Trend properties, cointegration, and diffusion of presale house prices in Taiwan: Can Taipei's house prices ripple out?

Habitat International 44 (2014) 432e441

Contents lists available at ScienceDirect

Habitat International journal homepage: www.elsevier.com/locate/habitatint

Trend properties, cointegration, and diffusion of presale house prices in Taiwan: Can Taipei's house prices ripple out? Ming-Te Lee a, Ming-Long Lee b, *, Shin-Hung Lin c a

Ming Chuan University, Taiwan National Dong Hwa University, Taiwan c National Yunlin University of Science and Technology, Taiwan b

a r t i c l e i n f o

a b s t r a c t

Article history: Available online

This study, which is the first to apply Vogelsang's (1998) t
PS1T test, shows that trends in presale house prices in Taiwan's largest urban areas are stochastic. Contrary to previous studies, this study's cointegration analysis clearly demonstrates that Taipei is not isolated from long-run price trends. Furthermore, as indicated by weak exogeneity tests, both Taipei and Kaohsiung are long-run leaders in house prices. Moreover, Granger causality test results reveal a strong lead-lag interdependence of house price changes among Taiwan's largest urban areas. These findings have important implications for housing policymakers, mortgage lending institutions, and property investors. © 2014 Elsevier Ltd. All rights reserved.

Keywords: House price Ripple effect Cointegration Global city Taipei Taiwan

Introduction Housing is many families' largest asset. Thus, house price dispersion among regions can cause a significant distortion in the wealth distribution (Shi, Young, & Hargreaves, 2009). Through the housing wealth effect, fluctuations in relative house prices have the potential to further influence relative regional economic activity (Holmes, 2007). Furthermore, those fluctuations could potentially influence labor mobility at the regional level due to housing affordability and relocation costs (Holmes, 2007). For these reasons, it is important for government policymakers to understand how regional house prices behave in relation to each other over time. The co-movements of regional house prices are also of interest to housing investors. A recent survey by Knight Frank LLP reports that investment is a key factor for housing acquisition in the U.K. (Knight Frank, 2007). Recent issues of the Institute for Physical Planning and Information's Housing Demand Survey report the same finding in the Taiwanese housing market (Chang & Lin, 2011). Specifically, in Taiwan, approximately 5% of housing sales were made to investors in 2002, a number that steadily increased to 16% in 2011 (Chang & Lin, 2011). Many studies in the real estate literature demonstrate that housing is an effective investment vehicle

* Corresponding author. E-mail addresses: [email protected] (M.-T. Lee), ming.long.lee@mail. ndhu.edu.tw (M.-L. Lee), [email protected] (S.-H. Lin). http://dx.doi.org/10.1016/j.habitatint.2014.09.003 0197-3975/© 2014 Elsevier Ltd. All rights reserved.

(Lee, 2008). Moreover, in 2006, the Chicago Mercantile Exchange introduced the first housing futures market in the U.S., trading housing futures contracts and options (Jud & Winkler, 2009). As a result of these significant developments, the phenomenon of house price diffusion across regions, known as the ripple effect, has long been an area of intense interest to researchers (Canarella, Miller, & Pollard, 2012; Clapp & Tirtiroglu, 1994; Luo, Liu, & Picken, 2007). In particular, the U.S. housing crash of 2007 has generated significant interest in exploring the ripple effect in Taiwan. Two studies examine Taiwan's market for secondhand housing, i.e., housing that is not newly built. Chien (2010) focuses on the longrun ripple effect of the regional/national house price ratio. Chen, Chien, and Lee (2011) examine the ripple effect in Taipei, New Taipei, Taichung, and Kaohsiung. To examine Taiwan's presale housing markets, Lee and Chien (2011) explore the long-run equilibrium among Taipei, New Taipei, TaoyuaneHsinchu, Taichung, and TainaneKaohsiung. Significantly, all of these studies document that Taipei, Taiwan's capital city, has long been isolated from Taiwan's other major urbanized areas in terms of house-price trends. These studies attribute this isolation to Taipei's status as a global city. However, this finding contrasts with the view of popular industrial and newspaper reports. These reports reveal increases in house prices in New Taipei, Taoyuan, Hsinchu, and even Taichung and Kaohsiung due to house price movements in Taipei (Cao, 2010; Liu, 2009; Ma, 2011; Taiwan Realty, 2010; Yu, 2010). In particular, located in northern Taiwan, Taipei is Taiwan's capital city and the country's most important economic center. As Taiwan is a relatively

M.-T. Lee et al. / Habitat International 44 (2014) 432e441

small and homogeneous country, house price movements in Taipei are likely to represent macroeconomic shocks that subsequently also affect other urbanized areas in Taiwan. Moreover, Taipei's house price movements are generally the focus of media reports on Taiwan's housing market. If individuals in other urban areas regard Taipei as a benchmark to evaluate what the “right” price change is, Taipei's house price changes should diffuse to these areas. In the face of contradictory information, this study aims to reexplore the issues of house price cointegration and diffusion in Taiwan's presale housing market. Similarly to forward or futures markets, presale housing markets are more likely to attract informed investors. Accordingly, a ripple effect caused by house prices in Taipei is likely to be more apparent in presale markets than in secondhand markets. Importantly, previous empirical studies on Taiwan devote little attention to the trend properties of house prices and neglect the consequence of misspecifying the deterministic components of a vector error correction (VEC) model. In contrast, the current study employs the Vogelsang (1998) t
PS1T test to formally justify the inclusion or exclusion of the linear deterministic time trend in an empirical model of Taiwan's house prices. Specifically, employing the Vogelsang (1998) t
PS1T test, unitroot tests, and a VEC model, this study seeks to answer the following question: In presale markets, is Taipei isolated from Taiwan's other major urban areas with respect to house price movements? The answer should interest Taiwanese government policymakers, property investors, and financial institutions. According to an Executive Yuan online poll, expensive housing in major cities, particularly in Taipei, is Taiwanese citizens' primary complaint (Wang, 2009). If Taipei's house prices can ripple out, then policies that are implemented in Taipei may either decrease or boost quickly rising house prices in other areas. Additionally, when planning regional development, those areas' governments must account for the influences of the ripple effect on local economic activity. If Taipei's house prices can ripple out, then investors who missed its housing boom may still have an opportunity to enter the market and invest in other areas that will experience future price surges (Real Estate Investar, 2012). However, if the ripple effect exists, it will be more difficult for financial institutions to diversify their mortgage-lending risks (Quigley & Van Order, 1991), and housing investors will find it more difficult to compile geographically diverse portfolios in Taiwan (Oikarinen, 2006). The study is also of interest to international property investorsdparticularly ethnic Chinesedbecause Taiwan is the dominant freehold-property market in the ethnic Chinese world. Taiwan has opened its property market to international investors. Since 1997, foreigners have been able to purchase property in Taiwan with official permission from the Department of Land Administration, and since 2001, based on the reciprocity principle, international investors have freely purchased properties in Taiwan (Lai & Fischer, 2007; Lee, 2009).1 That year, according to Department of Land Administration statistics, foreigners purchased approximately 60,000 m2 of land and 44,000 m2 of buildings. The numbers have increased dramatically since then. On average, foreigners have acquired an additional 582,000 m2 of land and 152,000 m2 of buildings every year from 2002 to 2010. Local developers and assetmanagement companies have also noticed and reported international investors' increasing interest international in Taiwan's housing markets (Pacific Assets Management Co. Ltd., 2011; Shining Group, 2011).

