Chapter 1 Introduction
1.3 Research Overview
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
5
1.3 Research Overview
1.3.1 Research Framework
This research is organized as follows. Chapter 1 is “Introduction,” which includes the general background and research purpose, research scope and method, and research overview. Chapter 2 is “Literature Review,” providing a review of the dynamics between REITs and direct real estate and the literature on agency problem in REITs. Chapter 3 is “Research Method and Data Information,” presenting the methodology employed and the data used in the empirical analysis. Chapter 4 is
“Empirical Results,” which illustrates the empirical results of the U.S. and Taiwan, and agency problem in T-REITs market. Finally, Chapter 5 is “Conclusion and Discussion,” summarizing the findings and implications of this research.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
6
1.3.2 Research Process
Figure 1-1 Research Process
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
7
Chapter 2 Literature Review
This chapter is divided into two sections. The first section explores the dynamic relationship between REITs and direct real estate markets. The second section discusses agency problem in REITs, as the basis for analyzing the performance of T-REITs market.
2.1 Dynamics between REITs and Direct Real Estate
The linkage between REITs, direct real estate, stock, and bond markets has been intensively studied since the late 1980s. Since REIT is the financial asset derived from real estate, much of the previous literature has focused on the correlation between REITs and direct real estate, and the conclusions are quite inconsistent. For example, Giliberto (1990) found that the residuals from regressions of REITs and direct real estate returns on financial asset returns are significantly correlated. This implies that both REITs and direct real estate returns are affected by a common real estate factor that links their performances together (Gyourko and Keim, 1992; Mei and Lee, 1994).
Instead, Goetzmann and Ibbotson (1990) indicated that both return and volatility of REITs were far above that of direct real estate, and two series were only weakly correlated. Since then, the low correlations between REITs and direct real estate in the U.S. have been confirmed in many studies (Ross and Zisler, 1991; Gyourko and Keim, 1992; Barkham and Geltner, 1995; Geltner and Kluger, 1998). Moreover, the same argument has been verified in several countries (Hoesli, Lekander, and Witkiewicz, 2004; Newell, Chau, Wong, and McKinnell, 2005).
In contrast with previous studies, the relationship between REITs and direct real estate has become more closely-related over the past two decades. This argument is supported by Clayton and MacKinnon (2001) who found that REIT returns exhibit an increasing sensitivity to real estate returns through time. Due to the dramatic growth and maturation of the REIT sector, REIT have been more like real estate and less like stock (Ghosh, Miles, and Sirmans, 1996; Ziering, Winograd, and McIntosh, 1997;
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
8
McIntosh and Liang, 1998). With better information about REITs available, REITs have begun to better reflect their “true” nature, stated by Clayton and MacKinnon (2001). More recently, Morawski, Rehkugler, and Füss (2008) found that correlations between REITs and direct real estate are clearly higher for longer holding periods.
Since direct real estate is deemed as a long-term investment, it should also influence the performance of REITs in a similar manner.
Other studies have focused on the lead-lag relation between REITs and direct real estate markets. For instance, Giliberto (1990) reported that the relationship between REITs and direct real estate returns is remarkably stronger when a lead in the REIT returns being considered. Moreover, Gyourko and Keim (1992) suggest that the correlation analysis between REITs and appraisal-based real estate indices seems to be deviated, since the latter is based on valuations conducted every two to four quarters.
Hence, the authors demonstrate a significant relationship between the adjusted returns of NCREIF and the one-year lagged returns of NAREIT indices. Other studies supporting this argument are conducted by Myer and Webb (1993) and Barkham and Geltner (1995), which employed Granger causality test. In more recent studies, Li, Mooradian, and Yang (2009) and Oikarinen, Hoesli, and Serrano (2011) indicated that NAREIT led both NCREIF and TBI indices after 1990. Myer and Webb (1994) and Newell et al. (2005), however, found no Granger causality between REITs and direct commercial real estate in short sample period.
In addition to the analyses of short-run volatility and lead-lag relations between REITs and direct real estate markets, some studies further examined the existence of cointegration through investigating the long-term dynamics between these two markets. Morawski et al. (2008) showed that there are cointegration relationships among NAREIT, NCREIF and the S&P 500 stock indices from 1978 to 2006. More recently, Oikarinen et al. (2011) presented that NAREIT are cointegrated with both NCREIF and TBI but not with the S&P 500 stock indices from 1977 to 2008. The results suggest that REITs and direct real estate are likely to have similar long-term diversification benefits in a stock portfolio.
