Time-varying spillovers among first-tier housing markets in China

Abstract

Chinese

Numerous efforts have over the last few years been devoted to studying spillovers (ripple effects) among cities as a means of evaluating overheated housing markets. What seems to be lacking, however, is the application of a rolling-window approach to further explore time-varying spillovers in a timely manner in order to look more closely at a housing market with Chinese characteristics; for example, a market with rapidly increasing prices and a sequence of policy recommendations. By focusing on total, directional and net spillovers, and using 2000–2017 monthly housing price data across six Chinese cities, this study’s results indicate that time-varying spillovers provide a better understanding of the interactions among first-tier cities. It is interesting to note that, following the downside risk faced by the economy in 2014, the spillovers among cities have been abruptly transformed into those exhibiting bilateral co-movements based on high total spillovers and low net spillovers, and these results are also confirmed by the frequency dynamics of spillovers. Based on the above, there is sufficient evidence to conclude that the housing frenzies in China, which have become a national-level issue, deserve a more explicit macro-control policy in relation to real estate assets.

Keywords

Chinese housing market housing frenzies spillovers time-varying ripple effects

Introduction

While irrational exuberance in China’s real estate market has attracted significant global interest (Glaeser et al., 2017), we should notice that this country’s real estate sector possesses its own features. First, compared with other nations, the real estate sector in China is a very young sector due to its privatisation and commercialisation after the 1997 Asian financial crisis (Chen et al., 2011a). Second, rapid economic growth has resulted in excessive domestic savings (Wei and Zhang, 2011), which has been correlated with the traditional concept of ‘land is wealth’ (Zhang et al., 2012), a strong credit boom (Zhang, 2013) and very few other investment goals (Chiang, 2016), ultimately culminating in one of the biggest housing frenzies in the world (Wu et al., 2012). Moreover, an urban-biased policy and a housing registration system in China have further given rise to skyrocketing housing prices in first-tier cities. In addition, the real estate sector plays an extraordinary role in China’s economic growth and investment (Glaeser et al., 2017). Thus, recent years have seen a renewal of interest in China’s housing market. However, it is reasonable to expect that the speedy growth and change in China’s housing market must rely on a time-varying estimation in order to fully track its evolution and development.

Another important fingerprint is China’s planned economy system, as opposed to a market economy, and so the importance of governmental policies must be emphasised, the real estate market being no exception. In fact, when confronted with the housing mania, China’s authorities have again and again demonstrated their desire to reduce the pressure of continuously increasing real estate prices. For example, they have initiated a succession of housing regulation measures. Therefore, when the irrational exuberance of China’s housing market is discussed, the effectiveness of related real estate policy arguments must constantly be kept in mind. However, the evaluation of the merits of real estate policy measures has surprisingly received little attention. To resolve this debate, a time-varying estimation can be used to further consider the effects of these policies on housing markets.

To sum up, regardless of whether considered from the high degree of fluctuations in housing prices caused by housing frenzies or the many housing policy measures implemented by this planned economy, the best way to explore a housing market with China’s characteristics is to search for a new time-varying estimation approach.

Nevertheless, it remains an unsettled question as to which concept can be introduced to analyse China’s housing frenzies. In real estate economics, research on overheated housing markets often considers the concept of a ripple effect, namely, spillovers among cities (Meen, 1999), on the grounds that high housing prices in one city can influence and push up housing prices in other cities, eventually resulting in housing frenzies on a national scale. That is to say, the ripple effect can help us know more about the housing price transmission mechanism and source cities in order to better understand China’s housing market.

From the above it is thus clear that we need a time-varying estimation to capture the ripple effect, namely, spillovers among urban housing markets with Chinese characteristics. Fortunately, one possible solution is the spillover index proposed by Diebold and Yilmaz (2009, 2012; hereafter, referred to as the DY spillover index), using a rolling-window approach to sequentially move the estimation window by adding new information to re-estimate the coefficients. Thus, the time-varying spillover effects computed from the DY spillover index can fully capture the interaction among housing markets with Chinese characteristics. Finally, we provide a new version of the DY spillover index based on the frequency domain proposed by Barunik and Krehlik (2018) to shed further light on the behaviour of spillovers in the short run and long run.

This article continues as follows. The second section provides a review of the ripple (spillover) effects of the housing market in different countries. The third section goes into detail about the general idea and calculation of the DY spillover index, using a rolling window and frequency domain, respectively, to explore the time-varying spillovers among cities in China. The fourth section describes and analyses the data regarding the economic conditions and housing prices of first-tier cities. The fifth section reports the estimation results and discusses some policy implications. The sixth section provides our conclusion.

Literature review

An overheated housing market is often accompanied by ripple effects among local housing markets, which makes the housing problem become both more complicated and more serious, because the ripple effect by implication suggests that high housing prices in one city may be transmitted to other cities.