1 The reciprocity principle allows investors from more than 40 countries to invest in Taiwanese real estate. These countries include the U.S., Canada, France, Switzerland, the Netherlands, Malaysia, and Singapore, among others.

433

This study contributes to the ripple effect literature as follows. First, no previous study has applied the Vogelsang (1998) t
PS1T test to check for deterministic trends in house prices. Ripple effect studies often employ unit-root tests, cointegration tests, and vector error correction models, which are well known to be sensitive to the inclusion or exclusion of deterministic time trends (Ahking, 2002). As a result, decisions based on deterministic trends could lead to different patterns and conclusions about regional house price diffusion. Luo et al. (2007) show that house price diffusion patterns in state capital cities in Australia clearly depend on the inclusion or exclusion of deterministic time trends in cointegration models. However, no existing studies have performed a form test to justify their inclusion or exclusion of deterministic time trends. By applying the t
PS1T test, this study can offer more convincing evidence related to deterministic trends. Second, the ripple effect implies that over the long term, regional house prices tend to return to a long-run fixed relationship (Meen, 1999, 2001). However, there is no consensus in the literature about house price cointegration among global cities (i.e., alpha cities) and other cities over the long term. On the one hand, Dublin, Johannesburg, and Melbourne and Sydney are linked to other regions in Ireland, South Africa, and Australia, respectively (Balcilar, Beyene, Gupta, & Seleteng, 2013; Luo et al., 2007; Stevenson, 2004). On the other hand, Madrid and Taipei display unique house-price behavior that is isolated from other regions (Chen et al., 2011; Chien, 2010; Larraz-Iribas & Alfaro-Navarro, 2008; Lee & Chien, 2011). Moreover, studies reveal contradictory evidence with respect to house price trends in London (Cook, 2003; Holmes & Grimes, 2008; Macdonald & Taylor, 1993). By offering new empirical evidence, this study enhances the understanding of house price behavior in global cities. The remainder of the paper is organized as follows. The next section describes Taiwan's major urban areas, along with the source and description of our data. The next section, which addresses trend properties and unit root tests, presents the t
PS1T test, its results, and the results of unit root tests. The VEC model section constructs the model and discusses the cointegrating vector and leadelag relationships. The final section provides our conclusions. Taiwan's major urban areas This study investigates whether house prices diffuse among the following major urban areas, shown in Fig. 1: Taipei, New Taipei, TaoyuaneHsinchu, Taichung, Tainan, and Kaohsiung. Located in northern Taiwan, Taipei is Taiwan's capital city and its political, economic, and cultural center. In 1967, Taipei was declared the country's first direct-controlled municipality. Since the 1990s, due to its highly diversified economy, Taipei has been transformed into the node of a high-technology knowledge center with the status of a regional global city (Wang, 2003). In 2008, the Globalization and World Cities Research Network classified Taipei as an alpha global city. Taipei is surrounded by its suburb, New Taipei, which is the most populous city in Taiwan and has been a directcontrolled municipality since December 2010. The TaoyuaneHsinchu area, which borders New Taipei, is home to many industrial parks and technology company headquarters. The area's counties, which are composed of Taipei's satellite cities, of which Taoyuan became a quasi-direct-controlled municipality in January 2011, are home to Taiwan's largest immigrant populations. Although Taichung was only declared a direct-controlled municipality in December 2010, it is the most important city in central Taiwan. The city has a diverse economy that incorporates traditional businesses, small family-run shops and factories, large industrial areas, and a thriving commercial sector. The city is nicknamed the “Mechanical Kingdom” for its successful development of precision

434

M.-T. Lee et al. / Habitat International 44 (2014) 432e441

Kaohsiung and Tainan are both located in southern Taiwan. The two cities' economies are linked by the Port of Kaohsiung, which is the largest harbor in Taiwan. Due to the two cities' geographical proximity, particularly for residents in border districts, commuting for work or shopping is common. Kaohsiung, which was declared Taiwan's second direct-controlled municipality in 1979, is the political, economic, transportation, manufacturing, refining, shipbuilding, and industrial center of southern Taiwan. Kaohsiung's northern neighbor, Tainan, is the oldest city in Taiwan and has been a direct-controlled municipality since December 2010. Once reliant on agriculture and the traditional manufacturing industry, the city has now become a major center of Taiwan's optoelectronics industry, with a complete supply chain. Nevertheless, retail and services is Tainan's largest employment sector. Data source and description To conduct an empirical examination of the ripple effect among the six major urban areas described above, this study uses Cathay Real Estate Development Co. Ltd.'s house price indices. That company develops Laspeyres price indices with hedonic house price models to integrate the average price, character, scale, and negotiated price margin of each individual case. These price indices represent price behaviors in Taiwan's presale housing markets (Ministry of the Interior, 2013). The study period runs from the first quarter of 2000 to the first quarter of 2013. Fig. 2 shows house price movements in the six metropolitan areas under study. The largest change in house prices during the study period occurred in New Taipei (139.4%), followed by Taipei (122.7%). On average, those two cities experienced 16.8% and 15.4% quarterly increases, respectively, in house prices from Q1 2000 to Q1 2013. The house price change in the TaoyuaneHsinchu area was 74.4% during the same period, for a 10.7% quarterly increase. Taichung's total house price increase was 41.6%, with a 6.7% quarterly increase during the investigated period. Tainan and Kaohsiung experienced house price changes of similar magnitudes: their total increases were 16.8% and 18.0% and their quarterly increases were 3.2% and 3.0%, respectively. Significantly, all house prices experienced increasing trends during the investigated period. Initially, the indices appear to have begun at many different levels. More specifically, at the beginning, Taipei and New Taipei began at similar levels, and the other areas were at levels close to one other. The indices then increased to similar levels. Their evolution reveals that among the six areas under study, house prices may have experienced cointegration.

Fig. 1. Taiwan's major metropolitan areas.

Fig. 2. House price indices for Taiwan's six major metropolitan areas.

machinery from machine tools to bicycles. Since 2000, when plans were first devised to declare Taichung Taiwan's third directcontrolled municipality, Taichung has competed with Taipei and Kaohsiung.

Trend properties and unit root tests As revealed in Fig. 2, there appear to be trends in house price indices. Therefore, this study first applied the Vogelsang (1998) t
PS1T test to check for deterministic trends in the logarithmic indices.

Table 1 Deterministic trend test. LKU 10% significance level t
PS1T 0.000 (1.331) 5% significance level 1 t
PST 0.000 (1.720) 1% significance level t
PS1T 0.000 (2.647)

LNTA

LTA

LTC

LTN

LTH

0.000 (1.331)

0.000 (1.331)

0.000 (1.331)

0.000 (1.331)

0.000 (1.331)

0.000 (1.720)

0.000 (1.720)

0.000 (1.720)

0.000 (1.720)

0.000 (1.720)

0.000 (2.647)

0.000 (2.647)

0.000 (2.647)

0.000 (2.647)

0.000 (2.647)

Notes: 1. LKU is the logarithm of the house price index for Kaohsiung. LNTA is the logarithm of the house price index for New Taipei. LTA is the logarithm of the house price index for Taipei. LTC is the logarithm of the house price index for Taichung. LTN is the logarithm of the house price index for Tainan. LTH is the logarithm of the house price index for Taoyuan-Hsinchu. 2. The critical values are in parentheses.

