Article Critique 2
angeliahunter
AnInternationalExaminationoftheDynamicLinkagesacrossREITMarkets.pdf
104 REAL ESTATE FINANCE WINTER 2018
Randy I. Anderson, PhD, CRE, is President of Griffin Capital Asset Management in el Segundo, CA. He may be contacted at randerson@ griffincapital.com.
Vivek Bhargava, PhD, CFA, is Associate Dean, Faculty and Administration & Professor of Finance for the Lutgert College of Business at Florida Gulf Coast University in Fort Myers, FL. He may be contacted at vbhargava@ fgcu.edu.
Akash Dania, PhD, is Associate Professor and Director of Graduate Business Programs at Alcorn State University in Natchez, MS. He may be reached at [email protected].
An International Examination of the Dynamic Linkages across REIT Markets By Randy I. Anderson, Vivek Bhargava, and Akash Dania
Individual and institutional investors increasingly are interested in includ-ing real estate equities in the form of REITs to their mixed asset portfolios. In the United States, the REIT index at the end of 2010 had a market capitalization of $390 billion, which is up nearly 10 fold from 1994. Domestically, REITs trade on all of the major stock exchanges and there is developing a publicly registered but non- traded REIT sector as well. Investors have been and continue to be attracted to REITs because REITs typically pay a relatively high dividend yield, and historically have had strong total return performance. REITs generally are considered to have a low cor- relation with stocks and bonds and hence have portfolio stabilizing features. Numerous studies have shown that the efficient frontier is enhanced by adding securities with the characteristics of REITS to their portfolios.1 The growth in publicly traded real estate
securities from other countries around the globe has been exponential as well.2 The ratio- nale for this growth is effectively the same as it is here in the United States in that public real estate allows many investors, both large and small, to add a real estate allocation to their mixed asset portfolio. From a US investor standpoint, an important question now becomes, should I include international real estate securities above and beyond my domestic allocation? The findings in Conover, Friday, and Sirmans3 would say yes as they find that foreign real estate has a lower cor- relation to US stocks than either US REITs
or international general equities. However, several recent studies have cast some doubts on the efficacy of including international REITs. For example, Zhou and Anderson4 show that volatility spikes between stock and REIT markets occur at nearly the same time. In other words, when there is a shock to the markets, the domestic real estate and domestic stock markets feel the shock in the same direc- tion and at the same time. They show that this event is globally synchronized as the shocks across countries in both equities and real estate occur across borders at the same time. Perhaps more importantly, the shocks, especially to the downside, actually were more acute for real estate than for general equities, this suggests that there actually may be some portfolio destabilizing effects of adding domestic or international REITs to mixed asset portfolios. Additionally, Anderson, Bony, and Guirguis5 find that unexpected shocks in monetary pol- icy have a much greater propensity to create volatility for real estate than for general equi- ties, once again suggesting potential additional risk and destabilizing effects. In this article, we go far beyond the
prior studies to understand the interre- lationships between the performances of numerous global REIT markets. In particu- lar, we perform VAR, GARCH, TGARCH, and MGARCH modeling to determine the return linkages, volatility spillover effects, and covariance between these global markets overtime. We divide the data into two groups: (1) major world REIT indices (Developed Market, European Union, and Far East); and
An International Examination of the Dynamic Linkages across REIT Markets
WINTER 2018 REAL ESTATE FINANCE 105
(2) major countries REIT indices (Australia, Belgium, Canada, France, Germany, Japan, Netherlands, and United Kingdom). Our initial findings show high correlations between the US market and the European Union, the Developed Markets composite index and with Canada, the Netherlands, and the United Kingdom among country indices. The low correlations were found with markets in the Far East and Japan. We go on, utilizing an unrestricted vector autoregressive analysis (VAR) and find that US REIT returns impact the returns in most markets and in fact Granger cause them. While there is some impact found from other markets on US REIT returns, the magnitude is much smaller. We then utilize a GARCH (Generalized AutoRegressive Conditional Heteroskedasticity) framework to test for volatility spill- overs from foreign countries to the United States and find that significant spillovers occur, increasing US volatility from nearly every studied market. TGARCH (Threshold GARCH) confirms these findings and shows the results to be asymmetric in that negative shocks abroad impact domestic REITs to a greater extent than good news. Finally, we utilize MGARCH (Multivariate GARCH) and find once again that the major developed markets are highly integrated with the US REIT market, except for the Far East, Japan, and Australia. In all, our study casts doubts for US investors to utilize REITs from many of the major developed markets in their mixed asset port- folios. However, the results emphasize that markets such as those found in the Far East might provide significant diversification opportunities and enhance portfolio risk- adjusted returns.
LITERATURE REVIEW The international real estate equities market has realized
significant growth over the past decade. Accordingly, the literature is now exploring issues and opportunities relating to these markets. The lion’s share of the work has focused on volatility, correlation analysis, returns, and risk-adjusted returns in these markets relative to the US REIT market. For example, Bond, Karolyi, and Sanders6 examine the index returns across countries and show significant variabil- ity in the average returns and standard deviations of those returns in the counties. They, however, do show evidence of a common or global factor that is present in all indices suggesting some existence of interdependence. Michayluk, Wilson, and Zurbruegg7 specifically focus on the relation- ship between the UK and US real estate markets and show
that the performance in one market certainly can influence the performance in another. They find that when there is an exogenous negative shock, the correlations between the two markets become even greater. Certainly, this reduces the benefits of diversification when it is needed the most. Zhou and Anderson8 examine issues of value at risk and
expected shortfalls for the four largest REIT markets. Their findings are consistent with those in Michayluk, Wilson, and Zurbruegg.9 In particular, they find that volatility spikes are highly synchronized within a country between the equity and REIT markets and in fact they are highly synchronized across countries. Perhaps more alarming is that the value at risk and expected shortfalls for all countries actually increases with the addition of REITs. Conover, Friday, and Sirmans10 use correlation analysis and efficient frontier construction techniques to examine the impact of adding both foreign equity and foreign real estate to the mixed asset portfolio. Their study comes to a differing conclusion in that foreign stock and foreign real estate have tremen- dous portfolio stabilizing characteristics. In fact, from a US investor perspective, adding foreign real estate is better than adding foreign stock or domestic REITs. In all, the jury is still out on the portfolio role of REITs
and international real estate on mixed assets portfolios and how REITs across countries interact with each other. This article fills this void by being the first to test these issues with timely data, numerous markets, and robust statistical techniques.
DATA DESCRIPTION AND ANALYSIS The purpose of this article is to investigate the dynamic
relations and volatility spillover among Global REITs, requiring we select major global return series that have been continuously tracked over the sample period. We have divided these series into two parts: (1) major world REIT indices and (2) individual country REIT indices. The vari- able of interest is the US real estate investment trust price index (USREIT) tracking changes in the value of the US real estate investment trust market. The major world indices used are
• Developed Market real estate investment trust price index (DVREIT) tracking changes in the value of the Developed Market real estate investment trust markets.
• European Union real estate investment trust price index (EUREIT) tracking changes in the value of the
An International Examination of the Dynamic Linkages across REIT Markets
106 REAL ESTATE FINANCE WINTER 2018
European mainland (other than the United Kingdom) real estate investment trust market.
• Far East real estate investment trust price index (FEREIT) tracking changes in the value of the real estate investment trust markets based in the Far East (representations from Hong Kong, South Korea, Indonesia, Malaysia, Singapore, Philippines and Thailand).
The major country indices used in the study include Australia (AUREIT), Canada (CAREIT), Belgium (BERIET), France (FRREIT), Germany(GRREIT), Netherlands (NTREIT), Japan (JPREIT), and UK (UKREIT). The above series indices are a quality proxy for the overall
performance of the global REIT market. These are among the largest in market size, depth and capitalization, and are thus an appropriate sample set for the study. All of the aforementioned data come from Thomson
Financials Datastream® real estate sector price index data. We use daily data from June 2003 to August 2008. This five year data-set is used because it is during this period that the real estate sector has gained increasingly widespread recog- nition in the public capital markets. The real estate market has displayed a considerably increased level of volatility during this period, especially since the latter half of 2006.
