

The same structure as equation (2) holds for the next test. However, in this regression I add the nominal value of real estate which is denoted by "#1*+M.. This is the value of PPE in millions of dollars and taking the logarithm of this number. The results from the linear regression will show if the nominal value of real estate has an effect on corporate investment. From literature of Chaney et al. (2012) I expect this relationship to be positive. As described in the literature review, Chaney et al. (2012) found that a “collateral channel” is the reason for this positive relationship. A similar test is conducted by Deng et al., (2017) as well.

(21)

21 | P a g e

/01I5 = =>+ =7 !"#"I5+ =J "#1*+M.I5+ =N !(0)K(+-I5+ LI5 (3)

The next regression is conducted to test the relationship between real estate risk and corporate investment. Risk is measured per country and year in regression (1). In this regression the systematic real estate market risk is defined by its volatility ("#OP) 1(+*5). Expected is that when risk is high, investment will be low. The proportion of real estate is important to control for the exposure to risk, therefore the CRER variable is included. /01I5 = => + =7 !"#"I5+ =N "#OP) 1(+*5+ =Q !(0)K(+-I5+ LI5 (4) Brounen and Eichholtz (2005) suggest that CRE ownership has different values across industries. To verify this, I made an overview of CRE holdings per industry in section 3. Brounen and Eichholtz find that effects from CRE holdings on performance differ among industries. From these findings I expect that the effect of CRE on investment also differs across industries and therefore I will control for these industry effects by using a dummy variable. Also the variable "#OP) 1(+* is included to control for risk and the CRER variable is used as measure the effect of CRE.

/01I5 = =I5+ =7 !"#"I5+ =J "#OP) 1(+*I5+ =N !(0)K(+-I5+ @IDRCS58E + LI5 (5)

The same test as regression (5) is used for regression (6). In regression (5) I expect to find results supporting the hypothesis that there are different results across industries. Therefore, I want to control in the last regression on investment for firm and year effects. This is done by a dummy for firm effects and a dummy for yearly effects. This test will control for all firm characteristics like industry, country, size, culture etc. Obviously, when controlling for all these aspects, this test should result in the most reliable statistics. Again, I expect to find a negative relation for CRER and "# OP) 1(+* with investment. The results of this regression will form the foundation of my conclusion.

/01_I5 = =_I5+ =₇ !"#"_I5 + =_J"#OP) 1(+*_I5 + =_N !(0)K(+-_I5+ @_<I8F+ @_E9T8+ L_I5 (6)

For the final test on the relationship between CRE and corporate investment, I will add the variable for the nominal value of real estate to regression (7). I will do this to verify consistency with current literature concerning CRE.

(22)

22 | P a g e

/01I5 = =I5+ =7 "# 1*+M.I5+ =J "#OP) 1(+*I5 + =N !(0)K(+-I5+ @<I8F+ @E9T8+ LI5 (7)

At last I will conduct a two-stage test to explore the effect of corporate real estate on firm stock performance. This test has also been done by Brounen and Eichholtz (2005). I will differentiate from their research by using another dataset and the inclusion of a new real estate market risk variable, where the latter is created in equation (1). I will combine these results with the results of the tests on investments to form a more extensive overview for the answer to the main research question and the coherent performance hypothesis. I will start with a beta estimator by means of the capital assets pricing model (CAPM) by Fama and French. Excel calculates each yearly individual beta for a firm automatically via the mentioned inputs of DataStream. By selecting the CAPM regression (8-I) within DataStream and manually including the required variables, DataStream determines a yearly beta for each firm. No results on the significance levels of this regression is documented, only the estimated = is used. The descriptive statistics of = can be found in Table 2. U)(VP".)MK0I5 = =>+ =I5 "2P5− "W5 + HI5 (8-I)

In equation (8-I) U)(VP".)MK0I5 represents the stock return for each firm. As in all regressions, i

represents the firm and t represents the time period for the sample. The stock return is calculated in excess over the risk-free rate (10-year German bond rate), "2P5 represents the return on the

country related stock market index, the beta of each firm is defined as =I5 and represents the

sensitivity of a stock to the movement of the related country market index return. => is the

intercept and can be interpreted as the excess stock return of firm i, to the country specific market return.

To test if investors are aware of the effect of CRE on investments and how they value this, I test the effect of CRE on the estimated stock beta from regression (1). Brounen and Eichholtz (2005) conducted a quite similar test but on a different sample of countries and period. However, they state in their methodology that firms with high levels of real estate should automatically have lower betas than their peers. This is assumed because real estate on itself has low systematic risk, which decreases the overall risk and subsequently the beta of firms with high ratios of CRE. However, Brounen and Eichholtz (2005) also suggest that firms with already a low beta, will have better facilities to gain capital because of their lower risks. For this reason, it is likely that those firms acquire more real estate anyway. Because of this, Brounen and Eichholtz (2005) suggest that low beta firms are more likely to have high ratios of CRE. This occurrence results in reverse causality and thus endogeneity. Therefore, it is suggested to use a two-stage least squared method. This is done

(23)

