Economies of scale : the case of REITs

(1)

Master Thesis of Maximilian Alleze

University of Amsterdam

MSc. Real Estate Finance

T h e s i s S u p e r v i s o r : M a r c e l T h e e b e J u l y 2 0 1 5

Economies of Scale: The Case

of REITs

(2)

Statement of Originality

This thesis is written by Student Maximilian Alleze, 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

(3)

Abstract

This thesis serves the purpose of analyzing the effect of size on several profitability measures of US Equity REITs. The topic is very controversial discussed and needs to be updated due to the fact that no extensive research was made after 2005. This thesis also tests how the financial crisis in 2007/2008 has affected REITs in general and especially the influence of a REITs size on its profitability. It was found that the macroeconomic environment seems to be very important for the relationship between a REITs size and profitability. Especially the financial crisis had a large impact on this relationship. Additional control variables were introduced to control for omitted variable bias, but also to test their effect on REIT profitability. The dataset was contains around 15 years of quarterly observations of 69 US Equity REITs in the years of 2001 until 2015. Therefore more than a full business cycle as well as the financial crisis were observed and tested.

The result of the analysis is that REITs can exploit economies of scale, especially in the aftermath of the financial crisis, but at a declining (diminishing) rate. This support already existing findings on the subject, but adds the aspect of different macroeconomic situations to the topic, in this case especially the financial crisis in 2007/2008.

(4)

1. Introduction

1.1 Introduction to Topic and Research Question

The topic of scale economies in Real Estate Investment Trusts (REITs) has come up quite some time after the introduction of REITs as a legal construct in the US in 1960. It was not until the 1980s when research started to examine possible economies of scale of REITs. This is most likely due to the fact that REIT market capitalization was too low to experience any kind of scale economies. Even after more research was done on the topic, outcomes were still very controversial as REIT growth accelerated after a period of rather moderate growth in the 1970s and 1980s. In the 1990s market capitalization increased almost fifteen fold (although the total number of REITs also almost doubled) and now very large REITs could be examined. This growth continues up until today with some back steps during the financial crisis in 2007 and 2008. It is just natural to assume that the answer to the question of whether economies of scale are present in REITs has changed over the last 30-40 years of research. Therefore the topic is very controversial as some researchers find proof of scale economies and others do not depending on the time period of their data set, but also depending on their methodology, which is quite different in across all accessible research. Due to this ongoing controversy the topic does not loose its interest as new variables are being introduced and larger datasets with newer data of larger REITs becomes accessible.

The main question that all researchers of past articles tried to solve and which the author of this thesis tries to answer is: Do Real Estate Investment Trusts experience benefits due to economies of scale or disadvantages from possible diseconomies of scale? The intuition behind this question is that from economical theory there might be evidence for scale economies, for example the higher bargaining power that a larger REIT might have and therefore the lower cost the REIT is exposed to when acquiring real estate. Other factors might be lower management costs for larger REITs (per asset), lower legal costs et cetera.

Additionally a second hypothesis will be examined: The possibility of scale effects for REITs are influenced by the macroeconomic environment and subject to change due to game changers like the financial crisis in 2007/2008.

(6)

To prove the economical intuition behind the topic the author will apply a linear ordinary least squared (OLS) regression to examine whether size has an influence on funds from operations (FFO), FFO per share or the ratio of revenue over costs. The size will be measured in the natural logarithm of total assets or alternatively measured in asset growth as this is a proxy for the natural logarithm of total assets. Suitable control variables will be included to control for omitted variable bias as good as possible. This thesis will contribute to the existing literature by providing an update of the data with similar methodology as Ambrose et. al. (2005) did, as well by providing regressions with different independent variables as well as new introduced dependent variables. The thesis will therefore not only be an update but rather contribute to the existing literature by introducing new variables of interest as well as to try to answer the question of whether economies of scale apply to REITs in the changed economic environment after the financial crisis.

For every dependent variable there will be three different sets of independent variables which will be tested in different combinations of random effects (yes/no), clustered standard errors (yes/no), and fixed effects (yes/no) as well as with or without dummy variables.

1.2 Course of Investigation

First a literature review will be given to assess the current state of the research on this topic as well as to give a short overview on the history of research on the topic of economies of scale in REITs. Within the literature review a brief overview on the statistical approaches up until today will be given. At the end of this chapter hypotheses will be derived and then tested in the following chapters.

In the next part the focus will be put on the methodology to test the hypotheses. The equations for the regressions will be given and the statistical model will be introduced. Afterwards the dependent variables will be introduced and the reasons for choosing these specific variables will be described. A closer look on the different sets of control variables will also be done.

In the data and descriptive statistics part the data that was used for said regressions will be shown as well as the data sources. Various data was retrieved from the SNL database as REITs provide the needed data due to the fact that they are public listed and traded.

(7)

Summary statistics for all dependent and selected independent variables will be given which will include mean, median, standard deviation, minimum, maximum and the number of observations.

The following part will present the main results of the master thesis by giving selected result tables. The results will then be interpreted economically and put into context of the general research area of economies of scale in REITs.

Robustness checks will be following the result section as well as additional results. The thesis will conclude with a discussion of the results and possible meanings for existing research as well as REITs itself. Limiting issues of the methodology and data will be discussed in this chapter as well as the implications that the findings of this thesis might have. The thesis will end with possible directions for future research.

2. Literature Review

2.1 Economies of Scale for Real Estate Investment Trusts

When it comes to the question whether REITs experience any degree of scale economics, the literature on the subject is not agreed. Especially older literature with older data being used seems to find evidence for economies of scale whereas recent literature finds even diseconomies of scale.

In the 1980s many researchers tried to test indirectly for economies of scale. Allen and Sirmans examine the price effects of REITs after a merger and find positive price effects due to better asset utilization. This finding suggests possible scale economics (Allen & Sirmans, 1987). Other indirect approaches by studying the wealth effects of property acquisition by REITs do not show any signs of scale economies (Corgel, McIntosh, & Ott, 1995).