Not many domestic studies have examined the existence of long-run dynamics
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
9
between T-REITs and other markets. Zheng, Chang, and Bai (2008) found that T-REITs index are not cointegrated with the stock index nor be the construction index in two years. The results imply that T-REITs have diversification benefits.
Overall, studies on the dynamics between REITs and direct real estate markets in different countries are extensive, especially on the U.S. However, empirical literature on this issue in Taiwan or other countries is relatively limited. Most studies have discussed the relationship between T-REITs and stock or construction stock indices as a proxy for the stock market, while few researches analyze the relationship between T-REITs and direct real estate markets. The purpose of this study is thus to explore the short- and long-run dynamics between T-REITs and direct commercial real estate markets.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
10
2.2 Agency Problem in REITs
There are two competing property management structures for the corporate organization of REITs: internal and external management. Since one notable characteristic of REIT is the separation of ownership and control, agency problem is likely to occur between shareholders and management. Jensen and Meckling (1976) defined the agency relationship as a contract when the principal engage the agent to perform some service on their behalf which involves delegating some decision making authority to the agent. If the incentive or reward mechanism is not well designed, then there is good reason to believe that the agent will not always act for the best interests of the principal. In this case, the agency cost is inevitable.5 In addition, the authors suggest that the agency conflicts will affect firm performance, and increasing management’s ownership can help mitigate agency problems. Therefore, agency theory implies that suppose agency inflicts appear in externally-managed REITs, their market performance will also be influenced by the ownership structure.
Conflicts of interest refer to situations where the interests for management and shareholders are misaligned: acting on their self-interests, managers make decisions that will not be in the best interests of shareholders. Sagalyn (1996) identified twelve types of conflicts of interest, which cut across all spheres of REIT decision making, i.e., offering formation, investment management, transaction activity, and property management.6 The author also argues that a misalignment of incentives exists for externally-managed REITs, while the potential for conflicts of interest will decline with internal management.
On the other hand, agency theory suggests that when corporate managers have a significant ownership stake, managerial incentives are more closely aligned with shareholders and agency costs are reduced (Jensen and Meckling, 1976). Cannon and Vogt (1995) found that self-administered REITs outperformed advisor REITs over the
5 Agency costs include the monitoring expenditures by the principal, the bonding expenditures by the agent, and the residual loss (Jensen and Meckling, 1976).
6Types of conflicts of interest (COI) contain allegiance, sponsor control, outside partners, over-compensation, resource allocation, competitive affiliates, tie-in business, captivity, tax timing, expense preference behavior, and malingering (Sagalyn, 1996).
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
11
1987 to 1992 sample period even after adjusting for the differences of market risks.
Ownership structure has considerably more effect on the performance of advisor REITs, but less effect on self-administered REITs. The authors suggest that self-administered REITs have been able to reduce agency problems effectively by other approaches, for instance, more standardized financial reporting or incentive-based compensation structures. The same findings of underperformance for externally-managed REITs are demonstrated by Howe and Shilling (1990), Hsieh and Sirmans (1991).
More recently, Capozza and Seguin (2000) exhibited that externally-managed REITs consistently underperformed internally-managed REITs due to the high financial leverage over 1985 to 1992. Ambrose and Linneman (2001) examine differences between externally-advised and internally-advised REITs with respect to operating structure, growth prospects, operating revenue and expenses, cash flow and profitability, equity returns, betas and capital costs. The results almost consistent with those found by Capozza and Seguin (2000), and indicate that internally-advised REITs continue to outperform externally-advised REITs. Furthermore, the authors found that internally-advised REITs have significantly higher betas than externally-advised REITs. It reflects the market’s perception of these firms as internally-advised (unproven) growth stocks.
In Taiwan, most of T-REIT managements are related to the originating companies (i.e. parent companies). It is likely to induce conflicts of interest and result in the loss of investors’ interests. By examining the trends of REIT price and Net Asset Value (NAV), Wang and Chang (2009) suggest that some T-REITs may exist conflicts of interest due to the close business relationships between property management and original owners. In more recent studies, Tsai, Chen, and Chang (2011) found that REITs in Taiwan are not defensive since investors have not yet been familiar with the characteristics of REITs market. However, we conjecture the potential agency problem may be the main reason for the limited development of T-REITs market. Since literature on the agency problem for T-REIT is relatively limited, this study attempts to empirically verify the hypothesis of agency problem.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
12
Chapter 3
Research Methodology and Data Information
This chapter is divided into two sections. The first section presents the econometric methodology applied in this research for empirical analysis. The second section introduces the current development of T-REITs market, describes the data used in empirical tests, and performs preliminary analyses by means of descriptive statistics and time-series graphs.
