Content uploaded by Rose Neng Lai

Author content

All content in this area was uploaded by Rose Neng Lai on Feb 09, 2021

Content may be subject to copyright.

Shadow Banking and Property Market 361

INTERNATIONAL REAL ESTATE REVIEW

2019 Vol. 22 No. 3: pp. 361 – 399

Shadow Banking and the Property Market in

China

Rose Neng Lai

Professor, Department of Finance and Business Economics, University of

Macau, Taipa, Macau, China. E-mail: RoseLai@um.edu.mo.

Robert Van Order

Oliver Carr Professor of Real Estate and Professor of Finance and Economics

at George Washington University, Washington D.C. E-mail: rvo@gwu.edu.

This paper studies the evolution of property values and the connections

between shadow banking and property markets in China. We use

Pooled Mean Group estimation to analyze Chinese house prices in 65

cities from 2007-2016, define the “fundamentals” of housing prices with

the Gordon dividend discount model, and use lagged rents, prices, real

and nominal interest rates, and shadow banking activity as short term

explanatory factors. We find that the cities tend to share long run

fundamentals and adjust relatively quickly to deviations from the

fundamentals. We do not find bubbles; rather houses are like growth

stocks with house prices rapidly chasing growing rents. More importantly,

we find that house prices increase more quickly with the availability of

shadow banking funds, which have grown rapidly.

Keywords

Chinese Housing Market, Shadow Banking, Pooled Mean Group Estimation

362 Lai and Van Order

1. Introduction

Property values have increased rapidly in China in the last decade, and one of

the contributing factors have been public policy changes. The Chinese

government claimed that a housing bubble had been successfully “deflated” at

the end of 2014, and subsequently relaxed restrictions on second-home

purchases upon the emergence of an economic slowdown in 2015, with the aim

to boost the property market as a means of supporting the dampened economy.

Chinese households have been since buying more apartments, and developers

have been borrowing more to fund their construction projects. The ensuing

demand for funding has boosted the expansion of the shadow banking system.

Our interest in this paper is the role of that system in the growth of property

values.

As in Lai and Van Order (2017), one of the ways to study the dynamics of house

prices is to apply the Gordon dividend discount model, which makes use of

rental and interest rates to explain for long run house prices, to which actual

prices adjust over time. This leads to a well- defined long run equilibrium, but

with a less restrictive adjustment process. We use Pooled Mean Group (PMG)

(and Mean Group (MG)) estimation to separate long run from short run. We

study the Chinese housing market with a focus on shadow banking as the source

of funding. Our specification forces shadow banking to only affect short run

property value adjustment (momentum), but not long run fundamentals. We

exploit the fact that shadow banking policies are made at the national level, but

we use them to estimate the determinants of city by city price changes, as a way

of avoiding endogeneity problems and suggesting causality from shadow

banking to price changes.

To the best of our knowledge, we are the first to formally test the effects of

shadow banking on the property markets in China and model Chinese house

prices by decomposing the effects into long run fundamentals and short run

adjustments. We find that the cities tend to share long run fundamentals and

adjust relatively quickly to deviations from the fundamentals, and we do not

find bubbles. We also find that housing growth in the short run is related to the

availability of shadow banking funds, which have also grown rapidly. A policy

suggestion for the Chinese government is to focus on regulatory monitoring in

this funding sector. Not only can its contraction hurt property markets, its non-

performing loans can trigger contagion to the main banking system and

therefore the economy as a whole.

2. Shadow Banking in China

The Financial Stability Board (FSB) (2015) defines shadow banking as “credit

intermediation involving entities and activities outside of the regular banking

system”. An important element of being outside the regular banking system is

Shadow Banking and Property Market 363

the absence of deposit insurance coverage. The FSB points out that non-bank

credits contribute to financing the real economy, simultaneously becoming a

source of systemic risk when they are highly interconnected with the regular

banking system.

1

In China, shadow banking can be broadly defined as non-bank financing, such

as trust and entrusted loans, bankers’ acceptances, interbank entrusted loan

payments, microfinance companies, financial leasing, special purpose finance

companies associated with e-commerce, guarantees, pawn shops and unofficial

lenders, bond markets, trust beneficiary rights, and wealth management

products (WMPs), and interbank market activities (see Elliott et al. (2015) for

detailed descriptions of each of these sources of shadow banking). Some of the

non-bank channels, such as bond markets and interbank market activities,

should not be classified as shadow banking. For instance, the bonds here refer

to corporate bonds, which are not generally traded in the bond markets like

those in the US, whereas interbank market activities are really large

corporations that are using finance company subsidiaries to act like banks. The

most common source of shadow banking funds that is collected from the

general public comes from the WMPs that pool assets together. Most of the

assets are loans. Sharma (2014) and Hsu and Li (2015) are some examples of

the literature on shadow banking in China.

In response to the Global Financial Crisis, the Chinese central government

initiated a stimulus package in 2009, which was followed by unprecedented

growth in fixed asset investments that were increasingly funded by shadow

banking. Infrastructure projects constituted 72% of the stimulus package, of

which 30% was funded by the central government, while the rest was from local

governments.

2

Funding was also available to riskier borrowers, typically real

estate developers and local government financing vehicles, the corporate arm

of local governments, which helped local governments generate their high

GDPs through infrastructure construction.

The growth of shadow banking has been fueled by the fact that the five largest

banks in China, all state-owned, are only allowed to lend to the large state-

owned enterprises (SOEs) but not corporates and small and medium sized

enterprises (SMEs). Hence, the shadow banking sector provides lending needs

outside regulations. On the supply side, the lack of investment opportunities

(made up of only the stock markets, the very small bond market, and the very

hot real estate market) stimulates all sorts of WMPs that can generate returns

higher than the very low (sometimes even negative in real terms) deposit rates.

Furthermore, average investors are under the impression that these products are

1

Various issues of the “Global Shadow Banking Monitoring Report” by the Financial

Stability Board.

2

Reported by Sarah Hsu in “The Rise and Fall of Shadow Banking in China – How

shadow banking became the catch-all for riskier” The Diplomat, 2015 11 09, available

at http://thediplomat.com/2015/11/the-rise-and-fall-of-shadow-banking-in-china/

364 Lai and Van Order

safe because they are mostly sold by big banks, which supposedly have implicit

guarantees from the People’s Bank of China (the Chinese central bank),

ignoring the fact that these big banks are only intermediaries and do not provide

any guarantees.

While boosting local incomes, over-investment in infrastructure has generated

“ghost towns”, with roads and bridges that very few people use. It is reported

that shadow banking makes up 20%-41% of on-balance bank lending, without

which total lending would have declined by 16-29%.

3

The downturn of the

Chinese property market resulted in the lack of liquidity for developers. Yet the

Chinese government is still optimistic that the shadow banking sector in China

is only a small problem.

Table 1 shows that shadow banking is only 26% of the GDP in China, which

ranks 13th among the 26 jurisdictions according to the Financial Stability Board

(2015), relative to 82% in the US. A comparison between Figures 1 and 2 shows

that, unlike the US (Figure 1) where funding comes from various sources, (with

“Other Financial Institutions” as the dominant sector) banks in China (Figure

2) dominate funding supply. It should be noted, however, that by referring to

only the “economic function-based” measures of shadow banking, the FSB

might have underestimated the proportion of loans in the total loan system.

As shown in Table 2, shadow banking in China has grown very rapidly, from

1.6% of all assets in financial intermediation to 7.7%, thus becoming the third

largest sector in terms of size. Along with that, according to Elliott and Yan

(2013), there are large pools of bad loans that are not acknowledged by banks.

An example is the situation in Wenzhou, a small city that eventually prospered

from profitable SMEs and was able to obtain financing through various

channels of shadow banking, subsequently followed by widespread defaults

(after 2012) because of the economic slowdown. Sheng et. al. (2015) report that

the real estate sector makes up 18% of shadow banking assets as of 2013, which

is the third largest industry. Also based on their calculations, an estimate of 22%

to 44% of the non-performing loans in shadow banking will be brought back to

the banking system. In fact, it is also reported that there are RMB1.19 trillion

(USD1 RMB7 as at mid-2019) bad loans at the end of September 2015, up

from RMB842.6 billion at the end of 2014

4

, and a 22% non-performing loan

rate in the whole financial system at the end of 2016.

5

It seems that even though

the central government is willing to stabilize the market through intervention,

3

Ibid.

4

“China’s December New Bank Loans Miss Expectations”, MarketWatch, January 15,

2016.

5

As reported by Peter Eavis in “Toxic Loans Around the World Weigh on Global Growth”

in The New York Times on February 3, 2016.

Shadow Banking and Property Market 365

there is still the risk of bad-debt that leads to contagion throughout the financial

sector.

6

Figure 1 FSB Assets of Financial Institutions and Economic Function-

Based Shadow Banking Measure – USA

Panel A: In Billions USD

Panel B: In Percentage

Source: Financial Stability Board (2015)

6

See, for example, the Bloomberg reports in “Be Scared of China's Debt, Not Its Stocks”

on January 7, 2016, available at http://www.bloombergview.com/articles/20160107/

bescaredofchinasdebtnotitscrashingstocks, and “Mid-tier Chinese banks piling up

trillions of dollars in shadow loans” of Thomson Reuters on January 31, 2016, available

at http://www.reuters.com/article/china-banks-investment-idUSL8N 156053.

