market forces and price dispersion: evidence from china’s...
TRANSCRIPT
1
Market Forces and Price Dispersion:
Evidence from China’s Passenger Car
Markets
Linhui Yu1 and Pinliang Luo2
1The University of Hong Kong and 2Fudan University
1. Introduction
A large body of literature has addressed price dispersion or violation of the Law of One Price
(LOP) in various markets. Previous findings have shown that price dispersion are more common in
international markets than in intranational markets (Frankel and Rose, 1996; Engel and Rogers, 1996;
Rodrik, 2000; Taylor and Taylor, 2004), and the LOP holds better in developed countries than in
developing countries (Parsley and Wei, 1996; Engel and Rogers, 2001; Roger, 2001; Cecchetti, Mark,
and Sonora, 2002; Ceglowski, 2003; Goldberg and Verboven, 2005). China’s economic miracle in
the past 30 years has aroused world-wide interests on Chinese-style market economy including the
market integration of China. In a seminal paper, Young (2000) finds substantial price dispersion of
selected consumer goods, agricultural products and industrial materials in China during 1980s-1990s.
This finding combined with some other evidences prompt her to conclude that there exists serious
market fragmentation in China, which may endanger sustainable economic development of China.
Poncet (2002) at the same time finds decreasing trend of inter-province trade flows in China, which
further strengthens Young (2000)’s findings. More recent work have shown evidences of improving
2
integration of China’s markets in later 1990s and early 2000s (Fan and Wei, 2007), but the realistic
situation according to survey papers by Li et al (2004a, 2004b) is still worrying, especially in some
industries where firms have considerable monopolistic powers, e.g. auto industries.(Wang, 2003; Qiu,
2005; Zhao and Anand, 2009).
Many scholars have also investigated the causes of market fragmentation in China. Bai et al
(2004) find that after China’s fiscal decentralization reform in 1994, local governments tend to
protect state-owned enterprises (SOEs) for tax purpose, which caused market fragmentation and slow
regional specialization across China’s provinces. Lin et al (2005) argue that Chinese central
government emphasizes too much on catch-up of provinces in economic growth, which results in
large amounts of redundant constructions and serious market fragmentation in China. Chen et al
(2007) argue that openness inequality of China’s regions caused by time and magnitude of openness
also accounts for the market disintegration in China.
A common feature of the above studies is they all focus on the destructive forces of government
to the market integration of China. Actually, except for government intervention, some spontaneous
market forces can also lead to serious market fragmentation. For example, Verboven (1996)
investigates the causes of price dispersion in European car markets and finds that demand elasticity,
import quota and collusion of car firms accounted for a considerable part of car price differences
across European countries. Goldberg and Verboven (2001), by estimating marginal costs and demand
functions of cars in European countries in the 1990s, find that demand elasticity combined marginal
costs well explain the car price gaps across European countries. In addition, O’Connell and Wei
(2002) investigate the price discrepancies of selected commodities in U.S markets. Their study
indicates that technology gaps and preference differences of customers across cities are among the
“market frictions” that lead to price disparity. These stream of works reveal the effect of market
forces on price dispersion in commodity markets, and provide us interesting angles to detect possible
reasons of violation of LOP. However, to our best knowledge, few studies have attempted to look
3
into the price dispersion in China from such angles.
This research, focusing on China’s locally-produced passenger car markets, aims to identify the
possible reasons for substantial car price dispersion across China’s regions detected by us. Different
from most existing works that attribute market fragmentation of China to government intervention,
this study relies on demand and supply analytical framework, and attempts to interpret car price
dispersion as the results of spontaneous market forces, i.e., income effects from demand side and
competition effects from supply side. We also empirically test this theoretical explanation. We
choose China’s locally-produced passenger car markets for investigation for the following reasons.
First, passenger car industry as one of the key industries to China’s economy has long been accused
of its substantial market fragmentation and price dispersion across regions (Harwit, 2001; Thun,
2006). Second, passenger car as a commodity has the advantages of uniform in quality, thus it is
much easier to identify their prices for different models, which is a major reason why car markets are
frequent investigated in literature on market integration.
By collecting the selling prices of 20 popular locally-produced passenger car models in China’s
36 big cities during 2004-2006, we find that car prices as a whole were experiencing rising
divergence across China’s big cities during that period. We further control for the costs in car prices
using hedonic pricing model and thus obtain the markup level of passenger cars in each city. The
results show that car price dispersion is still outstanding during the period examined, which reflects
the poor market integration of passenger cars in China. Based on the assumption that China’s
passenger car markets are oligopolistic, we illustrate how cross-region demand and supply changes
affect car prices, and eventually lead to price dispersion. We then build a simple oligopolistic pricing
model, based on which we interpret price differentials across cities as the results of income effect
from demand side and competition effect from supply side. As the last step of our research, we
empirically test our theoretical prediction using a comprehensive dataset containing both car prices
and car sales in China’s 36 big cities.
4
This research contribute to the existing literature in two ways. First, by providing new evidences
from China’s passenger car markets, we attempt to shed light on the ongoing debate about the quality
of market integration in China in recent years. Second, we also come up with two explanations for
car price dispersion in China from the perspective of market forces, which has been long been
ignored by existing studies on market integration in China.
The rest of the paper are organized as follows. In section 2, we briefly introduce the background
of China’s passenger car markets. In section 3, we describe the datasets and car price dispersion
across China’s cities. In section 4, we illustrate mechanism of price dispersion driven by demand and
supply changes in an oligopolistic market and come up with our prediction based on a simply
oligopolistic pricing model. In section 5, we empirically test our predictions from the theoretical
model. Section 6 are concluding remarks and policy implication.
