Chapter 4 Empirical Results
4.2 Results of Taiwan
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
35
4.2 Results of Taiwan 4.2.1 Structural Change
The results of CUSUM test are presented in Figure 4-3 and 4-4. Figure 4-3 shows that the W (CUSUM quantity) of T-REIT do not exceed the critical value at the 5% significance level (dashed line). It appears that the T-REIT series does not have a structural change during the study period. While the W (CUSUM quantity) of CTP seems to break above the 5% significance line but not be significant, which is shown in Figure 4-4.
Figure 4-3 Result of CUSUM Test for T-REIT -30
-20 -10 0 10 20 30
II III IV I II III IV I II III IV I II III IV I II III IV
2006 2007 2008 2009 2010
CUSUM 5% Significance
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
36
Figure 4-4 Result of CUSUM Test for CTP
4.2.2 Unit Roots Test
Table 4-6 presents the results of unit root tests for the T-REIT and CTP series by conducting ADF and PP tests. Both ADF and PP tests indicate that the null hypothesis of the unit root cannot be rejected for the level of each series, i.e., these time series are not stationary. However, these two tests reject the null hypothesis for the series which take first difference and then appear stationary. Therefore, the results suggest that the T-REIT and CTP series are integrated of order one, denoted as I(1) series.
Table 4-6 Test for Unit Roots on TREIT and CTP
ADF test PP test
Level 1st difference Level 1st difference TREIT -2.2309 -4.7356 *** -1.6475 -4.7223 ***
CTP -1.6190 -9.5501 *** -3.2220 -12.2347 ***
Note: 1. The null hypothesis is that the series has a unit root.
2. *** denotes significance at the 1% level.
-30 -20 -10 0 10 20 30
II III IV I II III IV I II III IV I II III IV I II III IV
2006 2007 2008 2009 2010
CUSUM 5% Significance
‧
In this study, the optimal lag length is selected based on the Akaike information criterion (AIC) and the likelihood ratio (LR) test statistic recommended by Sims (1980). As shown in Table 4-7, the second lag is appropriate for the equations in the VAR model. Then we consider the lag length selected by the VAR model to
Note: * denotes the lag order selected by Akaike Information Criterion (AIC) and adjusted likelihood ratio (LR) value.
According to the trace and maximum eigenvalue tests, the null hypothesis of no cointegrating vector (r=0) cannot be rejected at the 5% significance level, which is reported in Table 4-8. The results of the cointegration test indicate that there is no cointegration relationship between T-REIT and CTP in the sample period, i.e., these two markets do not move together in the long run. It suggests that T-REIT price could not reflect the fundamentals of commercial real estate market. On the other hand, there should be diversification function by including both REITs and commercial real estate in the investment portfolio.
Table 4-8 Test Statistics for the Cointegration between TREIT and CTP
Null hypothesis：
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
38
4.2.4 Vector Autoregressive Model
Since there is no cointegration relationship between T-REIT and commercial transaction price, we therefore apply VAR model to explore the short-run interrelationship between these two markets. Table 4-9 reports the coefficient estimates of the VAR analysis on the T-REIT and CTP series. The T-REIT series exhibit strong autocorrelation at the 1% level while it does not display a significant economic relation with the past commercial transaction price. On the other hand, the CTP series is positively related to the first lag of T-REIT at the 5% level, and to the first lag of itself at the 1% level. These results support the argument regarding the better informational efficiency in T-REITs markets, i.e., T-REIT price rapidly and accurately reflect the market information.
Table 4-9 VAR Analysis on TREIT and CTP
Variales ∆TREIT ∆CTP
Coefficient t-statistic Coefficient t-statistic
∆TREIT (-1) 1.3610 *** 10.9190 0.0066 ** 1.7602
∆TREIT (-2) -0.4965 *** -3.9243 -0.0032 0.8468
∆CTP(-1) 4.4978 0.9676 0.4503 *** 3.2385
∆CTP(-2) -1.5178 -0.3425 0.1541 1.1619 C 1.6678 0.1101 1.2514 *** 2.7618
Note: ***and ** denote significance at the 1% and 5% level, respectively.