0.0

10,000.0

20,000.0

30,000.0

40,000.0

50,000.0

60,000.0

70,000.0

80,000.0

90,000.0

100,000.0

2002 2005 2008 2011 2014

Shadow

banking

Other Financial

Intermediaries (OFIs)

Public Financial

Institutions

Pension Funds

Insurance Companies

Banks

Central Banks

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

2002 2005 2008 2011 2014

Shadow

banking

Other Financial

Intermediaries (OFIs)

Public Financial

Institutions

Pension Funds

Insurance Companies

Banks

Central Banks

366 Lai and Van Order

Figure 2 FSB Assets of Financial Institutions and Economic Function-

Based Shadow Banking Measure – China

Panel A: In Billions USD

Panel B: In Percentage

Source: Financial Stability Board (2015) and Lai and Van Order (2017)

0.0

5,000.0

10,000.0

15,000.0

20,000.0

25,000.0

30,000.0

35,000.0

40,000.0

45,000.0

2003 2005 2007 2009 2011 2013

Shadow

banking

Other Financial

Intermediaries (OFIs)

Pension Funds

Insurance Companies

Banks

Central Banks

0%

10%

20%

30%

40%

50%

60%

70%

80%

90%

100%

2003 2005 2007 2009 2011 2013

Shadow

banking

Other Financial

Intermediaries (OFIs)

Pension Funds

Insurance Companies

Banks

Central Banks

Shadow Banking and Property Market 367

Table 1 Shadow Banking, Other Financial Intermediaries (OFIs) and

Banks as Percentage of GDP of 26 Jurisdictions: End of 2014

Shadow banking

OFIs

Banks

Indonesia

1

8

54

Russia

4

5

109

Saudi Arabia

5

5

74

Argentina

6

6

30

Turkey

6

11

108

Singapore

10

90

607

Mexico

16

23

40

Italy

17

38

223

India

19

17

95

Hong Kong

20

85

817

Spain

21

69

267

Chile

23

31

106

China

26

29

271

South Africa

27

61

108

Australia

27

64

211

Brazil

33

60

91

Korea

48

100

205

Canada

58

147

228

Total

59

112

223

Japan

60

87

374

France

61

96

370

Germany

73

81

241

The Netherlands

74

838

326

United States

82

148

122

Switzerland

90

277

364

United Kingdom

147

326

601

Ireland

1190

1551

363

Notes: Banks = broader category of ‘deposit-taking institutions’; OFIs = Other Financial

Intermediaries; and Shadow Banking = economic function-based measure of

shadow banking.

Source: Financial Stability Board (2015)

368 Lai and Van Order

Table 2 Share of Shadow Banking Assets as Percentage of All

Financial Intermediations of 26 Jurisdictions: End of 2010

and 2014

End of 2010

End of 2014

United States

40.9

39.7

United Kingdom

13.0

11.4

China

1.6

7.7

Ireland

6.9

7.6

Germany

7.1

7.2

Japan

9.5

6.8

France

6.1

4.4

Canada

2.4

2.8

Brazil

2.0

1.9

Korea

1.3

1.8

The Netherlands

1.8

1.7

Switzerland

1.5

1.6

India

0.9

1.1

Australia

1.3

1.0

Italy

1.2

0.9

Spain

1.0

0.7

Mexico

0.5

0.5

South Africa

0.3

0.2

Hong Kong

0.1

0.2

Chile

0.1

0.2

Russia

0.1

0.1

Turkey

0.1

0.1

Saudi Arabia

0.1

0.1

Argentina

0.0

0.1

Singapore

0.2

0.1

Indonesia

0.0

0.0

Note: Shadow banking is based on an economic function-based measure.

Source: Financial Stability Board (2015)

3. The Chinese Housing Market

There have been many studies on the US housing bubble. Examples include

Black and Hoesli (2006), Chan et al. (2001), Chang et al. (2005), Coleman et

al. (2008), Hwang et al. (2006), and Wheaton and Nechayev (2008). However,

studies on the Chinese house price movements are scant, although studies on

the housing markets themselves are extensive. Deng et al. (2009) and Yang and

Chen (2014), among others, focus on the Chinese housing policy reform. Others

such as Wu et al. (2012) discuss the sustainability of its boom. The links

between house prices and land policies are studied in Cai et al. (2013), and Peng

and Thibodeau (2009). Ren et al. (2012) are one of the few to explicitly measure

Shadow Banking and Property Market 369

the extent of the Chinese house price run up. A very recent study is Glaeser et

al. (2017). Lai and Van Order (2018) compare the housing bubbles between

China and the US. To the best of our knowledge, there is no study on the effect

of shadow banking as a source of funding on Chinese house prices.

This paper studies the property markets in China over the past decade. We use

a variation of the Gordon dividend discount model as the common

representation of long run fundamentals, but the model allows a short run

momentum that can vary across cities. We are able to test and estimate a long

run fundamental model, as well as the short run adjustments and momentum

across cities. Associated with this is an estimation of how fast deviations from

the long run are corrected. For further analysis, we also classify cities into Tier

1 and Tier 2 cities (which are official classifications based on size and speed of

development) and coastal versus inland (because coastal cities are those that

have had more advanced development for a longer period of time). Our prior is

that the housing bubble would be larger in Tier 1 and/or coastal cities.

3.1 Data

We use monthly house price and rental series for 65 cities over the period of

2005 – 2014 obtained from CityRE Data Technology Co. Ltd,

7

which is the first

to compile comprehensive data on housing for sale and lease for over 290 cities

and areas in China starting from 2003. Numerous studies such as Ren et al.

(2012) use house price and rental indices from the CEIC Data. However, the

CEIC database only covers 35 cities. Monthly data on the sales price indices of

newly constructed residential buildings for 70 cities from 1997 onwards can be

obtained from the National Bureau of Statistics (NBS) of China; however, the

data were no longer collected after 2010. The property price index of the China

Real Estate Index System (CREIS) can also be used, especially after the

termination of the housing index in 2010. Note that it might be a challenge to

compare house price and rent series if the underlying representative housing

units are not comparable, for instance, due to improved quality over time.

Fortunately, such an effect is not likely to have significant impact on the

empirical results due to the relatively large volume of transactions, and the

relative homogeneity of housing within cities in China.

We use the 5-year Chinese government bond rates obtained from the National

Interbank Funding Center as a proxy of nominal long-term risk-free rate, and

from which we also obtain real interest rates by using the consumer price index

(CPI) from the NBS of China. We also proxy risk relative to government bonds

with the 5 year AAA corporate bond yields (obtained from the China Central

7

Details of CityRE Data Technology Co., Ltd. can be obtained from

http://www.cityre.cn/en/ or http://www.cityhouse.cn. They claim to operate the largest

real estate data set in China.

370 Lai and Van Order

Depository & Clearing Co., Ltd.), so that the yield spread is the 5-year corporate

bond yields minus the 5-year Chinese government bond yields.

3.2 Impact from Shadow Banking

The shadow banking data, classified as “loans from non-banking financial

institutions”, are reported by the NBS, with a monthly frequency that spans the

period of January 2006 to December 2015. These are nationwide data, and the

only public source of data on shadow banking. We cannot rule out the existence

of a significant amount of funds from informal banking via many informal

channels that are unfortunately not systematically and officially recorded. To

control for the basic factors in the real estate sector, we also include the house

price and the rental indexes from the CPI series published by the NBS as well

as interest rate, proxied by nominal lending rates for housing loans issued by

the People’s Bank of China.

In order to study how shadow banking affects investments in the real estate

sector in China, we include data on “housing completed”, “housing sold” and

“housing starts”, and land purchases and investment. All are aggregate monthly

data from the NBS, some of which start as early as March 1998, while data that

start as late as March 2007 and March 2008 are on land purchases and

investment. The “Housing Completed” category is proxied by “Floor Space

Completed”, “Floor Space Completed: Commodity Building – Residential”,

“Floor Space Completed – Residential”, and “Floor Space Completed: 40 cities

– Residential”. “Housing Sold” is proxied by “Floor Space Sold: Residential:

Presale”, “Floor Space Sold: Residential: Existing Units”, “Floor Space Sold:

Residential”, “Building Sold: Residential”, “Building Sold: Residential:

Presale”, and “Building Sold: Residential: Existing Units”. “Housing Starts”

are proxied by “Floor Space Started: Commodity Building: Residential” and

“Floor Space Started: 40 cities: Residential”. Finally, “Land Purchases and

Investment” is proxied by “Land Area Purchased”, “Real Estate Investments:

Residential”, “Real Estate Investments: New Increase”, “Real Estate

Investments”, “Real Estate Investments: Land Transactions”, “Land Area

Purchased: 40 cities” and “Real Estate Investments: Residential: 40 cities”.

All of the variables are presented in percentage changes. We perform unit root

tests and cannot reject that the log differenced (percentage change) data are

stationary. As expected, the rental rates do not have much influence on the

demand and supply measures of real estate because the rental market is very

small, and hence, we rerun the regressions without rental rates on the right-hand

side. The signs of the explanatory variables are mostly as expected. That is,

interest rates have negative effects while funds from shadow banks have

positive effects on the housing market. Other variables that do not seem to have

the expected signs are not significant anyway. Table 3 shows the results with

various proxies of real estate demand and supply.

Shadow Banking and Property Market 371

Table 3 Regression of Measures of Real Estate Supply and Demand

on Shadow Banking Loans

Panel A: Buildings and Floor Space Sold

Buildings Sold

Proxy 1

Proxy 2

Proxy 3

Housing Price

2.819***

4.79

8.443**

Shadow Banking

1.185***

1.361***

1.250***

Interest rates

-0.242**

-0.372**

-0.420***

Constant

0.022

-0.009

-0.008

Observations

100

70

70

Adjusted R-squared

0.664

0.638

0.689

Floor Space Sold

Proxy 1

Proxy 2

Proxy 3

Housing Price

2.731***

1.516

3.206***

Shadow Banking

1.312***

1.426***

1.277***

Interest rates

-0.248**

-0.192*

-0.268***

Constant

0.005

0.003

0.005

Observations

100

100

100

Adjusted R-squared

0.69

0.662

0.691

Panel B: On Floor Space Started and Completed

Floor Space Started

Proxy 1

Proxy 2

Housing Price

1.813***

2.776***

Shadow Banking

1.095***

1.275***

Interest rates

-0.169***

-0.218**

Constant

-0.003

-0.024

Observations

72

100

Adjusted R-squared

0.653

0.744

Floor Space Completed

Proxy 1

Proxy 2

Proxy 3

Proxy 4

Housing Price

-1.652

-2.062**

-1.961*

-3.424***

Shadow Banking

1.012***

0.878***

0.937***

0.373**

Interest rates

0.059

0.107

0.076

0.222**

Constant

0.106***

0.086***

0.094***

0.146***

Observations

100

100

100

72

Adjusted R-squared

0.552

0.509

0.534

0.195

372 Lai and Van Order

Panel C: On Land Purchases and Real Estate Investments

Land Purchases

Proxy 1

Proxy 2

Housing Price

2.338***

2.264***

Shadow Banking

1.404***

1.090***

Interest rates

-0.117

-0.157**

Constant

-0.033*

0.019

Observations

100

81

Adjusted R-

squared

0.779

0.558

Real Estate Investments

Proxy 1

Proxy 2

Proxy 3

Proxy 4

Proxy 5

Housing Price

3.320***

1.868**

-2.214**

3.387***

2.278***

Shadow Banking

1.313***

1.298***

0.668***

1.292***

1.032***

Interest rates

-0.248**

-0.202**

0.118

-0.245**

-0.164**

Constant

0.002

-0.008

0.108***

0.006

0.035*

Observations

100

79

100

100

81

Adjusted R-

squared

0.713

0.739

0.42

0.707

0.593

Note : *, ** and *** denote significance at the 10%, 5% and 1% levels respectively.