2. China’s Passenger Car Market
China’s passenger car industry started in 1950s but experienced very slow growth until 1980s
when Chinese government met the problem of ever-increasing needs for official cars. Volkswagen,
recognizing the huge potential of this market, grabbed the opportunity of establishing joint-venture
with China’s domestic car producer (Shanghai Automotive co., Ltd) in early 1980s. This JV occupied
more than half of China’s passenger car markets for over 10 years until other automobile titans
surged into China in early 2000s. We present in Figure 1 the total output of passenger cars in China
during 1978- 2006. It is shown that the first market boom appeared in the early 1990s when the
annual car output growth rate exceeded 100% in 1992 and 1993. Starting from 2000, this market
again experienced strong growth. E.g. the total output increased from less than 10 million in 2002 to
40 million in 2006, which is almost quadrupled within 4 years. Figure 2 shows the market shares of
car companies in China in 2006. It is shown that VW still ranks the first although it has lost most of
its previous markets in China (from market share of 80% in peak year to 17% in 2006). New entrants
5
including GM, Honda, Hyundai, Toyota, Nissan, etc., account for the vast majority of market shares
lost by VW. Interestingly, most passenger companies in China are JVs between local companies and
world auto titans. Local car companies only account for less than 26% of the total car output in China
in 2006. Local companies like Chery, BYD and Geely are growing rapidly in last decade, and have
played important roles in this market. The rapid growth of passenger car markets in China mirrors
the persistent and strong economic growth of China in the past 30 years, which can be seen in Figure
3 showing per capita disposable income of urban household in China from 1978 to 2006.
Figure 1. Production and Growth Rate of China’s Passenger Car Industry
(Source: China Automotive Industry Yearbook 2007)
China’s passenger car industry has long been strictly regulated by the government before 2001.
In the regulatory era, entry licenses, production, and even the selling prices were controlled by the
government. Most previous regulation terms were illuminated gradually after China’s joining WTO
in 2001. For example, Since 2001 the government no longer dictate prices of cars, and return pricing
0%
20%
40%
60%
80%
100%
120%
140%
160%
0
10
20
30
40
50
60
70
80
An
nu
al
gro
wth
ra
te
Pro
du
ctio
n (
Mil
lio
n)
Total auto production Passenger car production
Auto production growth Passenger car production growth
6
power to auto firms. This results in fierce price wars in car markets since then.1 Figure 4 shows the
price indices of locally-produced passenger cars in China in 2004-2006 based on selling prices of 20
car models in this research. It should be noted that the overall car price is declining sharply, almost
dropping 17% by 2006 comparing with 2004. Note that even until recently price cut looks still the
most effective and popular way for selling cars in China.
Figure 2. Market Share of Local-produced Passenger Car Brands in China (2006 sales)
(Source: China's auto market almanac, 2007)
There are two important characteristics of China’s passenger car markets. First, most auto
companies in China strictly prohibit their distributors from selling cars across regions. Distributors
violating this rule will be deprived the qualification of distribution. This measure actually facilitates
auto companies to charge different prices in different regions (Thun, 2006). Second, auto companies
mostly rely on local distributors to provide after sales services for their customers, which results in
1 Another reason for price wars is capacity surplus with new entrants enter and expansion of existing firms.
GM, 11%
Ford, 3%D.C., 0%Hyundai, 7%
Kia, 3%
Honda, 8%
Toyota, 6%
Nissan, 5%
Suzuki, 4%
Mazda, 3%
Mitsubishi, 1%
VW, 17%
Peugeot, 5%
Fiat, 1%
BMW, 1%
Volvo, 0%
Chery, 7%
Geely, 5%
FAW Xiali, 5%
BYD auto, 2%
Brilliance, 1%
Others, 6%
Domestic, 26%
7
discrimination of local distributors against users buying their cars non-locally. E.g., they may be
charged higher prices for repairing cars or changing parts (Qiu, 2005). Due to high costs and possible
troubles facing, people prefer buying cars from local sellers. As a result, passenger car markets are
actually fragmented in different regions of China. We therefore base our analysis on this important
assumption.
Figure 3. Annual Disposable Income of Urban Household in China (1978-2006)
(Source: China Statistical Yearbook, 2006)
Figure 4. Price Indices of China’s Locally-produced Passenger Cars (2004-2006)
0
2,000
4,000
6,000
8,000
10,000
12,000
1978 1980 1985 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006
RM
B
Average Disposalable Income of Urban Household
80
82
84
86
88
90
92
94
96
98
100
2004:Q1 2004:Q2 2004:Q3 2004:Q4 2005:Q1 2005:Q2 2005:Q3 2005:Q4 2006:Q1 2006:Q2 2006:Q3 2006:Q4
Car Price Indices Based on 20 models …
8
3. Description of Price Dispersion
a. Data
The price data used in this research is from China Price Information Network (CPIN)—a
subsidiary of National Development and Reform Commission (NDRC) of China. It contains monthly
pre-tax selling prices of 20 locally-produced passenger cars in China’s 36 big cities from 2004
January to 2006 December.2 Car models provided in the dataset represent the most popular cars from
high-end to low-end, including executive cars, large family cars, small family cars, super-mini cars,
and city cars. We show the detailed information of car models and cities in Appendix I and Appendix
II, respectively. Due to missing information, this price dataset is an unbalanced panel having 16114
observations. We also obtain from China Auto Circulation Association (CACA) a comprehensive
dataset containing sales information of all passenger cars sold in China during 2004-2006.3
Table 1 presents the summary statistics of prices and price differentials of car models. The
average prices of 20 car models range from 39,300 to 504,100 RMB (about 4,775-61,251 USD). The
average price differentials of these models across cities range from 2,500 to 30,800 RMB (about
304-3,743 USD). In Figure 5, we show the coefficient of variation (CV) of car prices cross cities. It
is seen that CVs of car prices are rising over time, which indicates that car price differentials are
getting larger and larger across regions. We further shows the CVs of prices for different groups of
cars classified by price ranges in Figure 6. It is seen that cars priced between 60,000-120,000 RMB
and below 60,000 RMB, typically economy cars, have the highest CVs of prices. In contrast, the cars
priced between 200,000-300,000 RMB have the lowest CV of prices. To exclude the possible impact
2 These 36 cities are 4 municipalities (Beijing, Shanghai, Tianjin and Chongqing), 27 provincial capital cities and 5
national-level big cities (Dalian, Qingdao, Ningbo, Xiamen and Shenzhen).