A large number of studies over the past few decades have shown ripple effects to exist in many cases. In the UK, for example, studies using Granger causality tests and cointegration have found that spillovers do exist and that increases in regional housing prices have spread from the greater London area to other regions (Alexander and Barrow, 1994; Drake, 1995; MacDonald and Taylor, 1993). In the case of Australia, Liu et al. (2008) applied VAR and impulse response functions to support the existence of ripple effects among eight capital cities, while Costello et al. (2011) first calculated the non-fundamental part of housing prices and then evaluated the related spillovers based on the VAR model to find that non-fundamental factors still reveal the existence of spillovers among these capital cities. Lee and Chien (2011) first applied cointegration tests and weak exogeneity to examine five cities in Taiwan, finding evidence of spillovers among all major cities, excluding Taipei City. Chen et al. (2011b) employed the Toda-Yamamoto causality test to search for the origin of the ripple effect among Taiwan’s cities. Fereidouni et al. (2016) applied the Granger causality test and cointegration across Malaysia’s major regions and Singapore to reveal that the ripple effect can be extended across the border due to a long-run equilibrium between Malaysia and Singapore. In the case of the US, Gupta and Miller (2012) estimated spillovers involving three cities, namely, Los Angeles, Las Vegas and Phoenix, using VAR and the cointegration approach to find that Los Angeles is identified as the source of rising housing prices. Yunus and Swanson (2013) in addition applied regional housing price indices to find that three regions in the US were the source of rising housing prices among a total of nine regions in that country. At the same time, stronger integration among regional housing markets based on a cointegration approach has recently been detected. Similarly, by dividing subsample periods in the presence of structural change, Brady (2014) showed that spillovers among states in the US have become stronger since the late-1990s. In the case of China, Chiang (2014) first applied the causality test to search for the source city of spillovers among six mega cities, finding the principal source city to be Beijing. Gong et al. (2016) utilised the cointegration approach and causality with spatial correlations to indicate that there is little evidence of spillovers among these cities within the Pan-Pearl River Delta region. Finally, Weng and Gong (2017) additionally considered spatial correlation and multivariate volatilities to prove that there are strong spillovers among 10 cities, including first-tier and second-tier cities, in China. As noted above, it is found that spillovers among local housing markets have been more prevalent in almost all countries, including China, so the concept of a ripple effect deserves particular attention.

Besides, as mentioned above, over the last few decades a great deal of effort has been made to investigate spillovers among regional housing markets and the directions of spillovers through traditional time-series econometrics, for example by utilizing VAR, co-integration and causality tests to determine the existence and direction of spillovers across regional housing markets. However, we believe that the time-varying process of spillovers is especially appropriate for the Chinese housing market. Moreover, this time-varying estimation has been applied in many fields, for example the integration of global and China’s stock markets (Zhou, 2012), spillovers among Greater China real estate markets (Liow, 2014), the relationship between national and regional housing markets (Tsai, 2015), spillovers among different assets (Chiang et al., 2017) and spillovers among G7 countries involving four assets, namely, stock, real estate, bond and foreign exchange (Liow et al., 2018). Therefore, we shall employ the DY spillover index to investigate the evolution of housing spillovers among China’s mega cities.

Time-varying spillover estimation and rolling windows

In accordance with the fluctuations in China’s housing markets and its housing policy recommendations, there is an urgent need to employ a new method in the dynamic evolution of spillovers, namely, the DY spillover index. In this section we will present this new idea.

DY spillover framework and different types of spillovers

Diebold and Yilmaz (2009) first proposed a spillover measure through the use of forecast error variance decompositions (FEVDs) from VAR. To solve the problem of variable ordering, Diebold and Yilmaz (2012) further suggested using the generalised VAR methodology, rather than VAR. However, the DY spillover index drives us to ask a new question: How can FEVD account for spillovers? This is simply because FEVD calculates the contribution of every variable shock to forecast error variance of a specific variable. In other words, higher explained powers of other variables to this variable’s variance intuitively imply that there are stronger spillovers of other variables to this variable. If this view is valid, then related DY spillover indices based on FVEDs can simultaneously be used to obtain the size and direction of spillovers.

In fact, the main purpose of the VAR model is to better understand the interaction among different variables, and so it is natural to introduce VAR as a starting point when considering spillovers. We first consider a covariance stationary VAR (p) of housing returns (Y) with N cities and p lags as follows:¹

Y_{t} = \sum_{i = 1}^{p} Φ_{i} Y_{t - i} + ε_{t}, ε_{t} ~ (0, Σ)

(1)

Here, $ε$ is a random error with a zero mean and covariance matrix, $Σ$ . Based on a generalised VAR system, we can compute the H-step-ahead FEVD by $φ_{ij}^{g} (H)$ as:

φ_{ij}^{g} (H) = \frac{σ_{jj}^{- 1} \sum_{h = 0}^{H} {[{(A_{h} \sum)}_{ij}]}^{2}}{\sum_{h = 0}^{H} {(A_{h} \sum A_{h}^{'})}_{ii}}

(2)

where $σ_{jj}$ is the standard deviation of the error term for the j-th equation and ${(B)}_{ij}$ denotes the i-th row and k-th column of matrix B.

According to the non-unity sum of FEVD, the normalisation of (2) can be operated by a new version of FEVD ( ${\tilde{φ}}_{ij}^{g}$ ) to sum up to 1 in a generalised VAR as follows:

{\tilde{φ}}_{ij}^{g} (H) = \frac{φ_{ij}^{g} (H)}{\sum_{j = 1}^{N} φ_{ij}^{g} (H)}

(3)

Since the magnitude of FEVDs can represent the effects of spillovers, we construct the first spillover index, the total spillover index ( $S P^{g}$ ), as:

S P^{g} (H) = \frac{\sum_{\begin{matrix} i, j = 1 \\ i \neq j \end{matrix}}^{N} {\tilde{φ}}_{ij}^{g} (H)}{\sum_{i, j = 1}^{N} {\tilde{φ}}_{ij}^{g} (H)}, where \sum_{i, j = 1}^{N} {\tilde{φ}}_{ij}^{g} (H) = N

(4)

The total spillover index is calculated based on the spillovers from or to other cities through the use of a variance decomposition of housing returns in a specific city minus the explained power from itself. Thus, the value of the total spillover between zero and unity can present the ‘average’ level of spillovers over all six cities in our study. A higher value of the total spillover index shows more active spillovers among cities and higher system risk (Diebold and Yilmaz, 2012, 2014). By contrast, past studies regarding ripple effects only focused on the direction of spillovers among cities. It is clear that the DY spillover index is a better approach than that used by past studies.