All variables except interest rates are differenced log values to obtain percentage

changes of the values. Interest rates are differenced values.

It can be seen from Panel A of Table 3 that high housing prices and availability

of shadow banking funds can increase the volume of buildings and floor space

sold, while interest rates have negative effects as expected. In particular, an

increase of 1% in the shadow banking funds increase, say, the floor space of

existing units (Proxy 2) by 1.426%, or the floor space of presale units (Proxy 3)

by 1.277%. Similarly, available funds and surging house prices can encourage

construction, as demonstrated by Floor Space Started in Panel B. While the

variables also exert effects on Floor Space Completed, they are negatively

affected by housing price, which is not reasonable. Nevertheless, note first that

they are mostly not significant, and more importantly, the house price variable

is not lagged, and therefore should not be a factor that would deter construction

which has already taken place. Panel C also shows interesting results in that

there will be real estate investments and land purchases as long as there is

funding and housing price continues to increase; therefore, interest rates do not

seem to be very important when considering investments by developers.

We use lagged shadow banking variables to represent situations where funding

was needed for construction one or two years beforehand in order for units to

be completed and sold . Alternatively, funds from shadow banking might be

needed immediately for purchases of housing units, the completion of which

would be triggered by high property prices one or two years ago. We repeat the

regression in two sets of tests with six-month, 1-year, 18-month, and 2-year lags,

Shadow Banking and Property Market 373

given that most housing buildings can be completed in around two years.

Similar lags are applied to all explanatory variables. The first set includes lags

for all variables, while the second set includes lags for all variables except

shadow banking funds. The rationale for the former is that current investment

decisions can be attributed to observations of a good market over the past period.

The rationale of the latter is that construction decisions are often made

beforehand, while funding is needed immediately to stimulate purchases. The

results are shown in Table 4 (tests with various lags generate similar results and

therefore not all are provided). Interestingly, shadow banking funds are the most

significant explanatory variable, regardless whether we use no lags, six-month,

1-year, 18-month, or two year lags. Moreover, when shadow banking is

considered, even a dominating factor such as housing price index sometimes

shows a negligible influence.

From Panels A and B of Table 4, it is clear that real estate investments and land

purchases are affected by the lagged availability of funds. Interestingly, housing

price is not a strong consideration for land purchases, probably because

developers always favor stacking up land banks for any option for construction

whenever housing prices become favorable. Panel C shows that the property

sales market is affected by house price and interest rate variables lagged by six

months, while shadow banking funds are not lagged; that is, testing whether

past information on house prices and interest rates trigger more purchases if

funds are available today. Again, while the variables are correctly signed, only

current funding availability is important. In terms of the decisions of the

developers, commencement of construction (i.e. Floor Space Started) and land

purchases are mostly based on availability of funding. While this is not very

convincing, it should be noted that our sample period covers a market boom

period when investors and developers were optimistic about the market, but

also when there were government restrictions and policies that curbed the

market, and the housing prices were once affected. Developers were apparently

willing to invest as long as there was funding.

We also study if there is an increased demand from shadow banking because of

increased demand and supply of real estate investments. The results are shown

in Table 5. All of the variables that represent real estate investments, building

and development, and housing completed and sold exert significant and positive

effects on shadow banking. In other words, both demand and supply of real

estate trigger demand for more shadow banking funds. While it is logical that

higher interest rates also attract a larger supply of shadow banking funds, it is

also logical to interpret that lower housing prices attract more buying and

therefore increase the demand for shadow banking funds.

In general, the results confirm that shadow banking might be an important,

although perhaps endogenous, factor in real estate investment in China. To

check for robustness, we also test for the presence of autocorrelation of the

regression residuals in the above tests. All of the regression results show no

autocorrelation. Next, we include shadow banking funds as a short term

374 Lai and Van Order

variable that affects the model for the pricing of housing units across cities in

China.

Table 4 Regression of Measures of Real Estate Supply and Demand

on Lagged Explanatory Variables, Including Shadow

Banking Loans

Panel A: Land Purchases and Real Estate Investments with 1-year Lag

Variables

Land Purchases

Proxy 1

Proxy 2

Housing Price

1.943

1.618*

Shadow Banking

1.218***

0.968***

Interest rates

-0.186

-0.202**

Constant

-0.001

0.038*

Observations

90

81

Adjusted R2

0.57

0.435

Real Estate Investments

Proxy 1

Proxy 2

Proxy 3

Proxy 4

Proxy 5

Housing Price

3.050***

1.146

-1.943*

3.010***

2.256***

Shadow Banking

1.296***

0.994***

0.632***

1.284***

0.960***

Interest rates

-0.225**

-0.213

0.136

-0.216*

-0.161*

Constant

0.004

0.049*

0.108***

0.006

0.045**

Observations

90

79

90

90

81

Adjusted R2

0.699

0.483

0.4

0.702

0.535

Panel B: Land Purchases and Real Estate Investments with 2-year Lag

Variables

Land Purchases

Proxy 1

Proxy 2

Housing Price

1.959

1.182

Shadow Banking

1.100***

0.814***

Interest rates

-0.12

-0.052

Constant

0.015

0.062**

Observations

80

72

Adjusted R2

0.576

0.336

Real Estate Investments

Proxy 1

Proxy 2

Proxy 3

Proxy 4

Proxy 5

Housing Price

2.926***

1.879

-2.245**

2.902***

2.369***

Shadow Banking

1.253***

0.930***

0.595***

1.246***

0.932***

Interest rates

-0.206*

-0.064

0.138

-0.204*

-0.167*

Constant

0.009

0.058**

0.110***

0.011

0.048**

Observations

80

79

80

80

72

Adjusted R2

0.707

0.509

0.41

0.708

0.531

Shadow Banking and Property Market 375

Panel C: Buildings and Floor Space Sold with 6-month Lag in Price Index

and Lending Rates but No Lag for Shadow Banking Funds

Buildings Sold

Proxy 1

Proxy 2

Proxy 3

Housing Price

0.217

0.591

0.094

Shadow Banking

1.153***

1.335***

1.221***

Interest rates

-0.169

-0.206*

-0.173

Constant

0.034*

0.007

0.019

Observations

100

70

70

Adjusted R2

0.642

0.611

0.624

Floor Space Sold

Proxy 1

Proxy 2

Proxy 3

Housing Price

0.536

1.091

0.356

Shadow Banking

1.282***

1.404***

1.244***

Interest rates

-0.155

-0.187

-0.146

Constant

0.016

0.009

0.018

Observations

100

100

100

Adjusted R2

0.669

0.66

0.659

Panel D: On Floor Space Started and Land Purchases with 6-month Lag

in Price Index and Lending Rates but No Lag for Shadow Banking Funds

Floor Space Started

Land Purchases

Proxy 1

Proxy 2

Proxy 1

Proxy 2

Housing Price

0.207

0.561

0.801

0.031

Shadow Banking

1.022***

1.251***

1.397***

1.017***

Interest rates

-0.031

-0.079

-0.119

-0.157*

Constant

0.013

-0.014

-0.029*

0.033

Observations

72

100

100

81

Adjusted R2

0.597

0.715

0.767

0.538

Note : *, ** and *** denote significance at the 10%, 5% and 1% levels respectively.

All variables except interest rates are differenced log values to obtain percentage

changes of the values. Interest rates are differenced values.

4. Modeling House Price Growth via Pooled Mean Group

This section follows Lai and Van Order (2017) in developing a model for

property value changes over time. In equilibrium, rent, which is essentially the

current dividend from the property, should equal the risk-adjusted interest rate

and expected capital gains over the period. Then, given an information set, t,

the equilibrium condition for holding the property at time t is given by

8

8

See Lai and Van Order (2010).

376 Lai and Van Order

1

/ / 1|

t t t t t t t ht

R P i E P P i

(1)

where Pt is the price of a constant quality house, Rt is net rental income, which

is the imputed net rent of the property in the case of owner-occupied housing,

it is the risk-adjusted hurdle rate, which can be thought of as a long term nominal

rate, α is constant depreciation, and πht is expected house price growth. Equation

(1) applies to a particular location. We suppress location notation until we

conduct the estimation later.

House price can thus be found from Equation (1) given expected future house

prices. Since future prices depend on future rents, current price depends on

future prices, which in turn depends on future rents through the expected

present value:

0( / | ) lim (1/ | )

t t i t i t t i t

i

P E R I E I

(2)

where the discount factor is It = 1+ it, such that It+i is the discount rate for an i-

period loan at time t. Assuming that the second term approaches zero and

dividing through by Rt, the expected present value formulation becomes:

0

/ (1/ | )

t t t i t

i

P R E D

(2’)

where

(1 )/ (1 *)

t i t i t i

Di

, and πt+i* is the expected rate of growth of

rent from period t to period i. If D and the growth rate of rent are constant in

the long run, then the reciprocal of Equation (2’) will converge to (1), which

gives the long run fundamentals.

The advantage of this approach is that it does not require the development of a

housing demand and supply model, but only a model of how the expectations

are formed. In other words, the model must allow transaction costs to facilitate

gradual adjustments of Equation (2’). In general, house prices adjust to shocks

slowly, and therefore are less efficient. Glaeser and Nathanson (2017) develop

a house pricing model in which traders are “almost” rational. That is, small

mistakes can lead to large forecasting errors, such that forecasted prices based

on past prices have short run momentum (positive feedback), long run mean

reversion, and excess volatility. Our estimated models generate all of these

phenomena, with the long run model being the Gordon dividend model.