Therefore this time horizon sets itself as an ideal labora- tory to understand the dynamics and drivers of volatility of returns in the real estate sector, and the impact of US REIT returns on global REIT returns. Exhibit 1 illustrates the summary statistics for the returns
on the sample REIT indices. From this exhibit it can be observed that the average daily returns on all the REIT markets except that of Australia are positive. However, these average returns are low. Further, all minimums are negative and all maximums are positive. The low average return for the indices may be explained by the strong performance of the REITs during the early portion of the time horizon of the data-set followed by a dramatic downturn from the latter half of 2006 onwards, swamp- ing the period of positive return performance. The skew- ness for Developed Markets, Far East, Australia, Belgium, Netherlands, Japan, and the US markets indices is negative indicating that the negative shocks in these series have a more pronounced effect than positive shocks. Also, kurto- sis for all series exceeds 3, indicating that the distribution are peaked (leptokurtic) relative to the normal which further justifies a time series analysis. Exhibit 2 gives the correlation matrix for the returns of
the various index returns. We are interested in the correla- tion between United States and various other REIT indi- ces. As can be seen, there is very low correlation among the
EXHIBIT 1—DESCRIPTIVE STATISTICS
This shows descriptive statistics for variables of interest. Included are Developed Markets (DV), European Union (EU), Far East (FE), Australia (AUS), Canada (CA), Belgium (BEL), France (FRA), Germany (GER), Netherlands (NET), Japan (JP), United Kingdom (UK) and the United States (US).
Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis
DV 0.0004 0.0007 0.0420 –0.0392 0.0089 –0.2852 5.2195
EU 0.0002 0.0001 0.0641 –0.0554 0.0120 0.2541 7.2733
FE 0.0002 0.0004 0.0637 –0.0829 0.0126 –0.3050 7.3559
AUS –0.0001 0.0000 0.0637 –0.1208 0.0120 –0.9677 15.4424
CA 0.0003 0.0000 0.0554 –0.0497 0.0099 0.2140 6.4185
BEL 0.0005 0.0008 0.0602 –0.0571 0.0113 –0.1699 5.6326
FRA 0.0009 0.0000 0.0953 –0.0605 0.0165 0.4838 6.3582
GER 0.0029 0.0000 0.3677 –0.1916 0.0336 1.7425 19.0307
NET 0.0003 0.0006 0.0527 –0.0579 0.0111 –0.2446 6.6686
JP 0.0001 0.0000 0.0650 –0.0782 0.0127 –0.3765 8.9309
UK 0.0003 0.0004 0.1040 –0.0630 0.0141 0.2491 7.5833
US 0.0004 0.0002 0.0795 –0.0737 0.0138 –0.0830 7.1125
An International Examination of the Dynamic Linkages across REIT Markets
WINTER 2018 REAL ESTATE FINANCE 107
EXHIBIT 2—CORRELATION MATRIX
This shows the correlation matrix for variables of interest. Included are Developed Markets (DV), European Union (EU), Far East (FE), Australia (AUS), Canada (CA), Belgium (BEL), France (FRA), Germany (GER), Netherlands (NET), Japan (JP), United Kingdom (UK), and the United States (US). DV EU FE AU CA BEL FRA GER NET JP UK US DV 1 EU 0.5561 1 FE 0.2368 0.1262 1 AU 0.3499 0.2198 0.3030 1 CA 0.4493 0.2559 0.0612 0.1364 1 BEL 0.5363 0.7632 0.1551 0.2359 0.2590 1 FRA 0.3055 0.5477 0.0208 0.1478 0.1533 0.3835 1 GER 0.1746 0.2535 0.0496 0.1072 0.0887 0.1960 0.1989 1 NET 0.4919 0.6361 0.0559 0.2420 0.2471 0.8731 0.4264 0.2029 1 JP 0.2516 0.1127 0.8930 0.3544 0.0840 0.1240 0.0910 0.0732 0.1334 1 UK 0.5540 0.6199 0.0107 0.1966 0.2696 0.5502 0.3910 0.2250 0.6260 0.0810 1 US 0.8418 0.2313 –0.0138 0.0091 0.3463 0.1852 0.1356 0.0692 0.2318 0.0219 0.2951 1
EXHIBIT 3—CONDITIONAL VARIANCE FOR MAJOR WORLD REIT INDICES
These graphs show conditional variance for variables of interest. Included are Developed Markets (DevMkts), European Union (EU), Far East (FE)
.00000
.00005
.00010
.00015
.00020
.00025
.00030
.00035
.00040
2003 2004 2005 2006 2007 2008 .0000
.0002
.0004
.0006
.0008
.0010
2003 2004 2005 2006 2007 2008
.0000
.0002
.0004
.0006
.0008
.0010
.0012
.0014
2003 2004 2005 2006 2007 2008
Conditional Variance - Dev Mkts
Conditional Variance - FE
Conditional Variance - EU
An International Examination of the Dynamic Linkages across REIT Markets
108 REAL ESTATE FINANCE WINTER 2018
United States and other sample indices except Developed Markets. Correlation between the United States and Far East is negative. Overall there is evidence of low or negative
correlation among the United States and other indices in the sample. The correlation between the United States and developed markets is very high and that can be attributed
EXHIBIT 4—CONDITIONAL VARIANCE FOR VARIOUS COUNTRIES MARKETS
These graphs show conditional variance for variables of interest. Included are Developed Markets (DevMkts), European Union (EU), Far East (FE), Australia (AUS), Canada (CA), Belgium (BEL), France (FRA), Germany (GER), Netherlands (NET), Japan (JP), United Kingdom (UK) and the United States (US).
.0000
.0002
.0004
.0006
.0008
.0010
.0012
.0014
2003 2004 2005 2006 2007 2008 .0000
.0001
.0002
.0003
.0004
.0005
.0006
2003 2004 2005 2006 2007 2008
.0000
.0004
.0008
.0012
.0016
.0020
2003 2004 2005 2006 2007 2008 .0000
.0001
.0002
.0003
.0004
.0005
.0006
.0007
.0008
.0009
2003 2004 2005 2006 2007 2008
.0000
.0001
.0002
.0003
.0004
.0005
.0006
.0007
2003 2004 2005 2006 2007 2008 .0000
.0004
.0008
.0012
.0016
.0020
.0024
2003 2004 2005 2006 2007 2008
Conditional Variance - AUS Conditional Variance - CA
Conditional Variance - JP Conditional Variance - U.K.
Conditional Variance - BEL Conditional Variance - FRA
An International Examination of the Dynamic Linkages across REIT Markets
WINTER 2018 REAL ESTATE FINANCE 109
EXHIBIT 4—CONTINUED
.000
.005
.010
.015
.020
.025
.030
2003 2004 2005 2006 2007 2008 .0000
.0001
.0002
.0003
.0004
.0005
.0006
.0007
.0008
2003 2004 2005 2006 2007 2008
.0000
.0002
.0004
.0006
.0008
.0010
.0012
.0014
2003 2004 2005 2006 2007 2008
Conditional Variance - GER Conditional Variance - NET
Conditional Variance - U.S.
to the fact that the US REIT market is a major component of the developed markets index. Exhibit 3 shows the conditional variance graphs for
Major World REIT indices, i.e., Developed Markets (DEV MKTS), European Union (EU), and the Far East (FE). Exhibit 4 shows the conditional variance graphs for vari- ous countries REIT markets, i.e., United Kingdom (UK), Japan (JP), Canada (CA), Australia (AU), Belgium (BEL), Germany (GER), France (FRA), and the Netherlands (NET). From these graphs it can be observed that all sample markets, except those for Canada, France, and Germany display marked volatility during the latter half of the sample period. This is consistent with the global downturn in real estate sector from 2007 onwards. Given the observations of variable regimes of volatility, it is important to test for sta- tionary characteristics in the included sample series. Next, to test for existence of a unit root, we perform
the Phillip-Perron test11 and Augmented Dickey Fuller
(ADF) test as it is necessary to determine whether the variables of interest are integrated in the same order. Stationary data should be used for the analysis because non-stationary data can lead to spurious results, especially when conducting an impact analysis. The null hypothesis is that the series is non-stationary or has a unit root. We conduct the test for two different states: (1) intercept and (2) intercept with trend. Unit root tests are presented in Exhibit 5. Irrespective of specification used in the Phillip- Perron test and Augmented Dickey-Fuller test the null hypothesis is rejected. Therefore it can be said that three series are stationary in nature and do not have unit roots and we can safely carry out the intended time-series analysis.