Aside from the total spillover index that gives us the magnitudes of total spillovers among mega cities in China, we can also understand more about the direction of the interactive relationships among urban housing markets, that is to say, directional spillovers by market i‘from’ other markets, as in (5), namely, ${SP}_{i .}^{g}$ ; and directional spillovers by market i‘to’ other markets, as in (6), namely, ${SP}_{. i}^{g}$ . Compared with past studies regarding spillover effects, which only discuss the final outcome, namely, the net spillovers, the DY spillover index provides more information.

{SP}_{i .}^{g} (H) = \frac{\sum_{\begin{matrix} j = 1 \\ j \neq i \end{matrix}}^{N} {\tilde{φ}}_{ij}^{g} (H)}{\sum_{i, j = 1}^{N} {\tilde{φ}}_{ij}^{g} (H)} = \frac{\sum_{\begin{matrix} j = 1 \\ j \neq i \end{matrix}}^{N} {\tilde{φ}}_{ij}^{g} (H)}{N}

(5)

{SP}_{. i}^{g} (H) = \frac{\sum_{\begin{matrix} j = 1 \\ j \neq i \end{matrix}}^{N} {\tilde{φ}}_{ji}^{g} (H)}{\sum_{i, j = 1}^{N} {\tilde{φ}}_{ji}^{g} (H)} = \frac{\sum_{\begin{matrix} j = 1 \\ j \neq i \end{matrix}}^{N} {\tilde{φ}}_{ji}^{g} (H)}{N}

(6)

According to the spillovers to other markets in (6) and the spillovers from other markets in (5), we can simply obtain a net index, referred to as ‘net spillovers’ ( ${SP}_{i}^{g}$ ), as in (7). Using net spillovers lets us know about the position of this urban housing market on the grounds that a positive net spillover index indicates that this city is a provider, and a negative value of this index shows that it is a receiver based on the viewpoint of housing spillovers.

{SP}_{i}^{g} (H) = {SP}_{. i}^{g} (H) - {SP}_{i .}^{g} (H)

(7)

Through the operational process of directional and net spillovers, it is clear that the DY spillover index comprehensively traces out the evolution of spillovers among urban housing markets in China.

Rolling-window approach

It is generally believed that many events can temporarily result in major changes in the conditions and environment of housing markets. Thus, we have no reason to believe that the spillover process in China’s housing market is stable and smooth. Unfortunately, traditional time-series econometrics relies on the stability of economic variables over the long run, so using the rolling-window approach to obtain time-varying coefficients is a good choice. In fact, as mentioned earlier when referring to Diebold and Yilmaz (2012), the rolling window spillover index can be used to monitor or track spillovers, especially financial crises. Moreover, China, as a model of a planned economy, often announces many housing-related measures intended to affect urban housing markets, and so it is suggested that by using a rolling window to obtain time-varying coefficients it may be possible to investigate a variety of spillover processes in the case of spillovers among China’s urban housing markets. The steps involved in the rolling-window approach are as follows. First, choose a rolling window size, for example m. Second, suppose that the number of increments between successive rolling windows is one period, and then partition the entire data set into K = T – m+ 1 subsamples. The first rolling window contains observations from period 1 to m, the second rolling window contains observations from period 2 to m+ 1 and a similar pattern continues until period T. Thirdly, estimate the model in (1) using each rolling window subsample to obtain a variance decomposition (2) and then calculate a value of the spillover index, for example the total spillover index in (4), the directional spillover index in (5) or (6) and the net spillover index in (7). Finally, plot each value of the spillover index over the rolling window to see how the estimates change over time, so that the parameters might be time varying.

Frequency domain approach

Although Diebold and Yilmaz (2012) constructed the time-varying variance decompositions of the generalised VAR to calculate the process of spillovers among markets, Barunik and Krehlik (2018) additionally applied the spectral representation of variance decompositions to set up a frequency dynamics of spillovers.² The latter emphasises that spillovers at different frequencies can reveal short-run and long-run behaviour. We believe that the spillover index from the viewpoint of the frequency domain can offer a richer picture of time-varying ripple effects compared with the DY spillover index.

Barunik and Krehlik (2018) first proposed the spectral representation of variance decompositions by means of a Fourier transformation from impulse responses to frequency responses, namely, $A (e^{- ih ω}) = \sum_{h} e^{- ih ω} A_{h}$ . Thus, we can rewrite (2) with a new FEVD, namely, $φ_{ij}^{g} (ω)$ on a specific frequency $ω$ , as in (8):

φ_{ij}^{g} (ω) = \frac{σ_{jj}^{- 1} \sum_{h = 0}^{\infty} {[{(A (e^{- ih ω}) \sum)}_{ij}]}^{2}}{\sum_{h = 0}^{\infty} {(A (e^{- ih ω}) \sum^{A} (e^{ih ω}))}_{ii}}

(8)

Similarly, we can rewrite (3) as the normalisation of this new FEVD in (9):

{\tilde{φ}}_{ij}^{g} (ω) = \frac{φ_{ij}^{g} (ω)}{\sum_{j = 1}^{N} φ_{ij}^{g} (ω)}

(9)

In a departure from Diebold and Yilmaz (2012), it is necessary for us to accumulate total spillovers over the frequency band $d = (a, b)$ , as in (10):

{({\tilde{φ}}_{d}^{g})}_{ij} = \int_{a}^{b} [{\tilde{φ}}_{ij}^{g} (ω)] d ω

(10)