4.1 Long Run Specification

Theory suggests that in the long run, prices and rents are expected to move

together and depend on real interest rates, which, according to the Gordon

model, should be:

Shadow Banking and Property Market 377

tt t t

t

Rir

P

(3)

where rt is the real rate. This suggests a coefficient of unity for the real rate. It

is possible that for tax reasons, money illusion or inability to borrow against

human capital that a coefficient of unity for the real rate is unlikely. Hence, we

consider a more general formula:

tt

t

Rc

P

(4)

where

t i t t i t r t

c i i r

is the “cap rate” for housing.

Our tests include both long-run and short-run behaviors. The long-run tests are

to check whether property prices converge to rent divided by the cap rate as in

Equation (4), and, if so, the speed of convergence. For the short-run, the tests

are about the nature of the deviations from the long-run phenomenon, and

whether the coefficients make sense. We use long run risk free rates for i; α

represents depreciation and long run expected future rent growth; and risk

adjustments with one for each city, which are assumed to be invariant over time.

We take Equation (4) as our representation of long run fundamentals. We do not

define short run fundamentals; rather we analyze how short run deviations

move over time. In general,

r

is expected to be close to 1, and

to be around

2%. We have no presuppositions about

i

.

4.2 Dynamic Heterogeneous Panel Estimation

We assume that R/P depends on a complicated lagged function of past levels of

R, P and i. We decompose the relationship into long-run and short-run effects

by using the PMG and MG estimation models developed in Pesaran et al. (1997,

1999).

9

Our hypothesis is that t contains only past rents, prices, interest rates

and shadow banking indicators, and that prices ultimately adjust to

fundamentals.

Traditionally, economic analysis has focused on long run relationships among

the dependent variables and the regressors. PMG estimation allows us to

identify long run relationships (Equation (4)) and short run dynamics separately;

the intercepts that reflect the fixed effect, short run coefficients and error

variances are allowed to differ across cities, but long run coefficients are

constrained to be the same. MG estimation is different in that the long run

coefficients are also allowed to vary across cities.

9

See Ott (2014) for a study that uses PMG on house price dynamics in the Euro area.

378 Lai and Van Order

Our model can be represented by:

,

, , , , ,

1 0 1

,,

q

ln

c t j

c,t kk

c j c j c t j c c t

j j k

c t c t j

R

Rx

PP

(5)

where

,

c,t

ct

R

P

is property rent to price ratio in city c at time t

c captures the city specific fixed effects

xkc,t-j is the kth of n regressors for city c

δkc, j is the coefficient of the kth regressor for city c

λc,j are scalars

εc,t are city specific errors

c represents panels or cities, i = 1,2,…,N

t represents time in quarters, t = 1,2,…,T

j is an indicator of lags;

j = 0,1,2,…, l for lagged dependent variable

j = 0,1,2,…, q lags for regressors

Letting

R

P

, Equation (5) can be written as:

, , 1 , , ,

01

qnkk

c t c c t c j c t j c c t

jk x

(6)

which when written in error correction form, yields:

, , 1 , , , ,

1 0 1

q

nn

k k k k

c t c c t c c t c j c t j c c t

k j k

β x δ x

(7)

where

(1 )

cc

,

,0

(1 )

k

c

k

ct

.

Equation (7) is used for the MG estimation model, which allows us to restrict

some of the parameters inside the brackets to be zero so that we can obtain

a long run specification that looks like the Gordon model, as given in Equation

(4), but with fewer restrictions on the short run adjustment parameters across

cities. Among the items inside the brackets in Equation (7) are long run fixed

effects, αc, and note that

/

c c c

.

The coefficients (one for each city) before the brackets,

c, denote the speed of

the reversion to the long run, after short run deviations. The adjustment outside

the brackets is the momentum (or mean reversion), which will disappear if the

model is not explosive.

Shadow Banking and Property Market 379

For the PMG, we assume homogeneous long run relations; i.e., βck = βk for all

cities, but we continue to allow long run adjustment speeds and constant terms

to vary across cities. Then:

, , 1 , , , ,

1 0 1

q

nn

k k k k

c t c c t c t c j c t j c c t

k j k

β x δ x

(8)

The double summation term in Equations (7) and (8) can include lagged

changes in the dependent variable, that is, in R/P. We measure the level of

momentum with the sum of these coefficients. If there is momentum, we expect

the sums of the coefficients to be positive; a negative sum implies short run

mean reversion.

Note that the model requires long run rents and prices to grow at a constant rate

within each city in the long run

10

, but

c allows the growth rates to vary across

cities in the long run, which in turn causes the long run level of R/P to differ

across cities. The long run equilibrium is given by:

1/

nkk

c c c c

k

βx

(9)

Recall that the last term in Equation (9), which is the negative of the ratio of the

constant term in Equation (8) (short run constant term) divided by the correction

speed (which is negative), is the long run constant term, αc. This allows for

differences in risk premia and growth.

Since the PMG and MG estimations are autoregressive distributed lag (ARDL)

models, the series in the models must be stationary or cointegrated. Hence, we

run unit root tests on our time series.

We perform cointegration analysis tests developed by Westerlund (2007) to

confirm the presence of long-run relationships among the time series. If long

run cointegration exists, then we can find the long-run and short-run effects

among the variables by using the MG and PMG models. All of the variables

pass the tests (lengthy results are omitted). The Hausman test can be used to

check if a common long run coefficient is present (that is, if the null hypothesis

of the common coefficients between the MG and PMG is not rejected, then the

common coefficients should be adopted).

4.3 Specifics of Data and Models

Before discussing the tests, we provide some observations of the data and an

explanation of the tests. Figure 3 shows the average rent and house price across

10

We also try to relax this condition by adding a linear time trend, common to all cities

inside the brackets in Equation (8). The results are similar, and therefore omitted here.

380 Lai and Van Order

cities over time. A key observation from Figure 3A is that in the aggregate, the

slopes of the two series are close to each other, which means that despite rapid

growth, there might not be a bubble. Figure 3B shows the rent divided by house

price, which is placed on the left side of our regression. The plotted curve does

fluctuate but not nearly as much as in the US, which is plotted in Figure 3C,

and shows a large departure of the house price from rent during the upswings

and downturns around the Great Recession. Note in Figure 3B that the raw data

have rents relative to prices as values like 0.003, which means 30 basis points

per month (data are in monthly frequency). In our regressions, we multiply the

rent to house price by 1200, so that the above is now 3.60 (% per year). This

makes those return data comparable to our interest rate data, and we can test for

whether the coefficient of the real or nominal interest rate is equal to one.

Figure 3 Aggregate Prices and Rents

Panel 3A: Averages of Monthly Rent and Average Price

Panel 3B: Ratio of Average Monthly Rent to Average Price (in %)

15

17

19

21

23

25

4000

5000

6000

7000

8000

9000

10000

11000

01/01/2008

01/06/2008

01/11/2008

01/04/2009

01/09/2009

01/02/2010

01/07/2010

01/12/2010

01/05/2011

01/10/2011

01/03/2012

01/08/2012

01/01/2013

01/06/2013

01/11/2013

01/04/2014

01/09/2014

01/02/2015

01/07/2015

01/12/2015

01/05/2016

price_avg rent_avg

2.0

2.5

3.0

3.5

4.0

4.5

01/01/2008

01/06/2008

01/11/2008

01/04/2009

01/09/2009

01/02/2010

01/07/2010

01/12/2010

01/05/2011

01/10/2011

01/03/2012

01/08/2012

01/01/2013

01/06/2013

01/11/2013

01/04/2014

01/09/2014

01/02/2015

01/07/2015

01/12/2015

01/05/2016

RPratio_avg

Shadow Banking and Property Market 381

Panel 3C: Rent to House Price in the US

We group the cities in our sample into different categories ─ bubble versus non-

bubble, coastal versus inland, and Tier 1 versus Tier 2 versus Other Tiers.

Bubble cities are those with a price growth over rent growth that is higher than

the 65-city average (2.59% for the 78-city average, and 2.75% for the 65-city

average). We classify coastal/inland cities because, according to Yang and Chen

(2014), for instance, there is a lower ownership rate in the eastern regions (i.e.

the coastal cities) because of the more expensive housing. This means that the

two groups of cities might be subject to different regimes. Lastly, Tier 1 cities

are made up of the four largest cities ─ Beijing, Shanghai, Guangzhou, and

Shenzhen. Other smaller and more remote cities are classified as other tier cities.

We try two sets of long run fundamental models. Model A includes various

combinations of real interest rates, 5-year bonds, and 5-year bonds minus rent

growth rates. Model B uses real interest rates as the only long run variable,

which forces the model to converge to a strong (no money illusion) version of

the Gordon Model. In both models, the lagged dependent variables are included

to capture momentum. Interest rates are shown in Figure 4.

Other variables used in the models are the lagged shadow banking funds and

the lagged yield spread. Lagged 5-year bonds, and 5-year bonds minus rent

growth rates alternate in different models. We use both monthly and quarterly

data. With our monthly data, we use up to six lags, a maximum of half a year.

We also try to omit shorter lags for shadow banking funds to omit immediate

funding effects. For the quarterly data, we include up to four lags, which

represents one whole year. We also try to omit shorter lags for shadow banking

funds. All tests show that the PMG outperforms the MG results, thus implying

that all of the cities share the same coefficients for the long run fundamental

variables. Hence, we only show the PMG results here (the MG results are

available upon request). The models that work best are those with monthly data

with three lags, that is, lags from one month up to one quarter.

0.5

0.55

0.6

0.65

0.7

0.75

0.8

0.85

0.9

0.95

1980 Q1

1981 Q3

1983 Q1

1984 Q3

1986 Q1

1987 Q3

1989 Q1

1990 Q3

1992 Q1

1993 Q3

1995 Q1

1996 Q3

1998 Q1

1999 Q3

2001 Q1

2002 Q3

2004 Q1

2005 Q3

2007 Q1

2008 Q3

2010 Q1

2011 Q3

2013 Q1

382 Lai and Van Order

Figure 4 Movement of Interest Rates and Housing Prices

Panel 4A: Various Rates (in %) Used for the Tests

Panel 4B: Plot of Real Interest Rates (in %), Housing Price (in RMB),

and Change in Housing Price (in %)

One of the concerns is that shadow banking is endogenous. In our model, the

dependent variables are by city, but the shadow banking variable is nationwide.