23 | P a g e by the following model, where the dependent variable is already estimated in regression (8-I) and the second stage is presented below: =I5 = X>+ X7!"#"I5+ XJ !(0)K(+-I5+ HI5 (8-II) =I5 = X>+ X7!"#"I5+ XJ"# OP) 1(+*I5+ XN !(0)K(+-I5+ HI5 (8-III)

=_I5 = X_>+ X₇!"#"_I5+ X_J"# OP) 1(+*_I5+ X_N !(0)K(+-_I5+ @_IDRCS58E+ H_I5 (8-IV) =_I5 = X_>+ X₇!"#"_I5+ X_J"# OP) 1(+*_I5+ X_N !(0)K(+-_I5+ @_<I8F+ @_E9T8+ H_I5 (8-V) The regressions relate the equity beta of a firm to the corresponding CRER, "# OP) 1(+*, control variables and include both firm, year and industry dummies. The latter is done to control for the different countries as well as different industries or other firm characteristics. First the = for each firm is obtained in equation 8-I. In the second stage, this beta is estimated with regressions containing real estate variables. If in the second stage regressions, the effect of real estate variables is significant. I assume that it affects the stock performance/returns of a firm. I will test the hypothesis that CRE has an effect on stock performance and I expect the relation between

REMvolatility to be positive. This because when market volatility increases it is expected that stock

prices become more volatile which results in a wider range of betas.

5. Empirical results

In this section I will report on the empirical results from the tests on investment and the two stage least square regression on stock returns. The results generally support the hypotheses clarified in the methodology section. The effect of CRE ownership on firm investment and stock performance is mostly in line with prior research. For all tests I conducted a vif-test to check for multicollinearity between variables. Here are no statistics than a vif-score of 105_{reported. Also all the variables are} tested on unit roots. All unit root tests resulted in no significant results to support the hypothesis for a unit roots, this implies that we can assume that the variables are stationary and the regressions are not spurious. Also, all tests are clustered on the standard errors of firms. This should account for serial correlation and heteroscedasticity which makes the tests more robust. The dependent variable

investment is reported as a ratio with a mean of 0.4776 or 47,76% (see Table 2).

5_{A statistic of 10 when conducting a vif-test is used as maximum acceptable value for a variable to be not}

(24)

24 | P a g e

5.1 Real estate and corporate investment

Table 4 represents an overview of the results from the effect of multiple CRE variables on the dependent variable corporate investment. In Column (1) the coefficients and significance level of the simplest regression are shown. This regression consists of the dependent variable and the control variables. The control variables explain 3.9% of corporate investment. Interesting remark is that, in previous literature the logarithm of assets is a common used control variable. However, after running the vif-test to test for multicollinearity, this variable showed a statistic of 45, which is way above 10. Because of this, I excluded assets and included the interest rate. Except for the variable interest rate, which is significant on a 5% level, all the control variables are significant on a 1% level in their effect on corporate investment. Column (2) shows the regression where the first real estate variable is included, namely the CRER (corporate real estate ratio). This variable has a negative coefficient, which implies that an increase in the proportion of real estate is associated with a decrease in corporate investments. This coefficient is significantly correlated with investment on a 1% level and is associated with a 0.00754 decrease in the ratio of corporate investment (ratio) for a 1%-unit increase in the CRER. In Column (3) I add the nominal value of CRE to the linear regression. As can been interpreted from Table 3, this results in a significant negative coefficient for REvalue which represents the logarithm of the nominal value of CRE in millions of dollars. However, this also results in no significance for the CRER variable. This is a logical result, since both variables represent the real estate value, only CRER is the ratio of CRE over total assets. Both measures seem to be significant on their own, but together seem to cause multicollinearity. In literature (Brounen and Eichholtz 2005; Chaney et al., 2012; and Deng et al. 2017) both measures are used. Since I control in the CRER for proportion of CRE relative to the size of the firm (total assets), I define this as a more reliable measure for CRE and will therefore mainly use this measure in the subsequent regressions. Also, from Column (3) I conclude that it’s not efficient to use both since they substitute each other. On the other hand is the "J_{of Column (3) higher than the "}J_{of Column (2), respectively 7.2% and}

10.7%. This implies that the equation with the nominal value of CRE added, (regression (3)) explains the dependent variable better, or has at least less variance in explaining the residuals. A higher "J

though, does not always imply a better regression. In this particular example I would add a non-significant variable to the regression which is likely to cause multicollinearity. In Column (4) the variable for systematic real estate market risk is added to the equation. The negative coefficient stays nearly the same for CRER as in Column (2). This indicates that there is no effect of REMvolatility in CRER. However, the REMvolatility is associated with a significant negative relation with investment with a coefficient of -0.353. This implies that a general increase in the risk of real estate of 1% is associated with a decrease in corporate investment of 0.00353. In their paper, Brounen and

(25)