Thus, we can obtain a new total spillover within the frequency band d, namely, ${SP}_{d}^{g}$ , as in (11):

{SP}_{d}^{g} = \frac{\sum_{\begin{matrix} i, j = 1 \\ i \neq j \end{matrix}}^{N} {({\tilde{φ}}_{d}^{g})}_{ij}}{\sum_{i, j = 1}^{N} {({\tilde{φ}}_{d}^{g})}_{ij}} = 1 - \frac{\sum_{i = 1}^{N} {({\tilde{φ}}_{d}^{g})}_{ij}}{\sum_{i, j = 1}^{N} {({\tilde{φ}}_{d}^{g})}_{ij}}

(11)

When this new total spillover index is close to unity (zero), it implies that overall the spillovers are very strong (weak) within the band. As far as directional spillovers are concerned, we can obtain directional spillovers for market i‘from’ other markets, as in (12), and from market i‘to’ other markets, as in (13):

({SP}_{i .}^{g})_{d} = \sum_{\binom{j = 1}{j \neq i}}^{N} {({\tilde{φ}}_{d}^{g})}_{ij}

(12)

({SP}_{. i}^{g})_{d} = \sum_{\binom{j = 1}{j \neq i}}^{N} {({\tilde{φ}}_{d}^{g})}_{ji}

(13)

Since the directional spillovers for market i‘from’ and ‘to’ other markets have been obtained, it is easy to compute the net spillovers, namely, $(S_{i}^{g})_{d}$ , as in (14):

{({SP}_{i}^{g})}_{d} = ({SP}_{. i}^{g})_{d} - ({SP}_{i .}^{g})_{d}

(14)

If this net spillover index is positive (negative), this implies that market i transmits more information to (from) other markets.

Data description and related policies

This section outlines the data sources, the economic development of these mega cities and related housing policy recommendations. This information will help us inquire into the status of China’s housing markets.

Data sources

Our urban housing price indices are based on monthly data gathered from the China Real Estate Index System (CREIS), which shows many kinds of real estate price indices (Chiang, 2016; Hui and Yue, 2006; Weng and Gong, 2017). For example, the urban composite price index is the weighted-average value of residential, office and retail price indices. From the real estate market frenzies that are now concentrated in China’s housing submarkets, we select the residential price index as our sample rather than office and retail real estate assets, which is calculated using the Laspeyres index with a base of 1000 points in Beijing in December 2000. Moreover, in this database system, all housing price data are collected based on actual (primary) real estate transactions and only the new-dwelling type of housing is surveyed to immediately observe the development of the housing market. In contrast, the combined size of the office and retail real estate markets is much smaller than that of the housing market, while at the same time the data for these commercial assets may only be traced back to 2003 or later.

We choose the first-tier cities of Beijing, Tianjin, Shanghai, Guangzhou, Shenzhen and Chongqing to evaluate time-varying spillovers, since exploring their spillovers is very useful in seeking to comprehend China’s real estate problem. This article collects data for the period from December 2000 (the basis period) to July 2017, comprising a total of 200 observations. At the same time, we apply 60-month rolling samples, that is, five years being the size of an estimation window, to implement all kinds of spillovers using the rolling-window approach. As a result, we obtain 141 estimated parameters to assess the time-varying evolution of spillovers.

Real estate policy from 2000 to the present

Following the 1997 Asian financial crisis, the commercialisation of China’s housing was implemented nationally over the years 1998–2000 (Chen et al., 2011a) in order to ensure the sustainability of the country’s growth through a new channel of economic development from the real estate sector. Because housing prices rose very dramatically thereafter, China’s government decided to apply a macro-control policy (2004–2007) in order to stabilise economic development, especially housing prices. For example, the State council announced eight directives and six directives, in 2005 and 2006, respectively, in order to slow down the appreciation in housing prices. Nevertheless, the global financial crisis of 2008 again compelled the government to change its mind to boost domestic consumption, especially in rural areas, by expanding credit quotas and implementing an expansionary monetary policy. However, this policy to revitalise the economy still spilled over into the housing markets. With higher and higher housing prices in many cities, the State council again issued four directives to prevent the excessive liquidity from permeating the real estate markets in 2009. Subsequently, the State council announced 11 and eight new directives, in 2010 and 2011, respectively, in order to restrain overly-high housing prices by strongly curbing lending. We thus come to the conclusion that past housing policy actions were intended to control housing price inflation, but housing prices in many cities still continued to rise.

The negative impacts of a succession of anti-housing policies during 2009–2011, however, soon began to emerge, and China’s economy started to experience a downturn in 2014. The fear of further declines or even a collapse pervaded China’s housing market. Because the real estate industry plays an absolutely essential role in China’s economy, after 2014 the authorities suddenly relaxed mortgage lending and implemented tax cuts to aggressively spur the domestic housing market. This strategy immediately eliminated the possibility of a bear market in housing assets, but the repercussions in the housing market were far greater than expected. From 2015 to the present day, a much more overheated and frenzied housing market seems to be unstoppable. In 2016, unprecedentedly high housing prices again forced China’s government to propose the most serious house purchasing restrictions (HPRs) to date, that is, it limited loans and purchases, associating them with restricted selling periods of between five and eight years. At the same time, the central government ordered every local or city government to implement a local version of its ‘against real estate speculation’ policy.

To sum up, according to the spirit of a planned economy, the China government created and then controlled the development of its urban housing markets through its real estate policy recommendations. It is thus indispensable to know more about real estate policies when examining China’s housing issues.