M.-T. Lee et al. / Habitat International 44 (2014) 432e441 Table 2 Univariate unit root tests.

Table 4 VAR lag selection. ADF

Panel A: level series LKU LNTA LTA LTC LTN LTH Panel A: differenced series DLKU DLNTA DLTA DLTC DLTN DLTH

435


0.885 1.148 0.779
1.800 1.108 0.821
8.865***
9.928***
7.723***
9.538***
8.411***
10.231***

PP
1.076 1.136 0.774
1.800 0.744 0.877
8.865***
9.928***
9.536***
9.538***
8.525***
10.232***

Notes: 1. LKU is the logarithm of the house price index for Kaohsiung. LNTA is the logarithm of the house price index for New Taipei. LTA is the logarithm of the house price index for Taipei. LTC is the logarithm of the house price index for Taichung. LTN is the logarithm of the house price index for Tainan. LTH is the logarithm of the house price index for Taoyuan-Hsinchu. 2. DLKU is the first difference of LKU. DLNTA is the first difference of LNTA. DLTA is the first difference of LTA. DLTC is the first difference of LTC. DLTN is the first difference of LTN. DLTH is the first difference of LTH. 3. *** indicates significance at the 1% level.

The test is valid both for errors integrated of order zero (I(0)) and for errors integrated of order one (I(1)) (Vogelsang, 1998). A priori knowledge of house price innovations and testing whether those innovations are I(0) or I(1) was not required. The t
PS1T test was based on the following specification:

Ht ¼ a0 þ a1 t þ mt

(1)

where Ht is the logged house price index level in one studied area, a0 is the initial level of Ht , a1 is the average slope of time trend in Ht , and mt is a serially correlated random process. Testing for a time trend in the house price index was essentially a test of whether the parameter a1 was different from zero. The t
PS1T test statistic was specified as:

  t
PS1T ¼ T
1=2 ts exp
cJT1

(2)

where T is the sample size, ts is the set of t statistics for testing the null hypothesis that the individual parameters in the partial-sums regression of Eq. (2) are zero, c is a constant, and JT1 is the unit root statistic proposed by Park and Choi (1988) and Park (1990) (Chiang, Lee, & Wisen, 2005). Because it is not clear whether the innovations were I(0) or I(1), c was chosen so that the critical values of the t
PS1T test statistics were the same whether mt was I(0) or I(1). Therefore, different values for c were employed for different levels of statistical significance. Statistical inferences were based upon the critical values tabulated in Vogelsang (1998) because the asymptotic distribution of the t
PS1T statistic was non-normal. Table 3 Panel unit root tests. Method

Level series

Differenced series

LLC IPS Fisher-ADF Fisher-PP

4.515 4.733 2.475 2.670


19.236***
21.241*** 253.430*** 267.594***

Notes: 1. LLC and IPS are the panel unit root tests of Levin, Lin, and James Chu (2002) and Im, Pesaran, and Shin (2003), respectively. Fisher-ADF and Fisher-PP are the panel unit root tests of Maddala and Wu (1999) and Choi (2001), respectively. 2. *** indicates significance at the 1% level.

Lag

Log L

LR

FPE

AIC

SC

HQ

1 2 3 4

472.864 507.278 544.209 587.653

233.176** 50.569 45.221 42.557

9.39e-16** 1.07e-15 1.23e-15 1.33e-15


17.586
17.522
17.560
17.863**


15.965**
14.510
13.158
12.072


16.971**
16.379
15.890
15.666

Notes: 1. **indicates lag order selected by the criterion. 2. LR is the sequential modified likelihood ratio test statistic; FPE is final prediction error; AIC is Akaike's information criterion; SC is Schwarz's information criterion; HQ is Hannan-Quinn's information criterion.

According to Vogelsang (1998), the values of c should be specified as 0.494, 0.716, and 1.501 for the 10%, 5%, and 1% levels of significance, respectively. Table 1 presents the resulting t
PS1T test statistics for logarithmic house prices of Taiwan's six geographic areas. The resulting test statistics are all very low with a value of 0.000, and they are not statistically significant at any level. The statistics provide strong evidence that there are no deterministic trends in logarithmic presale house prices in Taiwan. The evidence makes sense because deterministic trends imply that house prices are constrained to increase forever. Specifically, a deterministic trend would not be reasonable because the price indices have adjusted for variations in housing quality over time (Chen, Kawaguchi, & Patel, 2004). Importantly, this finding has implications for the unit-root tests and the VEC modeling, which are well known to be sensitive to the inclusion or exclusion of deterministic time trends (Ahking, 2002). Specifically, we are able to conclude that the house prices are not trend stationary and that the VEC model contains no deterministic trends. Next, we checked the stationarity of logarithms of house prices. Using evidence from the t
PS1T test, the unit root tests did not include deterministic time trends in their specifications. This study began testing for the presence of unit roots in house prices with the ADF (augmented DickeyeFuller) unit root test (Dickey & Fuller, 1979) and PP (PhillipePerron) test (Phillips & Perron, 1988). Panel A of Table 2 presents the results of these univariate tests. The results of the ADF and PP tests cannot reject the unit-root null hypothesis for house prices in the studied areas at any conventional significance level. Panel B of Table 2 reports the results of the univariate tests on logarithmic house price changes. The ADF and PP tests clearly reject the unit-root null hypothesis at the 1% significance level. The results shown in Table 2 indicate that house prices are first difference stationary and are series integrated of order one, or I(1). To provide more evidence about the stationarity of the logarithms of house prices, we also conducted corresponding panel data tests. Table 3 reports the results of the panel unit-roots tests. Supporting the univariate tests, the panel data tests clearly show that house prices are first difference stationary and thus, are I(1) series. The above results are contrary to those of Lee and Chien (2011), who conclude that Taipei's house price index is trendstationary. The results of this study echo the argument advanced by Nelson and Plosser (1982) that the trends in most macroeconomic time-series are stochastic. Moreover, based on this finding, modeling and identifying trend shocks are critical to conducting and evaluating house price policies in Taiwan (Murray & Nelson, 2000).

The VEC model Constructing the VEC model The first step in constructing the VEC model was to select the optimal lag length in the VAR (vector autoregression) model for

436

M.-T. Lee et al. / Habitat International 44 (2014) 432e441

Table 5 VAR residual autocorrelation Lagrange multiplier tests. Order of correlation Panel A: VAR(1) 1 2 3 4 5 6 7 8 Panel B: VAR(4) 1 2 3 4 5 6 7 8

LM-Stat

P-value

51.525** 43.408 45.050 48.470 31.857 47.089 20.585 32.872

0.045 0.185 0.143 0.080 0.666 0.102 0.982 0.618

40.174 31.647 41.618 45.341 25.653 26.994 32.286 36.394

0.291 0.676 0.239 0.137 0.900 0.861 0.646 0.450

Notes: 1. LM-Stat denotes Lagrange multiplier test statistics. 2. ** indicates significance at the 5% level.

logarithmic house prices. This study computed five selection criteria: the sequential modified likelihood ratio (LR) test, the final prediction error (FPE) method (Akaike, 1969), Akaike's information criterion (AIC) (Akaike, 1974), the Schwarz information criterion (SC) (Schwarz, 1978), and the HannaneQuinn criterion (HQ) (Hannan & Quinn, 1979). According to Table 4, LR, FPE, SC, and HQ recommend one lag and AIC selects four lags to include in the VAR model. To judge the optimal lag length, this study takes into account autocorrelation of the VAR residuals. Table 5 reports the autocorrelation Lagrange multiplier test statistics for the VAR(1) and VAR(4) models. VAR(1) clearly suffers from an autocorrelation problem, and VAR(4) does not. Therefore, the VAR model in this study includes four lags, which are considered as four quarters. According to the unit-root test results, house prices in all studied cities are I(1) processes. To explore their long- and short-run relationships, we first reformulated the VAR(4) into a VEC model with three lags in differenced house prices and employed Johansen's (1988) multivariate maximum likelihood cointegration test to check the existence of common trends. The VEC model contained only intercepts, not deterministic trends, according to the t
PS1T test statistics described in the previous section.