Bers and Springer were one of the first to apply direct approaches towards scale economies. Their direct approach started the research on efficient cost frontiers - which is a frontier where the optimal combinations of output (as high as possible) and input (as low as possible) are located - in the context of economies of scale. The rationale behind this approach is that if a firm operates on the efficient cost frontier then there are no more economies of scale to exploit, but if there is evidence that REITs do not operate on the efficient cost frontier then this might allow for possible economies of scale when

(8)

optimizing the input/output relationship. They used REIT data from the National Association of Real Estate Investment Trusts (NAREIT) to estimate a translog cost function. A translog cost function (Transcendental Logarithmic Cost Function) is a cost function with a flexible functional form, which permits partial elasticities of substitution between inputs to vary, i.e. labor, material or capital. It is usually used to estimate average costs, which is then a sign for possible economies of scale (if average costs decrease with an increase in output) and widely used in the literature regarding economies of scale. Bers and Springer then examine the elasticity of the average costs with respect to output to determine economies of scale (Bers & Springer, 1998). But according to Anderson et al. this statistical approach lagged inefficiency estimates and their traditional regression-based technique has just a single error-term. By implementing just one error term Bers and Springer assume that every REIT operates at efficient cost frontier. On this frontier scale Berger et al define economies. which is unrealistic since Anderson et al. found that firms usually operate off of the efficient cost frontier, as operating at full efficiency is a theoretical construct (Berger, Hunter, & Timme, 1993) (Anderson R. F., 1998). Therefore losses from operating off the efficient cost frontier or performing x-inefficient (the difference between the theoretical efficient behavior and the real observed behavior of a firm) have been shown to be more substantial than losses resulting from failure to be scale efficient, i.e. having no additional scale economies to exploit (Berger, Hunter, & Timme, 1993). Nevertheless Bers and Springer estimated significant scale economies. Anderson et al. choose to apply Data Envelopment Analysis (DEA) to construct efficient cost frontier to measure overall technical efficiency, i.e. the ability of a firm to generate an optimal amount of output with a given amount of input. DEA examines x-inefficiencies and does not require choosing a functional form. X-inefficiencies have been found to exist in the REIT industry mainly because of poor utilization of inputs. Those inefficiencies are smaller in large REITs compared to smaller REITs and therefore Anderson et al. conclude that there are economies of scale (Anderson, Fok, Springer, & Webb, 2002). Ambrose et al. are the first to use a full business cycle in their data set ranging from 1990 to 2001 with 187 equity REITs and a total of 1,508 yearly observations. They take a look back on the history of research of scale economies of REITs and come to a similar conclusion as other papers. In the 1970s to 1980s scale economies did not exist for REITs as the legal regulations for those companies were quite large and their size

(9)

was very small compared to todays REITs. Therefore they just did not have the size to experience any economies of scale (Ambrose, Highfield, & Linneman, Real Estate and Economies of Scale: the Case of REITs, 2005). In the 1990s REITs became larger and therefore the possibility of scale economies arose. The literature on this time frame usually finds evidence for smaller inefficiencies in larger REITs and therefore concludes that there is some degree of economies of scale (Anderson, Fok, Springer, & Webb, 2002) (Bers & Springer, 1998). With REITs continuing to grow at a rapid pace at the end of the 1990s and the beginning of the 2000s it was possible to apply the research on newer data of very large REITs. In the early 1980s the average size of a REIT in the US was around $28 million, in 1990 $73.4 million to $95 million, in 2000 an average REIT had a size of $733.9 million and finally in 2014 the average size for REITs was $4.201 billion (NAREIT, 2015).

Table 1: Number of REITs and total market capitalization of all US-REITs This table looks at the market capitalization of all REITs in the respective year and is measured in million dollars. Additionally the number of REITs is stated.

Table 1: NAREIT data from reit.com, table by the author (NAREIT, 2015)

The table shows the rapid growth of the market capitalization over the years with a moderate increase of the number of REITs. Therefore the average size per REIT also increased at rapid pace. With the new data on those very large REITs new results were generated. Miller et al. find that there is no evidence for scale economies for large REITs and Yang even suggests diseconomies of scale for REITs, i.e. with increasing size of REITs efficiency decreases (Yang, 2001).

(10)

In a paper from 2000 Ambrose et al. argue that the reason for larger REITs outperforming their smaller counterparts could be that REITs with high growth expectations receive high multiples, which then creates a chain reaction of continued growth (Ambrose, Ehrlich, Hughes, & Wachter, 2000). They combine their findings with the previous results of Zell that larger REITs tend to have higher earnings growth potential and conclude that when REITs grow larger, their multiples rise and thereby funding further growth and consolidation (Zell, 1997) (Ambrose, Ehrlich, Hughes, & Wachter, 2000). In their statistical analysis however they found no evidence that larger REITs generate a higher income growth. Nevertheless the reason for not finding supporting evidence could be the data set as it is rather old (1997) and small (4 years) and only based on residential REITs (Ambrose, Ehrlich, Hughes, & Wachter, 2000). In their next paper Ambrose et al. argue that REITs might experience economies of scale as their largest input cost – capital costs – can be lowered by expanding. The used data set is more extensive than the previous data set from the study conducted by Ambrose et al. The data sets consists of 187 REITs that are traded on the Nasdaq, New York Stock Exchange or the American Stock Exchange from January 1990 until December 2001. Additionally all REIT-related press releases in the time frame from January 1988 through December 2001 were examined (Ambrose, Highfield, & Linneman, Real Estate and Economies of Scale: the Case of REITs, 2005). REITs have easy access to capital compared to other firms and the cost for accessing capital decreases with an increase in REIT size (Ambrose, Highfield, & Linneman, Real Estate and Economies of Scale: the Case of REITs, 2005). Ambrose et al. find evidence for economies of scale on their whole sample. They test for scale economies by examining growth prospects, revenue and expense measures, profitability ratios, systematic risk and cost of capital measures (Ambrose, Highfield, & Linneman, Real Estate and Economies of Scale: the Case of REITs, 2005). They examine that smaller REITs have possible efficiency gains in the area of growth and that large REITs manage to cut down their costs in particular their G&A costs and thereby increasing their profit margin. A conclusion is that since there is a direct relationship between return on equity and firm size, REITs experience scale economies (Ambrose, Highfield, & Linneman, Real Estate and Economies of Scale: the Case of REITs, 2005).

Calculating a translog cost function to estimate the stochastic-frontier is a widely accepted method to examine for economies of scale in REITs. Miller et al. used this

(11)

method but instead of applying just one profitability measure to the model they have chosen to add a second output measure – revenue and assets. Plus instead of the typical cross-sectional analysis this study uses panel data from the time sample 1995 until 2003 (Miller, Clauretie, & Springer, 2006). As many researchers before them found evidence of scale economies, but did not give sufficient reasons for potential gains in operating efficiency they wanted to alter their statistical method by enlarging their model. One result was that revenue, as an profitability measure, seems to be better suited than assets as when measuring in revenue self-management at a REIT seems to increase efficiency opposed to measuring in assets (Miller, Clauretie, & Springer, 2006). Possible reasons for this could be that when a REIT is not self-managed, external management might be incentivized to increase the asset position to earn a higher compensation. Miller et al. find no evidence of scale economics on the whole sample whereas older research does find such evidence. The authors explain this by the use of newer data and the application of panel data instead of cross-section analysis. Nevertheless if the sample is divided into three subsets – 1995-1997, 1998-2000, and 2001-2003 – the paper concludes that for the old data scale economies can be found but as REITs grew dramatically this growth might have exhausted possible scale economies so that in the newest sample (2001 – 2003) no scale economies can be found (Miller, Clauretie, & Springer, 2006). This might imply that there is a certain size of REITs that annihilates possible economies of scale, i.e. REITs can become too large.