METHODOLOGY The purpose of this article is to understand the inter-
actions between the return and volatility of US REITS
An International Examination of the Dynamic Linkages across REIT Markets
110 REAL ESTATE FINANCE WINTER 2018
with REITS from various global markets. We also are interested in understanding whether during periods of high volatility such as today, if the volatility from global markets spills over to the domestic REIT returns. We use a GARCH (1, 1) model to determine whether there is any volatility spillover and TGARCH model to deter- mine whether the spillover is asymmetric. We also use multivariate GARCH models to see whether there are any co-movements in the conditional variances and cova- riance of the markets. We use an unrestricted vector autoregression (VAR)
analysis to analyze whether there is any dependency of Global REIT index returns on the returns of the US REITs or their own previous returns and vice versa. The VAR approach used in this study was developed by Sims.12 This model provides a multivariate framework where changes in a particular variable are associated to the changes in its own lags and to changes in other vari- ables in the model. The VAR treats all variables as jointly
endogenous and imposes no a-priori restrictions on the structural relationship between the variables of interest. This makes the VAR approach appropriate for under- standing unknown or relatively unexplored relationships like the ones examined here. Because the variables of interest have been found stationary in nature, we can safely apply this approach without concern over spurious results. Specification for VAR is as follows:
REITt = α1 + ∑ k
j= 1 β1 jREITt- j + ∑
k
j= 1 γ1jXt-j + ε1t (1a)
Xt = α2 + ∑ k
j= 1 β2 j Xt- j + ∑
k
j= 1 γ2j REITt-j + ε2 t (1b)
Where, REITt is the return on individual sample Global REIT index, Xt is the return on the US REIT index, αi are the constants, βi j are coefficients for the lagged regres- sors of the dependent variable, γi j are the coefficients for the lagged independent variable, and εi t are the random error terms. The Akaike information criterion (AIC) and Schwarz information criterion (SIC) of the VAR model were employed to identifying the appropriate lag lengths for our analysis. To test for volatility spillover effects, we utilize a GARCH
approach. GARCH models enable estimation of the vari- ance of the sample series at a particular point of time. GARCH is more parsimonious and avoids model over- fitting, which accordingly, is less likely to violate non-neg- ativity constraints.13 We use a GARCH (1, 1) model to test for volatility spillover effect. A GARCH (1, 1) allows the conditional variance to be dependent not only on its past conditional variance, but also past innovations. We can test for the existence of volatility spillovers from global REIT indexes to US REIT by extracting individual residuals from their returns and inserting them as the independent vari- ables in the conditional volatility equation for US REITs. Hence, the basic GARCH (1, 1) specification can be given as following:
yt = c + τyt-1 + εt , εt∼N (0,σt2) (2)
ht = α0 + α1ε2t-1 + β1 ht -1 (3)
Equation (2) is the mean equation and Equation (3) is the conditional variance equation. yt is the return on the REIT index, c is the intercept, yt-1 is the previous period
EXHIBIT 5—UNIT ROOT TESTS
Included are variables of interest in this study; i.e., test results for Developed Markets (DV), European Union (EU), Far East (FE), Australia (AU), Canada (CA), Belgium (BEL), France (FRA), Germany (GER), Netherlands (NET), Japan (JP), United Kingdom (UK) and the United States (US).
PP ADF
Intercept Trend and Intercept Intercept
Trend and Intercept
DV –30.276 –30.356 –30.747 –30.85
EU –37.747 –37.867 –24.354 –37.375
FE –37.081 –37.361 –36.821 –36.904
AU –35.618 –35.769 –35.493 –35.528
CA –33.428 –33.441 –33.584 –33.596
BEL –36.048 –36.048 –36.021 –36.060
FRA –39.869 –40.561 –39.634 –39.696
GER –35.028 –35.118 –34.640 –34.868
NET –36.710 –36.799 –36.710 –36.783
JP –34.41 –34.558 –34.498 –34.573
UK –38.697 –39.336 –38.146 –38.265
US –39.381 –39.621 –38.624 –38.661
1% level –3.435 –3.965 –3.435 –3.965
5% level –2.863 –3.413 –2.863 –3.413
10% level –2.568 –3.129 –2.568 –3.129
An International Examination of the Dynamic Linkages across REIT Markets
WINTER 2018 REAL ESTATE FINANCE 111
REIT index return and ε t is the white noise error term. ht is the conditional variance. Equation (3) gives the basic volatility model that captures heteroscedasticity in REIT returns, but to incorporate the spillover effects from Global REIT returns, the residuals are extracted from these global REIT indices and are incorporated as regressors in the volatility equation. Therefore, the specification for the GARCH (1, 1) spillover equation is as follows:
ht(Spillover) = α0 + α1ε2t-1 + β1 ht -1 + ψξ2t-1 (4)
Where α0 > 0, β1 ≥ 0, α1 ≥ 0. ε2t-1 is the lagged squared shock of the US REIT index and provides the news about volatility from the previous period. It is measured as the lag of the squared residual from the mean equation, and ξ2t-1 is the lagged squared shock extracted from the global REIT indices. The coefficient ψ represents the volatility spillover coefficient measuring the extent and behavior of the vola- tility spillover effect. Despite the apparent success of GARCH parameteriza-
tion, this model has been criticized to be a symmetric vari- ance process where the sign of the disturbance is ignored and hence it fails to capture the asymmetric news effect. This is because of the fact that the shocks are squared and hence are forced to be positive. The asymmetric effect (or leverage effect) reflects the usual observation that down- ward movements in returns are followed by higher vola- tilities than upward movements of the same magnitude. To overcome this shortcoming, a TGARCH (p, q) approach is used as the model captures the leverage effect in quadratic form. The specification for a TGARCH (1, 1) is given as following:
yt = c + τyt-1 + εt (5)
ht = α0 + β1ht -1 + α1ε2t-1 + α2ε2t-1dt-1 (6)
Where ε2t-1 is the lagged square residual of US REIT returns, and dt is the dummy variable such that dt=1 if εt-1<0 (bad news) and dt=0 if εt-1>0 (good news). α2 is the coefficient that takes leverage effect into consideration. In this model, if α2 is positive, it implies that a negative shock or bad news has a higher impact on volatility than when compared to positive shock. Similarly if it is negative, the volatility will increase more with a positive shock, than it would with a negative shock. To capture the spillover effect
using TGARCH model the conditional volatility equation takes the following form
ht = α0 + β1ht -1 + α1ε2t-1 + α2ε2t-1dt-1 + ψ1ξ2t-1 + ψ
2 ξ2t-1dt-1 (7)
Where ξ2t-1 is the lagged squared residual from Global REIT returns. ψ2 captures the leverage effect on the vola- tility of the US REIT index from other REITS. If ψ2 is positive, then a large negative shock or bad news in the return of other REITs will result in the higher volatility of US REIT returns than compared to positive shocks or good news. One of the problems with univariate garch models is
that they model only the return and the volatility of the series. If one is interested in modeling and understanding the co-movements between the series, one has to employ Multivariate GARCH models. To that effect in this article we use the Diagonal VECH and the constant conditional correlation (CCC) MGARCH models. We use these mod- els to keep the number of variables low. For example in the bivariate VECH (1, 1) model 21 parameters need to be estimated where as in DVECH (1, 1) only nine parameters need to be estimated. This makes the model restrictive in the sense that spillover effects cannot be measured, but we already have measured the spillover effects in univariate GARCH and TGARCH models. We are interested in the co-movements from the MGARCH models. A basic MGARCH model with time varying means, vari-
ances and co-variances is given below.