39:728LKUt þ 7:598LNTAt
29:759LTAt
17:294LTCt þ 22:797LTNt þ 65:516LTHt  Ið0Þ (3) where LKU is Kaohsiung's logged house price index, LNTA is New Taipei's index, LTA is Taipei's index, and LTC, LTN, and LTH are Taichung's, Tainan's, and TaoyuaneHsinchu's indices, respectively. This study also performs Johansen and Juselius's (1990) cointegrated vector coefficient significance test. As shown in Table 7, all of the coefficients are significant at the 1% level except for New Taipei's, which is significant at the 5% level. This result clearly shows that all six urbanized areas' house prices should be included in the cointegrating equation. This evidence further supports the existence of a long-run stable relationship among all six areas' house prices. This finding contradicts the study of Lee and Chien (2011), which finds that Taipei and TaoyuaneHsinchu do not share a stable long-run relationship with Taiwan's other major urban areas. Our finding may diverge from that of Lee and Chien (2011), as the latter include deterministic time trends in their unit-root testing and VEC modeling. More significantly, our findings clearly show that Taipei is not isolated from Taiwan's other major urban areas in terms of long-run house price trends. This finding is consistent with those for Dublin, Johannesburg, and Melbourne and Sydney, which are linked to other regions in Ireland, South Africa, and Australia, respectively (Balcilar et al., 2013; Luo et al., 2007; Stevenson, 2004). However, this study's findings contradict other findings for Madrid and Taipei, which display unique house price behavior isolated from other regions due to their status as global cities (Chen et al., 2011; Chien, 2010; Larraz-Iribas & Alfaro-Navarro, 2008; Lee & Chien, 2011). When the coefficient of Taipei's logged house price index in Eq. (3) is normalized to unity, the cointegrating relationship may be expressed as:

LTAt ¼
1:335LKUt þ 0:255LNTAt
0:581LTCt þ 0:766LTNt þ 2:202LTHt þ mt where mt  Ið0Þ (4) Analogously, when Kaohsiung's coefficient in Eq. (3) is normalized to unity, the cointegrating relationship may be expressed as:

The cointegrating vector Table 6 presents the trace and maximum eigenvalue test statistics and their sample-size-adjusted 5% critical values. As predicted by the ripple effect, both tests support the convergence of house prices over the long run. More specifically, the two tests show the existence of only one cointegrating vector driving the series with common stochastic trends in the data. Thus, these results indicate

LKUt ¼ 0:191LNTAt
0:749LTAt
0:435LTCt þ 0:574LTNt þ 1:649LTHt þ yt where yt  Ið0Þ (5) In this study, the normalized cointegrating coefficients can be interpreted as long-run cross-elasticities because all of the indices Table 7 Cointegration vector coefficient exclusion test.

Table 6 Johansen's cointegration test. Hypothesized

Trace

5%

Max-Eigen

5%

No. of CE(s)

Statistic

Critical value

Statistic

Critical value

None At most At most At most At most At most

142.838** 72.500 40.674 22.720 7.729 0.476

141.740 103.155 71.073 44.682 23.199 5.661

70.337** 31.826 17.955 14.991 7.253 0.476

59.270 50.373 40.753 31.570 21.182 5.661

1 2 3 4 5

the existence of a long-run stable relationship for house prices in Taiwan's six studied areas for the Q1 2000 to Q1 2013 period. Specifically, the cointegrating vector is the following combination:

Notes: 1. No. of CE(s) denotes the number of cointegration vectors. 2. **denotes rejection of the hypothesis at the 5% level.

LKU

LNTA

LTA

LTC

LTN

LTH

Coeff.
39.728 7.598
29.759
17.294 22.797 65.516 LR 38.508*** 5.196** 38.457*** 22.678*** 16.667*** 36.983*** P-value 0.000 0.023 0.000 0.000 0.000 0.000 Notes: LKU is the logarithm of the house price index for Kaohsiung. LNTA is the logarithm of the house price index for New Taipei. LTA is the logarithm of the house price index for Taipei. LTC is the logarithm of the house price index for Taichung. LTN is the logarithm of the house price index for Tainan. LTH is the logarithm of the house price index for Taoyuan-Hsinchu. 2. Coeff. denotes coefficient and LR denotes likelihood ratio statistics. 3. *** and ** indicate significance at the 1% and 5% levels, respectively.

M.-T. Lee et al. / Habitat International 44 (2014) 432e441

are expressed as natural logarithms (Johansen, 2005; Lee & Chiang, 2010; Tuluca, Myer, & Webb, 2000). In conjunction with Fig. 1, the coefficients reveal two regional clusters in which house prices in neighboring urbanized areas move in the same direction as their regional centers over the long run. Specifically, the positive coefficients of New Taipei and TaoyuaneHsinchu in Eq. (4) reveal that they move in the same direction as Taipei, the regional center of Northern Taiwan, with respect to long-run price movements. Similarly, Eq. (5) indicates that Tainan's house prices move in the same direction as those in Kaohsiung, the southern center of Taiwan, in the long run. Because New Taipei's net number of immigrants is highly negatively correlated with that of Taipei (Executive Yuan, Directorate General of Budget, Accounting and Statistics, 2014), it is unlikely that this long-run association is attributable to migration. Nevertheless, migration may play a role in the long-run association between Taipei and TaoyuaneHsinchu, as the two areas' net numbers of immigrants are positively correlated. As discussed previously, Taipei is the node of a high-technology knowledge center, New Taipei is its suburb, and TaoyuaneHsinchu is home to many industrial parks and technology company headquarters. Therefore, the observed positive co-movements are also likely driven by the region's internal economic factors. In contrast to the northern cluster, it is more clear that migration may play a role in house price co-movements in the southern cluster because Tainan's net immigrants are positively correlated with those of Kaohsiung (Executive Yuan, Directorate General of Budget, Accounting and Statistics, 2014). Because the two cities are linked economically, this region's internal economic factors may also be an underlying cause of the cities' positive, long-run house price co-movement. Moreover, Taipei, Taichung, and Kaohsiung, which are the regional centers of Taiwan's northern, central, and southern regions, respectively, have negative coefficients in Eqs. (4) and (5) and thus indicate a negative association among regions in terms of house prices over the long run. It is unlikely that interregional migration causes this association because net immigrants are positively correlated or nearly uncorrelated among the three leading cities (Executive Yuan, Directorate General of Budget, Accounting and Statistics, 2014). However, the pattern is consistent with the interregional competition among the cities of northern, central, and southern Taiwan. According to Hill, Wolman, Kowalczyk, and Clair (2012), cities in different regions compete for production factorsdincluding labor, capital, and knowledgedfor their economic development.2 Because housing is partly an investment asset, house prices should reflect current economic conditions and expectations related to future economic health (Zhu, Füss, & Rottke, 2013). Therefore, interregional competition could lead to geographical movements of housing investment capital and, thus, could result in a long-run negative association of house prices among regions. Notably, the significant coefficients of cities across regions indicate that over the long run, the ripple effect might spread nationally across northern, central, and southern Taiwan. Similar findings are also reported in Australia, Ireland, and Malaysia (Hui, 2010; Luo et al., 2007; Stevenson, 2004). Nevertheless, Shi et al. (2009) find little evidence of house price cointegration among regional centers in New Zealand and conclude that the ripple effect is primarily caused by regional economic factors. This study's findings do not support such a conclusion in Taiwan.