3.1 Research Methodology
7In order to detect the existence of long-run equilibrium relationship between REITs and direct real estate, we employ cointegration test proposed by Johansen (1988). If there exists a cointegration relationship between these two variables, we could analyze the short-term relation by estimating Vector Error Correction Model (VECM). If there is no cointegration relationship, however, we should examine the interrelation between the variables through Vector Autoregressive (VAR) model.
Finally, Granger causality test is applied in this research to clarify the lead-lag relation between REITs and direct real estate.
3.1.1 Cointegration
The concept of cointegration was first introduced by Engle and Granger (1987).
According to Engle and Granger’s original definition, cointegration refers to variables that are integrated of the same order. More specifically, if a time series is non-stationary, it could become stationary after taking d time difference, which means to be integrated of the d order, i.e., a I(d) variable. When two non-stationary time series are integrated of the same order and a linear combination relationship of them is stationary, the time series are cointegrated. In other words, there exists a long-run equilibrium relationship between the variables. Engle and Granger detect whether variables are cointegrated by testing the stationary of the residuals. If the residuals are
7 The econometric methods applied in this research are referred to Enders (2004), p. 264–310; 320–372.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
13
stationary, then the two variables are said to be cointegrated. If the residuals are non-stationary, however, then the two variables are not cointegrated.
However, the Engle and Granger cointegration approach still have several important defects. First, the results of cointegration test may be contrasting depending on the choice of the variable selected for normalization. In other words, the results may not be consistent. Second, when using three or more variables in cointegration tests, we expect that there may be more than one cointegrating vector. This approach, however, has no systematic procedure for indicating multiple cointegration relationships. Finally, since the Engle and Granger procedure relies on a two-step estimator, any error introduced by the researcher in Step 1 is carried into Step 2.
Therefore, Johansen cointegration test is employed in this research, which can avoid aforementioned problems.
The Johansen cointegration approach is a maximum likelihood estimation of a fully specified error correction model, which is based on VAR model. This method is more robust for interpreting the multiple long-run equilibrium relationship between variables. Assuming a VAR model of order p and n variables can be expressed as:
X= AX+ AX+ ⋯ + AX+ ε (1) where: X= the (n．1) vector (X, X, ⋯ , X );
ε= an independently and identically distributed n-dimensional vector with zero mean and variance matrix ∑
After adding and subtracting AX to the right-hand side, we can continue in this fashion to obtain
ΔX = πX+ ∑πΔX
+ ε (2)
where π = −I − ∑ A
and π = − ∑ A
The key feature to note in equation (2) is rank of the matrix π, which is equal to the number of independent cointegrating vectors. If rank π = 0, the matrix is null and equation (2) is the usual VAR model in first difference. If rank π = 1, the system exists a single cointegrating vector.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
14
The number of distinct cointegrating vectors can be obtained by checking the significance of the characteristic roots of π. In practice, we can obtain only estimates ofπ and its characteristic roots. In order to determine whether there exists cointegration relationship, we can test the number of characteristic roots by using the following two test statistics:
λ$%&' r = −T∑ $ln (1 − λ*) (3)
λ+%,(r, r + 1) = −Tln(1 − λ*$) (4) where: T = the number of usable observations;
λ*= the estimated values of the characteristic roots (i.e. eigenvalues) obtained from the estimated Π matrix
The trace statistic tests the null hypothesis that the number of cointegrating vectors is less than or equal to r. On the other hand, the maximum eigenvalue statistic tests the null hypothesis that the number of cointegrating vectors is equal to r.