yt = Ht 1/2ηt (8)
Where yt is a N × 1 matrix and Ht is a N × N matrix of y and ηt such that E ηt' ηt = I. For the VECH model
VECH (Ht) = (h11t, h21t, h22t, h31t, ….. hNNt)’ (9a)
or VECH(Ht ) = c + ∑ p
j= 1 Aj VECH(εt −j ε't −j)
+ ∑ q
j= 1 Bj VECH(Ht −j ) (9b)
A simplified version of the model is Diagonal VECH model but it does not allow for interactions between the
An International Examination of the Dynamic Linkages across REIT Markets
112 REAL ESTATE FINANCE WINTER 2018
variances and covariances and hence does not allow for spillover effects. For a bivariate DVECH (1, 1) model, the Ht is specified as below
h11t = c1 + a11ε21,t−1 + b11h11,t−1
h12t = c2 + a12ε1,t−1 ε2,t−1 + b12h12,t−1
h22t = c3 + a22ε22,t−1 + b22h22,t−1 (10)
As can be seen from above, only nine parameters need to be estimated. h11, h12, h22 represent the conditional volatil- ity of the US REIT market, the conditional covariance between the United States and the global market under consideration, and the conditional variance of the global market. The h12 represent the conditional covariance which depends on the past co-movement of shocks and the past conditional covariance. We next employ the constant conditional correla-
tion (CCC) multivariate GARCH model proposed by Bollerslev.14 This model is used to see the conditional correlation between the US and the global REIT mar- kets. In this model, first, univariate GARCH models are estimated for each asset, and then the correlation matrix is estimated. This is done by transforming the residuals using their estimated conditional standard deviations. The assumption of constant correlation makes estimating a large model feasible and ensures that the estimator is posi- tive definite, simply requiring each univariate conditional variance to be non-zero and the correlation matrix to be of full rank. The conditional variance covariance is given by
Ht = DtRtDt
Dt = diag (h11t 1/2 …. HNNt
1/2)
Rt = R=(ρij) = ρii = 1 (that is CCC)
hijt = (hiit hijt) 1/2 for all i ≠ j (11)
EMPIRICAL ANALYSIS Exhibit 6 provides the results for the VAR analysis. The
advantage of a VAR analysis is that it acts as a system. A VAR model provides a multivariate framework where changes in a particular variable are associated to the changes in its own lags and to changes in other variables
in the model. The VAR treats all variables as jointly endogenous and imposes no a-priori restrictions on the structural relationship between the variables of interest. This makes the VAR approach appropriate for under- standing unknown or relatively unexplored relationships like the ones examined here. The Akaike information criterion (AIC) and Schwarz information criterion (SIC) of the VAR model identify the appropriate lag lengths to be four. We can analyze the impact of US REIT returns on
Global REIT indices and vice versa, besides understanding whether these markets impact their own returns. We first analyze the impact of US REIT returns on Major World REIT indices. From this exhibit it can be observed that US REIT
returns do impact the sample Major REIT indices REIT returns. For the case of Developed Markets (DV), US REITs displays a positive (and significant) impact, for three lags. DV market’s own lags (i.e., lag two and three) have a negative impact followed by a positive and significant lag four impact. So the Developed market seems to display a tendency to go up, immediately after it has observed a negative performance. The impact of US REIT returns on other Global samples is observed to be positive, significant, and what may be termed as persistent. Results for European Union (EU) REIT show a lag one and two dependen- cies on US REIT returns that are positive and significant. Similarly for Far East (FE) a positive and significant depen- dence is observed for lags one, two, and three. This tells us that when the REIT index returns in the United States go up, they are followed by the REIT returns going up in the major world indexes. Next we evaluate the impact of other markets on the
US market when US REIT returns are the dependent variable in the VAR analysis. US REITs demonstrate an impact by the DV (lag one which is positive and signifi- cant), EU (lag one which is positive and significant, albeit weak), and a delayed response to FE (positive and signifi- cant in lag two and a negative and significant during lag 4). A negative response indicates that the US REIT index moves up when the FE index moves down. This could be an indication of cross border investment transfer from one index to another based on investor perception of indi- vidual world markets. Ease with which global investors can invest in US REITs has transpired the willingness of global investors in REIT assets that they perceive would provide higher returns.
An International Examination of the Dynamic Linkages across REIT Markets
WINTER 2018 REAL ESTATE FINANCE 113
Exhibit 7a and Exhibit 7b report the VAR results for various countries REIT markets and the US REIT market returns. From these exhibits we can say that the majority of sample countries REIT indices returns demonstrate a posi- tive and significant dependence on the US REIT returns. For example, UK (one lag) Japan (two lags), Canada (four lags), Australia (four lags) and Germany (one lag) demon- strate an immediate positive and significant dependence to the US REIT returns. France has a delayed (lag two) dependence on the US REIT returns which is positive and significant. However, Belgium and Netherlands do not pro- vide evidence of a dependence on the US REIT returns. These findings have interesting implications for these REIT markets that tend to follow the performance trend
of the US REIT market returns. Though REIT returns are known to follow domestic economic conditions, global performance and demand for underlying properties and income thereof seems to explain the observed linkages among Global REITs. One cannot deny a possible link between individual countries domestic monetary policy, consumer confidence, and REIT performance. We also analyze the impact of the Global REIT markets
on US REITs. Though, intuitively, one would consider US REITs as a bigger market than other Global markets, given the nature of REITs as an asset class and how accessible Global markets are, it is important to understand whether there is an impact of Global REITs on US REIT perfor- mance. Like equity markets, investors are faced with choice
EXHIBIT 6—VAR RESULTS FOR MAJOR WORLD REIT INDICES
This exhibit shows the VAR analysis results for the US and World REIT indices. Included are Developed Markets (DV), European Union (EU), Far East (FE), and the United States (US). US (–1), US (–2), US (–3), US (–4) are lagged variables for the US REIT, IR (–1), IR (–2), IR (–3), IR (–4) are lagged variables for individual major world indices.
US DV US EU US FE
US (–1) –0.058 0.118794*** –0.057858** 0.197659*** –0.048128** 0.238173***
[–1.04113] [ 3.39617] [–2.06058] [ 8.28265] [–1.77879] [ 9.98525]
US (–2) 0.014 0.056462* –0.032 0.04068* 8–0.032687* 0.058769**
[ 0.25338] [ 1.57401] [–1.11814] [ 1.65045] [–1.16562] [ 2.37720]
US (–3) 0.047 0.092842*** –0.021 –0.005 –0.048576** 0.055212**
[ 0.83661] [ 2.59070] [–0.73634] [–0.19041] [–1.73179] [ 2.23276]
US (–4) –0.042 –0.019 –0.02 0.025 –0.039272* 0.011
[–0.81303] [–0.57412] [–0.68118] [ 1.03116] [–1.39838] [ 0.43755]
IR (–1) 0.018294** 0.023 0.051719* –0.070685** 0.013 –0.021
[ 0.20922] [ 0.42297] [ 1.56808] [–2.52159] [ 0.42309] [–0.76800]
IR (–2) –0.088 –0.083009* –0.038 0.061237** 0.061182** –0.073682***
[–1.00752] [–1.50674] [–1.15792] [ 2.17600] [ 1.97828] [–2.70247]
IR (–3) –0.123667* –0.14943*** –0.038 –0.023 0.04 –0.034
[–1.42052] [–2.71499] [–1.15591] [–0.80373] [ 1.29997] [–1.25614]
IR (–4) 0.069 0.067655* 0.013 –0.055652** –0.08125*** –0.002
[ 0.88497] [ 1.36925] [ 0.40520] [–2.03995] [–2.71279] [–0.05943]
C 0 0 0 0 0 0
[ 1.08968] [ 1.37957] [ 1.06621] [ 0.30407] [ 1.07919] [ 0.44608]
***, **, and * are significance levels at 1%, 5%, and 10%, respectively.
An International Examination of the Dynamic Linkages across REIT Markets
114 REAL ESTATE FINANCE WINTER 2018
E X
H IB
IT 7
A —
V A
R R
E S
U LT
S F
O R V
A R
IO U
S C
O U
N T
R IE
S R
E IT
M A
R K
E T
S
T hi
s sh
ow s
th e V
A R
a na
ly sis
r es
ul ts
fo r
th e
U S
an d
W or
ld R
E IT
in di
ce s.
In cl
ud ed
a re
U ni
te d
K in
gd om
( U
K ),
Ja pa
n (J
P) , C
an ad
a (C
A ),
A us
tr al
ia (
A U
) an
d th
e U
ni te
d St
at es
( U
S) . U
S (–
1) , U
S (–
2) , U
S (–
3) , U
S (–
4) a
re la
gg ed
v ar
ia bl
es fo
r th
e U
S R
E IT
, I R
( –1
), IR
( –2
), IR
( –3
), IR
( –4
) ar
e la
gg ed
v ar
ia bl
es fo
r in
di vi
du al
m aj
or w
or ld
in di
ce s.