Hence, while the shadow banking variable might be influenced by national data,

it is unlikely that individual cities (65 to 78 of them) have influenced the

national level of shadow banking. As a result, we feel justified in assuming that

shadow banking is exogenous in the city-by-city equations.

-6

-4

-2

0

2

4

6

8

5Yr Government Bond 5Yr Corporate Bond

Real Rates

-5

0

5

10

15

0

2000

4000

6000

8000

10000

2007M3

2007M9

2008M3

2008M9

2009M3

2009M9

2010M3

2010M9

2011M3

2011M9

2012M3

2012M9

2013M3

2013M9

2014M3

2014M9

Housing Price (Primary Axis)

Real Rates (Secondaryd Axis)

Change in Price (Secondaryd Axis)

Shadow Banking and Property Market 383

5. Results

The results of Model A are shown in Panel A of Table 5, while those of Model

B are in Panel B. All of the variables, both long term and short term, have

coefficients with the correct signs as expected. All of the long term variables

are significant, thus implying that the proposed fundamental model works in

the case of the Chinese housing markets. The error correction coefficients,

which show the speed of reverting to the long term fundamental from the short

term deviation, range between about -0.16 and -0.21. This is very fast. Since

this is monthly data, the coefficients imply correction from short term deviation

takes about 5 to 6 months to return to a long run relationship. Short term lagged

yield spreads do not show a very strong and persistent effect, while 5-year

bonds and 5-year bonds minus rent growth are mostly significant. Lagged rent

to price ratios actually have negative effects which not only suggest that there

is no bubble but many mean reversions during the short run. Hence, our model

is not one of bubbles; rather it appears that prices chase rents and adjust rather

quickly. Note that this does not mean that house prices are stable; the stability

depends on the variations in rents.

All of the models show significant short run effects of shadow banking on price

changes.

11

We can also test whether the long run effects of the real rates on rent

to price is unity. For instance, the effects are around 1.2 in Model A, and close

to one in Model B in the first two panels and around a half in the third panel.

We note, however, the very strong long run effects of the nominal rates in Model

A. Hence, while our model is somewhat consistent with the Gordon Model of

the long run dependence on real rates, it is too ambiguous to warrant serious

consideration. Perhaps this is because the data set does not cover a very long

time period even though many cities are included.

We next compare and contrast how different cities react to the availability of

shadow banking funds. In particular, we group cities into bubble versus non-

bubble cities, coastal versus inland cities, as well as Tiers 1, 2, and others (see

Appendix A for the list of cities with various classifications). We identify bubble

cities in two ways. First, they have to have housing price growth rates higher

than the mean growth rate for the period of 2007-2014 (the housing boom

period). Second, they are the cities in which housing price growth minus rent

growth rates are above the mean for the period of 2007-2014. These three

categories overlap in that many bubble cities are also coastal cities, and the Tier

1 cities fall in the former two groups. The sum of the coefficients of three lags

of change in shadow banking funds as the short run variables for different city

classifications is shown in Table 7. Note that negative coefficients imply

positive effects on price relative to rent, as the dependent variables in the PMG

estimations are rent to price ratios.

11

Note that because the dependent variable is the reciprocal of the price to rent ratio,

negative shadow banking effect means increasing effects on house price growth.

384 Lai and Van Order

Table 5 Regression of Shadow Banking on Various Independent Variables

Part A

Model 1

Model 2

Model 3

Model 4

Model 5

Model 6

Model 7

Model 8

Model 9

Model 10

House Price

-1.801

-1.485

-3.906*

0.615

0.859

0.81

-0.111

-1.640**

-0.923

-1.936***

Interest Rate

0.198

0.219

0.267

0.051

0.032

0.045

0.144*

0.188***

0.152**

0.203***

Building Sold

Total Residential

0.564***

Existing units

0.473***

Presale

0.545***

Floor Space

Completed

Total

0.545***

Residential &

Commercial

0.571***

Residential

0.566***

40-city Residential

0.186**

Floor Space Sold

Total Residential

0.529***

Existing units

0.470***

Presale

0.543***

Constant

0.052***

0.071***

0.064***

0.029*

0.048***

0.038**

0.133***

0.056***

0.062***

0.056***

Observations

100

70

70

100

100

100

72

100

100

100

Adj. R-Squared

0.667

0.633

0.672

0.551

0.5

0.529

0.125

0.694

0.67

0.693

Note : *, ** and *** denote significance at the 10%, 5% and 1% levels respectively.

All variables except interest rates are differenced log values to obtain percentage changes of the values. Interest rates are differenced values.

384 Lai and Van Order

Shadow Banking and Property Market 385

Part B

Model 11

Model 12

Model 13

Model 14

Model 15

Model 16

Model 17

Model 18

Model 19

House Price

-1.363***

-1.787***

-1.439**

-1.540***

-1.988***

-1.125**

0.957

-2.041***

-1.659***

Interest Rate

0.170***

0.175***

0.106**

0.154***

0.188***

0.160***

0.040

0.189***

0.162***

Floor Space Started

40-city Residential

0.600***

Residential &

Commercial

0.585***

Land Area Purchased

Total

0.555***

40-Cities

0.516***

Real Estate

Investments

Total

0.544***

Land Transactions

0.573***

New Increase

0.604***

Residential-Total

0.548***

Residential-40 Cities

0.604***

0.577***

Constant

0.052***

0.071***

0.064***

0.029*

0.048***

0.038**

0.133***

0.056***

0.056***

Observations

100

70

70

100

100

100

72

100

100

Adj. R-Squared

0.667

0.633

0.672

0.551

0.5

0.529

0.125

0.694

0.693

Note : *, ** and *** denote significance at the 10%, 5% and 1% levels respectively.

All variables except interest rates are differenced log values to obtain percentage changes of the values. Interest rates are differenced values

Shadow Banking and Property Market 385

386 Lai and Van Order

It can be seen from all three PMG models with the best explanatory power (i.e.

highest likelihood) that shadow banking has its largest impacts in the coastal

and Tier 1 cities. Exceptions are the results from the “bubble” cities defined as

price growth rates higher than mean growth rates, which can nevertheless be

ignored. This is because “bubble” cities (defined as cities with a house price

growth minus rent growth that is greater than the mean) would be more

meaningful since the rental markets in major cities are much larger, and

therefore rents tend to grow faster. Another observation is that the variations of

the sums of coefficients (i.e. maximum minus minimum) are in general larger

in non-bubble cities as well as non-tier cities. This implies that shadow banking

as a source of investment funds in these cities tends to vary a lot. The fact that

the means are generally higher in “bubble”, coastal, or Tier 1 cities is mostly

due to more investments drawn into those cities, where normal banking funds

would be available to only very large developers or State Owned Enterprises.

That is why other developers and investors have to rely on shadow banking. As

a result, more shadow banking funds of all forms would be available in those

cities, which subsequently push up housing prices. The sum of all short run

coefficients are provided in the Appendix B for reference.

Lai and Van Order (2017) mainly focus on testing whether the bubbles of the

housing markets in the US were explosive by checking if the residuals from the

regression estimates are highly autocorrelated, and the variances of the residual

autoregression equations differ between bubble and non-bubble cities. Since the

Chinese housing markets have also been described as having large bubbles that

have not exploded yet, we repeat their tests. In particular, we refer to PMG

Models A3, A4, and B3 as reported in Table 6, and attempt the autoregressive

modeling of the residuals with various lags. None of the equations show much

autocorrelation in the residuals (results are shown in Panel A of Table 8). This

and the low sums of the coefficients of lagged rent to price mean that bubbles

are not evident in the Chinese housing markets. Rather, there is momentum in

the short run that is nowhere near explosive. It should be noted that the Chinese

government has undertaken several policy changes to boost and curb the

markets at different stages of the market boom and bust. Such effects on

housing markets have been studied by, for instance, Cao et al. (2018). They

could be largely responsible for the lack of observed bubbles. Since we test the

effects of shadow banking on housing markets, and not the overall stabilization

policy, we do not try to separate stabilization policy from stabilization due to

the underlying market structure.

We further test if the variances of these models are different. Note that since the

residuals are not autocorrelated, the variances of these autoregressive models

are really the variances of the PMG models. We show the sums of the

coefficients of the lagged error terms in Panel B of Table 8. Also listed in the

same panel are the variances of the residuals from these autoregressive

equations. While large residual variances might be sources of bubbles in

housing markets, their small magnitudes show that there are no such sources of

bubbles in our sample cities. Nevertheless, bubble cities have smaller variances

Shadow Banking and Property Market 387

than non-bubble cities, coastal cities have larger variances than inland cities,

and Tier 1 cities have smaller variances than Tier 2 cites which are in turn

smaller than the other tier cities. This shows that our models are able to explain

those major cities (bubble and Tier 1 cities) with higher precision. To further

test if the variances from the models with different lags are indeed different, we

run the Goldfeld-Quandt test as shown in Panel D. Finally, we check the

differences in variances across city classifications with the Goldfeld-Quandt

test again, and the results are also shown in Panel D. Both show that these

models are different both across lags and cities, which imply that cities in

different categories do possess unique characteristics.