Estimation results and policy implications

This section calculates the DY spillover index by using a generalised VAR with 1 lag based on the Bayesian information criterion (BIC). Using a 60-month (five-year) window, we can then obtain time-varying spillovers among the six cities according to total, directional and net spillovers.³ Alongside this, we provide a new spillover method, namely, the frequency dynamics of spillovers from Barunik and Krehlik (2018), to provide a richer analysis of the spillover effects. Finally, we present policy implications for pursuing a healthy and stable housing market in China.

Total spillovers

From the above section ‘DY spillover framework and different types of spillovers’, we can begin to estimate a generalised VAR based on the six cities’ housing returns during the full-sample period (2000–2017) in order to calculate the total spillovers, as seen in Table 1.⁴ First, the total spillover index based on (4) averages 48.8%, implying that ripple effects among the six cities in China do exist. Second, the outcome of net spillovers reveals that Beijing and Shenzhen are both source cities of housing markets in China over the full sample period (2000–2017). Finally, Chongqing’s housing market, where the former is 16% and the latter is 30.4%, is a net receiver of housing spillovers. At the same time, Chongqing is also the most isolated housing market due to the lowest spillovers ‘to’ or ‘from’ other cities.

Table 1.

Total spillovers during 2000–2017.

	Beijing	Shanghai	Tianjin	Chongqing	Shenzhen	Guangzhou	Contribution from others
Beijing	46.19	9.83	5.35	5.21	19.11	14.30	53.8
Shanghai	14.30	51.62	5.27	2.85	16.17	9.79	48.4
Tianjin	13.27	7.68	49.68	5.89	13.67	9.81	50.3
Chongqing	11.02	5.59	3.41	69.62	4.31	6.04	30.4
Shenzhen	15.27	12.41	12.07	1.79	45.95	12.51	54.1
Guangzhou	15.35	8.62	9.32	1.16	16.70	48.84	51.2
Contribution to others	69.2	44.1	35.4	16.9	70.0	52.5	288.1
Contribution including own	115.4	95.8	85.1	86.5	115.9	101.3	48.8 (288.1/600)

Note: Unit: %.

We further apply the rolling-window approach to help us better understand the process of spillovers, as seen in Figure 1, which well illustrates the time-varying evolution of total spillovers. In fact, Figure 1 presents two things: one that spillovers among cities start to increase following the 2008 global financial crisis, and the other that spillovers abruptly and substantially increase like a ‘big bang’ after 2014, and this high level of spillovers has remained up to the present. This is sufficient evidence to prove the authenticity of the time-varying nature of spillovers. The most important point is that the increasing total spillovers indicate that China’s urban housing market is very active, especially after 2014. Finally, the fact that increasing housing prices totally correspond to higher spillovers implies that spillovers are inextricably intertwined with housing prices, and so our study is essential to evaluating China’s housing market frenzies.

Figure 1.

Total return spillovers.

Directional and net spillovers among urban housing markets

While total spillovers only describe the average level of spillovers over all six cities, individual city-level spillovers ‘from’ or ‘to’ other cities can provide more information about the interactions among urban housing markets. In fact, by further using directional spillovers, which are divided into two types as in (6) and (7), we see many critical points of spillovers among the six cities using the rolling-window approach. First, from Figure 2(a) with directional spillovers from other cities, we investigate spillovers based on three intervals: the first period (2006–2010), when increasing spillovers from other cities prevailed over most cities; the second period (2010–2014), when decreasing spillovers from other cities began to appear in some cities, especially Shanghai, Chongqing and Shenzhen under two waves of housing control policies; and the third period (since 2015), which reveals that almost every city has begun to enjoy more spillovers from other cities than ever before. Second, Figure 2(b), which depicts directional spillovers to other cities, illustrates the ability of one city to affect other cities. We note that for the anti-housing price policies of 2009–2011, decreasing spillovers to other cities occurred in Tianjin, Shenzhen, Chongqing and Guangzhou. Thus, these policies have a small impact on urban housing markets. Shenzhen and Beijing are both at the centre of spillovers moving outwards to other cities, but since 2015 their influence has apparently been decreasing, while the spillovers moving outwards from the other four cities, especially Shanghai and Guangzhou, have been elevated. Therefore, spillovers to other cities are now coming from the other four cities, instead of from Beijing and Shenzhen.

Figure 2.

Directional return spillovers: (a) spillovers from other cities; (b) spillovers to other cities.

From these two types of spillovers, it is obvious that control policies during the period 2009–2011 had a negative impact on the directional spillovers (both ‘from’ and ‘to’) of other cities. Conversely, the policy after 2014 has resulted in much larger spillovers, especially those from other cities. When comparing the policies to control and to stimulate the housing markets, it is clear that the former has been less successful than the latter.

Having discussed the two types of directional spillovers, net spillovers based on spillovers to other cities minus spillovers from other cities should not be overlooked on the grounds that the concept of net spillovers is similar to that of ripple effects. For example, a positive value of net spillovers implies that a city has a spillover effect to other cities, and so this city can be identified as a source city. Similarly, a negative value of net spillovers reveals that a city is a receiver of spillover effects. It is more interesting to note that near-zero net spillovers denote bilateral interactions (or co-movement) among cities. Figure 3 again indicates that a structural change appears after 2014. Before 2014, significantly positive net spillovers point to Beijing and Shenzhen, while there are three main receivers of housing returns spillovers in Shanghai, Tianjin and Chongqing with significantly negative net spillovers. However, after 2014, the roles of source cities based on net spillovers in Beijing and Shenzhen have been decreasing.

Figure 3.

Net return spillovers.