U
S U
K U
S JP
U S
C A
U S
A U
U S
(– 1)
–0 .0
47 *
0. 23
56 29
** *
–0 .0
45 52
6* 0.
25 62
99 **
* –0
.0 61
96 **
0. 13
26 46
** *
–0 .0
45 89
8* 0.
37 42
64 **
*
[– 1.
65 92
7] [
8. 27
40 4]
[– 1.
68 21
9] [
10 .7
01 4]
[– 2.
13 40
9] [
6. 48
58 0]
[– 1.
68 97
3] [
17 .5
85 3]
U S
(– 2)
–0 .0
29 0.
00 5
–0 .0
23 0.
05 35
16 **
–0 .0
29 0.
04 17
17 **
–0 .0
14 0.
05 90
73 **
[– 0.
97 21
0] [
0. 16
47 3]
[– 0.
82 93
0] [
2. 14
46 7]
[– 0.
96 90
1] [
1. 99
10 9]
[– 0.
45 97
0] [
2. 50
02 9]
U S
(– 3)
–0 .0
31 –0
.0 24
–0 .0
54 68
6* *
0. 03
2 –0
.0 23
0. 03
91 86
** –0
.0 31
0. 06
87 75
** *
[– 1.
05 53
5] [–
0. 82
51 3]
[– 1.
94 04
0] [
1. 29
88 0]
[– 0.
75 73
2] [
1. 86
99 3]
[– 1.
02 52
2] [
2. 90
89 9]
U S
(– 4)
–0 .0
35 –0
.0 08
–0 .0
37 86
5* 0.
02 3
–0 .0
48 60
1* 0.
03 85
65 **
–0 .0
18 0.
03 63
03 *
[– 1.
18 65
0] [–
0. 27
63 9]
[– 1.
34 33
6] [
0. 92
65 3]
[– 1.
64 54
6] [
1. 85
35 7]
[– 0.
59 48
7] [
1. 53
50 8]
IR (
–1 )
0. 00
1 –0
.1 04
41 1*
** –0
.0 26
0. 03
74 61
* 0.
06 70
25 *
0. 02
1 –0
.0 50
06 6*
0. 00
0
[ 0.
01 83
4] [–
3. 65
91 3]
[– 0.
83 92
4] [
1. 36
88 8]
[ 1.
62 29
9] [
0. 73
26 9]
[– 1.
44 49
4] [
0. 01
17 7]
IR (
–2 )
–0 .0
1 –0
.0 17
0. 07
46 38
** –0
.0 53
44 9*
* –0
.0 38
–0 .0
48 27
6* 0.
00 2
–0 .0
57 11
**
[– 0.
35 97
2] [–
0. 60
45 0]
[ 2.
41 21
3] [–
1. 95
19 1]
[– 0.
91 10
7] [–
1. 66
10 7]
[ 0.
04 63
9] [–
2. 10
98 0]
IR (
–3 )
–0 .0
02 –0
.0 2
0. 04
08 04
* –0
.0 36
81 6*
–0 .0
42 –0
.0 65
68 **
–0 .0
16 –0
.0 79
19 8*
**
[– 0.
08 60
2] [–
0. 70
56 7]
[ 1.
31 60
5] [–
1. 34
17 9]
[– 1.
02 33
4] [–
2. 26
16 5]
[– 0.
46 46
5] [–
2. 92
63 4]
IR (
–4 )
0. 03
1 –0
.0 37
–0 .0
91 35
1* **
0. 00
6 0.
11 54
99 **
* 0.
00 0
0. 01
1 0.
01 1
[ 1.
09 34
3] [–
1. 32
40 9]
[– 3.
06 54
8] [
0. 21
21 3]
[ 2.
86 73
8] [
0. 01
25 7]
[ 0.
36 21
1] [
0. 45
46 4]
C 0.
00 00
0. 00
00 0.
00 00
0. 00
00 0.
00 00
0. 00
00 0.
00 00
0. 00
00
[
1. 05
33 5]
[ 0.
72 68
2] [
1. 09
35 0]
[ 0.
06 80
7] [
1. 01
91 6]
[ 0.
77 28
2] [
1. 02
49 0]
[– 0.
95 00
4]
* **
, * *,
a nd
* a
re s
ig ni
fic an
ce le
ve ls
at 1
% , 5
% , a
nd 1
0% , r
es pe
ct iv
el y.
An International Examination of the Dynamic Linkages across REIT Markets
WINTER 2018 REAL ESTATE FINANCE 115
E X
H IB
IT 7
B —
V A
R R
E SU
LT S
F O
R V
A R
IO U
S C
O U
N T
R IE
S R
E IT
M A
R K
E T
S
T hi
s sh
ow s
th e V
A R
a na
ly sis
r es
ul ts
fo r
th e
U S
an d
W or
ld R
E IT
in di
ce s.
In cl
ud ed
a re
B el
gi um
( B
E L)
, G er
m an
y (G
E R
), Fr
an ce
( FR
A ),
N et
he rla
nd s
(N E
T )
an d
th e
U ni
te d
St at
es (
U S)
. U S
(– 1)
, U S
(– 2)
, U S
(– 3)
, U S
(– 4)
a re
la gg
ed v
ar ia
bl es
fo r
th e
U S
R E
IT , I
R (
–1 ),
IR (
–2 ),
IR (
–3 ),
IR (
–4 )
ar e
la gg
ed v
ar ia
bl es
fo
r in
di vi
du al
m aj
or w
or ld
in di
ce s.
U S
B E L
U S
G E R
U S
F R
A U
S N
E T
–0 .0
57 95
8* *
0. 20
88 22
–0 .0
46 26
4* 0.
18 72
84 *
–0 .0
49 89
2* 0.
15 47
91 –0
.0 55
49 1*
* 0.
20 20
09
[– 2.
09 07
4] [
9. 45
16 0]
[– 1.
70 12
9] [
2. 88
50 6]
[– 1.
81 86
3] [
4. 78
95 7]
[– 1.
98 11
6] [
9. 25
44 5]
–0 .0
40 09
9 0.
07 24
11 –0
.0 31
50 6
0. 03
49 08
–0 .0
33 23
1 0.
06 94
63 **
–0 .0
40 11
0. 07
66 28
[– 1.
39 08
3] [
3. 15
13 0]
[– 1.
15 15
0] [
0. 53
44 6]
[– 1.
19 66
7] [
2. 12
33 2]
[– 1.
37 72
1] [
3. 37
61 3]
–0 .0
30 15
3 0.
01 79
57 –0
.0 36
57 5
–0 .0
20 38
2 –0
.0 33
64 6
–0 .0
00 95
4 –0
.0 35
52 9
–0 .0
00 85
6
[– 1.
04 36
4] [
0. 77
98 3]
[– 1.
33 63
7] [–
0. 31
19 6]
[– 1.
21 03
6] [–
0. 02
91 2]
[– 1.
21 54
0] [–
0. 03
75 6]
–0 .0
28 96
5 0.
03 70
02 –0
.0 24
86 6
0. 02
04 38
–0 .0
22 04
7 0.
04 98
32 –0
.0 40
39 9
0. 03
38 47
[– 1.
01 00
5] [
1. 61
89 5]
[– 0.
90 90
8] [
0. 31
30 0]
[– 0.
79 45
2] [
1. 52
43 9]
[– 1.
39 36
9] [
1. 49
83 0]
0. 08
15 11
** –0
.0 42
36 –0
.0 05
09 7
0. 03
84 96
0. 01
78 19
–0 .0
90 29
8* **
0. 05
03 66
–0 .0
69 43
6* *
[ 2.
34 66
2] [–
1. 53
01 0]
[– 0.
44 97
4] [
1. 42
30 9]
[ 0.
76 70
4] [–
3. 29
95 9]
[ 1.
40 04
7] [–
2. 47
74 5]
–0 .0
32 30
7 –0
.0 07
24 9
0. 00
42 32
0. 03
77 89
–0 .0
14 24
2 0.