Statistical Approaches:

The statistical approach to test for scale economies has changed over time and there are different ways to proceed. In the following a short summary of the different approaches will be made.

The standard approach as used by Bers and Springer in 1997 and 1998 as well as by Pennington-Cross in 2000 is to estimate the cost function and then an examination of how larger REITs cost structure differs from smaller REITs. Total costs are depending on the asset and financing structure of the REIT (and therefore the type of the REIT) and the type of management. Geographic location, as well as diversification will also have an important impact on the cost function (Bers & Springer, 1997). This method does not allow the possibility of inefficient production as it relies on the efficient cost frontier. Therefore it was found to be not realistic as almost no firm can operate on this frontier.

(12)

Different approaches are the frontier studies of REIT operating efficiency. Anderson et al. use the data envelopment analysis (DEA) and calculate the scale economies and inefficiency for REITs using this approach for 1992 to 1996 (Anderson R. F., 1998). Anderson et al. then applied this method to a newer data set of 1995 to 1999 and used the measure as an indicator for portfolio selection (Anderson, Fok, Springer, & Webb, 2002). Both studies found large inefficiencies.

Another frontier approach is to use a stochastic frontier, which is a efficient cost frontier with a stochastic variable which allows to account for possible inefficiencies of a firm and therefore seems to be more realistic with Bayesian statistics to calculate economies of scale and inefficiency. Lewis et al use this approach with 1995 to 1997 data and find lower inefficiency, but find the sources of this inefficiency – management type, degree of diversification, and leverage ratio (Lewis, Springer, & Anderson, 2003). Self management is found to be efficient for some of the years, but not for all years which is a contradiction to previous work, e.g. Bers & Springer 1998b, and Anderson et al. 2002. Ambrose et al. also use the stochastic-frontier approach with a dataset of 1990 to 2001, but their description of the model and its estimation are regarded as not scientifically satisfying (Miller, Clauretie, & Springer, 2006).

Ambrose et al. also use a frontier approach to measure the translog cost function but their main focus lies on an OLS regression. Their regression measures the effect of size on REIT growth prospects, net operating income over sales, rental revenues over sales and G & A (General and Administrative) expenses over sales. Their size variable is the logarithm of total capitalization measured in dollars.

The profitability is very often measured in assets or by dividing assets in subcategories, even though newer research suggests revenue to be the better profitability measurement (Miller, Clauretie, & Springer, 2006; Yang, 2001). As real estate is a very heterogeneous good, it cannot be easily measured by physical characteristics such as square foot of all assets et cetera. Scale effects for other industries are more easily estimated as they have a certain physical measurement unit, e.g. gallons of beer for a brewery (Ambrose, Highfield, & Linneman, Real Estate and Economies of Scale: the Case of REITs, 2005).

For DEA input is total cost, divided into interest, operating, general and administrative expenses and in management fees. The stochastic-frontier models usually use input

(13)

prices; Lewis et al. only include output in the translog cost function (Lewis, Springer, & Anderson, 2003). Ambrose et al. make only use of input costs rather than input prices in the stochastic-frontier model (Ambrose, Highfield, & Linneman, Real Estate and Economies of Scale: the Case of REITs, 2005).

3. Methodology

3.1 Econometric Models

To adequately assess the developed hypothesis over a larger time frame, it is necessary to collect a vast amount of quantitative data on accounting and performance data of US REITs. Although a larger proportion of research has applied frontier analysis to estimate REIT operating efficiency, this paper will apply an OLS regression on panel data to benefit from the advantages of panel data regression methodologies. The rationale behind the OLS regression is based on Ambrose et. al, but slightly edited to account for the possibility of heterogeneous effects over the time, especially over the course of the financial crisis in 2007/2008 and in the aftermath.

Another very important aspect when examining economies of scale is how to measure performance of REITs. As already discussed physical measurements do not seem to be fit for measurement as real estate is a very heterogeneous good. Many studies use revenue as the measurement for REIT performance and intuitively revenue might seem to be the best suited, but since revenue does not include the cost structure of the REIT – despite the fact that it might be a possible source of scale benefits as larger REITs might be able to decrease some aspects of their costs – it seems not to be suitable for scale economies research. Including the costs would result in the net operating income that still has one remaining flaw when measuring REIT scale economies. When looking at income statements of REITs, depreciation is a large position that affects the NOI to a large degree. This might be suitable for many businesses as depreciation is an important aspect when dealing with machinery, cars et cetera. But real estate is different: property rarely looses value and often even appreciates over time. This would (and very often does) result in different book values than real values. This difference will not be shown on the NOI as long as the property is not being sold and as soon as it is being sold a very high NOI will be the result. Therefore NOI is an inadequate measurement of REIT performance. Newer research suggests that FFO might be better suited. FFO is

(14)

calculated by taking the NOI and adding amortization, depreciation and subtracting the gain on sale of property, as this gain is not recurring and in conclusion not contributing to the sustainable dividend-paying capacity of the REIT.

Taking FFO as dependent variable will be statistically inaccurate, since there might be an underlying trend that would not be accounted for. To avoid this issue, the author has chosen to take FFO growth as dependent variable.

The model used for testing the hypothesis whether REITs experience any kind of scale economies is given in the following, other models will be given and explained in the robustness chapter.

Model 1: FFO Growth

The author of this paper uses the Hausman specification test to decide whether to take fixed or random effects to estimate the model. Each model in the main part as well as in the robust section was tested with time-invariant property type and management dummies as well as with fixed or random effects after dropping those time-invariant variables (Hausman, 1978). All different combinations of models were once tested with clustered standard errors and without since standard errors allow for heteroskedasticity

(15)

and for arbitrary correlation within a cluster, but in return treat the errors uncorrelated across all entities (Stock & Watson, 2007). To check the results for their robustness two additional models are introduced in the robustness section with the dependent variables of FFO per Share as well as Revenue over Costs. Both dependent variables are also regressed with the growth of assets as well with the natural logarithm of total assets. For the model with Revenue over Costs the additional independent variable of FFO

growth is added, but besides that the set of independent variables remains the same.