3 It reports province-level sales information of each model at monthly intervals. Although sales are mostly reported at car
model level, we are still unable to obtain accurate sales information of several car models because their sales information
is provided at more aggregated level. E.g. at brand level. We finally have exact sales information for 14 models.
9
of outlying observations, we also present in Figure 7 the interquantile range (IQR) of prices for each
group of cars. Consistent with above findings, the IQRs are still increasing for all car groups, which
indicates that our findings are not driven by outliers.
Table 1. Summary Statistics of Car Prices (2004-2006)
Selling price Price differential No. of Obs. Mean S.D. No. of Obs. Mean S. D. Model 1 1120 71.7 0.66 17296 5.8 0.56 Model 2 1173 236.6 2.11 18874 10.6 1.29 Model 3 1160 40.2 0.34 18258 2.5 0.29 Model 4 1106 112.9 1.35 16725 7.6 0.72 Model 5 1001 92.2 1.21 13882 10.0 0.79 Model 6 1119 250.3 1.45 17036 8.4 1.35 Model 7 557 140.3 1.42 4801 10.6 0.95 Model 8 997 94 1.16 13405 8.9 0.72 Model 9 1166 196.7 1.05 19132 8.4 0.83 Model 10 1185 104.7 0.83 19034 6.5 0.77 Model 11 1078 39.2 0.49 15845 3.2 0.40 Model 12 1145 504.1 3.66 18678 30.8 3.11 Model 13 972 102.6 1.01 12889 7.4 0.66 Model 14 1136 169.9 1.52 18279 12.9 1.03 Model 15 1132 79.3 1.37 17481 10.5 0.98 Model 16 933 134.9 2.01 12123 14.4 2.31 Model 17 978 39.2 0.45 13142 3.8 0.35 Model 18 876 89.8 1.06 10607 11.4 0.93 Model 19 1007 199.6 1.92 14118 17.3 1.83 Model 20 1176 41.1 0.39 18768 3.7 0.29 Note: unit of price is 1,000RMB (121.5USD)
10
Figure 5. CV of Car Prices for 20 Models (2004-2006)
0
2
4
6
8
10
12
14
16
18
2004:Q1 2004:Q2 2004:Q3 2004:Q4 2005:Q1 2005:Q2 2005:Q3 2005:Q4 2006:Q1 2006:Q2 2006:Q3 2006:Q4
Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7 Model 8 Model 9 Model 10
Model 11 Model 12 Model 13 Model 14 Model 15 Model 16 Model 17 Model 18 Model 19 Model 20
11
Figure 6. CV of Car Prices by Category (2004-2006)
Figure 7. Interquantile Range of Car Prices by Category (2004-2006)
0
1
2
3
4
5
6
7
8
9
2004:Q1 2004:Q2 2004:Q3 2004:Q4 2005:Q1 2005:Q2 2005:Q3 2005:Q4 2006:Q1 2006:Q2 2006:Q3 2006:Q4
<60,000RMB 60,000-120,000RMB 120,000-180,000RMB
180,000-300,000RMB >300,000RMB
0.9
0.95
1
1.05
1.1
1.15
1.2
1.25
2004:Q1 2004:Q2 2004:Q3 2004:Q4 2005:Q1 2005:Q2 2005:Q3 2005:Q4 2006:Q1 2006:Q2 2006:Q3 2006:Q4
<60,000RMB 60,000-120,000RMB 120,000-200,000RMB
200,000-300,000RMB >300,000RMB
12
b. A formal test of Price Dispersion
The car prices dispersion presented above give us an impression that China’s passenger car
markets were experiencing market disintegration during 2004-2006. But to reach an conclusion, we
need a more rigorous test. We follow Goldberg and Verboven (2001, 2005) to formally test the price
dispersion using a hedonic pricing model. The aim of this approach is to obtain markup level of cars
in each region by eliminating cost components from car prices. Usually cost components of a car
include manufacturing costs that can be well captured by its physical characteristics (e.g.
displacement, horse power, weights, length, etc.), transport costs from assembly plants to destination
markets measured by distance, and sales costs (e.g. advertisement expenses, promotion expenditure,
etc.) which varies with car models and corresponding firms and can be controlled for by full set of
model dummies and firm dummies. Therefore, we employ the following hedonic pricing regression
model:
, ,a a a ai t i i i f t t a i tp X dγ µ θ θ θ θ θ ε= + + + + + + + , (1)
where, ,ai tp is log selling price of the car model i in city a at time t , iX is a vector of physical
characteristics of model i , aid is distance between assembly plant of model i and city a , fθ
and iθ are firm dummies and model dummies, atθ is full sets of city-time dummies capturing
markup levels of car prices in each city at time t . We also allow for the possible economic cycle,
sales season and region specific factors that affect the car prices by controlling for time dummies tθ
and city dummies tθ in the regression equation.