25 | P a g e

Eichholtz (2005) suggest that the ratio of CRE ownership differs between industries, which is also confirmed by the data statistics in section 3. In regression (5) I control for this effect, of which the results are shown in Column (5). As can be seen from the dummy statistic in the appendix, there are some industries which turn out to be significant. This could indicate that the results of the first regressions are biased and that when controlling for industries, the results are significant different. The results of this test are shown in Column (5). When compared to Column (4) there are however, not the differences as expected. Still both the variables show a negative coefficient and both are significant at the same level as in Column (4). The standard error for CRER has increased and the coefficient stayed nearly the same. This means there is higher variance in the residuals when controlling for industry fixed effects. When controlled for industry effects, the real estate variables seem to be negatively correlated with corporate investment. This is more in line with the research of Liow (2004) empirically tested in his paper the relationship between the ratio of corporate real estate ownership and the parameter corporate performance. In his research he found this relationship to be negative for the whole sample and within all industries. Column (6) shows results from the same regression as in Column (5), but in this regression I controlled for firm and year fixed effects. Both the explanatory variables, CRER and REMvolatility, are still significantly negative associated with the dependent variable corporate investment. However, the coefficient of REMvolatility becomes significant on a 1% level. If I control for firm and year, the new coefficient of

REMvolatility indicates that a 1% increase in market risk leads to a 0.01224 decrease in investment

instead of the 0.00499 decrease in Column (5). This means that when controlling for firm specific characteristics as industry and country, the effect becomes of higher significance and more profound. The "J_{increases from 7.2% to 9.1% which indicates that the regression with a dummy for}

industry explains the residuals better. The CRER is affected the opposite way. It stays negative, but the coefficient becomes smaller and the standard error increases, which indicates that the variable becomes less significant in explaining corporate investment although it is still significant on a 1% level. Corporate investment is associated with a 0.00502 decrease when the CRER increases with a 1% unit.

In the last column (Column (7)) of Table 4, I again control for firm and year effects and add the variable REvalue. I do this to verify consistency with existing literature. All variables are significant on a 1% level. Also the variables REMvolatility and CRER become more significant when adding the

REvalue variable. Also, the coefficient of the latter becomes positive, which is in line with literature

(Chaney et al., 2012). This indicates that a one unit increase in the value of real estate measured in logarithm (in millions of dollars) is associated with an increase by 2.4% in corporate investments. Which is relatively comparable to results found in Chaney et al. (2012) and Deng et al. (2017). In

(26)

26 | P a g e

Column (7) the "J_{increases from 3.2% to 3.8%. Which again indicates that adding the REvalue to}

the equation is associated with a slightly better explanation of the residuals. It is a strange outcome that when controlling for industry effects the "J_{increases, but when controlling for year and firm}

fixed effects the "J_{decreases relatively to regressions without year fixed effects or industry fixed}

effects. These "J_{statistics are relatively low, however the variable coefficients are significant. This} means that the variables have a significant effect although the model is not explaining the residuals very well. Table 4. Results from tests on the panel data for 6709 firms with a maximum of 66474 observations for time series between 1995-2016 (unbalanced). In this Table investment is estimated by regressions (1) to (7). The variables of interest are CRER and REMvolatility and the independent variables are Interest rate, Profit,

Leverage and Investment. In Column 1,2,3 and 4 linear regressions are used to test for the effect of real estate variables on investment. In Column 5,6 and 7 cross-sectional regressions results are controlled for industry, firm and year effects respectively. All regressions are checked on robustness by using cluster analysis on firms. Statistic significance at the 1%, 5%, and 10% levels is respectively represented by ***, **, and *. (1) (2) (3) (4) (5) (6) (7) REvalue -0.117*** _0.024*** (0.005) (0.007) CRER -0.754*** -0.038 -0.756*** -0.772*** -0.316*** -0.502*** (0.025) (0.027) (0.025) (0.107) (0.065) (0.070) REMvolatility -0.353** _-0.499** _-1.224***_-1.314*** (0.152) (0.187) (0.154) (0.154) Log(cash) -0.024*** -0.025*** 0.069*** -0.025*** -0.022*** 0.034*** 0.020*** (0.001) (0.002) (0.004) (0.002) (0.002) (0.004) (0.005) Interest rate 0.005** 0.016*** 0.025*** 0.016*** 0.020*** 0.050*** 0.051*** (0.002) (0.002) (0.002) (0.002) (0.003) (0.002) (0.002) Profit(%) -0.474*** _-0.326*** _-0.092*** _-0.328*** _-0.252** _0.266*** _0.262*** (0.024) (0.031) (0.033) (0.031) (0.088) (0.045) (0.046) Leverage(%) -0.599*** -0.325*** -0.052* -0.324*** -0.265*** -0.221*** -0.247*** (0.019) (0.028) (0.029) (0.028) (0.046) (0.037) (0.038) Industry fixed effects No No No No Yes No No Firms fixed effects No No No No No Yes Yes Year fixed effects No No No No No Yes Yes N 66474 66474 66214 66434 66355 66434 66174 r2 0.039 0.072 0.107 0.072 0.091 0.032 0.038

Standard errors in parentheses *_{p < 0.10,}**_{p < 0.05,}***_{p < 0.01}

Empirical evidence on corporate real estate ownership, firm investment and stock performance