3.2 Variable Description

Total Capitalization (logtotalcap and logtotalcap2):

The size variable of total capitalization is the value of a REITs total amount of shares (i.e. market capitalization) measured by the natural logarithm of the dollar value. A positive sign of the coefficient might suggest that REITs experience scale economies. The square of the natural logarithm of the total capitalization is also included to test whether a possible effect of the capitalization on the FFO growth is non linear, i.e. diminishing or increasing. Since this variable relies on the market capitalization rather than the value of total assets, perfect multicollinearity can be avoided.

Growth of assets (growthassets):

The growth of assets is the independent variable of interest, the rationale behind this is that if a REITs assets have grown over a period of time and this has an effect on the FFO growth of said REIT then this will strongly suggest economies of scale. Therefore this coefficient is expected to be positive in the case of economies of scale or negative in the case of diseconomies of scale.

Market Leverage:

Market Leverage is included in the regression to differentiate between the effect of size on the FFO growth and the effect of financial leverage, as this is also a very important factor for FFO. As opposed to book leverage market or financial leverage is calculated by dividing the (book) value of debt by the shareholders’ equity. The reference paper of Ambrose et. al. shows a positive sign of market leverage nevertheless the author of this paper would expect a negative sign as the financing itself does not affect the FFO, REITs with a high leverage ratio might not be as flexible as REITs with lower leverage ratio regarding the acquisition of new properties as they are not as dependent on

(16)

external financing. Additionally it could be easier for REITs to get financing if they have a lower leverage ratio and therefore a lower risk for financial distress.

Proportion of Short-Term Debt (shorttotaldebt):

The proportion of the short-term debt with respect to the total debt amount is calculated by taking the dollar value of the short-term debt and dividing it by the total amount of debt of a REIT. As a higher ratio of short-term debt induces most likely financial distress, the expected sign will be negative, i.e. more short-term debt will reduce the FFO growth rate of a REIT.

Tobin’s Q:

Tobin’s Q is derived by subtracting the book value of equity from total assets and then adding the market value of equity, and then divide by total assets (Feng, McKay, & Sirmans, 2011).

Tobin’s Q is the ratio between the market value of a firms assets and its replacement value and therefore often referred to as possible growth opportunities of a REIT. This is justified by the fact that the ratio puts the trust of the financial market into the management of the REIT in relation to the real physical value of the REITs assets. Therefore a higher ratio means that the management of the REIT is trusted by the market participants concluding that sign of the coefficient is expected to be positive as higher market trust in the management allows for higher FFO growth in the future. This ratio plays a larger role in other financial firms than in REITs as REITs are almost completely invested into real estate and for real estate the book value is often very similar to the replacement value, nevertheless it might play a role in REITs profitability and was therefore included in the regression.

Cash and Total Assets Ratio (Cash/Total Assets):

When including the short-term debt ratio in the regression it seems only logical to also include the cash to total assets ratio as a higher cash position might neutralize possible financial distress of a REIT. But more importantly a higher cash to total assets ratio also means that a larger proportion of the equity of a REIT is not generating any profit, therefore a negative sign of the coefficient is expected. The ratio is calculated by dividing the sum of US GAAP cash and cash equivalents by the dollar value of a REITs total assets.

(17)

As different property type foci might cause potential constant heterogeneity in the profitability of a REIT, dummies for the different REIT investment foci were created. Dummies for the different types residential, retail (including shopping center, regional malls or other retail REITs), office and industrial, specialty (e.g. health care, specialty or self-storage REITs), diversified and hotel were created.

Dummy for Management Type:

As the type of management has probably an effect on the profitability of a REIT the author accounted for this effect by implementing a dummy variable that indicates whether a REIT is externally or internally managed. Internally managing the REIT might reduce the moral hazard of that REIT as the management would have to face the consequences that are attributed to their decisions, e.g. bad managing will result in loss of money and their jobs as when managed externally the risk is not associated with the management.

Dummies for During Crisis and After Crisis periods:

Unlike the reference paper of Ambrose et. al. the author of this paper wants to account for possible “game changing” effects of the financial crisis in 2007/2008. Therefore two dummy variables DuringCrisis (Q3 2007 – Q2 2009) and AfterCrisis (Q3 2009 – Q1 2015) were created in order to account for constant differences in the profitability of REITs during financially distressed times and in the aftermath of the financial crisis. Interaction Terms:

As the time frame of this paper exceeds a 10-year period in which the largest financial crisis since the great depression in 1929 took place it seems only logical to allow for heterogeneous effects in all variables of interest, i.e. the size variables. Therefore interaction terms for all three size variables with the Crisis dummies were included in the regression.

4. Data

The data required for the empirical analysis was mainly retrieved from the SNL database in quarterly frequency. The examined time horizon was from 2001 until 2015 (= 59 quarters), which made it possible to examine the longest time period that has ever been examined with respect to economies of scale in REITs. Due to the elimination of outliers, e.g. leverage (debt) ratios of more than 100%, not all available data from the SNL database was used in the analysis. The final panel dataset contains 4,419

(18)

observations with 78 different entities, i.e. REITs. The data set is partially unbalanced due to the fact that not all REITs reported all accounting data continuously over the whole time period from 2001 to 2015. Nevertheless the majority of the REITs reported a complete set of accounting data, which are required for the proposed statistical model. To avoid potential bias regarding different currencies and different governmental regulations only US equity REITs were selected, since this sample has the largest amount of observations and was used in the previous research of scale economies in REITs.

Table 2: Descriptive Statistics of REIT Characteristics

This table provides the different characteristics that were used to separate REITs for this empirical analysis and the number of REITs of that special characteristic that were contained in the dataset

By Investment Focus Number of REITs

contained in dataset Retail (= Shopping Centers, Regional Mall & Other Retail) 21

Office & Industrial 15

Specialty (= Specialty, Health Care, Self-Storage) 12

Diversified 11

Hotel 6

Residential 13

Management Characteristics

Self-Advised 73

Total Number of REITs 78

Table 2 shows the distribution of the REITs regarding investment focus and whether the REITs advised/managed themselves. REITs with an investment focus on shopping centers, regional malls and other retail were merged to `Retail`, office and industrial REITs were merged to `Office & Industrial` and specialty, health care and self-storage REITs were summarized in the investment focus group of `Specialty`.

Only 6 of the 78 examined REITs invest in Hotels whereas the majority of the REITs (=21) invest in Retail followed by Office & Industrial (15) and Residential (13). Specialty (12) and Diversified (11) REITs complete the different investment foci. The vast majority of the examined REITs were internally managed and advised with 73 of the 78 REITs being internally managed.