The regression results of equation (1) are reported in Table 2. We first regress log prices on all
covariates except city-time dummies and report the results in column 1. It is shown that distance has
significantly positive impact on selling prices of cars. Besides, the prices of cars can be well
13
explained by their various physical characteristics. Note that R-square of regression in column 1 is
0.88, which indicates that all covariates in the regression account for about 88% of total car prices.
As a second step, we further put city-time dummies into the regression and show the results in
column 2. In addition to consistent estimates in column 1 and column 2, more importantly, R-square
of column 2 increases to 0.98, which implies that markups captured by city-time dummies account
for 10% of car prices. So, we can utilize this information to detect variation of regional markup level
of cars.
Table 2. Hedonic Pricing Regression
Dependent: Log price Hedonic(1) Hedonic(2)
Log distance 0.02043 *** 0.00469 ***
(0.00143) (0.00086)
Displacement (L) 1.38216 *** 0.91262 ***
(0.00769) (0.01174)
A/T 0.00956 ** 0.08862 **
(0.00525) (0.00449)
Acceleration (SECs/100km) -0.00087 *** -0.04375 ***
(0.00021) (0.00215)
Fuel consumption (L/100km) -0.11955 *** -0.11476 ***
(0.00316) (0.00275)
Weight (Ton) 0.08794 *** 0.08435 ***
(0.00184) (0.00998)
Firm dummies Yes Yes
Model dummies Yes Yes
City dummies Yes Yes
Time dummies Yes Yes
City-time dummies (markups) No Yes
R-squared 0.8846 0.9809
Observations 21017 21017
Notes: Robust standard errors are reported in the parenthesis. ***, **, and * indicate
statistical significance at the 1%, 5%, 10% level, respectively.
14
Figure 8. Markup Level of Passenger Car Prices in China’ Cities (2004-2006)
-25%
-20%
-15%
-10%
-5%
0%
5%
10%
15%
beijin tianjin shijiazhuang taiyuan huhehaote shenyang dalian changchun harbin
shanghai nanjing hangzhou ningbo hefei fuzhou xiamen jinan qingdao
nanchang zhengzhou wuhan changsha guangzhou shenzhen nanning haikou chongqing
chengdu kunming guiyang lasa xian lanzhou xining yinchuan wulumuqi
15
We set Beijing as benchmark in each period and plot relative value of all estimated city-time
dummies in Figure 8. This figure indicates that markup levels of cars across regions were still
widening during the period examined. More exactly, the markup differentials across regions are up to
less than 10% (lowest: Zhengzhou; highest: Shenzhen) in January of 2004,4 but has evolved to
around 20% (lowest: Xian; highest: Beijing) in three years by 2006 December. Since we use the
markups of Beijing as benchmark for comparison, the declining trend of the most lines indicates that
markups of cars in Beijing move gradually from moderate level at the beginning to the highest
among all 36 cities by the end of 2006. By contrast, Chongqing, Jinan and Taiyuan have been among
the cities with lowest markup level of car prices since 2004. The information delivered by Figure 8
further verify the existence of price dispersion in China’s passenger car markets.
4. Theoretical model
a. Demand and Supply Analytical Framework
Once the existence of price dispersion in China’s passenger car markets have been verified, we
go further to investigate what drives increasing car price differentials across cities in China. Previous
studies have carefully examined car price dispersion in European countries. For example, Goldberg
and Verboven (2001) decompose car prices into costs and markup, and find that markup levels of
cars vary substantially across European countries, which may account for the considerable car price
differentials across European countries around 1990. Lutz (2004) also looks into the price
differentials in European car markets and finds that exchange rates, transport costs, and market size
are major reasons for car price dispersion in Europe. Similar to Europe countries, China’s passenger
car markets, although exempt from the problems like exchange rate, import quota, trade barriers, etc.,
are actually segmented due to the reasons previously mentioned. In this circumstances, market forces
4 Car prices in Lhasa, the capital city of Bibet, are considered as outliers because Lhasa is very far away from most other cities in
China, making its transport costs prohibitively high and the demand there is also distinct from other places.
16
should play important roles in driving car price dispersion in China as shown in existing studies of
European car markets. Therefore, we attempt to discover the reasons for car price dispersion in China
under demand and supply framework. Following IO literature, equilibrium prices are determined by
demand of buyers and supply of producers. In Figure 8, we illustrate the evolution of car prices
facing demand and supply changes, which eventually forms price dispersion across regional markets.
Figure 9. Impacts of Demand and Supply Changes on Price Dispersion
For a car producer in an oligopolistic market, the demand curve they face in a region is mainly affected
by the income level of the region and the preference of customers in the region. And the supply curve they
face is mainly affected by his input costs and optimal output strategies (also his competitors’ output strategies).
In the left panel of Figure 9, we show the impact of cross-region demand changes on car price. Assume D1
and D2 are demand curves of a car model in city 1 and city 2, respectively, and S is supply curve of the car
model. Price gap between city 1 and city 2 is thus Gd. Given supply do not change, demand curves of city 1
and city 2 move to D1* and D2*, respectively, which enlarges price gap to Gd*. Similarly, in the right panel of
Figure 9, we show the impact of cross-region supply change on car price. Given demand do not change, if
supply curve of two cities change from S1 and S2 to S1* and S2*, respectively, the price gap between them
enlarges from Gs to Gs* accordingly. In sum, increase of demand gap and increase supply gap between two
cities both lead to increase of price gap, but their effects offset when they move in the same direction, so the
eventual price gap between them depends on their net effects.