As shown in Figure 2, there are relatively high spillovers ‘to’ other cities compared with those ‘from’ the other four cities during the years 2010–2014 for Beijing and Shenzhen. This is consistent with the case of significant positive net spillovers, as illustrated in Figure 3, in which Beijing and Shenzhen are definitely the source cities of other cities that have negative net spillovers. However, it is interesting to note that there has been a much higher value of total spillovers after 2014, confirming that there exist strong spillover effects among China’s urban housing markets, and this intense interaction has pushed housing prices higher and higher. Moreover, when spillovers from other cities are increasingly close to spillovers to other cities after 2014, this not only reveals the rise in total spillovers, but it also reflects the near-zero net spillovers across the six cities. This situation implies that spillover behaviour is a bilateral interaction among cities that have remarkable and equal spillovers ‘from’ and ‘to’ other cities – that is, there is co-movement among the six cities’ housing returns (Chiang, 2014). In fact, the co-movement among all urban housing markets fully indicates that very high system risk appears and the overheated housing market is now a national and not a local debate.

Frequency dynamics of spillovers

As mentioned in the ‘Frequency domain approach’ section, it is clear that Barunik and Krehlik’s (2018) frequency dynamics of spillovers is a frequency domain analogue of the Diebold–Yilmaz approach in both the short run and long run in the case of total, directional and net spillover indices. For this reason, we now analyse the frequency dynamics of spillovers among urban housing markets in China.

For the empirical analysis of this new method, we choose 1–3 months (quarterly) as the short run and more than three months as the long run to investigate different spillover behaviours. First of all, total spillovers over the total sample using quarterly and longer frequencies are shown in Table 2. It is found that spillovers among urban housing markets are higher in the long run (55.56%) than in the short run (44.14%). This result tells us that higher spillovers among housing markets reveal higher system risk in the long run; on the other hand, housing investors can implement a portfolio strategy to reduce housing market risk in the short run.

Table 2.

Total spillovers by frequency dynamics during 2000–2017.

(a) Frequency 1: 1–3 months
	Beijing	Shanghai	Tianjin	Chongqing	Shenzhen	Guangzhou	Contribution from others
Beijing	28.72	6.26	3.00	3.49	9.68	7.66	30.09
Shanghai	8.24	40.14	3.68	1.33	11.42	6.68	31.35
Tianjin	6.16	4.14	40.61	4.41	7.82	5.77	28.30
Chongqing	3.88	2.27	2.60	44.87	1.50	2.25	12.50
Shenzhen	10.99	9.34	7.69	1.65	30.19	9.43	39.10
Guangzhou	9.28	6.70	6.20	0.95	10.21	36.55	33.34
Contribution to others	38.55	28.71	23.17	11.83	40.63	31.79	174.68
Contribution including own	67.27	68.85	63.78	56.7	70.82	68.34	44.14 (174.68/395.76)
(b) Frequency 2: More than three months
	Beijing	Shanghai	Tianjin	Chongqing	Shenzhen	Guangzhou	Contribution from others
Beijing	17.47	3.57	2.35	1.72	9.43	6.64	23.71
Shanghai	6.06	11.47	1.59	1.53	4.75	3.11	17.04
Tianjin	7.12	3.54	9.06	1.48	5.85	4.04	22.03
Chongqing	7.16	3.32	0.81	24.72	2.82	3.80	17.91
Shenzhen	4.28	3.07	4.38	0.15	15.75	3.08	14.96
Guangzhou	6.07	1.93	3.11	0.22	6.49	12.29	17.82
Contribution to others	30.69	15.43	12.24	5.1	29.34	20.67	113.47
Contribution including own	48.16	26.9	21.3	29.82	45.09	32.96	55.56(113.47/204.23)

However, why are there higher spillovers in the long run? This question can further refer to total spillovers using the rolling window estimation in Figure 4. From this figure, it is suggested that, before 2014, spillovers in the long run are apparently higher than the ones in the short run on the grounds that housing investors exhibit a persistent belief that housing prices can steadily increase. Based on market efficiency, this outcome indicates that housing markets in China are not efficient, due to the price changes in housing markets not adjusting quickly (Krehlik and Barunik, 2017). Finally, after 2014, the extent of spillovers regardless of whether from a short- or long-run perspective is approximately the same as the result of suddenly increasing total spillovers in the short run, which fully reveals that systemic risk emerges in both the short run and long run, and so portfolio management among these urban housing markets may not have its desired effect. In short, after 2014, high total spillovers imply a high degree of systemic risk (Diebold and Yilmaz, 2012, 2014); moreover, the futility of portfolio management again proves the emergence of systemic risk and the national problem of housing frenzies in China. This conclusion is consistent with that for the DY spillover index.

Figure 4.

Total return spillovers of frequency dynamics.

As for directional spillovers depicted in Figure 5, spillovers regardless of whether ‘from’ or ‘to’ other markets in the long run exceed those in the short run. These results are similar to those for the total spillovers in Figure 4. Moreover, there are larger differences between the short run and long run in the case of spillovers ‘to’ other markets. Finally, net spillovers in Figure 6 show that before 2014, Beijing and Shenzhen are both the main causes and sources of high housing prices, and this result is consistent with those for the DY spillover index and Chiang (2014). However, after 2014, net spillovers decrease rapidly over time regardless of whether in the short run or long run. As mentioned earlier, the high total spillovers with near zero net spillovers imply that there is bi-literal interaction or co-movement. That is to say, we again prove that housing frenzies have become a full-scale phenomenon, rather than only for some specific cities.

Figure 5.

Directional return spillovers of frequency dynamics.

Figure 6.

Net return spillovers of frequency dynamics.