3.1.2 Vector Error Correction Model
A critical characteristic of cointegrated variables is that their time paths are influenced by the extent of any deviation from long-run equilibrium. After all, if the system is to return to long-run equilibrium, the movements of at least some of the variables must respond to the magnitude of the disequilibrium. Hence, if cointegration relationship exists between two series, according to Granger representation theorem, an error correction term must be added to correct the short-term dynamics influenced by the deviation from the long-run relationship. VECM is a special form of VAR model for I(1) that are cointegrated, making the variables move toward to the direction of long-run equilibrium. To examine the relationship between cointegration and error correction, it is important to study the properties of the simple VAR model:
Y= aY+ aZ+ ε0 (5) Z= aY− aZ+ ε1 (6) where ε0 and ε1 are white-noise disturbances that may be correlated with each other and, for simplicity, intercept terms have been ignored.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
15
To ensure that the variables are cointegrated of order (1,1), we must place following restrictions on the coefficients of equation (5) and (6):
a= 2(1 − a) − aa3/(1 − a) (7)
a > −1 (8)
aa+ (a) < 1 (9) To see how these coefficient restrictions bear on the nature of the solution, write equation (5) and (6) as
7∆Y
∆Z8 = 7a− 1 a
a a− 18 7Y
Z8 + 9ε0
ε1: (10)
After a bit of manipulation, equation (10) can be written in the form
∆Y = −2aa/(1 − a)3Y+ aZ+ ε0 (11)
∆Z = aY− (1 − a)Z+ ε1 (12) Equation (11) and (12) form an error-correction model. If both a and a
differ from zero, we can normalize the cointegrating vector with respect to either variables. Normalizing with respect to Y, we get
∆Y = α0(Y− βZ) + ε0 (13)
∆Z = α1(Y− βZ) + ε1 (14) where: α0= −aa/(1 − a);
β = (1 − a)/a; α1= a
Notice that α0 and α1 have the interpretation of speed of adjustment parameters. The larger α0 is, the greater the response of to the previous period’s deviation from long-run equilibrium. At the opposite extreme, very small values of α0 imply that the short-term of the variable Y is unresponsive to last period’s equilibrium. If both α0 and α1 are equal to zero, the long-run equilibrium relationship does not appear and the model is not one of error-correction or cointegration.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
16
3.1.3 Vector Autoregressive Model
If those series are not cointegrated, the vector autoregressive model is a general framework to explore the dynamic interrelationships among economic variables. All the variables in a VAR model are treated symmetrically. In particular, each variable has an equation explaining its evolution based on its own lags and the lags of all the other variables in the model. In this case, VAR model can identify the lags short-term impact on the dependent variable by analyzing the correlation between the lags of the dependent variable and of other variables. Therefore, this study applies the VAR approach to examine the reactions of REIT to direct real estate and the reactions of direct real estate to REIT.
In the bivariate case, we can let the time path of Y be affected by current and past realizations of the Z sequence and let the time path of Z be affected by current and past realizations of the Y sequence. Based on this concept, we estimate a VAR in the standard form:
Y= α;+ α<Y+ α<Z<+ e (15)
Z= α;+ α<Y+ α<Z<+ e (16)
It is assumed that (1) both Y and Z are stationary; (2) the error term (i.e. e
and e) are composites of the two shocks ε> and ε?.
In addition, there are two useful techniques employed by VAR analysis to understand the interrelationship between variables. One is impulse response function which can quantify and graphically depict the time path of the short-term impact varies under the long-run fluctuations. In other words, it will present how the variables react to shocks. The other is variance decomposition which allows us to assess the relative contributions of different shocks to the forecast error variance, that is, it will be informative to present the sources of volatility.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
17
3.1.4 Granger Causality
In addition to cointegration test, we can gain some additional insights into the interrelation between two series by performing Granger causality tests both of REIT on direct real estate and of direct real estate on REIT. The main purpose of this methodology is to examine the existence of lead-lag relations between two variables.
In other words, it can investigate the ability of one series to predict another, conditional on its own past value.8 For instance, if current and past value of Y is helpful to forecast future values of Z, it is said that Y does Granger cause Z, alternatively called Y leads Z. Moreover, if there is an interaction between the two variables, then the result indicates the feedback relation between variables.
Suppose two variables in VAR model are stationary, but does not have a cointegration relationship, the Granger causality equation is defined as:
∆Yt= α0+ ∑ αp i∆Zti
i1 + ∑ βp j∆Ytj
j1 + εt, (17)
where Y is the dependent variable; Z is independent variable, p is lag terms. The null hypothesis is α = α = ⋯ ⋯ = αA = 0. If the results reject the null hypothesis that Z sequence does not lead Y sequence, then the inclusion of Z sequence in the equation is useful in predicting Y sequence.
If there is a cointegration between two variables, the result of causality test would be biased by using equation (17) directly. In order to avoid the distortion, the deviation from the long-run equilibrium level should be taken into consideration.
Hence, we employ VECM to estimate by adding error correction term λµB into the above VAR model, becoming equation (18).
∆Y= α;+ ∑ α ∆Z+ ∑ β ∆Y+ λµB+ ε (18)
8 Such causality, based on predictability, is not to be confused with causality based on cause and effect, which can only be tested by performing controlled experiments (Myer and Webb, 1993).
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
18
3.2 Data Information
This section is divided into three subsections, including the introduction of T-REITs market, data source and data analysis. The first subsection introduces the current development of T-REITs market and general information of REITs launched in Taiwan. Next, the source and information of data applied in this study are described in
This section is divided into three subsections, including the introduction of T-REITs market, data source and data analysis. The first subsection introduces the current development of T-REITs market and general information of REITs launched in Taiwan. Next, the source and information of data applied in this study are described in