D1
D2
D1*
D2* Gd*
Gd
S
P
Q
S1
S2
S1*
S2*
Gs Gs*
D
P
Q
17
b. An Oligopolistic Pricing Model
Like European car markets, China’s passenger car markets can also be characterized as oligopolistic markets
that are fragmented in different regions. Assuming that all car producers sell in nation-wide markets, and they
price differently in each region for each car model to maximize their profits. Let us assume the demand
function of car model i as
a a a ai i ip e rq qη −= − − , (2)
where aip denotes selling price of car model i in city a , and ae , a
iq and aiq− are income level of city
a , supply level of car model i in city a , and supply level of model i ’s competitors in city a , respectively.
This demand function make sense because selling price increases with income level and decreases with supply
level (both own and competitors’). Therefore, the profits of car model i is
N N Na a a a a
i i i i i i ia a a
p q t q c q Fπ = − − −∑ ∑ ∑ , (3)
where ait is transport costs of car model i from its assembly plant to city a , ic is marginal costs of
model i , F is fixed costs of production, and N is the total number of cities. The first order condition of
profit function is
2 ( ) 0a a a aii i i ia
i
e rq q t cq
π η −∂ = − − − + =∂
. (4)
Accordingly, we obtain the optimal output and price of model i in city a as
* 1( ) ( )
2a a a ai i i iq e q t c
rη −= − − −
* 1( ) ( )
2a a a ai i i ip e t c qη −= + + − . (5)
And the price differential between city a and city b is
, * * * 1 1 1( ) ( ) ( ) ( ) ( ) ( )
2 2 2a b a b a b a b a bi i i i i i ip p p e e t t q qη − −= − = − + − − − . (6)
Therefore, first-differenced term of ,( ) *a bip is
18
, * , , , , , ,1 , , 1 ,
1 1( ) ( ) ( )
2 2 2 2a b a b a b a b a b a b a bi t t i t i t t i tp e e q q e q
η η− − − − −∆ = − − − = ∆ − ∆ . (7)
As seen in equation (7), the marginal costs ic and the transport costs ait are cancelled out, thus the price
differentials of car model i across cities is determined by income differentials and its competitors’ supply
differentials across cities. Positive sign of ,a bte∆ implies the positive driving force of income differentials on
car price differentials from demand side (i.e. income effect), while the negative sign of ,,
a bi tq−∆ implies
negative driving force of competition on car price differentials from supply side (i.e. competition effect). We
will empirically test for these two kinds of forces in the next section.
5. Empirical Model and Estimation Strategy
Based on equation (7), we arrive at the following dynamic panel regression model for estimation:
5
, 0 , , , ,i t i t i g i g t i tg
DPD DID DSDα γ β ε− −= + + +∑ , (8)
where the dependent variable ,i tDPD denotes first-differenced price differentials of model i between cities
within city-pairs in time t , ,i tDID denotes first-differenced urban household disposable income differentials
between cities within city-pairs, and , ,i g tDSD− is first-differenced supply differentials of all car models other
than i in car group g 5, and ,i tε is error terms. We differentiate competition effect of different car groups
because cars in one group don’t compete directly with cars in other groups due to their different market
positioning. E.g. supply change of executive cars like Auto A6 suppose to have little impact on selling prices
of city cars like Suzuki series. Following the typical classification of passenger cars in China, we classify all
20 car models into five groups:
Group 1: Executive cars/Full-size cars (price >300,000 RMB)
Group 2: Large family cars/mid-size cars (priced180,000-300,000 RMB)
5 We calculate within-group competitive supply by deducting the car model’s supply from the total supply of its group,
and use total supply as competitive supply of other groups.
19
Group 3: Small family cars/Compact cars (priced 120,000-180,000 RMB)
Group 4: Supermini cars /Subcompact cars (priced 60,000-120,000 RMB)
Group 5: City cars (price <60,000 RMB)
As car sales information is unavailable at city-level, we thus disaggregate provincial-level sales to
city-levels by GDP share of cities in the province.6 We obtain urban household disposable income in
each quarter from china statistical yearbook, and aggregate monthly price and sales information at
quarterly level accordingly. As we has mentioned before, we only have accurate sales information of 14 car
models.
As to the econometrical issues of empirical part, although first-differenced variables are put into our
regression, there are still some concerns about this approach. For one, although all unobservable
time-invariant factors are eliminated in first-differenced regression, independent variables may still be
correlated with unabsorbed time-varying variables hidden in error terms, which causes bias of estimation. For
another, , ,i g tDSD− is endogenous variable because of its potential reverse causality with dependent variable
, ,i g tDPD− . Leaving these endogeneity issues unsolved will make our estimation results biased. To obtain
consistent estimates in our regression model, we use two-step system GMM estimator (Arellano and Bover,
1995; Blundell and Bond, 1998), which uses lagged first-differences as well as levels as instruments for
estimation of equation (8). This estimator has significant merits over alternative estimators, which can help us
easily deal with endogeneity concerns in estimation. Although system GMM approach is more advantageous
than most other estimation approaches, we also need to be careful that it easily produces invalid estimates. So
correct application of this method strictly reply on some important test statistics for autocorrelation,
overidentification and exogeneity of instruments (Roodman, 2006).