To sum up, using the frequency dynamics proposed by Barunik and Krehlik (2018) enables us to better understand the short run and long run behaviours of spillovers among urban housing markets in China. Based on this new version of total spillovers on a static basis, it is found that spillovers in the long run are stronger than those in the short run. A portfolio strategy (Barunik and Krehlik, 2018; Corbet et al., 2018; Krehlik and Barunik, 2017) tells us that overly-high spillovers lead to systemic risk, so a portfolio plan is more useful in the short run than in the long run. What is more, higher spillovers in the long run show that a persistent belief in the housing market needs a longer period of time, and this result is often cited by a general view on housing, namely, the housing price myth: when housing prices are continuously rising over a long period of time, the public usually expects that higher housing prices can last forever. Unfortunately, after the economic downturn of 2014, this belief turned into an even bigger expectation of rising housing prices whether from a short- or long-run perspective, so the authorities need to more seriously manage housing frenzies.

Policy implications

As stated above, the spillovers of housing returns among these six cities in China can be divided into three stages: 2006–2009, 2010–2014 and 2015–present. In the first stage, total spillovers and directional spillovers are both relatively small, and so high housing prices are concentrated in specific cities and there is little evidence of spillovers among cities. After the 2008 global financial crisis, the overheated housing markets must be evaluated by spillovers based on increasing total spillovers and increasing net spillovers. Beijing and Shenzhen are clearly identified as source cities, and local authorities should control housing prices in these two cities. If the housing prices of these two source cities are properly controlled, then other cities will not encounter the pressure of rising housing prices.

However, a strong bilateral interaction, namely, co-movement across China’s urban housing market, has been detected after 2014 as a result of observing the increasing total spillovers and the close-to-zero net spillovers. This result suggests that the housing problem is not confined to the source cities. In other words, the housing frenzy in China has become a national problem due to high systemic risk. As a result, the central government should initiate a nationwide policy for overheated urban housing markets, such as a new macro-control policy. To sum up, the interaction of China’s housing market is changing from specific local cases, that is, source cities with interaction among cities, to the overall housing market. That is to say, the scope of the housing problem is continuously magnifying.

On the other hand, the frequency dynamics of spillovers show that there are stronger spillovers in the long run, rather than the short run. What is worse is that, after 2014, a rapid increase in total spillovers in the short run seems to support an approximate equality between short- and long-run spillovers. This implies that regardless of whether considered from a short-run or long-run perspective, housing investors cannot propose any portfolio opportunity to diversify housing fluctuations; at the same time, this result suggests the possibility of short-run speculation across all first-tier housing markets. In other words, it is imperative to implement an effective policy to control housing frenzies in China now, and a macro control policy would be a good choice.

Why has China’s housing problem become a national debate after 2014? Zhang (2013) suggested that credit booms and increasing money supply are the main reasons for the soaring housing prices. It would be better to say that an excess of liquidity is the main factor for the overheated housing market with stronger spillovers and source cities. Second, the new policy after 2014 associated with the economic downturn has led to China’s housing problem becoming a national rather than a local concern. This is because this new policy is the first recommendation to cancel mortgage credit restrictions and to spur the housing market in order to prevent the economy from slowing down. However, because the new policy makes the public believe that real estate development is indispensable to economic growth, it has become very difficult for the authorities to cool down the housing market. Finally, since the housing frenzy has become a national problem, a new macro-control policy for the housing market is necessary right now.

Conclusions

While housing market frenzies have increased to such an extent as to become a remarkable characteristic of China’s economy, how to correctly evaluate them remains an open question. This article applies the DY spillovers index using a rolling window and frequency domain spillovers to evaluate this overheated sector for six first-tier cities during the period 2000–2017 on a monthly basis. In a departure from earlier studies, the DY spillover index employs the rolling-window approach to capture the time-varying process of spillovers among urban housing markets. At the same time, time-varying spillovers can be used to fully explain two critical issues in China’s housing market: the rapidly-increasing housing prices as well as policy interventions.

As for policy recommendations, ever since real estate commercialisation was launched in 1998 in China, there have been more than 10 initiatives, including one policy designed to stimulate the housing market after 2014. Using three types of spillover indices, it is clear that time-varying spillovers are the best way to describe China’s housing market. At the same time, the country’s housing market development can be divided into three phases: individual and isolated events (2006–2009), source cities based on positive net spillovers (2010–2014) and co-movement among cities (2015–present). In actual fact, the sequence of policy directives to cool down the housing market has been less successful than the one policy adopted to stimulate the market after 2014. This result shows that China’s housing market is still exhibiting an upwards trend, and so any policy directed against the housing sector will not be very effective. On the contrary, a policy that spurs the housing market can easily go out of control. Moreover, after 2014, this new policy to stimulate the housing market has altered the spillovers among cities, further exacerbating the housing market problem to the national level, rather than it just remaining a source-city issue. Similarly, the frequency dynamics of spillovers still suggest that, after 2014, housing frenzies have become a serious national problem due to the high total spillovers and near-zero net spillovers, regardless of whether from a short- or long-run perspective. These findings confirm that any macro-control policy aimed at China’s housing market must have explicit directives and goals in the face of exceptionally high housing prices.

Footnotes

Appendix

Acknowledgements

We would like to thank the editor and three anonymous referees for their most valuable comments to greatly improve the earlier version of this paper. Professor Chiang is indebted to Shang-hsien Chiang, his father, for inspiring and supporting his research.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Financial support is offered by Ministry of Science and Technology, Taiwan (MOST 108-2410-H-033-030).

ORCID iD

Shu-hen Chiang

Notes

References

Alexander

Barrow

(1994) Seasonality and cointegration of regional house prices in the UK. Urban Studies 31: 1667–1689.

Barunik

Krehlik

(2018) Measuring the frequency dynamics of financial connectedness and systemic risk. Journal of Financial Econometrics 16: 271–296.