(19)

Table 3: Descriptive Statistics of Dependent and Independent Variables Table 3 provides the descriptive data of selected dependent and independent variables regarding their Mean, Standard Deviation, Minimum, Maximum, as well as the number of observations (=N), the number of entities, i.e. REITs (=n) and the T-bar value.

Table 3: Descriptive Statistics of Dependent and Independent Variables

Variable Mean Std. Dev. Min Max Observa

tions

Growth in FFO overall -0,0320168 2,435552 -63,77778 38,37994 N = 4215

(in %) between 0,3946465 -2,094752 1,150009 n = 78

within 2,405954 -61,71504 38,91655 T-bar = 54,0385

Size 1 overall 14,80265 1,428533 6,927336 18,34788 N = 4420

ln(Total Capitalization) between 1,35373 10,19084 17,52348 n = 78

within 0,5168954 11,53915 16,5497 T-bar = 56,6667

Size 2 overall 221,1587 40,90886 47,98798 336,6445 N = 4420

ln(Total Capitalization)2 between 38,57385 105,311 307,3406 n = 78

within 15,06149 160,9869 278,207 T-bar = 56,6667

Size 3 overall 0,0254845 0,0989419 -0,5051383 1,741223 N = 4333

Growth of Assets between 0,0174864 -0,0241195 0,077064 n = 78

within 0,0974216 -0,4555342 1,699982 T-bar = 55,5513

Market Leverage overall 39,44482 15,11458 0 98,519 N = 4420

(in %) between 12,04741 2,536417 66,40889 n = 78

within 9,251564 -14,49225 83,05657 T-bar = 56,6667

Proportion of Short-Term overall 0,0506271 0,0969225 0 1 N = 2743

Debt of Total Debt between 0,0531724 0 0,3699386 n = 69

within 0,0890349 -0,3193115 0,9067828 T-bar = 39,7536 Tobins Q overall 2,002396 0,4190794 0,8619761 4,202201 N = 4420 between 0,3278189 1,38939 3,146669 n = 78 within 0,2633049 0,9283442 3,713018 T-bar = 56,6667

Cash over Assets Ratio overall 0,0237609 0,0451721 0 0,5637501 N = 4420

between 0,0331634 0,0011823 0,2410448 n = 78

within 0,0310523 -0,2132318 0,4892529 T-bar = 56,6667

Revenues over Cost Ratio overall 1,131291 0,5213895 -1,233289 15,12397 N = 4201

between 0,3484721 0,0181739 2,014721 n = 76

within 0,3913934 -2,116719 14,63421 T-bar = 55,2763

(20)

between 0,4402331 -0,139 3,315614 n = 78

within 0,6067249 -16,3652 10,1048 T-bar = 54,8333

In order to understand the used dependent and independent variables Table 3 depicts the mean, standard deviation, minimum, maximum and the number of observations of said variables. Additionally the mean between and within the respective groups, i.e. the different REITs are given.

The T-bar value shows for how many time periods the respective variable was recorded. For most variables nearly all 57 time periods were recorded, except for the Proportion of Short-Term Debt of Total Debt (approximately 40). This is due to the fact that this ratio was not recorded by 9 out of the 78 REITs.

The dependent variable Growth in FFO is measured in percent and has a mean of -0.03%, which implies that there was no sustainable growth in FFO over the data horizon of 15 years, i.e. that possible gains in FFO before and after the crisis where neutralized by the losses during the financial crisis. Nevertheless the standard deviation is rather high with 2.436, but not extremely high. This suggests that there is a relatively high variation in between the time periods – recall that the FFO was not measured annually but quarterly. The minimum of -63.78% is very high and most likely the result of the financial crisis when tremendous cuts of FFO were suffered by the REITs. The maximum is not as large as the minimum with 38.38%, which was probably in the aftermath of the crisis when some REITs recovered quite rapidly. The variable itself was not recorded by the REITs but created for this empirical analysis by subtracting the FFO of time period t from the FFO of time period t +1.

The mean of the Size 1 and Size 2 variable is difficult to interpret since it is the natural logarithm of total market capitalization. The range of values for the Size 1 variable is 6.9 to 18.3 and for Size 2 it is 48.00 to 336.6, which is not very surprising since Size 2 is the square of Size 1.

Growth of Assets, which is the independent variable of interest ranges from -0.505 to 1.74, which means that the growth in assets from one quarter to the next quarter was between -50.5% and +174% in the observed time periods. Since the growth of assets depends on the asset value the seemingly large range is not unusual. During the financial crisis losses of 50% in asset value were possible as well as increases by more

(21)

than 150% in the booming periods that followed the financial crisis. The standard deviation of almost 10% means a high degree of fluctuation during the observed time periods.

The average market leverage ratio is about 39-40%, which seems reasonable for real estate investments. For this variable outliers of more than 100% were dropped, as a leverage ratio of more than 1 or 100% means that the book value of debt is larger than the equity of the shareholders, which results in the REIT being bankrupted. The minimum value of leverage was 0%, which means that all assets of the REIT were financed through equity. The maximum was 98.5%, which seems very large as only 1.5% of the REITs assets were financed by equity. This value was most probably reached during the financial crisis and means that the REIT had a high degree of financial distress during that period.

The Proportion of Short-Term Debt of Total Debt has a mean of 5%, which means that on average only 5% of the total debt of a REIT was financed by short-term debt. This seems reasonable, as real estate is usually a long-term investment, which in conclusion calls for long-term debt financing. The range of this ratio ranges from 0 – no short-term debt – to 1, which means nothing but short-term debt. A ratio of 1 was probably achieved during the financial crisis when banks did not want to lend out long-term loans to REITs in financial distress.

Tobin’s Q has a mean of 2, which can be interpreted as that the market value of REITs is on average double of the REITs total assets. Put differently a ratio of 2 means that the market values a REIT twice as much as the replacement costs of its total assets, even though the ratio ranges from 0.86 to 4.2.

The Cash over Asset Ratio has a mean of 2.3%. This value is not unexpected as a large cash position does not generate returns for the REIT and is therefore being avoided by REIT management. Nevertheless some cash has to be retained as it keeps the REIT liquid. The ratio ranges from 0 to 56.37%, but as the standard deviation is 4.5% a ratio of over 50% is an outlier. A ratio of 0 means that the REIT does not have a cash position, which is very dangerous as REITs could be forced to quickly sell their assets during a crisis when they cannot get bank loans fast enough.

The first additional dependent variable Revenues over Cost has a mean of 1.131 with a standard deviation of 0.52. The value is rather low, but not highly unexpected. The

(22)

value of 1.131 means that on average REITs had a revenue that was 13.1% higher than their costs, i.e. they were profitable. The minimum is -1.233, which means that the REIT had a negative return (as negative costs are not realistic) probably during the crisis and the maximum of 15.124 means that a REIT had revenue that was more than 15 times as large as the cost of that REIT.