Logarithm of variables are used in the regression, and we expect positive estimate of iγ (income effect
of model i ) and negative estimate of , ii gβ− (competition effect of all car models other than i ) as predicted
in previous theoretical analysis.
6 We also tried the method of using population share to disaggregate the province-level data to city-level, and the results are very similar.
20
6. Empirical Results
We report in Table 3 the two-step system GMM estimation results for the 14 car models.7 Instead of
pooling all car models together for estimation, we estimate equation (8) one by one for each car model, and all
the results are shown in column 1-14. Our reason for doing so is that different car model belongs to different
groups which may have different demand function from other groups, so the estimation results obtained from
pooled sample regression may be mixed or even misleading. Separately estimating the regression model help
us clearly disentangle differential income effects and competition effects for different cars.
As DSD is an endogenous variable, we use its lag 2 and deeper lags of all independent variables as
instruments for the first-differenced equations. And we also use lagged first-differences as instruments for
level equations. We rely on Hansen J statistic to test the validity of instruments, and rely on p-value of
Arellano-Bond m1 and m2 test to ensure the absence of autocorrelation of order 2 in the residuals. The
corresponding tests results are reported at the bottom of Table 3. As we can see from it, the Hansen J statistic
of all estimation equations do not reject the null hypothesis of no misspecification, and all estimations pass
Arellano-Bond m1 and m2 tests, i.e. reject zero first-order correlation in m1 and accept zero second-order
correlation in m2, suggesting the validity of moment conditions we adopted in system GMM estimation.
The first row of Table 3 shows the estimated coefficients for the lagged dependent variables, which are all
significant and positive for all car models, which indicate the route dependence of price differentials over time.
This result is actually consistent with the empirical evidence of rising car price differentials across China’s
regions during 2004-2006. The second row of Table 3 reports the estimated coefficients for Log DID. We find
all estimates are significant and positive, which implies that income differentials across regions bring out car
price differentials across regions. This makes perfect sense intuitively and is consistent with existing economic
theories. The magnitude of estimates (i.e. price elasticity of income) range from 0.041 to 0.259, which means
that 1% income differential change will result in 4%-26%’s changes of car price differentials in the same
direction. Note that car models with smaller price elasticity are those subcompact cars or city cars that have
cheaper prices, while car models with highest price elasticity are those large family cars or mid-size cars
7 The first row of each column shows the number of car models. Due to data unavailability, estimation results of model 12, 16, 17, 18,
19, and 20 are missing. For the name and detailed description of each car model, please refer to Appendix I.
21
which are relatively expensive and specially target high-end users who are insensitive to prices. Row 2-5
shows the estimated coefficients of Log DSD, which reflect the competition effect on car prices (i.e. price
elasticity of substitute supply). Since different car models belong to different groups, we underline the
corresponding estimates obtained from their own groups. It is found that all estimates underlined are negative
and significant, which suggests that for a car model increase of supply from its competitors in a city will result
in decreased price differentials between this city and other cities. This is in line with our expectation that
increasing competition (reflected as competitors increase their supply) in a car market will subsequently
decrease the car prices in the market. The magnitude of estimated coefficients range from -0.043 to -0.138,
indicating that 1% increase of supply differentials of the model’s substitute models between two cities will
result in 4.3%-13.8% decreases of price differentials between these two cities. Different from income effect,
competition effect show that high-price cars are more sensitive to competition than cheaper ones, i.e., their
prices tend to drop more substantially once their competitors increase supply.
It is also interesting to look at the competition effect from other groups (estimated coefficients not
underlined). We find in most cases that supply differentials of a car’s neighboring groups have some negative
effect, but the negative effect disappears or even become positive when the groups are “far away” from the car
model. Our explanation for this interesting finding is that two cities with substantial differentials in supply of a
certain car model may have distinct characteristics (e.g. income level, preference, etc.), which in turn fuel the
price differentials. Overall, the empirical findings obtained from two-stage system GMM estimation are in line
with our expectation and well support our predictions from theoretical analysis.
7. Conclusions and Policy Implication
In this research, using prices of cars in China’s big cities, we first verify the car price dispersion in China
during 2004-2006. Moreover, we also investigate the causes of price dispersion from the perspective of market
forces, which has been ignored by existing studies on market disintegration in China. Our theoretical
prediction suggests that income differentials and supply differentials across China’s cities may drive car price
differentials in China. And this prediction is strongly supported by the empirical evidences.