Brady

(2014) The spatial diffusion of regional housing prices across U.S. states. Regional Science and Urban Economics 46: 150–166.

Chen

Guo

(2011a) One decade of urban housing reform in China: Urban housing price dynamics and the role of migration and urbanization, 1995–2005. Habitat International 35: 1–8.

Chen

Chien

Lee

(2011b) Dynamic modeling of regional house price diffusion in Taiwan. Journal of Housing Economics 20: 315–332.

Chiang

Sing

Tsai

(2017) Spillover risks in REITs and other asset markets. Journal of Real Estate Finance and Economics 54: 579–604.

Chiang

(2014) Housing markets in China and policy implications: Co-movement or ripple effect. China & World Economy 22: 103–120.

Chiang

(2016) Rising residential rents in Chinese mega cities: The role of monetary policy. Urban Studies 53: 3493–3509.

Corbet

Meegan

Larkin

, et al. (2018) Exploring the dynamic relationships between cryptocurrencies and other financial assets. Economics Letters 165: 28–34.

10.

Costello

Fraser

Groenewold

(2011) House prices, non-fundamental components and interstate spillovers: The Australian experience. Journal of Banking and Finance 35: 653–669.

11.

Diebold

Yilmaz

(2009) Measuring financial asset return and volatility spillovers with application to global equity markets. Economic Journal 119: 158–171.

12.

Diebold

Yilmaz

(2012) Better to give than to receive: Predictive directional measurement of volatility spillovers. International Journal of Forecasting 28: 57–66.

13.

Diebold

Yilmaz

(2014) On the network topology of variance decompositions: Measuring the connectedness of financial firms. Journal of Econometrics 182: 119–134.

14.

Drake

(1995) Testing for convergence between UK regional house prices. Regional Studies 29: 357–366.

15.

Fereidouni

Al-Mulali

Lee

JYM

, et al. (2016) Dynamic relationship between house prices in Malaysia’s major economic regions and Singapore house prices. Urban Studies 50: 657–670.

16.

Glaeser

Huang

, et al. (2017) A real estate boom with Chinese characteristics. Journal of Economic Perspectives 31: 93–116.

17.

Gong

Boelhouwer

(2016) Spatial interrelations of Chinese housing markets: Spatial causality, convergence and diffusion. Regional Science and Urban Economics 59: 103–117.

18.

Gupta

Miller

(2012) ‘Ripple effects’ and forecasting home prices in Los Angeles, Las Vegas, and Phoenix. Annals of Regional Science 48: 763–782.

19.

Hui

ECM

Yue

(2006) Housing price bubbles in Hong Kong, Beijing and Shanghai: A comparative study. Journal of Real Estate Finance and Economics 33: 299–327.

20.

Krehlik

Barunik

(2017) Cyclical properties of supply-side and demand-side shocks in oil-based commodity markets. Energy Economics 65: 208–218.

21.

Lau

Vigne

Wang

, et al. (2017) Return spillovers between white precious metal ETFs: The role of oil, gold and global equity. International Review of Financial Analysis 52: 316–332.

22.

Lee

Chien

(2011) Empirical modeling of regional house prices and the ripple effect. Urban Studies 48: 2029–2047.

23.

Liow

(2014) The dynamics of return co-movements and volatility spillover effects in Greater China public property markets and international linkages. Journal of Property Investment and Finance 32: 610–641.

24.

Liow

(2016) Global financial crisis and cyclical co-movements of Asian financial markets. Journal of Property Investment and Finance 34: 465–495.

25.

Liow

Yue

(2017) Return and co-movement of major public real estate markets during global financial crisis: A frequency domain approach. Journal of Property Investment and Finance 35: 489–508.

26.

Liow

Liao

Huang

(2018) Dynamics of international spillovers and interaction: Evidence from financial market stress and economic policy uncertainty. Economic Modelling 68: 96–116.

27.

Liu

Luo

, et al. (2008) Identifying house price diffusion patterns among Australian state capital cities. International Journal of Strategic Property Management 12: 237–250.

28.

MacDonald

Taylor

(1993) Regional house prices in Britain: Long-run relationships and short-run dynamics. Scottish Journal of Political Economy 40: 43–55.

29.

Meen

(1999) Regional house prices and the ripple effect: A new interpretation. Housing Studies 14: 733–753.

30.

Tiwari

Cunado

Gupta

, et al. (2018) Volatility spillover across global asset classes: Evidence from time and frequency domains. Quarterly Review of Economics and Finance 70(C): 194–202.

31.

Tsai

(2015) Spillover effect between the regional and the national housing markets in the UK. Regional Studies 49: 1957–1976.

32.

Wei

Zhang

(2011) The competitive saving motive: Evidence from rising sex ratios and savings rates in China. Journal of Political Economy 119: 511–564.

33.

Weng

Gong

(2017) On price co-movement and volatility spillover effects in China’s housing markets. International Journal of Strategic Property Management 21: 240–255.

34.

Gyourko

Deng

(2012) Evaluating conditions in major Chinese housing markets. Regional Science and Urban Economics 42: 531–543.

35.

Yunus

Swanson

(2013) A closer look at the U.S. housing market: Modeling relationships among regions. Real Estate Economics 41: 542–568.

36.

Zhang

(2013) Money, housing and inflation in China. Journal of Policy Modeling 35: 75–87.

37.

Zhang

(2012) What drives China’s house prices: Marriage or money? China & World Economy 20: 19–36.

38.

Zhou

(2012) Volatility spillovers between the Chinese and world equity markets. Pacific-Basin Finance Journal 20: 247–270.