Table 6 Pooled Mean Group Estimation for Rent to Price Ratio

Panel A: Model A

Model A1

Model A2

Model A3

Model A4

Long run variables

Real Interest Rates

1.20***

1.20***

0.96***

1.20***

Tbond_5y

2.88***

31.2***

3.62***

15.7***

T5y_rentg

-10.68***

Short run variables

Error Correction

-0.17***

-0.17***

-0.14***

-0.14***

ΔR/Pt-1

-0.25***

-0.21***

-0.24***

-0.23***

ΔR/Pt-2

-0.17***

-0.15***

-0.15***

-0.153***

ΔR/Pt-3

0.015

-0.02

0.003

-0.00

Δ Shadow Bankt

-3.48**

-2.26*

-0.60

-0.60

Δ Shadow Bankt -1

-3.73*

-4.56***

-3.72***

-0.4.08***

Δ Shadow Bankt -2

-4.80***

-5.04***

-4.20***

-4.32***

Δ Shadow Bankt -3

-1.68**

-2.04***

-2.04***

-2.26***

ΔYield Spread t

0.48

0.36

-0.12

-0.12

ΔYield Spread t-1

0.48

0.36

-0.48

-0.48

ΔYield Spread t-2

-0.60

-0.96***

-1.32***

-1.32***

ΔYield Spread t-3

-1.32**

-1.32***

-1.32***

-1.32***

Δ5Yt

-1.68***

9.00***

7.68***

Δ5Yt-1

0.24

3.48***

2.64**

Δ5Yt-2

-0.84**

2.64**

1.80

Δ5Yt-3

-1.08***

-0.24

-0.72

Δ5Yt - RentGt

-2.26***

-10.2***

-9.12***

Δ5Yt-1 - RentGt-1

0.24

-3.48***

-2.64**

Δ5Yt-2 - RentGt-2

-0.84***

-3.48***

-2.76***

Δ5Yt-3 - RentGt-3

-0.72***

-0.72

-0.24

Constant

0. 37***

0. 35***

0. 27***

0. 22***

Observations

3,113

3,007

3,007

3,007

Number of groups

65

65

65

65

Log likelihood

16111

15980

16603

16606

Hausman Test

0.19

1.08

1.01

2.32

p-value

0.9102

0.5831

0.6022

0.5078

Note : *, ** and *** denote significance at the 10%, 5% and 1% levels respectively.

388 Lai and Van Order

Panel B: Model B

Model B1

Model B2

Model B3

Long run variables

Real Interest Rates

0.96***

0.96***

0.48***

Short run variables

Error Correction

-0.16***

-0.16***

-0.13***

ΔR/Pt-1

-0.21***

-0.24***

-0.22***

ΔR/Pt-2

-0.15***

-0.16***

-0.15***

ΔR/Pt-3

-0.01

0.01

0.02

Δ Shadow Bankt

-2.16*

-3.48**

-0.60

Δ Shadow Bankt -1

-4.08***

-3.48*

-3.12**

Δ Shadow Bankt -2

-4.92***

-4.56***

-3.84***

Δ Shadow Bankt -3

-1.56***

-1.32*

-1.44**

ΔYield Spread t

0.24

0.38

-0.24

ΔYield Spread t-1

0.48

0.60

-0.36

ΔYield Spread t-2

-0.84**

-0.48

-1.08***

ΔYield Spread t-3

-1.44***

-1.32**

-1.44***

Δ5Yt

-1.44***

9.00***

Δ5Yt-1

0.48

3.36***

Δ5Yt-2

-0.60

2.76**

Δ5Yt-3

-0.84***

-0.36

Δ5Yt - RentGt

-2.04***

-10.08***

Δ5Yt-1 - RentGt-1

0.48*

-0.3.12***

Δ5Yt-2 - RentGt-2

-0.48**

-3.24***

Δ5Yt-3 - RentGt-3

-0.60***

-0.36

Constant

0. 47***

0. 49***

0. 40***

Observations

3,007

3,113

3,007

Number of groups

65

65

65

Log likelihood

15945

16092

16562

Hausman Test

1.13

0.07

0.24

p-value

0.2876

0.7973

0.622

Note : *, ** and *** denote significance at the 10%, 5% and 1% levels respectively.

Table 7 Sum of Short Run Coefficients for Shadow Banking Loans

from PMG Three-lag Model (without Adjustment Rent to

Price Units)

Panel A: Model A3

Bubble Price

Bubble

Price-Rent

Coastal City

Non-

bubble

Bubble

Non-

bubble

Bubble

Non-

coastal

Coastal

Mean

-0.0106

-0.0063

-0.0066

-0.0109

-0.0080

-0.0105

Maximum

0.0224

0.0204

0.0205

0.0224

0.0224

0.0205

Minimum

-0.0606

-0.0498

-0.0606

-0.0498

-0.0474

-0.0606

No. of cities

37

28

32

33

44

21

(Continued…)

Shadow Banking and Property Market 389

(Panel A Continued)

Tier

Tier 1

Tier 2

Other Tiers

Mean

-0.0117

-0.0063

-0.0110

Maximum

-0.0022

0.0224

0.0186

Minimum

-0.0202

-0.0474

-0.0606

No. of cities

4

31

30

Panel B: Model A4

Bubble Price

Bubble

Price-Rent

Coastal City

Non-

bubble

Bubble

Non-

bubble

Bubble

Non-

coastal

Coastal

Mean

-0.0115

-0.0067

-0.0075

-0.0112

-0.0082

-0.0120

Maximum

0.0222

0.0197

0.0197

0.0222

0.0222

0.0174

Minimum

-0.0610

-0.0499

-0.0610

-0.0499

-0.0480

-0.0610

No. of cities

37

28

32

33

44

21

Tier

Tier 1

Tier 2

Other Tiers

Mean

-0.0121

-0.0070

-0.0115

Maximum

-0.0032

0.0222

0.0170

Minimum

-0.0200

-0.0480

-0.0610

No. of cities

4

31

30

Panel C: Model B3

Bubble Price

Bubble

Price-Rent

Coastal City

Non-

bubble

Bubble

Non-

bubble

Bubble

Non-

coastal

Coastal

Mean

-0.0087

-0.0060

-0.0051

-0.0099

-0.0071

-0.0085

Maximum

0.0274

0.0273

0.0274

0.0228

0.0273

0.0274

Minimum

-0.0536

-0.0478

-0.0536

-0.0490

-0.0490

-0.0536

No. of cities

37

28

32

33

44

21

Tier

Tier 1

Tier 2

Other Tiers

Mean

-0.0112

-0.0058

-0.0089

Maximum

-0.0001

0.0228

0.0274

Minimum

-0.0199

-0.0490

-0.0536

No. of cities

4

31

30

Notes: “Bubble” cities are those with price growth greater than the mean during 2007-

2014.

“Bubble Price-Rent” cities are those with price growth minus rent growth greater

than the mean during 2007-2014.

390 Lai and Van Order

Table 8 Residual Autoregressive Models from Model A4

Panel A Coefficients of the Autoregressive Models

Overall

Bubble (Price-Rent)

Non-Bubble (Price-Rent)

Residual

1 lag

3 lags

5 lags

1 lag

3 lags

5 lags

1 lag

3 lags

5 lags

1 Lag

0.0176

0

-0.0099

0.0464*

-0.0178

-0.0629

-0.0078

0.015

0.0231

2 Lags

0.0293

0.0303

0.0251

0.0942

0.0371

-0.03

3 Lags

-0.0367

0.0018

-0.0633*

-0.1175*

-0.0123

0.1046*

4 Lags

0.0247

0.0009

0.0971

5 Lags

0.1249**

-0.0022

0.2860***

Obs

2,501

1,493

495

1,287

769

255

1,214

724

240

Adj. R2

-9.74E-05

0.000181

0.0051

0.00134

0.000642

0.0075

-0.00077

-0.00222

0.0696

Tier 1

Tier 2

Other Tiers

Residual

1 lag

3 lags

5 lags

1 lag

3 lags

5 lags

1 lag

3 lags

5 lags

1 Lag

0.0358

-0.0094

0.1399

0.0500*

0.0192

-0.0935

-0.0113

-0.0158

0.0255

2 Lags

-0.0736

0.0115

0.0587*

0.0772

0.0132

0.003

3 Lags

-0.0236

0.1371

-0.0188

-0.0457

-0.0583

0.0439

4 Lags

-0.480**

0.1473**

-0.0514

5 Lags

0.0285

0.0135

0.220***

Obs

158

94

31

1,227

735

244

1,116

664

220

Adj. R2

-0.00513

-0.0279

0.0737

0.00162

0.000441

0.0197

-0.00077

-0.00089

0.0283

Coastal

Non- Coastal

Residual

1 lag

3 lags

5 lags

1 lag

3 lags

5 lags

1 Lag

-0.005

-0.046

-0.103

0.0294

0.0213

0.0573

2 Lags

-0.0332

0.1175

0.0591**

-0.0305

3 Lags

-0.0897*

-0.0766

-0.0137

0.0225

4 Lags

0.0112

0.0463

5 Lags

0.0838

0.158***

Obs

824

492

163

1,677

1,001

332

Adj. R2

-0.00119

0.00337

0.00228

0.000248

0.00182

0.015

Note: *, ** and *** denote significance at the 10%, 5% and 1% levels respectively.

390 Lai and Van Order

Shadow Banking and Property Market 391

Panel B: Sum of Coefficients and Variances of the Autoregressive Models

Overall

Bubble (Price-Rent)

Non-Bubble (Price-Rent)

Sum

1 lag

3 lags

5 lags

1 lag

3 lags

5 lags

1 lag

3 lags

5 lags

Coeff.

0.0176

-0.0074

0.1718

0.0464

-0.056

-0.0875

-0.0078

0.0398

0.4808

Sig. Coef

0

0

0.1249

0.0464

-0.0633

-0.1175

0

0

0.1046

Variance

1.24E-06

1.22E-06

1.12E-06

1.12E-06

1.12E-06

1.06E-06

1.37E-06

1.33E-06

1.10E-06

Tier 1

Tier 2

Other Tiers

Sum

1 lag

3 lags

5 lags

1 lag

3 lags

5 lags

1 lag

3 lags

5 lags

Coeff.

0.0358

-0.1066

-0.1631

0.05

0.0591

0.0988

-0.0113

-0.0609

0.2411

Sig. Coef

0

0

-0.4801

0.05

0.0587

0.1473

0

0

0

Variance

1.05E-06

1.35E-06

6.56E-07

1.09E-06

1.01E-06

1.08E-06

1.43E-06

1.44E-06

1.17E-06

Coastal

Non- Coastal

Sum

1 lag

3 lags

5 lags

1 lag

3 lags

5 lags

Coeff.

-0.005

-0.1689

0.0329

0.0294

0.0667

0.2538

Sig. Coef

0

-0.0897

0

0

0.0591

0

Variance

1.30E-06

1.39E-06

1.29E-06

1.21E-06

1.13E-06

1.02E-06

Notes: “Coeff.” means sum of coefficients; “Sig. Coef” means sum of significant coefficients.