The FFO per Share dependent variable has a mean of 0.65, which means that on average per share of a REIT 0.65$ of FFO was reached. The value itself of this variable has not very much meaning as the REIT can change the number of shares and therefore very high or very low values can be reached without having any meaningful interpretation regarding the economic situation of said REIT. Nevertheless the sign of this value is of importance as a negative sign for example for the minimum of -13.7, means that the FFO was negative as the number of shares cannot be negative. A positive sign of 12.77 as it is in the maximum means that the FFO is positive.

5. Results

In order to stay within the focus of this thesis the author will mainly focus on the size variables, but will give explanations to other explanatory variables as well. For the dependent variable of FFO growth, two different models will be given and interpreted. The findings will then be put in context to existing literature on the topic especially on the reference paper of Ambrose et. al.

5.1 Regression Results

The regression results will be listed in the following starting with the main regression with the dependent variable of FFO growth. First the model with the independent size variable of asset growth will be interpreted starting with the relevant size variables and then continuing with other independent control variables. Afterwards the same model with the independent size variable of the natural logarithm of total assets will be examined following the same structure as the model before. In chapter 6, the robust section the same approach will be taken for the other two dependent variables to support the findings of chapter 5.

(23)

5.1.1 Growth of Assets as Dependent Variable

The following panel shows the regression output of the proposed model:

Regression Output Panel 1: Model with growth of assets as size indicator

Dependent variable: FFO growth

The Panel shows the independent variables in the left column and their coefficient in blank. The p-‐value of the coefficient is given in bold and below the respective coefficient. The difference between the models are given in the last three columns and are whether the model includes fixed-‐effects, random effects and/or clustered standard errors. R-‐values as well as the number of observations/groups and the p-‐value of the whole model are also given in the last section of this panel.

Model A B C D E F

Coefficient Coefficient Coefficient Coefficient Coefficient Coefficient

P-Value P-Value P-Value P-Value P-Value P-Value

logtotalcap -0,1732161 -0,1732161 1,053452 1,053452 -0,3111202 -0,3111202 0,8948121 0,7703571 0,6551642 0,6657257 0,8108206 0,5961941 logtotalcap2 0,0069543 0,0069543 -0,0376227 -0,0376227 0,0118184 0,0118184 0,8748369 0,7248008 0,6366367 0,639726 0,7873265 0,5460188 growthassets -2.938716*** -2,938716 -2.92036*** -2,92036 -2.917451*** -2,917451 0,00000536 0,4248526 0,0000082 0,428369 0,00000589 0,4269854 leverage -0.0084078** -0,0084078 0,0070231 0,0070231 -0.0077965** -0,0077965 0,0315762 0,1278682 0,3405283 0,5726859 0,0342076 0,1062298 shorttotaldebt -0,2997806 -0,2997806 -0,5836837 -0,5836837 -0,204242 -0,204242 0,5984167 0,5238402 0,3212258 0,2864721 0,7171216 0,6230947 tobinsq -0.1056298*** -0,1056298 0.0778727*** -0,0778727 0.0552078*** -0,0552078

4,24E-01 0,403119 7,64E-01 0,8055631 6,56E-01 0,5817892

cashoverassets -2.407192*** -2,407192 -4.158286** -4,158286 -2.190752*** -2.190752**

5,01E-02 0,1126482 0,0250001 0,1763795 4,67E-02 0,0368132

d_selfadvised 0,1140656 0,1140656 omitted omitted 0,1324462 0,1324462

0,5568882 0,4733151 0,4741283 0,2510263 d_officeindustrial -0,1466469 -0,1466469 0,2938488 0,2236404 d_diversified -0,0118577 -0,0118577 0,9485172 0,9669312 d_resi -0,1096447 -0,1096447 0,5339195 0,2315028 d_specialty -0,1728703 -0,1728703 0,3246807 0,2711798 d_hotel -0,2426971 -0,2426971 0,2833584 0,5764195 d_duringcrisis -6,018691 -6,018691 -5,051256 -5,051256 -7,039239 -7,039239 0,7601531 0,604029 0,8057514 0,7068756 0,7206431 0,5180815 d_postcrisis -27.7352** -27,7352 -28.35437** -28,35437 -29.62024** -29,62024 0,0237518 0,1918832 0,0499075 0,2883965 0,0147054 0,1885807 it_size_duringcrisis 10.79971*** 10.79971** 11.76727*** 11.76727** 10.65002*** 10.65002** 0,0004578 0,028228 0,0001956 0,0237287 0,0005162 0,0310543 it_size_postcrisis 3.773247*** 3,773247 3.377744*** 3,377744 3.722614*** 3,722614 0,0014487 0,3194672 0,0052403 0,3832232 0,0015661 0,3294897 it_sizel_duringcrisis 0,9314173 0,9314173 0,7902815 0,7902815 1,058114 1,058114 0,7236761 0,5243338 0,7728099 0,6427579 0,6875238 0,4416357 it_sizel_postcrisis 3.483289** 3,483289 3.542643* 3,542643 3.7303** 3,7303 0,0346553 0,2057705 0,0655973 0,2956689 0,0223567 0,1991206 it_sizeq_duringcrisis -0,0352808 -0,0352808 -0,0306746 -0,0306746 -0,0391968 -0,0391968 0,6878299 0,4467354 0,7360268 0,5707248 0,6548812 0,3697532 it_sizeq_postcrisis -0.109076** -0,109076 -0.1100417* -0,1100417 -0.1171692** -0,1171692 0,048672 0,2208797 0,0852914 0,3050529 0,0325609 0,2105425 _cons 1,658552 1,658552 -7,250975 -7,250975 2,393756 2,393756

(24)

0,864742 0,7149038 0,6791789 0,6919437 0,8043129 0,5900615 R^2 (overall) 0,0273847 0,0273847 0,0169412 0,0169412 0,0264958 0,0264958 R^2 (within) 0,0146255 0,0146255 0,0183589 0,0183589 0,014266 0,014266 R^2 (between) 0,2411985 0,2411985 0,1058735 0,1058735 0,2387394 0,2387394 Number of observations 2644 2644 2644 2644 2644 2644 Number of groups (REITs) 69 69 69 69 69 69 P-Value of F-Statistic 0,0000 0,5077 P-Value of Chi-Squared Statistic 0,0000 0,0000 0,0000 0,0012

Fixed-Effects No No Yes Yes No No

Random Effects Yes Yes No No Yes Yes

Clustered SE No Yes No Yes No Yes

*= Significant at 90% Confidence Level / **= Significant at 95% Confidence Level / ***= Significant at 99% Confidence Level

The main observation of this model is that the sign of growthassets is negative and significant at a 99% confidence level at least for non-clustered standard errors. This means that a growth in asset value correlates with a decrease in FFO growth, therefore this result is evidence for existing diseconomies of scales in REITs. In the robust section a closer look on different models will be taken to control the reliability of this result and an economic interpretation with possible reasons will be discussed in the conclusion section of this paper.