04 22
27 –0
.0 01
84 7
–0 .0
14 53
8
[– 0.
92 73
8] [–
0. 26
11 0]
[ 0.
37 31
6] [
1. 39
60 1]
[– 0.
61 02
8] [
1. 53
59 8]
[– 0.
05 12
0] [–
0. 51
71 9]
–0 .0
36 05
0. 01
11 99
0. 00
67 7
0. 02
61 64
0. 00
43 35
0. 00
40 35
–0 .0
19 90
8 0.
02 99
96
[– 1.
03 97
3] [
0. 40
52 8]
[ 0.
60 18
0] [
0. 97
42 9]
[ 0.
18 61
5] [
0. 14
70 9]
[– 0.
55 49
8] [
1. 07
30 2]
0. 07
37 51
** –0
.0 70
57 9*
** –0
.0 15
85 4
0. 02
47 09
–0 .0
22 36
2 –0
.0 75
95 9*
** 0.
09 77
4* **
–0 .0
68 22
3* *
[ 2.
20 15
7] [–
2. 64
35 3]
[– 1.
41 58
1] [
0. 92
43 7]
[– 0.
97 13
3] [–
2. 80
07 6]
[ 2.
81 94
4] [–
2. 52
52 4]
0. 00
03 66
0. 00
03 94
0. 00
04 27
0. 00
21 7*
* 0.
00 04
12 0.
00 09
17 **
0. 00
03 73
0. 00
02 27
[ 0.
97 94
9] [
1. 32
22 8]
[ 1.
12 94
7] [
2. 40
22 1]
[ 1.
09 52
1] [
2. 06
85 6]
[ 0.
99 90
7] [
0. 78
08 1]
** *,
* *,
a nd
* a
re s
ig ni
fic an
ce le
ve ls
at 1
% , 5
% , a
nd 1
0% , r
es pe
ct iv
el y.
An International Examination of the Dynamic Linkages across REIT Markets
116 REAL ESTATE FINANCE WINTER 2018
of where to invest, given their perception of growth and performance. Ease of cross border investment in security alternatives have made investment choices such as REITs interconnected globally. On this issue, we observe mixed results. For example, Japan, Far East, Dev Markets, and Canada are the only markets where we observe a signifi- cant impact on US REITs (significance of 5%, 1%). Japan is positive and significant for lag two and lag three, and nega- tive and significant for lag four. Far East is positive and sig- nificant for lag two and negative and significant for lag four. Belgium is positive and significant for lag one and lag four. Dev Markets is positive and significant for lag one and three, and Canada has positive and significant impact for lag one and four. For the AU and EU markets a weak significant impact also is observed. However, we can safely say that the
US REITs seem to have a bigger impact on Global REIT performance than vice versa. A Granger causality also was analyzed (results not reported).
US grangers cause every Global REIT market at 99 percent level (p value less than .01). However, none of the Global markets Granger cause United States except for Japan at 90 percent. We use GARCH approach to investigate whether there
is any volatility spillover from Global REIT markets to US REITs. In order to achieve this, we estimate a GARCH (1, 1) (Equations 2 and 3) on all global REIT indices and extract the residuals to introduce in the variance equation of US REITs. From the basic GARCH model we find that both the GARCH term, i.e., β and the ARCH term, i.e., α are significant for all series, world indices as well as major countries REIT indices. Furthermore, α+β is very close to 1 in all series except developed markets, indicating that volatility shocks in these series are persistent.15 The results for developed markets could be because US markets are a large part of that index. We now estimate the GARCH model by introducing
the residuals in the global REIT indices in the variance equation of US REITs. Exhibit 8 reports the estimation results for this volatility analysis for major world indi- ces and Exhibit 9 gives the results for major countries. Ψ represents the volatility spillover coefficient. It can be observed from Exhibits 8 and 9 that α1 and β1 coefficients are significant at for all indices indicating evidence of ARCH and GARCH structures. Except for the European Union in major indices and France in individual coun- tries, Ψ is positive and significant for all other global REIT markets. This implies that the volatility shocks originating
EXHIBIT 8—VOLATILITY SPILLOVER USING GARCH FOR MAJOR WORLD REIT INDICES
Coefficients Developed Markets
European Union Far East
C 0.0011a (0.00)
0.0008a (0.001)
0.0008a (0.0008)
α0 1.18E-05 a
(0.00) 1.25E-06b
(0.0159) 8.71E-07c
(0.0934)
α1 0.0443 b
(0.0256) 0.0911a
(0.00) 0.0814a
(0.00)
β1 0.1058 a
(0.00) 0.902707a
(0.00) 0.8977a
(0.00)
ψ 1.8362a (0.00)
0.0050 (0.1792)
0.0232a (0.0003)
a, b, and c are significance levels at 1%, 5%, and 10%, respectively. P-values are given in ( )
EXHIBIT 9—VOLATILITY SPILLOVER USING GARCH FOR VARIOUS COUNTRIES REIT MARKETS
Coefficients Australia Canada Belgium France Germany Netherland Japan UK
C 0.0008a (0.0013)
0.0008a (0.001)
0.0008a (0.001)
0.0008a (0.000)
0.0004 (0.580)
0.0009a (0.00)
0.0008a (0.0008)
0.0008a (0.001)
α0 1.45E-06 b
(0.0366) 8.00E-07
(0.1612) 9.11E-07
(0.120) 1.02E-06c
(0.089) 0.0001a
(0.002) 8.77E-06a
(0.00) 1.54E-06a
(0.0078) 1.65E-06b
(0.0346)
α1 0.0949 a
(0.00) 0.0953a
(0.00) 0.091a
(0.00) 0.0932a
(0.00) 0.131a
(0.009) 0.043a
(0.00) 0.0830a
(0.00) 0.1001a
(0.00)
β1 0.8806 a
(0.00) 0.8930a
(0.00) 0.899a
(0.00) 0.9045a
(0.00) 0.556a
(0.000) 0.037a
(0.00) 0.8960a
(0.00) 0.8574a
(0.00)
ψ 0.0353a (0.003)
0.0217a (0.0021)
0.022b (0.043)
0.0012 (0.306)
–0.002a (0.00)
0.891a (0.00)
0.0202a (0.0003)
0.0372a (0.00)
a, b, and c are significance levels at 1%, 5%, and 10%, respectively. P-values are given in ( )
An International Examination of the Dynamic Linkages across REIT Markets
WINTER 2018 REAL ESTATE FINANCE 117
from major world indices such as the developed markets and far east, as well as from countries such as Australia, Canada, Belgium, Germany, Netherlands, Japan, and United Kingdom carry a spillover effect on domestic REITs, and these effects are positive, i.e., return shocks or innovations originating in these markets increase the volatility in US REIT returns.