2 Taiwanese cities must compete not only for capital investment from private enterprises but also, due to their lack of tax autonomy, for grants from the Taiwanese central government for local development activities (Menifield, 2011; Wang, Chan, & Wu, 2012).

437

Long-run leadelag relationships To explore long-run leadelag relationships among house prices in Taiwan's major urban areas, this study estimated their speeds of adjustment to the long-run equilibrium and examined their weak exogeneity, which means there was no error-correction of their deviations from the long-run equilibrium. Table 8 clearly shows that weak exogeneity cannot be rejected for Taipei and Kaohsiung at any conventional significance. Therefore, this finding implies that house prices in the two cities do not error-correct their deviations and thus, they lead the house price trend of the studied urban areas over the long run. The weak exogeneity statistics for New Taipei and TaoyuaneHsinchu are statistically significant at a 10% level. The weak exogeneity statistics for Taichung and Tainan can be rejected at a 1% significance level. This finding indicates that with respect to house price movements over the long run, these urban areas are followers. The above results indicate a unidirectional Granger causality from Taipei and Kaohsiung to Taiwan's other urban areas in the long run. That these two cities lead the house price trend is not surprising because Taipei and Kaohsiung are Taiwan's two most important economic centers and have different economic structures (Chen et al., 2011; Lin, 2006; Lin & Liaw, 2000). Because business cycles usually have an impact on Taiwan's primary economic centers first (Oikarinen, 2006), house prices in Taipei and Kaohsiung change first, and that movement later diffuses to other urban areas. In particular, Taichung responds very quickly and corrects 93.6% of its resulting deviation within a quarter. Tainan, New Taipei, and TaoyuaneHsinchu react relatively slowly and correct 66.2%, 43.1% and 41.0% of their deviations, respectively, within one quarter. The quick response of Taichung likely indicates that informed investors are particularly interested in their housing investments in Taiwan's regional centers. This observation is similar to the case in Finland (Oikarinen, 2006).

Short-run leadelag relationships Although long-run, lead-lag price relationships should reflect spatial patterns in economic fundamentals, short-run changes in lead-lag patterns may also be caused by non-fundamental forces such as representative heuristics (Füss, Zhu, & Zietz, 2011; Meen, 1999; Oikarinen, 2006). As a result, causality associated with adjustments to long-run relationships might move in different directions from causality associated with short-run disturbances (Andersson, 1999). Therefore, in addition to exploring price adjustments to the long-run relationship, this study also investigates short-run lead-lag relationships in price changes among the major urbanized areas in Taiwan. Table 9 reports short-run block Granger causality tests on the house price changes in the VEC model discussed previously. As indicated in the bottom row of that table, the joint tests for shortrun causality from the remaining markets are significant in all Table 8 Speed of adjustment coefficients and weak exogeneity test.

jCoeff.j LR P-value

LKU

LNTA

LTA

LTC

LTN

LTH

0.215 0.422 0.516

0.431 2.902* 0.088

0.146 2.331 0.127

0.936 6.967*** 0.008

0.662 11.094*** 0.001

0.410 3.400* 0.065

Notes: 1. LKU is the logarithm of the house price index for Kaohsiung. LNTA is the logarithm of the house price index for New Taipei. LTA is the logarithm of the house price index for Taipei. LTC is the logarithm of the house price index for Taichung. LTN is the logarithm of the house price index for Tainan. LTH is the logarithm of the house price index for TaoyuaneHsinchu. 2. jCoeff.j denotes the absolute value of coefficient and LR denotes likelihood ratio statistics. 3. *** and * indicate significance at the 1% and 10% levels, respectively.

438

M.-T. Lee et al. / Habitat International 44 (2014) 432e441

Table 9 Short-run block Granger causality test results. Dependent variable Lagged Lagged Lagged Lagged Lagged Lagged

DLKU

DLKU DLNTA DLTA DLTC DLTN DLTH

DLNTA

DLTA

3.046 (0.385) 2.089 2.298 1.054 4.195 1.733

Lagged DALL

(0.554) (0.513) (0.788) (0.241) (0.630)

15.026 (0.450)

5.105 1.615 3.101 4.718

1.964 (0.580) 7.077* (0.070)

DLTC

DLTN

1.052 (0.789) 1.468 (0.690) 14.530*** (0.002)

25.843*** 1.592 4.376 7.051*

DLTH (0.000) (0.661) (0.224) (0.070)

(0.164) (0.656) (0.376) (0.194)

5.723 (0.126) 1.979 (0.577) 1.208 (0.751)

0.499 (0.919) 4.047 (0.256)

10.139*** (0.017)

36.065*** (0.002)

35.915*** (0.002)

24.459* (0.058)

42.904*** (0.000)

1.840 1.970 3.817 4.749 1.463

(0.606) (0.579) (0.282) (0.191) (0.691)

22.454* (0.096)

Notes: 1. DLKU is the first-differenced logarithm of the house price index for Kaohsiung. DLNTA is the first-differenced logarithm of the house price index for New Taipei. DLTA is the first-differenced logarithm of the house price index for Taipei. DLTC is the first-differenced logarithm of the house price index for Taichung. DLTN is the first-differenced logarithm of the house price index for Tainan. DLTH is the first-differenced logarithm of the house price index for TaoyuaneHsinchu. DALL denotes the first-differenced logarithm of all house price indices in the same column. 2. P-values are in parentheses. 3. *** and * indicate significance at the 1% and 10% levels, respectively.

urban areas except Kaohsiung. The insignificance of Kaohsiung is perhaps indicative of that city's dominant role within southern Taiwan's housing market. As shown in individual test results, Kaohsiung's house price change significantly predicts Tainan's price change, but not vice versa. Notably, New Taipei's house price change Granger-causes Taipei's price change unidirectionally. A similar finding has also been documented in Helsinki, Finland's metropolitan area (Oikarinen, 2006). The short-run leading role of New Taipei is most likely the result of New Taipei experiencing a larger price change than Taipei during the sample period. Therefore, the relative attractiveness of Taipei increases, and individuals move from New Taipei to Taipei, and thus the city's house price changes with a lag. It is also likely that, based on the house price change observed in New Taipei, individuals form positive expectations concerning house price growth in Taipei, even if no shock affecting Taipei has actually occurred. The latter explanation is particularly suited to the short-run causal relationship between areas geographically distant from one another (Oikarinen, 2006). Concerning the results of this study, this explanation may also apply to the finding that house price change in Taipei significantly predicts change in Taichung and that house price changes in Taichung and TaoyuaneHsinchu Grangercause change in Tainan. In Taiwan, the media typically concentrate on reporting price movements in Taipei. It is likely that individuals in Taichung regard Taipei as a benchmark to evaluate what the “right” price change is. In recent years, Taichung and TaoyuaneHsinchu have also often received attention, partially because of large-scale development in re-planning zones and transportation projects in the two areas. Similarly, Tainan residents may regard the house prices of these two closest northern major urbanized areas as benchmarks in evaluating what the “right” changes are in Tainan house prices. Robustness checks