In addition to the VAR analysis, impulse response function and variance decomposition are helpful to analyze the dynamic relation between variables. Figure 4-5 shows the impulse response functions of T-REIT and CTP to both types of one-standard deviation shocks (alternatively called innovations). The effect of a T-REIT shock is to cause an immediate increase in T-REIT and CTP about three months. In particular, the degree of jump in T-REIT is larger than that of CTP. On the other hand, the effect of a CTP shock is to cause an immediate rise in price while it sharply drops and returns to its long-run value. The response of T-REIT to the CTP shocks, however, seems to be relatively insignificant. Since the system is stable, both sequences eventually converge to zero in about 20 months.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
39
Figure 4-5 Impulse Response of T-REIT and CTP -1
0 1 2 3 4 5
5 10 15 20 25 30 35 40 45 50 55 60
TREIT CTP
Response of TREIT to Cholesky One S.D. Innovations
-.02 .00 .02 .04 .06 .08 .10
5 10 15 20 25 30 35 40 45 50 55 60
TREIT CTP
Response of CTP to Cholesky One S.D. Innovations
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
40
The results of variance decomposition are presented in Table 4-10. It is apparent that the T-REIT shocks explain almost of the forecast error variance of T-REIT at any forecast horizon. This result suggests that the performance of T-REITs may not be significantly affected by transaction price of commercial real estate. On the other hand, the CTP shocks account for 94 percent of the forecast error variance of CTP initially while the contribution decreases to 69 percent in five months.
Table 4-10 Variance Decomposition Results
Variance Decomposition of TREIT Variance Decomposition of CTP
Period S.E. TREIT CTP Period S.E. TREIT CTP
1 2.93 100.00 0.00 1 0.09 5.85 94.15
5 8.28 99.17 0.83 5 0.13 30.62 69.38
10 9.14 98.94 1.06 10 0.14 38.85 61.15
15 9.17 98.92 1.08 15 0.14 39.39 60.61
20 9.17 98.92 1.08 20 0.14 39.41 60.59
25 9.17 98.92 1.08 25 0.14 39.41 60.59
30 9.17 98.92 1.08 30 0.14 39.41 60.59
35 9.17 98.92 1.08 35 0.14 39.41 60.59
40 9.17 98.92 1.08 40 0.14 39.41 60.59
45 9.17 98.92 1.08 45 0.14 39.41 60.59
50 9.17 98.92 1.08 50 0.14 39.41 60.59
55 9.17 98.92 1.08 55 0.14 39.41 60.59
60 9.17 98.92 1.08 60 0.14 39.41 60.59 Overall, the current T-REIT is only significantly affected by its past performance.
As discussed in previous part, the explanatory power of CTP to T-REIT is insignificant during the sample period. However, the current CTP is influenced by the past realization of T-REIT and itself. The explanatory power of T-REIT to CTP is almost 40 percent. This result suggests that T-REIT price seems to serve as a leading indicator to forecast the commercial real estate markets. In order to specifically detect the lead-lag relation between these two markets, we should conduct the following test, i.e., Granger causality test.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
41
4.2.5 Granger Causality Test
Based on the VAR analysis, we conduct Granger causality tests to investigate the possible linear causality between T-REIT and CTP, in which no cointegration relationship was found. In Table 4-11, the result cannot reject the null hypothesis, or T-REIT does not Granger cause CTP. It implies that commercial transaction price cannot be predicted by T-REIT performance. The possible explanation is that the REITs market in Taiwan might be too immature or inadequately capitalized to lead the commercial real estate market.
Table 4-11 Granger Causality Test Results
Independent Variable Dependent Variable
TREIT CTP
TREIT — 0.0631
CTP 0.7554 —
Note: 1. The table shows the p-values of the Granger causality tests.
2. The null hypothesis is that of no Granger causality.
Compared with the empirical results of the U.S., there are several possible explanations for the different results between these two REITs markets. The first reason is the difference in sample period. Since it has been only seven years since the first REIT was launched in Taiwan, the data available for empirical analysis is limited.
The results of cointegration test may be distorted due to the lack of observation. The second reason is the difference in market capitalization. The market capitalization of U.S. REITs is substantially greater than that of T-REITs. The long-term dynamics is likely to be insignificant as a result of the small-scale REITs market.
The third reason may be the difference in the concentrated risk. In contrast to the sound diversification in the U.S. REITs, T-REIT may confront the concentrated risk in terms of the type and location of REIT properties, which are mostly commercial office buildings in Taipei City. Hence, the performance and volatility of T-REITs seem to be influenced by the concentrated risk. Finally and most importantly, we suggest that the agency problem may exist in T-REITs markets. Since most T-REIT managements are the related parties of original owners or subsidiaries established by the parent companies. With the impact of agency problem, T-REIT price do not reflect the
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
42
fundamentals of commercial real estate market. Therefore, this study further attempts to explore this hypothesis in the following section, and proposes to improve the efficiency of T-REITs market.