22
Table 3. Two-step System GMM (Dependent variable: Log differenced price differentials between cities; Lag 2 for FD equations, and lag 1 for level equations)
No. of Car Model 1 2 3 4 5 6 7 8 9 10 11 13 14 15
L.DPD 0. 343** 0. 375*** 0. 338*** 0. 341*** 0. 337*** 0. 366*** 0. 359** 0. 342*** 0. 357*** 0. 331*** 0. 324*** 0. 332*** 0. 343*** 0. 356**
(0. 104) (0. 129) (0. 113) (0.096) (0. 094) (0. 106) (0. 110) (0. 102) (0. 109) (0. 093) (0. 094) (0. 101) (0. 105) (0. 108)
Log DID 0. 069** 0. 259*** 0. 061*** 0. 089*** 0. 085*** 0. 206*** 0. 110** 0. 082*** 0. 185*** 0. 086*** 0. 041*** 0. 087*** 0. 143*** 0. 162**
(0. 034) (0. 079) (0. 015) (0. 029) (0. 031) (0. 063) (0. 048) (0. 026) (0. 068) (0. 033) (0. 012) (0. 037) (0. 052) (0. 076)
Log DSD _G1 0. 045*** -0. 274** 0. 037*** 0. 048*** 0. 051*** -0. 252* 0. 074 0. 045*** -0. 119 0. 049*** 0. 023*** 0. 044*** 0. 086 0.076***
(0.017) (0. 132) (0. 012 ) (0. 018) (0. 023) (0. 149) (0. 072) (0. 017) (0. 083) (0. 018) (0. 009) (0. 016) (0. 081) (0. 025)
Log DSD _G2 0. 039 -0. 138** 0. 017** 0. 024 0. 021 -0.127** -0. 064 -0. 022 -0. 132** 0. 038* 0. 021*** -0. 039* -0. 068 0. 057**
(0. 034) (0. 053) (0. 007 ) (0. 027) (0. 025) (0. 051) (0. 082) (0. 026) (0. 063) (0. 021) (0. 008 ) (0. 022) (0. 078) (0. 027)
Log DSD _G3 -0. 037 -0. 049* 0. 025* -0. 043* -0. 046** -0. 055* 0. 092*** -0. 069** -0. 077 -0. 078** 0. 019 -0. 073** -0.103*** 0. 058
(0.027) (0. 029) (0. 014) (0. 022) (0. 023) (0. 032) (0. 027) (0. 033) (0. 054) (0. 038) (0. 013) (0. 035) (0. 039) (0. 050)
Log DSD _G4 -0. 064** 0. 038** -0. 023* -0. 052*** -0. 068*** 0. 041*** -0. 035** -0. 072*** 0. 062*** -0. 078** -0. 027 -0. 075*** -0. 057*** -0. 084***
(0. 027) (0. 014) (0. 013) (0. 017 ) (0. 025) (0. 016) (0. 014) (0. 026) (0. 031) (0. 031) (0.017) (0. 029 ) (0. 016) (0. 028)
Log DSD _G5 -0. 035*** 0. 114*** -0. 043*** -0. 042** -0. 085*** 0. 137*** 0. 106*** -0. 064** 0. 148*** -0. 069** -0. 046*** -0.072 0. 117** -0. 109***
(0. 016) (0. 035) (0.014) (0. 018) (0. 027 ) (0. 045) (0. 033) (0. 029) (0. 049) (0. 035) (0. 014) (0. 062) (0. 046) (0. 038)
p-value of Hansen J 0.179 0.386 0.265 0.117 0.672 0.426 0.275 0.127 0.221 0.235 0.297 0.621 0.448 0.134
p-Value of m1 test. 0.004 0.001 0.000 0.003 0.013 0.006 0.047 0.008 0.001 0.001 0.003 0.021 0.036 0.010
p-Value of m2 test. 0.310 0.428 0.232 0.382 0.511 0.247 0.139 0.372 0.351 0.206 0.396 0.218 0.499 0.381
NO of Obs. 4610 5233 5568 4675 3659 4242 1050 4295 5018 5425 4541 1552 1937 4428 Notes:
All variables in the regression are logged.
***, **, and * denote statistical significance of the 1%, 5% and 10%, respectively. Robust standard errors are in the parenthesis.
G1-G5 are five groups of car models classified by prices from high to low. Estimates underlined are effects from the corresponding models’ own groups.
Estimation results for car model 12, 16, 17, 18, 19, 20 are unavailable due to missing sales information.
23
There are two major findings of this research. First, we find rising price dispersion in China’s passenger
car markets during the examination periods, which indicates the poor integration level of China’s passenger
car markets. Second, we find that income effect from demand side and competition effect from supply side
significantly account for the price dispersion in China’s passenger car markets, which sheds light on the
existing explanations of market disintegration in China in recent years.
The findings of this research have some meaningful implications for China’s existing economic policies.
Firstly, rising dispersion in China’s passenger market has proven to be partly driven by income inequality
across China’s regions, which is mostly induced by development imbalance in China. Leaving this issue
unsolved, China is taking the risks of heading to serious market disintegration which will definitely endanger
sustainable growth of China’s economy. Secondly, it is essential that implicit trade barriers created by big
firms with monopolistic power, which are prevailing in some industries like passenger industry, should be
restricted in order to facilitate better market integration in China.
7. References
Arelleno, M. and O. Bover (1995), Another Look at Instrumental Variable Estimation of Error Component Models, Journal of Econometrics, 68, pp.29–51.
Bai, Chong-En, Yingjuan Du, Zhigang Tao, and Sarah Tong (2004), Local Protectionism and Regional Concentration: Evidence for Chinese Industries, Journal of International Economics, 63, pp.397–417.
Blundell, R. S. and S. Bond (1998), Initial Conditions and Moment Restrictions in Dynamic Panel Data Models, Journal of Econometrics, 87, pp.115–143.
Cecchetti, S. G., N. C. Mark, and R. J. Sonora (2002), Price Index Convergence among United States Cities, International Economic Review, 43, pp.1081-99.
Ceglowski, Janet (2003), The Law of One Price: Intranational Evidence for Canada, Canadian Journal of Economics, 36, pp.373-400.
Chen, M., Q. Gui, M. Lu, and Z. Chen (2007), How Do China Keep Scale Economy in Economic Growth—An Empirical Research on Economic Openness and Market Fragmentation in China’s Domestic Commodity Market, China Economic Quarterly, 10, pp.125-150. (in Chinese)
Engel, Charles and John H. Rogers (1996), How Wide is the Border, American Economic Review, 86, pp.1112-1125.
Engel, Charles and John H. Rogers (2001), Violating the Law of One Price: Should We Make a Federal Case Out of It? Journal of Money, Credit, and Banking, 33, pp.1–15
Fan, C. S. and X. Wei (2006), The Law of One Price: Evidence from the Transitional Economy of China, Review of Economics and Statistics, 88, pp. 682-697.