52:474LKUTNt þ 9:224LNTAt
37:269LTAt þ 38:844LTCt þ 75:047LTHt  Ið0Þ (6) Table 11 reports the results of Johansen and Juselius's (1990) cointegrated vector coefficient significance test. As indicated in this table, all of the coefficients are significant at least at the 5% level except for New Taipei's. As will be demonstrated below, the insignificance is likely due to multicollinearity (Dennis, Hansen, Johansen, & Juselius, 2006; Kivedal, in press), and thus, New Taipei should still be included in the cointegrating equation. More important, Taipei's coefficient is significant at the 1% level. This finding again clearly demonstrates that, with respect to long-run house price trends, Taipei is not isolated from Taiwan's other major urban areas. This finding lends strong support to the robustness of the finding based on the more recent set of price indices. To gain a clear understanding of the long-run association between Taipei and other urbanized areas, Eq. (6) can be re-expressed as:

LTAt ¼
1:408LKUTNt þ 0:247LNTAt þ 1:042LTCt

As a robustness check, we conduct further analyses based on an old set of price indices, also provided by Cathay Real Estate

Table 10 Johansen's cointegration test. Hypothesized

Trace

5%

Max-Eigen

5%

No. of CE(s)

Statistic

Critical value

Statistic

Critical value

None At most At most At most At most

104.003** 48.776 25.444 11.246 1.713

85.163 58.677 36.889 19.153 4.673

55.227** 23.332 14.198 9.532 1.713

41.587 33.645 26.063 17.487 4.673

1 2 3 4

Development Co. Ltd., for the major urban areas. This dataset has a single price index for Kaohsiung and Tainan. As a result, and in contrast to the more recent set of price indices, this set of price indices includes data on Taipei, New Taipei, TaoyuaneHsinchu, Taichung, and KaohsiungeTainan. The data period is from the first quarter of 1993 to the fourth quarter of 2010.3 During this period, Taipei and Kaohsiung are the only two direct-controlled municipalities in Taiwan. Table 10 reports the trace and maximum eigenvalue test statistics and their sample-size-adjusted 5% critical values. Both tests confirm the existence of a long-run, stable relationship for house prices in Taiwan's major urban areas for the Q1 1993 to Q4 2010 period. The following is the cointegrating vector:

Notes: 1. No. of CE(s) denotes the number of cointegration vectors. 2. **denotes rejection of the hypothesis at the 5% level.

þ 2:014LTHt þ εt

where εt  Ið0Þ

(7)

Specifically, Eq. (7) again reveals that New Taipei and TaoyuaneHsinchu move in the same direction as the regional center of Northern Taiwan, Taipei, with respect to long-run price movements. The equation also reveals a negative association between Taipei and KaohsiungeTainan. These results are consistent with those based on the more recent set of house prices reported in the previous sections. However, in contrast from the result obtained using the more recent set of house prices, Taichung is

3 The t
PS1T test and unit root tests indicate that the logarithmic house prices are I(1) without deterministic trends. The optimal lag length criteria and the autocorrelation Lagrange multiplier test statistics suggest that VAR(4) model should be adopted. The results are available from the authors upon request.

M.-T. Lee et al. / Habitat International 44 (2014) 432e441

LKUTN

LNTA

LTA

LTC

LTH

KaohsiungeTainan, and house price changes in KaohsiungeTainan and New Taipei Granger-cause change in Taichung. Admittedly, the short-run leadelag patterns are not identical to those based on the more recent set of house price indices. This dissimilarity may be partly the result of there only being a single price index for Kaohsiung and Tainan in the older set of house price indices. This dissimilarity is also consistent with the notion that the causal relationship among house price changes depends on the net effect of various forces during a particular period of time (Oikarinen, 2006). Specifically, New Taipei does not experience a larger price change than Taipei during the sample period of the more recent set of house price indices. Thus, New Taipei does not lead Taipei in house price changes. Similarly, Taichung and TaoyuaneHsinchu are not hot markets that experience substantial growth in house prices and thus do not play short-run leading roles before 2010.

0.002 0.001 0.978

0.215 3.149* 0.076

0.015 0.014 0.904

0.362 9.756*** 0.002

0.282 13.606*** 0.000

Conclusion

Table 11 Cointegration vector coefficient exclusion test.

Coeff. LR P-value

LKUTN

LNTA

LTA

LTC

LTH


52.47431 31.163*** 0.000

9.223567 0.934 0.334


37.26889 31.875*** 0.000

38.84363 5.751** 0.016

75.04716 14.894*** 0.000

Notes: LKUTN is the logarithm of the house price index for KaohsiungeTainan. LNTA is the logarithm of the house price index for New Taipei. LTA is the logarithm of the house price index for Taipei. LTC is the logarithm of the house price index for Taichung. LTH is the logarithm of the house price index for TaoyuaneHsinchu. 2. Coeff. denotes coefficient and LR denotes likelihood ratio statistics. 3. *** and ** indicate significance at the 1% and 5% levels, respectively.

Table 12 Speed of adjustment coefficients and weak exogeneity test.

jCoeff.j LR P-value

439

Notes: 1. LKUTN is the logarithm of the house price index for Kaohsiung-Tainan. LNTA is the logarithm of the house price index for New Taipei. LTA is the logarithm of the house price index for Taipei. LTC is the logarithm of the house price index for Taichung. LTH is the logarithm of the house price index for TaoyuanHsinchu. 2. jCoeff.j denotes the absolute value of coefficient and LR denotes likelihood ratio statistics. 3. *** and * indicate significance at the 1% and 10% levels, respectively.

positively associated with Taipei. A possible explanation is that the strong competition between Taichung and Taipei did not begin until Taichung was designated as Taiwan's third direct-controlled municipality, as discussed in a previous section. Table 12 clearly indicates that weak exogeneity cannot be rejected for Taipei and KaohsiungeTainan at any conventional level of significance. These test results imply that house prices in the two areas lead the house price trend in the urban areas studied here in the long run. This finding supports the robustness of the conclusion that Taipei and, arguably, Kaohsiung lead the house prices in the long run. The weak exogeneity statistics for New Taipei, TaoyuaneHsinchu, and Taichung are statistically significant at least at the 10% level. This finding indicates that these urban areas are followers with respect to house price movements in the long run. In addition, the finding supports the contention that New Taipei should be included in the cointegrating equation because the city's housing market error-corrects its deviations from the long-run price equilibrium shared by the studied urbanized areas. Table 13 reports short-run block Granger causality tests on the house price changes. The tests results again confirm that there is a strong leadelag interdependence in quarterly house price changes among Taiwan's major urban areas. In particular, house price change in Taipei significantly predicts change in