Shadow Banking and Property Market 391

392 Lai and Van Order

Panel C: Goldfeld-Quandt Tests of Variance of Residuals from Autoregression Models

Overall

Bubble

(Price-Rent)

Non-Bubble

(Price-Rent)

Coastal

Non-

Coastal

Tier 1

Tier 2

Other Tiers

Model A4

1 & 5 lags

1.646***

1.678***

1.627***

1.789***

1.568***

2.205***

1.550***

1.6907

3 & 5 lags

4.584***

4.853***

4.143***

5.151***

4.324***

3.758***

5.063***

4.236***

1 & 3 lags

2.786***

2.892***

2.546***

2.879***

2.757***

1.705***

3.267***

2.506

Model B3

1 & 5 lags

1.677***

1.742***

1.630***

1.796***

1.616***

2.223***

1.584***

1.724

3 & 5 lags

4.626***

5.023***

4.071***

4.973***

4.484***

3.647***

5.224***

4.210***

1 & 3 lags

2.758***

2.884***

2.498***

2.769***

2.775***

1.641***

3.299***

2.442

Model A3

1 & 5 lags

1.660***

1.703***

1.634***

1.801***

1.584***

2.217***

1.565***

1.705

3 & 5 lags

4.551***

4.750***

4.148***

5.107***

4.296***

3.746***

4.999***

4.240***

1 & 3 lags

2.741***

2.790***

2.539***

2.837***

2.713***

1.690***

3.194***

2.488

392 Lai and Van Order

Shadow Banking and Property Market 393

Panel D: Goldfeld-Quandt Tests of Variance of Residuals of Different City

Classifications

Bubble (p-r)

vs. non-

Bubble (p-r)

Coastal vs.

non-

Coastal

Tier 1 vs.

Tier 2

Tier 1 vs.

Other

Tiers

Tier 2 vs.

Other

Tiers

Model A3

1 lag

1.3135***

2.1625***

7.4781***

5.2374***

1.4278***

3 lags

1.2602***

2.4586***

10.5935***

6.8152***

1.5544***

5 lags

1.1471

2.5708***

5.6036***

4.6272***

1.211

Model A4

1 lag

1.2959***

2.1877***

7.5468***

5.2216***

1.4453***

3 lags

1.2569***

2.4957***

10.7368***

6.8087***

1.5769***

5 lags

1.1064

2.6060***

5.6016***

4.6322***

1.2093*

Model B3

1 lag

1.3369***

2.1429***

7.3524***

5.0559***

1.4542***

3 lags

1.2511***

2.3823***

10.3197***

6.5190***

1.5830***

5 lags

1.0836

2.3769***

5.1327***

4.3800***

1.1718

Notes: The Goldfeld-Quandt Test is a test for statistical differences between two

fundamental equations.

*, ** and *** denote significance at the 10%, 5% and 1% levels respectively

(compared to an F-value of 1.3).

6. Conclusions

This is perhaps the first study that incorporates the availability of shadow

banking funds on real estate prices in China, as well as using PMG estimations

of house price dynamics to analyze the role of fundamentals and adjustment to

them. The question lies in determining their causality. The issue is similar to the

case in the US on whether the increase in private label securities is the cause,

or result, of the house price bubble. That we use aggregate shadow banking data

but local house prices is an attempt to manage the endogeneity of shadow

banking policy with respect to economy-wide variables. The strong link

between shadow banking and housing price movements suggests important

implications for the effects of a collapse in the shadow banking market.

We further analyze pricing dynamics by classifying cities into bubble versus

non-bubble, coastal versus inland, and Tier 1 versus Tier 2 and other tier cities.

We find that shadow banking indeed helps to improve the liquidity of

developers who cannot easily borrow from the major banking channels,

particularly in non-bubble cities, which are mostly also inland cities/non-Tier 1

cities. This is particularly remarkable as anecdotal evidence suggests that

shadow banking is an essential source of funding in second tier cities. As for

the long run, we find that housing in China is priced like a growth stock (high

P/E ratio) with different expected growth rates across cities, as in Lai and Van

Order (2018), and cities tend to share common long run fundamentals and

394 Lai and Van Order

adjust relatively quickly to deviations from them, without bubbles. Prices

appear to be rapidly chasing growing rents. Indicators of shadow banking

activity have a positive effect on house price growth. The data are consistent

with the notion that a one-percentage point increase in the real rate leads to

approximately a one-percentage point change in the rent to price ratio, but the

data are too thin to take this argument seriously.

That there is no evidence of bubbles does not mean Chinese property is not

risky. Growth stocks are risky because small changes in expected growth of

earnings (in this case rents) can lead to large changes in value. Alternatively,

one might argue that there is a rent bubble, or perhaps a lending bubble, that is

causing high prices. However, rents and lending activity are not prices of traded

assets; they are at best considered as factors in risk assessment. Fear of shadow

banking collapse is reasonable; nevertheless, the property market in China is

risky but not doomed to crash.

Acknowledgement

We thank Brent Ambrose and John Glascock and the participants at the

American Real Estate and Urban Economics Association 2017 Annual

Conference and the Fellows Forum at the 2019 Asian Real Estate Society

Annual Conference for their helpful comments. This research is funded by the

Research Committee of the University of Macau with grant number: MYRG

2015-00085-FBA.

References

Black, P.A. and Hoesli, M. (2006). House Prices, Fundamental and Bubbles.

Journal of Business Finance and Accounting, 33, 1535-1555.

Cai, H., Henderson, V. and Zhang, Q. (2013). China's Land Market Auctions:

Evidence of Corruption? The RAND Journal of Economics, 44(3): 488-521

Cao, X., Huang, B. and Lai, R.N. (2018). The Impact of Exogenous Demand

Shock on the Housing Market: Evidence from the Home Purchase Restriction

Policy in the People’s Republic of China, Asian Development Bank Institute

(ADBI) Working Paper Series, No. 824 (March 2018).

Chan, I., Lee, S. and Woo, K. (2001). Detecting Rational Bubbles in the

Residential Housing Markets in Hong Kong. Economic Modelling, 18, 61-73.

Shadow Banking and Property Market 395

Chang, Y., Cutts, A. and Green, R. (2005). Did Changing Rents Explain

Changing House Prices During the 1990s? Geroge Washington University

School of Business Working Paper.

Coleman, M.I., LaCour-Little, M. and Vandell, K.D. (2008). Subprime Lending

and the Housing Bubble: Tail Wags Dog? Journal of Housing Economics, 17(4):

272-290

Deng, L., Shen, Q. and Wang, L. (2009). Housing Policy and Finance in China:

A Literature Review. Retrieved from

http://www.chinaplanning.org/Publications/Lan Deng - Housing Policy and

Finance In China.pdf

Elliott, D. J. and Yan, K. (2013). The Chinese Financial System: An

Introduction and Overview. John L. Thornton China Center at Brookings.

Elliott, D., Kroeber, A. and Qiao, Y. (2015). Shadow Banking in China: A

Primer. Economics Studies at Broookings. The Brookings Institution.

Financial Stability Board. (2015). Global Shadow Banking Monitoring Report

2015 (November 2015).

Glaeser, E., Huang, W., Ma, Y. and Shleifer, A. (2017) A Real Estate Boom with

Chinese Characteristics. Journal of Economic Perspective, 31(1): 93-116

Glaeser, E.L. and Nathanson, C. (2017) An Extrapolative Model of House Price

Dynamics. Journal of Financial Economics, 126(1): 147-170

Hsu, S. and Li, J. (2015). The Rise and Fall of Shadow Banking in China.

Political Economy Research Institute, Working Paper Series Number 375.

University of Massachusetts Amherst.

Hwang, M., Quigley, J. and Son, J. (2006). The Dividend Pricing Model: New

Evidence from the Korean Housing Market. Journal of Real Estate Finance

and Economics, 32, 205-228.

Lai, R.N. and Van Order, R. (2010). Momentum and House Price Growth in

the U.S: Anatomy Of A Bubble. Real Estate Economics, 38(4), 753-773.

Lai, R.N. and Van Order, R. (2017). U.S. House Prices over the Last 30 Years:

Bubbles, Regime Shifts and Market (In)Efficiency. Real Estate Economics,

45(2), 259-300.

Lai, R. N. and Van Order, R. (2018). A Tale of Two Countries: Comparing the

US and Chinese Bubbles Housing Markets, Journal of Real Estate Finance and

Economics, forthcoming.

396 Lai and Van Order

Ott, H. (2014). Will Euro Area House Prices Sharply Decrease? Economic

Modelling, 42, 116-127.

Peng, L. and Thibodeau, T.G. (2012). Government Interference and the

Efficiency of the Land Market in China, Journal of Real Estate Finance and

Economics, 45(4), 919-938.

Pesaran, M.H., Shin, Y. and Smith, R.P. (1997). Pooled Estimation of Long-

Run Relationships in Dynamic heterogeneous Panels. Working Paper

Pesaran, M.H., Shin, Y. and Smith, R.P. (1999). Pooled Mean Group Estimation

of Dynamic Heterogeneous Panels. Journal of the American Statistical

Association, 94, 621-634.

Ren, Y., Xiong, C. and Yuan, Y. (2012). House Price Bubbles in China. China

Economic Review, 23, 786-800.

Sharma, S.D. (2014). Shadow Banking, Chinese Style. Economic Affiars, 34(3),

340-352.

Sheng, A., Edelmann, C., Sheng, C. and Hu, J. (2015). Bringing Light Upon

The Shadow: A Review of the Chinese Shadow Banking Sector. Oliver Wyman

& Fung Global Institute.

Westerlund, J. (2007). Testing for Error Correction in Panel Data. Oxford

Bulletin of Economics and Statistics, 69, 709–748.

Wheaton, W. and Nechayev, G. (2008). The 1998-2005 Housing ‘Bubble’ and

the Current ‘Correction’: What’s Different this Time? Journal of Real Estate

Research 30:1-26.

Wu, J., Gyourko, J. and Deng, Y. (2012). Evaluating Conditions in Major

Chinese Housing Markets. Regional Science and Urban Economics, 42(3):

531-543

Yang, Z. and Chen, J. (2014). Housing Reform and the Housing Market in

Urban China. In Housing Affordability and Housing Policy in Urban China (pp.