To control for changes of the effect of asset growth on FFO growth during and after the financial crisis in 2007/2008, interaction terms with all three size variables (asset growth / logarithm of total capitalization / square of logarithm of total capitalization) and the time dummies of during the crisis and after the crisis were included. Remarkably the interaction term of asset growth during the crisis (size_duringcrisis) was significant at the 99% confidence level with a positive sign of more than 10. This means that as the interaction term has a positive sign of more than three times the value of the variable of growth of assets that there would be evidence for economies of scale, but only during and after the crisis as both interaction terms are larger than the original coefficient of the independent variable of asset growth. The results of this regression confirm the findings of Ambrose et. al. who have found evidence for economies of scale due to a positive impact of total capitalization on REIT profitability measures. Their variable of asset growth is also negative (for the pre crisis period), but since their data only reaches from 1990 to 2001 they cannot examine the impact of the financial crisis on the regression model. This problem has been solved in the regression of this paper by implementing the time dummy variables and the interaction terms, which has brought very interesting new findings:

(25)

A possible scale effect of asset growth on FFO growth seems to be highly dependent on the macroeconomic situation. For this model that means that before the crisis there was evidence for diseconomies of scale as the variable of growth of assets is significant and negative, whereas during the crisis the interaction terms are highly positive and significant which changes the effect from diseconomies to economies of scale. The same holds true for the after crisis period. So before the crisis REITs might have gotten too big to exploit any economies of scale and where in fact too big so that there were diseconomies of scale. During the crisis REIT value decreased dramatically so that they were back in a region where there was room for economies of scale. This effect slowed then after the crisis as REITs began to grow again. This all hints towards a perfect size for REITs from where possible economies of scale turn into diseconomies of scale. The results for the interaction terms will also be examined more closely in the robust section where a different set of dependent variables with mainly the same independent variables will be examined.

To estimate whether the regressions with fixed effects or with random effects allowed for meaningful interpretation the hausman test was run on regression C and E, as this test is only applicable when using non clustered standard errors. The p-value of the test was 0.0163 and therefore below the critical threshold of 0.05, which means that in this model the regression with fixed effects had a higher degree of validity. Even though the clustered standard errors did not allow for the hausman test to be applied, the fixed effect model for clustered standard errors is preferred due to the findings of the test for non-clustered standard errors. The interpretation of the results does not change though, since all significant values have the same sign for both regressions.

Leverage is also negatively correlated with FFO growth meaning that a high leverage

ratio slows the growth of the FFO. This result is statistically significant at a 95% confidence interval for model A and E. Ambrose et. al. come to a different result, in their model Leverage has a positive effect on the FFO yield. No economic interpretation of this result was given in the reference paper, nevertheless a possible reason for the negative sign could be that, as the financing itself does not affect the FFO, REITs with a high leverage ratio are not as flexible for acquiring new properties as REITs with lower leverage ratio as they are not as dependent on external financing. Additionally it could be easier for REITs to get financing if they have a lower leverage ratio and therefore a lower risk for financial distress.

(26)

A large Tobin’s Q ratio means that the market value of a REIT is higher than its actual market value of assets. Reasons could be a higher trust in the REITs management. Tobin’s Q has a negative influence on the FFO growth for model A, C, and E, all of those at a 99% confidence level. The negative sign of the variable could mean that the amount of trust that exceeds actual asset value (the difference between market and actual book value) is larger than the actual gains of FFO growth that the management can realize. Or put differently, market trust in the management exceeds the potential FFO growth on average. The measurement of TobinsQ was not implemented in the reference paper of Ambrose et. al.

The next independent variable – the ratio of cash over assets – also has a significant negative coefficient for model A, C, E and F at the 99% confidence level, except for model C, which significance is only at the confidence level of 95%. The negative sign of this variable is also reasonable as a larger cash position means that a smaller proportion of the firms assets is actually generating revenue as cash usually generates little to no return. A reason for REITs to have some cash on their bank accounts might be to keep the REIT liquid in times of financial distress like the financial crisis in 2007/2008. Another implication could be that having just sold some properties increases the cash position but decreases the FFO as the earnings from selling properties do not affect the FFO, but the missing of the NOI over the next months does affect the FFO negatively.

The last significant independent variable is the time dummy of after crisis. The coefficient is negative which means that after the crisis FFO growth were in general lower than before the crisis. This might be due to the fact that different market behavior and market regulations cause the market participants – the REITs – to act more carefully and not to seek aggressive FFO growth as they did before the crisis. As this is not the main aspect of this paper the interpretation of this sign is only speculation. For all models except for model D the p-value is below 0.01, which allows for meaningful interpretation of the significant variables.

5.1.2 Logarithm of Total Assets as Dependent Variable

Another possibility to measure the size of a REIT is to take the value of total assets. One problem that might arise from this is by implementing the variable total assets with

(27)

an actual dollar value, results might be biased due to a trend. To correct for this possible bias the author of this paper uses the natural logarithm of total assets.

Regression Output Panel 2: Model with ln of total assets as size indicator

Dependent variable: FFO growth

The Panel shows the independent variables in the left column and their coefficient in blank. The p-‐value of the coefficient is given in bold and below the respective coefficient. The difference between the models are given in the last three columns and are whether the model includes fixed-‐effects, random effects and/or clustered standard errors. R-‐values as well as the number of observations/groups and the p-‐value of the whole model are also given in the last section of this panel.