This paper examines trend properties, cointegration, and diffusion of house prices in six major urban areas: Taipei, New Taipei, TaoyuaneHsinchu, Taichung, Tainan, and Kaohsiung. This study first applied the Vogelsang (1998) t
PS1T test to re-examine previous studies on trend specification in the unit-root tests and VEC models of house prices in Taiwan's urban areas. The study further employed Johansen's (1988) cointegration test to inspect whether house prices share a long-run equilibrium relationship and then constructed a VEC model to investigate how they leadelag one another. Overall the empirical results indicate that, as the most important economic center and the focus of media attention, Taipei's house prices can ripple out to other major urbanized areas in Taiwan. The study's main findings and implications are as follows. First, the trends in presale house prices in Taiwan's major urban areas, including Taipei, are stochastic. This finding has far-reaching implications for the management and assessment of house price policies. More specifically, the finding implies that both Taiwan's central government and Taipei's city government must identify trend shocks to effectively curb Taipei's soaring house prices. Second, the results support house price cointegration, which involves all of Taiwan's major urban areas. Taipei is not isolated in terms of long-run house price trends. Therefore, when designing house price policies, the central government should take into account that policies targeted at the capital city will also affect house prices in other urban areas. Third, Taipei and Kaohsiung are clearly leaders in terms of house price movements over the long run. Therefore, investors who missed the housing boom in those two cities still have an opportunity to enter the market by investing in other urban areas such as New Taipei, TaoyuaneHsinchu, and Tainan. This finding also has planning implications for local governments in those urban areas.

Table 13 Short-run block Granger causality test results. Dependent variable Lagged Lagged Lagged Lagged Lagged

DLKUTN DLNTA DLTA DLTC DLTH

Lagged DALL

DLKUTN 0.005 13.170*** 4.999 0.714

DLNTA

DLTA

0.229 (0.973)

3.495 (0.321) 0.042 (0.998)

DLTC 10.879** (0.012) 9.928** (0.019) 2.419 (0.490)

(1.000) (0.004) (0.172) (0.870)

0.496 (0.920) 2.466 (0.482) 0.524 (0.914)

1.272 (0.736) 0.837 (0.841)

3.608 (0.307)

32.575*** (0.001)

5.402 (0.943)

5.155 (0.953)

39.147*** (0.000)

DLTH 5.867 3.357 3.102 1.477

(0.118) (0.340) (0.376) (0.668)

18.387 (0.104)

Notes: 1. DLKUTN is the first-differenced logarithm of the house price index for KaohsiungeTainan. DLNTA is the first-differenced logarithm of the house price index for New Taipei. DLTA is the first-differenced logarithm of the house price index for Taipei. DLTC is the first-differenced logarithm of the house price index for Taichung. DLTH is the firstdifferenced logarithm of the house price index for TaoyuaneHsinchu. DALL denotes the first-differenced logarithm of all house price indices in the same column. 2. P-values are in parentheses. 3. *** and ** indicate significance at the 1% and 5% levels, respectively.

440

M.-T. Lee et al. / Habitat International 44 (2014) 432e441

Fourth, there is a strong leadelag interdependence of quarterly house price changes among Taiwan's major urban areas. Due to the leadelag relations, static quarterly correlations are likely to exaggerate geographical diversification opportunities. Therefore, when diversifying their housing-related portfolios, financial institutions and housing investors should be aware of this finding and adopt conditional investment strategies. Acknowledgments Grant support for this study from Taiwan's Ministry of Science and Technology (NSC 101-2410-H-259-022-) is gratefully acknowledged. References Ahking, F. W. (2002). Model mis-specification and Johansen's co-integration analysis: an application to the US money demand. Journal of Macroeconomics, 24(1), 51e66. Akaike, H. (1969). Fitting autoregressions for prediction. Annals of the Institute of Statistical Mathematics, 21, 243e247. Akaike, H. (1974). A new look at the statistical model identification. IEEE Transactions on Automatic Control, AC-19, 716e723. Andersson, B. (1999). On the causality between saving and growth: Long-and shortrun dynamics and country heterogeneity. Balcilar, M., Beyene, A. D., Gupta, R., & Seleteng, M. (2013). “Ripple” effects in South African house prices. Urban Studies, 50(5), 876e894. Canarella, G., Miller, S., & Pollard, S. (2012). Unit roots and structural change an application to US house price indices. Urban Studies, 49(4), 757e776. Cao, Y. (2010). Northern buyers southward, Q1 Taoyuan price index rose the most. Nownews. Chang, C.-O., & Lin, V. C. C. (2011). Housing demand survey of Second Quarter 2011. In C.-O. Chang (Ed.), Housing demand survey (pp. 1e159). Taipei: Institute for Physical Planning and Information. Chen, P. F., Chien, M. S., & Lee, C. C. (2011). Dynamic modeling of regional house price diffusion in taiwan. Journal of Housing Economics, 20, 315e332. Chen, M. C., Kawaguchi, Y., & Patel, K. (2004). An analysis of the trends and cyclical behaviours of house prices in the Asian markets. Journal of Property Investment & Finance, 22(1), 55e75. Chiang, K. C., Lee, M.-L., & Wisen, C. H. (2005). On the time-series properties of real estate investment trust betas. Real Estate Economics, 33(2), 381e396. Chien, M.-S. (2010). Structural breaks and the convergence of regional house prices. Journal of Real Estate Finance and Economics, 40, 77e88. Choi, I. (2001). Unit root tests for panel data. Journal of International Money and Finance, 20(2), 249e272. Clapp, J. M., & Tirtiroglu, D. (1994). Positive feedback trading and diffusion of asset price changes: evidence from housing transactions. Journal of Economic Behavior & Organization, 24(3), 337e355. Cook, S. (2003). The convergence of regional house prices in the UK. Urban Studies, 40(11), 2285e2294. Dennis, J. G., Hansen, H., Johansen, S., & Juselius, K. (2006). CATS in RATScointegration analysis of time series, Version 2. Evanston: Estima. Dickey, D. A., & Fuller, W. A. (1979). Distribution of the estimators for autoregressive time series with a unit root. Journal of the American statistical Association, 74(366a), 427e431. Executive Yuan. (2014). Directorate General of Budget, Accounting and Statistics. National Statistics. Retrieved April 30, 2014, from http://www.dgbas.gov.tw. Füss, R., Zhu, B., & Zietz, J. (2011). US regional housing bubbles, their co-movements €t. and spillovers. EBS Universita Hannan, E. J., & Quinn, B. G. (1979). The determination of the order of an autoregression. Journal of the Royal Statistical Society, 41, 190e195. Hill, E. W., Wolman, H. L., Kowalczyk, K., & Clair, T. S. (2012). Forces affecting city population growth or decline: the effects of interregional and inter-municipal competition. In Rebuilding America's legacy cities: New directions for the industrial heartland. Holmes, M. J. (2007). How convergent are regional house prices in the United Kingdom? Some new evidence from panel data unit root testing. Journal of Economic and Social Research, 9(1), 1e17. Holmes, M. J., & Grimes, A. (2008). Is there long-run convergence among regional house prices in the UK? Urban Studies, 45(8), 1531. Hui, H. C. (2010). House price diffusions across three urban areas in Malaysia. International Journal of Housing Markets and Analysis, 3(4), 369e379. Im, K. S., Pesaran, M. H., & Shin, Y. (2003). Testing for unit roots in heterogeneous panels. Journal of Econometrics, 115(1), 53e74. Johansen, S. (1988). Statistical analysis of cointegration vectors. Journal of Economic Dynamics and Control, 12(2/3), 231e254. Johansen, S. (2005). Interpretation of cointegrating coefficients in the cointegrated vector autoregressive model. Oxford Bulletin of Economics and Statistics, 67(1), 93e104.