Exhibit 10 and Exhibit 11 give the estimates for the TGARCH model. This model not only investigates whether the effect of the shock is from its returns is asym- metric (α2), but also whether the effects of shocks from the other return series are asymmetric (ψ2). ψ1 gives the spillover effect from the global REITs and ψ2 captures the asymmetric effect of shocks or innovations from the global REITs markets. α2 is significant for European Union in Exhibit 10, and for Australia, Belgium, France, Germany, Netherlands, Japan, and United Kingdom in Exhibit 11, indicating a presence of leverage effect, i.e., that the impact of past innovations has been asymmetric. The coefficient is positive indicating that the volatility increases more with bad news when compared with good news of the same magnitude. ψ2 indicates whether the spillover effect from world REIT markets is asymmetric. For developed markets, Australia, Germany, and Netherlands, the coefficient is not significant, which indicated that the previously observed spillover effect is not asymmetric. For European Union and Fareast among major indices and Canada, Belgium, France, Japan, and United Kingdom among major countries, ψ2 is observed as positive and significant. This indicates that the volatility spillover effect from these countries is asymmetric and, as such, a negative shock or bad news in these markets is likely to increase the volatility in domestic REIT returns. The observed results for European Union and France are interesting. For the GARCH estimation, we observed no spillover effect however; with TGARCH estimation
EXHIBIT 10—VOLATILITY SPILLOVER USING TGARCH FOR WORLD REIT INDICES
Coefficients Developed Markets
European Union Far East
C 0.001232a (0.00)
0.00077a 0.0043
0.000777a 0.0044
α0 1.19E-05 a
(0.00) 1.84E-06a 0.0014
1.92E-06a 0.0054
β1 0.10793 a
(0.00) 0.893673a
(0.00) 0.886596a
(0.00)
α1 0.042557 c
(0.081) 0.067236a
(0.00) 0.06383a
(0.00)
α2 0.002562 (0.940)
0.047898b 0.0245
0.038441c (0.080)
ψ1 1.639075 a
(0.00) –0.01047a (0.00)
0.001705 (0.853)
ψ2 0.377357 (0.2065)
0.04085a (0.0004)
0.043581b (0.0147)
a, b, and c are significance levels at 1%, 5%, and 10%, respectively. P-values are given in ( )
EXHIBIT 11—VOLATILITY SPILLOVER USING TGARCH FOR VARIOUS COUNTRIES REIT MARKETS
Coefficients Australia Canada Belgium France Germany Netherland Japan UK
C 0.000694b (0.0127)
0.000869a (0.0014)
0.0006b (0.011)
0.0006b (0.012)
0.0006b (0.047)
0.0008a (0.00)
0.000767a 0.0051
0.000753a 0.0068
α0 2.01E-06 a
(0.005) 1.45E-06b
(0.017) 1.18E-06b
(0.039) 1.67E-06b
(0.013) 4.99E-05a
(0.00) 1.39E-05a
(0.00) 2.67E-06a 0.0002
1.80E-06a 0.008
β1 0.884487 a
(0.00) 0.887806a
(0.00) 0.904a
(0.00) 0.895a
(0.00) 0.400a
(0.00) 0.011b
(0.015) 0.884683a
(0.00) 0.874786a
(0.00)
α1 0.058911 a
(0.0025) 0.082345a
(0.00) 0.054a
(0.000) 0.068a
(0.000) 0.273a
(0.00) –0.011a (0.006)
0.061799a (0.0001)
0.060166a (0.0008)
α2 0.053046 b
(0.0223) 0.01659
(0.477) 0.048b
(0.018) 0.049b
(0.019) 0.170a
(0.007) 0.065a
(0.00) 0.040939c
(0.056) 0.04728b
(0.039)
ψ1 0.008913 (0.6629)
–0.01422b (0.1433)
–0.005 (0.726)
–0.003 (0.205)
–0.0008a (0.00)
0.832a (0.00)
–0.00543 (0.535)
0.009828 (0.271)
ψ2 0.043695 (0.1316)
0.087176a (0.0001)
0.064b (0.046)
0.013c (0.059)
–0.0006 (0.826)
0.094 (0.585)
0.053259a (0.0028)
0.042017a (0.0092)
a, b, and c are significance levels at 1%, 5%, and 10%, respectively. P-values are given in ( )
An International Examination of the Dynamic Linkages across REIT Markets
118 REAL ESTATE FINANCE WINTER 2018
EXHIBIT 12—BIVARIATE DVECH (1, 1) ESTIMATION RESULTS WITH US AND GLOBAL REIT INDICES
US/FE US/EU US/Dev Mkt
C11 1.3E-06 a
(0.01) 9.5E-07a
(0.01) 2.5E-06a
(0.00)
C12 1.3E-06 (0.53)
2.9E-07b (0.03)
1.7E-06a (0.00)
C22 2.8E-06 a
(0.00) 1.1E-06a
(0.00) 1.8E-06a
(0.00)
A11 0.091 a
(0.00) 0.121a
(0.00) 0.071a
(0.00)
A12 0.037 c
(0.08) 0.066a
(0.00) 0.063a
(0.00)
A22 0.115 a
(0.00) 0.109a
(0.00) 0.087a
(0.00)
B11 0.907 a
(0.00) 0.889a
(0.00) 0.912a
(0.00)
B12 0.494 (0.38)
0.933a (0.00)
0.914a (0.00)
B22 0.872 a
(0.00) 0.888a
(0.00) 0.886a
(0.00)
a, b, and c are significance levels at 1%, 5%, and 10%, respectively. P-values are given in ( )
EXHIBIT 13—BIVARIATE DVECH (1, 1) ESTIMATION RESULTS WITH US AND VARIOUS REIT COUNTRIES
US/UK US/Japan US/BL US/FR US/GR US/NT US/Canada US/Australia
C11 1.8E-06 a
(0.001) 1.4E-06a
(0.01) 1.65E-06a
(0.003) 1.7E-06a
(0.003) 1.52E-06a
(0.005) 1.39E-06a
(0.000) 1.6E-06a
(0.002) 1.4E-06a
(0.008)
C12 1.3E-07 (0.18)
–2.4E-07 (0.85)
9.82E-08 (0.34)
1.43E-07 (0.248)
1.16E-06 (0.325)
1.11E-06a (0.000)
2.3E-07c (0.1)
1.1E-07 (0.43)
C22 1.0E-06 a
(0.01) 1.1E-06a
(0.00) 3.82E-07a
(0.00) 6.37E-05a
(0.00) 7.84E-05a
(0.00) 1.17E-06a
(0.000) 1.2E-06a
(0.00) 2.2E-07
(0.22)
A11 0.090 a
(0.00) 0.092a
(0.00) 0.095a
(0.00) 0.093a
(0.00) 0.093a
(0.00) 0.090a
(0.000) 0.084a
(0.00) 0.092a
(0.00)
A12 0.014 c
(0.09) 0.042
(0.88) 0.022b
(0.039) 0.020a
(0.005) 0.043b
(0.040) 0.091a
(0.000) 0.039a
(0.00) 0.005
(0.58)
A22 0.048 a
(0.01) 0.160a
(0.00) 0.043a
(0.00) 0.221a
(0.00) 0.180a
(0.00) 0.092a
(0.0) 0.072a
(0.00) 0.058a
(0.00)
B11 0.904 a
(0.00) 0.906a
(0.00) 0.899a
(0.00) 0.903a
(0.00) 0.902a
(0.00) 0.905a
(0.0) 0.911a
(0.00) 0.905a
(0.00)
B12 0.980 a
(0.00) 0.413
(0.51) 0.956a
(0.00) 0.975a
(0.00) 0.885a
(0.00) 0.906a
(0.0) 0.950a
(0.00) 0.971a
(0.00)
B22 0.949 a
(0.00) 0.849a
(0.00) 0.952a
(0.00) 0.560a
(0.00) 0.774a
(0.00) 0.907a
(0.0) 0.918a
(0.00) 0.946a
(0.00)
a, b, and c are significance levels at 1%, 5%, and 10%, respectively. P-values are given in ( )
an evidence of volatility spillover from EUREIT over to domestic REITs is observed. We also observe evidence of the spillover to be asymmetric. Because TGARCH provides a more robust specification of the relationship, we accept the TGARCH findings. Finally, in this article, Multivariate GARCH is used to
determine the co-movements between the United States and the global REIT markets. Exhibit 12 and Exhibit 13 give the results from Bivariate DVECH MGARCH (1, 1) model. The analysis is performed with two variables, US and other global market index. In all the three pairs in Exhibit 12 and eight pairs in Exhibit 13, A11, A22, B11 and B22 are statistically significant and the sum of Aii and Bii is less than and close to one indicating that MGARCH is the right specification. Lagged shocks and lagged conditional volatilities impact the current conditional volatilities. A12 and B12, which are the coefficients for the conditional cova- riance equation are significant at 99 percent and positive for EU and developed markets among major world indices and for Canada, France and Netherlands among countries, indicating that the covariance between these markets and US REIT markets are impacted by past co-movements and covariances and these markets are highly integrated with the US market. For Belgium and Germany, A12 is significant
An International Examination of the Dynamic Linkages across REIT Markets
WINTER 2018 REAL ESTATE FINANCE 119
EXHIBIT 15—BIVARIATE CCC (1, 1) ESTIMATION RESULTS OF US WITH VARIOUS REIT COUNTRIES
US/UK US/Japan US/BL US/FR US/GR US/NT US/CN US/AUS
C11 1.8E-06 a
(0.002) 1.4E-06a
(0.008) 1.59E-06a
(0.004) 1.59E-06a
(0.003) 1.45E-06a
(0.006) 1.55E-06a
(0.001) 2.0E-06a
(0.00) 1.4E-06a
(0.007)
A11 0.095 a
(0.00) 0.092a
(0.00) 0.095a
(0.00) 0.094a
(0.00) 0.092a
(0.00) 0.066a
(0.00) 0.087a
(0.00) 0.092a
(0.00)
B11 0.899 a
(0.00) 0.905a
(0.00) 0.900a
(0.00) 0.901a