15-43). Springer Briefs in Economics.

Shadow Banking and Property Market 397

Appendices

Appendix A List of 78 Cities in the Sample

City

Bubble

Coastal

Tier

City

Bubble

Coastal

Tier

Anqing

Shantou

*

*

Baoding

*

Shaoxing

Beihai

*

2

Shenyang

*

2

Beijing

*

1

Shenzhen

*

*

1

Bengbu

Shijiazhuang

2

Changchun

*

2

Suzhou

*

Changde

Taiyuan

2

Changsha

2

Taizhou

Changzhou

Tangshan

*

Chengdu

2

Tianjin

*

*

2

Chongqing

*

2

Urumqi

*

Dalian

*

2

Weifang

*

Dongguan

*

*

Weihai

*

Foshan

*

Wenzhou

*

*

2

Fuzhou

*

*

2

Wuhan

*

2

Guangzhou

*

*

1

Wuxi

*

2

Guiyang

2

Xiamen

*

*

2

Haikou

*

*

2

Xi'an

*

2

Hangzhou

2

Xining

2

Harbin

2

Xuzhou

*

Hefei

*

2

Yancheng

*

*

Huzhou

Yangzhou

*

Jiaxing

Yantai

*

Jilin

*

Zhengzhou

*

2

Jinan

*

2

Zhuhai

*

Jinhua

Zibo

Kunming

2

Shantou

Lanzhou

2

Shaoxing

*

Nanchang

2

Shenyang

Nanchong

Shenzhen

Nanjing

*

2

Shijiazhuang

Nanning

2

Suzhou

Nantong

*

Taiyuan

*

2

Ningbo

*

*

2

Taizhou

*

Qingdao

*

*

2

Tangshan

Qinhuangdao

*

Tianjin

Quanzhou

*

Urumqi

*

2

Rizhao

Weifang

Shanghai

*

*

1

Weihai

*

*

Note: Due to missing data, our tests are also based on 65 cities, which do not include the

13 cities on the bottom right of the table. Shaded are the four Tier 1 cities.

398 Lai and Van Order

Appendix B Sum of Significant Short-run Coefficients for Individual

Cities from PMG Estimation of Model A4

Bubble versus Non-bubble Cities (classified by price growth minus rent growth)

Panel A: Non-Bubble Cities

City

Error

Correct-

ion

Sum of

Δ(R/P)

Sum of Δ

Shadow

Bank

Sum of

Δ

Spread

Sum of

Δ5Y

rate

Sum of Δ

(5Y Rate

– Rent

Growth)

Constant

Changsha

-0.1126

-0.7644

-0.0063

-0.0015

0.0024

-0.0078

0.0029

Guiyang

-0.1466

-1.3375

-0.0095

-0.0217

0.0643

-0.0651

0.0046

Yantai

-0.2121

-0.7142

-0.0105

-0.0073

-0.0014

-0.0006

0.0025

Beihai

-0.0096

0.2451

-0.0275

-0.0012

0.0271

-0.0297

-0.0012

Bengbu

-0.2321

-0.429

0.0151

-0.0043

0.0104

-0.0159

0.004

Chengdu

-0.0903

-0.0528

-0.0199

0.0013

0.0033

-0.0073

0.0013

Dalian

-0.2779

0.0789

0.0174

-0.0006

0.0095

-0.011

0.0065

Harbin

-0.3119

-0.7486

-0.0094

-0.0193

0.0174

-0.0195

0.0091

Jinhua

-0.1282

-0.8079

-0.0232

0.0041

0.0055

-0.0124

0.0005

Kunming

-0.2546

-0.9534

-0.0107

-0.0037

0.0144

-0.0159

0.0044

Lanzhou

-0.0256

-0.1349

-0.006

-0.0035

0.014

-0.0164

0.0002

Nanchang

-0.0742

-0.7484

0.0036

0.0075

0.0127

-0.0182

0.0011

Nanchong

-0.2512

-0.0579

0.017

-0.0013

0.007

-0.0136

0.0037

Nanning

-0.0902

-0.2988

0.0002

-0.0082

0.0062

-0.0094

0.0023

Nantong

-0.0352

-1.1987

-0.0099

-0.0015

0.0139

-0.013

0.0005

Qinhuangdao

-0.5426

0.0583

0.0005

-0.0011

-0.0195

0.018

0.0054

Quanzhou

-0.0601

-1.2077

-0.0090

-0.0081

0.0258

-0.0268

0.0016

Shijiazhuang

0.0099

-0.6356

0.0197

-0.0096

0.0042

-0.0041

-0.0003

Weihai

-0.2452

0.0892

-0.061

0.0027

0.0158

-0.0173

0.0015

Zibo

-0.2032

-0.4254

-0.0021

0.0014

0.0026

-0.0067

0.0022

Anqing

-0.0725

0.4811

-0.0084

-0.0006

-0.0085

0.009

0.0007

Changde

-0.2479

0.3327

-0.0024

0.0002

0.0112

-0.0162

0.0058

Changzhou

-0.0948

-0.8634

-0.0270

0.0019

0.0161

-0.0228

0.0021

Hangzhou

-0.2147

0.9152

-0.0244

0.0015

0.0007

-0.0075

0.0014

Huzhou

-0.1731

0.0787

-0.0254

-0.0034

0.0005

-0.0006

0.0021

Jiaxing

-0.0796

-0.3421

-0.0099

-0.0028

-0.0109

0.0048

0.0015

Rizhao

-0.1669

-0.4414

-0.0233

-0.0016

0.001

-0.0031

0.0006

Shaoxing

-0.4128

-0.6861

0.0025

-0.0044

-0.0038

-0.003

0.0036

Taiyuan

-0.1203

-0.1278

0.0055

-0.0023

0.0107

-0.0118

0.0022

Taizhou

-0.1686

-1.3656

0.0123

-0.0105

0.021

-0.0195

0.0016

Xining

-0.4072

-0.358

0.0161

-0.0144

0.0128

-0.0154

0.0114

Zhuhai

-0.0418

-0.6482

-0.0243

-0.0012

-0.0178

0.0157

0.0004

Shadow Banking and Property Market 399

Panel B: Bubble Cities

City

Error

Correct-

ion

Sum of

Δ(R/P)

Sum of

Δ

Shadow

Bank

Sum of

Δ

Spread

Sum of

Δ5Y

rate

Sum of Δ

(5Y Rate

– Rent

Growth)

Constant

Qingdao

-0.0560

0.1124

-0.0010

0.0018

0.0033

-0.0070

0.0005

Changchun

-0.0395

-0.3366

-0.0152

-0.0011

0.0072

-0.0105

0.0002

Jinan

-0.1337

-0.1401

-0.0204

0.0011

0.0169

-0.0194

0.0013

Shantou

-0.1559

-1.8098

-0.0499

0.0072

0.0363

-0.0476

0.0036

Suzhou

-0.0422

-0.3962

-0.018

-0.0109

0.0301

-0.0286

0.0005

Weifang

-0.0718

-0.7851

-0.0162

-0.0002

0.0024

-0.0065

0.0008

Xiamen

-0.0322

-0.3512

-0.012

-0.0049

-0.0077

0.0050

-0.0001

Xi'an

-0.0685

-0.6129

-0.0073

-0.0032

0.0346

-0.0354

0.0014

Chongqing

-0.3375

-1.4195

-0.0228

-0.0083

0.0803

-0.0849

0.0076

Dongguan

-0.0893

-0.5404

0.0161

-0.0071

0.0207

-0.0277

0.0025

Foshan

-0.2111

-0.7477

-0.0293

-0.0023

-0.0305

0.0275

0.0042

Fuzhou

-0.1081

-0.3706

-0.0164

-0.0004

0.0074

-0.0092

0.0011

Guangzhou

-0.0185

-0.0815

-0.0032

-0.0022

0.0107

-0.0072

-0.0002

Haikou

-0.0304

-0.4868

0.0077

-0.0081

0.0292

-0.0291

0.0013

Jilin

-0.1305

-0.3070

-0.0055

0.0027

0.0052

-0.0091

0.0025

Nanjing

-0.2480

-0.6407

0.0050

-0.0126

0.0224

-0.0221

0.0026

Ningbo

-0.0664

0.4866

-0.0137

0.0029

0.0007

-0.0038

0.0005

Shanghai

-0.0495

-0.4857

-0.0200

-0.0028

0.0068

-0.0093

0.0000

Shenyang

-0.0415

-1.1229

0.0222

-0.0077

0.0262

-0.0272

0.0016

Shenzhen

-0.0101

0.5447

-0.0166

0.0096

-0.0343

0.0323

-0.0002

Tangshan

0.0073

0.1926

0.0164

-0.0059

0.0138

-0.0148

-0.0004

Tianjin

-0.0133

-0.2947

-0.0196

0.0079

0.0115

-0.0160

-0.0003

Wenzhou

-0.1267

0.0563

-0.0043

-0.0020

-0.0088

0.0066

0.0003

Wuhan

-0.2260

-0.1896

-0.0100

0.0000

0.0016

-0.0078

0.0046

Urumqi

-0.1436

-0.7475

-0.0387

-0.0027

0.0122

-0.0159

0.0018

Wuxi

-0.2133

-0.4083

-0.0078

-0.0070

0.0172

-0.0169

0.0033

Xuzhou

-0.1794

-0.6709

-0.0101

-0.0068

0.0214

-0.0213

0.0023

Zhengzhou

-0.0961

-0.1679

-0.0018

-0.0050

0.0167

-0.0175

0.0016

Baoding

-0.2461

-0.2766

-0.0362

0.0078

-0.0069

1E-04

0.0021

Beijing

-0.0725

0.4200

-0.0087

0.0012

0.0090

-0.0127

0.0004

Hefei

-0.1805

-0.7416

-0.0480

0.0005

0.0308

-0.0293

0.002

Yancheng

-0.2243

-0.1542

0.0059

-0.0024

0.0116

-0.0141

0.0037

Yangzhou

-0.1106

0.3070

0.0088

-0.0015

-0.0515

0.0459

0.0014