Johansen, S., & Juselius, K. (1990). Maximum likelihood estimation and inference on cointegration: with applications to the demand for money. Oxford Bulletin of Economics and Statistics, 52(2), 169e210. Jud, G. D., & Winkler, D. T. (2009). The housing futures market. Journal of Real Estate Literature, 17(2), 181e203. Kivedal, B. K. (2013). A DSGE model with housing in the cointegrated VAR framework. Empirical Economics, 1e28 (in press) Advance online publication. http:// dx.doi.org/10.1007/s00181-013-0765-7. Knight Frank. (2007). In K. F. LLP (Ed.), 07 Annual Wealth Report: Prime Residential Property (pp. 1e30). London. Lai, P. Pi-Y., & Fischer, D. (2007). The determinants of foreign real estate investment in Taiwan. Pacific Rim Property Research Journal, 13(3), 263e279. Larraz-Iribas, B., & Alfaro-Navarro, J. L. (2008). Asymmetric behaviour of Spanish regional house prices. International Advances in Economic Research,14(4), 407e421. Lee, C. L. (2008). Housing in Australia as a portfolio investment. International Journal of Housing Markets and Analysis, 1(4), 352e361. Lee, M.-L. (2009). Non-diversifiable risk and quantity discounts in urban land markets. Pacific Rim Property Research Journal, 15(1), 122e139. Lee, M.-L., & Chiang, K. (2010). Long-run price behaviour of equity REITs: become more like common stocks after the early 1990s? Journal of Property Investment & Finance, 28(6), 454e465. Lee, C. C., & Chien, M. S. (2011). Empirical modelling of regional house prices and the ripple effect. Urban Studies, 48(10), 2029e2047. Levin, A., Lin, C.-F., & James Chu, C.-S. (2002). Unit root tests in panel data: asymptotic and finite-sample properties. Journal of Econometrics, 108(1), 1e24. Lin, J. (2006). The dynamics of labor migration in Taiwan: Evidence from the 1990 and 2000 Taiwan population censuses. Paper presented at the Geography Research Forum. Lin, Ji-P., & Liaw, K.-L. (2000). Labor migrations in Taiwan characterization and interpretation based on the data of the 1990 census. Environment and Planning A, 32(9), 1689e1709. Liu, P. (2009). Suite hot! Northern buyers southward earn three times. Smart Rich, 131. Retrieved September 15, 2013, from http://smart.businessweekly.com.tw/ webarticle.php?id¼37129. Luo, Z. Q., Liu, C., & Picken, D. (2007). Housing price diffussion pattern of Australia's state capital cities. International Journal of Strategic Property Management, 11, 227e242. Ma, W. (2011). Favored by self-residing purchasers, ten-million dollar houses in New Taipei city are hot. Commercial Times. Retrieved September 20, 2013, from http://u-trust.housefun.com.tw. Macdonald, R., & Taylor, M. P. (1993). Regional house prices in Britain: long-run relationships and short-run dynamics. Scottish Journal of Political Economy, 40(1), 43e55. Maddala, G. S., & Wu, S. (1999). A comparative study of unit root tests with panel data and a new simple test. Oxford Bulletin of Economics and Statistics, 61(S1), 631e652. Meen, G. (1999). Regional house prices and the ripple effect: a new interpretation. Housing Studies, 14(6), 733e753. Meen, G. P. (2001). Modelling spatial housing markets: Theory, analysis, and policy (Vol. 2). Springer. Menifield, C. E. (2011). Comparative public budgeting: A global perspective. Sudbury, Massachusetts: Jones & Bartlett Learning. Ministry of the Interior. (2013). Real estate information platform. Retrieved September 23, 2013, from http://pip.moi.gov.tw/NET/E-Statistics/E2.aspx. Murray, C. J., & Nelson, C. R. (2000). The uncertain trend in US GDP. Journal of Monetary Economics, 46(1), 79e95. Nelson, C. R., & Plosser, C. R. (1982). Trends and random walks in macroeconmic time series: some evidence and implications. Journal of Monetary Economics, 10(2), 139e162. Oikarinen, E. (2006). The diffusion of housing price movements from center to surrounding areas. Journal of Housing Research, 15(1), 3e28. Pacific Assets Management Co. Ltd. (2011). Foreingers buy 30 thousands square meters of land in March. Retrieved November 8, 2013, from http://www.pamc. tw. Park, J. Y. (1990). Testing for unit roots and cointegration by variable addition. In T. Fomby, & F. Rhodes (Eds.), Advances in econometrics. Cointegrationi, spurious regressions and unit roots (pp. 107e134). London: JAI Press. Park, J. Y., & Choi, B. (1988). A new approach to testing for a unit root. CAE Working Paper. Cornell University. Phillips, P. C., & Perron, P. (1988). Testing for a unit root in time series regression. Biometrika, 75(2), 335e346. Quigley, J. M., & Van Order, R. (1991). Defaults on mortgage obligations and capital requirements for US savings institutions: a policy perspective. Journal of Public Economics, 44(3), 353e369. Real Estate Investar. (2012). Finding the right suburb to invest in e Using the idea of the ripple effect. Retrieved October 25, 2013, from https://www.realestateinvestar. com.au. Schwarz, G. (1978). Estimating the dimension of a model. Annals of Statistics, 6, 461e464. Shi, S., Young, M., & Hargreaves, B. (2009). The ripple effect of local house price movements in New Zealand. Journal of Property Research, 26(1), 1e24. Shining Group. (2011). Real estate foreign investors optimistic about Taiwan. Retrieved December 18, 2011, from http://www.shininggroup.com.tw. Stevenson, S. (2004). House price diffusion and inter-regional and cross-border house price dynamics. Journal of Property Research, 21(4), 301e320. Taiwan Realty. (2010). Taiwan house price index rose 6.42% in first quarter of 2010. Industrial Reports. Retrieved May 25, 2010, from http://www.ippi.org.tw.

M.-T. Lee et al. / Habitat International 44 (2014) 432e441 Tuluca, S. A., Myer, Fc N., & Webb, J. R. (2000). Dynamics of private and public real estate markets. Journal of Real Estate Finance and Economics, 21(3), 279e296. Vogelsang, T. J. (1998). Trend function hypothesis testing in the presence of serial correlation. Econometrica, 123e148. Wang, C.-H. (2003). Taipei as a global city: A theoretical and empirical examination. Urban Studies, 40(2), 309e334. Wang, F. (2009, November 30). Top complaint in online poll is expensive housing. Taipei Times, 2.

441

Wang, C.-L., Chan, Fu-Y., & Wu, C.-L. (2012). Consolidating support or targeting the opposition? The impact of political factors on the allocation of intergovernmental grants. Taiwan Journal of Political Science, 51, 51e90. Yu, Z. W. (2010). Northern buyers southward, major house purchasing power in Taichung. United Evening News. Retrieved September 10, 2013, from http:// money.udn.com. Zhu, B., Füss, R., & Rottke, N. B. (2013). Spatial linkages in returns and volatilities among US regional housing markets. Real Estate Economics, 41(1), 29e64.