Model G H I J K L

Coefficient Coefficient Coefficient Coefficient Coefficient Coefficient

P-Value P-Value P-Value P-Value P-Value P-Value

logtotalcap 0,6178278 0,6178278 2,897396 2,897396 0,4364707 0,4364707 0,736477 0,7206777 0,2999956 0,4486919 0,8077912 0,8084936 logtotalcap2 0,0070418 0,0070418 -0,0498283 -0,0498283 0,0100256 0,0100256 0,873921 0,682458 0,5339499 0,5112797 0,8199246 0,5534348 lntotalassets -0,8034603 -0,8034603 -1,525759 -1,525759 -0,7011654 -0,7011654 0,5218433 0,6513147 0,3429265 0,5555605 0,5598042 0,6864401 leverage -0,0035328 -0,0035328 0,0136647 0,0136647 -0,0032183 -0,0032183 0,7205996 0,8030715 0,313644 0,5382863 0,7344246 0,8196591 shorttotaldebt -0,3625798 -0,3625798 -0,6453969 -0,6453969 -0,2511992 -0,2511992 0,5262892 0,4280116 0,275617 0,2208443 0,6574516 0,5333977 tobinsq -0.5792256*** -0,5792256 -0.969718*** -0,969718 -0.480099*** -0,480099

3,24E-01 0,4739558 2,27E-01 0,3784233 3,95E-01 0,5439295

cashoverassets -2.124494*** -2,124494 -4.048965** -4,048965 -2.0703*** -2.0703**

8,51E-02 0,113524 0,0304979 0,1172988 6,28E-02 0,0279345

d_selfadvised 0,0691133 0,0691133 omitted omitted 0,1132442 0,1132442

0,7236282 0,6803545 0,5430332 0,3267834 d_officeindustrial -0,1153821 -0,1153821 0,4128327 0,3497681 d_diversified -0,0638394 -0,0638394 0,7285996 0,8310188 d_resi -0,1025663 -0,1025663 0,5695148 0,1843292 d_specialty -0,2028463 -0,2028463 0,2582683 0,1868849 d_hotel -0,2876977 -0,2876977 0,2073949 0,5021763 d_duringcrisis -2,661766 -2,661766 -3,087692 -3,087692 -3,98653 -3,98653 0,8937684 0,8438696 0,8818545 0,8402813 0,8412153 0,7537774 d_postcrisis -27.3189** -27,3189 -27.36058* -27,36058 -29.17934** -29,17934 0,0268374 0,1747194 0,0610705 0,2712632 0,0169473 0,1728382 it_size2_duringcrisis 0,3884124 0,3884124 0,1447523 0,1447523 0,3297039 0,3297039 0,5231572 0,5555658 0,823451 0,7934351 0,5849911 0,5964486

(28)

it_size2_postcrisis -0,5052678 -0,5052678 -0,4548113 -0,4548113 -0,4867282 -0,4867282 0,2768942 0,1948541 0,3781075 0,1860476 0,2905082 0,2262214 it_sizel_duringcrisis 0,1078724 0,1078724 0,3952717 0,3952717 0,3300075 0,3300075 0,9691659 0,9599599 0,8920625 0,8621725 0,9056595 0,8698557 it_sizel_postcrisis 3.937328** 3,937328 3.863743* 3,863743 4.157078** 4,157078 0,0236629 0,162696 0,0583414 0,2498164 0,0158684 0,1664586 it_sizeq_duringcrisis -0,0201405 -0,0201405 -0,0216751 -0,0216751 -0,0252427 -0,0252427 0,8205981 0,7066159 0,8139044 0,720472 0,7759268 0,6162969 it_sizeq_postcrisis -0.1077171* -0,1077171 -0,1056181 -0,1056181 -0.1153425** -0,1153425 0,0530903 0,2037181 0,1021326 0,2934606 0,0366734 0,1963443 _cons 2,274981 2,274981 -8,326299 -8,326299 2,476213 2,476213 0,8162123 0,5994196 0,6382472 0,6213038 0,798768 0,5369018 R^2 (overall) 0,0182174 0,0182174 0,00868 0,00868 0,0172995 0,0172995 R^2 (within) 0,004854 0,004854 0,008324 0,008324 0,0045808 0,0045808 R^2 (between) 0,2580731 0,2580731 0,1245911 0,1245911 0,2540486 0,2540486 Number of observations 2644 2644 2644 2644 2644 2644 Number of groups (REITs) 69 69 69 69 69 69 P-Value of F-Statistic 0,1230 0,6805 P-Value of Chi-Squared Statistic 0,0006 0,0000 0,0001 0,0081

Fixed-Effects No No Yes Yes No No

Random Effects Yes Yes No No Yes Yes

Clustered SE No Yes No Yes No Yes

*= Significant at 90% Confidence Level / **= Significant at 95% Confidence Level / ***= Significant at 99% Confidence Level

The size variable of ln of total assets is also negative in this model, but it is not statistically significant. Therefore making any interpretations regarding economic implications of this coefficient are speculative and not reliable.

The hausman test of regressions I and K gives a p-value of 0.0588 meaning that the random effect models seems to have a higher validity. Again only the size of the coefficient differs between the models but not the sign itself, the following interpretation of the control variables will therefore not distinguish between random and fixed effects models.

As no size variable is significant in this model, closer attention towards the control variables will be directed. As the models I and J have a p-value of more than 0.10 the reliability of the models, especially for model J with a p-value of 0.68 is very limited.

The measurement of TobinsQ is also negative and significant at a 99% confidence level. The economic implications of this finding are the same as previous stated. The variable is significant for all models without clustered standard errors (G, I and K).

(29)

The same holds true for the ratio of cash over assets. The coefficient of this variable is negative (-2.12) and significant at a 99% confidence level. This enforces the previous findings that a larger cash position means a lower FFO growth as a larger proportion of the company value is invested in cash and therefore not generating any revenue (or very low revenue) a lower FFO growth is the consequence. Here it is important whether the dependent variable is FFO growth or FFO itself: A high cash position means that there is a higher proportion of the assets not invested to generate higher (i.e. growth) FFO in the future, nevertheless a higher cash position means that right now in the time period the FFO (not the growth of it) is larger – this will be closer examined in the robust section. In the models without clustered standard errors (G, I and K) the variable is significant as well as in model L.

The time dummy variable of post crisis is also negative with a similar coefficient as in the previous model (-27.3) and also significant at a 95% confidence level. This is evidence that the financial crisis in 2007/2008 has changed market behavior and/or market regulations, which had in general a lower FFO growth as consequence. Again this variable is significant in all models without clustered standard errors (G, I and K) even though in model I the significance is only at the 90% confidence level.

Regarding the interaction terms of the time dummies and the size variables only the terms of total capitalization x post crisis and total capitalization^2 x post crisis are significant at different confidence levels (95% or 90%).

The p-values for models G, H, K, L are below 0.10, therefore meaningful interpretation of the significant variables can be made. For models G, H, and K the p-values are lower than 0.01.

6. Robustness of the Results

6.1 Alternative Dependent Variables

To test the robustness of the results two other dependent variables that also depict profitability ratios were implemented and regressed with the set of independent variables. The first of the new dependent variables is FFO per share, which can be compared to the earnings per share ratio of stocks. As REITs are publicly listed they are usually traded per share. Therefore the ratio of FFO per share is a very important measurement to estimate the profitability of a REIT of interest for potential investors.