(0.00) 0.904a
(0.00) 0.926a
(0.00) 0.904a
(0.00) 0.905a
(0.00)
C22 1.2E-06 a
(0.009) 1.1E-06a
(0.00) 3.71E-07a
(0.000) 6.50E-05a
(0.00) 7.85E-05a
(0.00) 5.20E-07c
(0.098) 1.7E-06a
(0.00) 2.3E-07
(0.19)
A22 0.055 a
(0.00) 0.160a
(0.00) 0.044a
(0.00) 0.233a
(0.00) 0.179
(0.00) 0.078a
(0.00) 0.081a
(0.00) 0.059a
(0.00)
B22 0.942 a
(0.00) 0.848a
(0.00) 0.952a
(0.00) 0.547a
(0.00) 0.774
(0.00) 0.926a
(0.00) 0.904a
(0.00) 0.945a
(0.00)
ρ12 0.241 a
(0.00) 0.019
(0.493) 0.106a
(0.000) 0.132a
(0.00) 0.074c
(0.016) 0.826a
(0.00) 0.308a
(0.00) 0.042
(0.114)
a, b, and c are significance levels at 1%, 5%, and 10%, respectively. P-values are given in ( )
at 95 percent and B12 at 99 percent also showing impact of past co-movements and covariances. For United Kingdom A12 is significant at 90 percent and B12 at 99 percent show- ing that these markets also are integrated although the impact of shocks may not be as pronounced. For Japan and Far East the coefficients are not significant implying that in these markets, the shocks and previous covariance do not
impact present covariance and these markets are not inte- grated. The results for Australia are mixed where A12 is not significant, but B12 is significant. This can mean that shocks due to co-movement do not impact the covariance but lagged covariance does. In short, there is strong evidence of co-movements between the REIT markets of the United States with developed markets, United Kingdom, Canada, and the European Union. Exhibit 14 and Exhibit 15 give the results for Bivariate
CCC (1, 1) model. It can be seen that the sum of αii and βii are close to and less than one indicating a high level of persistence in conditional variances. The mean value of conditional correlation coefficients show that the US market is highly integrated with Developed markets and the European Union among major world indices and the United Kingdom, Belgium, France, Germany, Netherlands, and Canada among country REIT indices. Such level of conditional correlation and integration is not found with Far East, Japan, and Australia.
SUMMARY AND CONCLUSIONS In this study, we go well beyond the existing literature
to determine the interrelations between the US REIT market and other major Global REIT markets and assess these results in the context of portfolio construction. The major findings of our study suggest that Global REIT markets are demonstrating signs of convergence with the
EXHIBIT 14—BIVARIATE CCC (1, 1) ESTIMATION RESULTS OF US WITH GLOBAL REIT INDICES
US/FE US/EU US/Dev Mkt
C11 1.4E-06 a
(0.008) 1.5E-06a
(0.005) 1.9E-06a
(0.00)
A11 0.093 a
(0.00) 0.095c
(0.00) 0.095a
(0.00)
B11 0.905 a
(0.00) 0.901a
(0.00) 0.898a
(0.00)
C22 2.8E-06 a
(0.00) 1.0E-06a
(0.001) 1.1E-06a
(0.00)
A22 0.117 a
(0.00) 0.098a
(0.00) 0.097a
(0.00)
B22 0.870 a
(0.00) 0.897a
(0.00) 0.890a
(0.00)
ρ12 0.015 (0.576)
0.197a (0.00)
0.858a (0.00)
a, b, and c are significance levels at 1%, 5%, and 10%, respectively. P-values are given in ( )
An International Examination of the Dynamic Linkages across REIT Markets
120 REAL ESTATE FINANCE WINTER 2018
domestic REIT market in the United States and that the US REIT market really is a driver of all the global REIT markets. In all, our study casts doubts on US investors’ ability to benefit from adding major developed markets REITs in their mixed asset portfolios. The results of this article show that the domestic real
estate market significantly influences world markets, and with few exceptions world markets do not lead domestic returns. The Garch and TGARCH estimations suggest that world markets increase volatility of domestic markets and that the additional volatility is asymmetric, more sig- nificant for negative surprises than for positive news. The results emphasize that newer REIT markets such
as those found in Canada are highly correlated with domestic markets and unlikely to provide significant diversification opportunities. The markets of the Far East, Japan, Germany and Belgium, which are less integrated might provide significant diversification opportunities and enhance portfolio risk-adjusted returns. Timing of asset allocation also is important. These findings have interesting implications for investors
and policy makers interested in REITs. Although REIT returns are known to follow domestic economic condi- tions, global performance and demand for underlying prop- erties and income thereof seems to explain the observed linkages among Global REITs. However, one cannot deny a possible link between
individual countries domestic monetary policy, con- sumer confidence and REIT performance. The linkages between monetary policy, consumer confidence and
domestic REIT performance and portfolio diversifica- tion are left as a future study. Future research can use these findings, such as the leading relationship between the US REIT market and others and the differing cova- riance structures to build portfolios that truly enhance the efficient frontier and provide investors with superior risk-adjusted returns.
NOTES 1. For example, see the studies of Grauer, R.R. and N.H. Hakansson, “Gains from
Diversifying into Real Estate: Three Decades of Portfolio Returns Based on the Dynamic Investment Model,” Real Estate Economics, (23) 117–159 (1995); and Muller, G.R., Pauley, K.R., and Morrill, W.K., “Should REITs be Included in a Mixed-Asset Portfolio?,” Real Estate Finance, 11(1), pp 23–28 (1994).
2. Mueller, G., Boney, V., and A. G. Mueller, “International Real Estate Volatility: A Tactical Investment Strategy,” Journal of Real Estate Portfolio Management, 14, Number 4, December 2008, 415-423.
3. Conover, C. M., Friday, H. S., and G. S. Sirmans, “Diversification Benefits from Foreign Real Estate Investments,” Journal of Real Estate Portfolio Management 8(1):17-25, 2002.
4. Zhou, J. and R. Anderson, “Extreme Risk Measures for International REIT Markets,” The Journal of Real Estate Finance and Economics, (2010), Forthcoming, Doi: 10.1007/ s11146-010-9252-5.
5. Anderson, R., Boney, V. and Guirguis, H., “The Impact of Switching Regimes and Monetary Shocks: An Empirical Analysis of REITs,” Journal of Real Estate Research, (2010). Forthcoming.
6. Bond, S. A., Karolyi G. A., and A. B. Sanders, “International Real Estate Returns: A Multifactor, Multicountry Approach,” Real Estate Economics, 31:3, 481–500 (2003).
7. Michayluk, D., Wilson, P. J., and R. Zurbruegg, “Asymmetric Volatility, Correlation and Returns Dynamics Between the U.S. and U.K. Securitized Real Estate Markets,” Real Estate Economics, American Real Estate and Urban Economics Association, vol. 34(1), pages 109-131 (2006).
8. Zhou and Anderson, supra n.4. 9. Michayluk, Wilson, and Zurbruegg, supra n.7. 10. Conover, Friday, and Sirmans, supra n.3. 11. Phillips, P.C.B, and Perron, P., “Testing for a Unit Root in Time Series regression,”
Biometrika, 75, 335-46 (1998). 12. Sims, C., “Macroeconomic and reality,” Econometrica, 48, 1-49 (1980). 13. Enders, W., Applied Econometrics Time Series, John Wiley and Sons Inc. (2009). 14. Bollerslev, T., “Modelling the Coherence in Short-run Nominal Exchange Rates:
A Multivariate Generalized ARCH Model,” The Review of Economics and Statistics, vol. 72(3), 498-505 (1990).
15. Basic GARCH(1,1) results are not reported as the focus is on spillover effect, but are available on request.
Reproduced with permission of copyright owner. Further reproduction prohibited without permission.