Frankel, J. A. and Andrew K. R. (1996), A Panel Project on Purchasing Power Parity: Mean Reversion Within and Between Countries, Journal of International Economics, 40, pp.209-224.
24
Goldberg, P. and Verboven, F. (2001), The Evolution of Price Dispersion in the European Car Market, Review of Economic Studies, 68, pp.811-848.
Goldberg, P. and Verboven F. (2005), Market Integration and Convergence to the Law of One Price: Evidence from the European Car Market, Journal of International Economics, 65, pp.49-73.
Harwit, Eric (2001), The Impact of WTO Membership on the Automobile Industry in China, The China Quarterly, 167, pp.655-670.
Li, S., Y. Hou, Y. Liu, and B. Chen (2004a), A Survey Study on China’s Local Protectionism—Evidence from Enterprises Sample, Economic Research References, 6, pp. 2-18 (in Chinese)
Li, S., Y. Hou, and B. Chen (2004b), A Survey Study on China’s Local Protectionism—Preliminary Result from Non-Enterprise Samples, Economic Research References, 18, pp. 31-38. (in Chinese)
Lin, Y. and P. Liu (2005), Local Protectionism and Market Disintegration: Investigation From Development Strategy, working paper, Economic Development Forum. No. FC20050015
Lutz, M. (2004), Pricing in Segmented Markets, Arbitrage Barriers and the Law of One Price: Evidence from the European Car Market, Review of International Economics, 12, pp. 456-475.
O'Connell, Paul G. J. and Shang-Jin Wei (2002), The Bigger They Are, the Harder They Fall: Retail Price Differences across U.S. Cities, Journal of International Economics, 56, pp.21-53.
Parsley, David and Shang-jin Wei (1996), Convergence to the Law of One Price without Trade Barriers or Currency Fluctuations, Quarterly Journal of Economics, 111, pp.1211-1236.
Poncet, S. (2002), Is China’s Market Approaching to Disintegration? A Comparison on China’s Intranational and International Market Integration, Essays on World Economy, No.1, pp.3-17. (in Chinese)
Rodrik, D. (2000), How Far Will International Economic Integration Go? Journal of Economic Perspectives, 14, pp.177–186.
Roodman, D. (2006), How to Do xtabond: An Introduction to ‘‘Difference’’ and ‘‘System’’ GMM in Stata, Center for Global Development Working Paper No. 103. Washington, DC: Center for Global Development.
Qiu, Larry D. (2005), China’s Automotive Industry, MBA case, HKU.
Taylor, Alan M. and Taylor, Mark P. (2004), The Purchasing Power Parity Debate, Journal of Economic Perspectives, 18 , pp.135–158.
Taylor, Alan M. (2001), Potential Pitfalls for the Purchasing-Power-Parity Puzzle? Sampling and Specification Biases in Mean-Reversion Tests of the Law of One Price, Econometrica, 69, pp.473-498.
Thun, Eric (2006), Changing Lanes in China: Foreign Direct Investment, Local Governments, and Auto Sector Development, Cambridge University Press, New York
Verboven, F. (1996), International Price Discrimination in the European Car Market, Rand Journal of Economics, 27, pp.240-268.
Wang, Hua (2003), Policy Reforms and Foreign Direct Investment: The Case of the Chinese Automobile Industry, Journal of Economics and Business, 6, pp.287 –314.
Young, Alwyn (2000), The Razor’s Edge: Distortions and Incremental Reform in the People’s Republic of China, Quarterly Journal of Economics, 115, pp.1091-1135.
Zhao, J. Z. and Jaideep Anand (2009), A multilevel Perspective on Knowledge Transfer: Evidence from the Chinese Automotive Industry, Strategic Management Journal, 30, pp.959-983.
25
Appendix Appendix I. Description of the 20 Car Models
Model No Model Name Plant Location Group Disp AT/MT Accel. Fuel Cons
Weight (Ton)
Horse PW
No. of Cylin
Length (Meter)
Max Spd
Price (04-06)
1 Chevrolet_SaiL1.6L_mt_std Jinan, Shandong G4 1.6 M 12.7 5.3 1 66 4 4 170 7.17
2 Buick_Regal2.5GL Shanghai G2 2.5 A 12.9 6.7 1.6 112 6 4.9 173 23.66
3 Suzuki_Alto_klwzSC7081 Chongqing G5 0.8 M 37 4 0.7 26.2 4 3.2 120 4.02
4 Citroen_ElyseeSX16L_mt Wuhan, Hubei G4 1.6 M 11.3 6.1 1.2 78 4 4.3 185 11.29
5 Fukang1.4L_mt Wuhan, Hubei G4 1.4 M 13.2 5.9 1 55 4 4.1 170 9.22
6 Honda_Accord2.4L_at_std Guangzhou, Guangdong G2 2.4 A 9.3 7.3 1.5 119 4 4.8 200 25.03
7 Santana3000_18L_mt_std Shanghai G3 1.8 M 13 7.6 1.2 74 4 4.5 185 14.03
8 Santana_GLi1.8L_mt_std Shanghai G4 1.8 M 13.4 7.7 1.1 72 4 4.5 165 9.4
9 Passat_ly2.0L_mt Shanghai G2 2 M 12.2 6.2 1.4 85 4 4.8 198 19.67
10 Polo1.4L_at_std Shanghai G4 1.4 A 14.8 4.6 1.1 55 4 3.9 170 10.47
11 Xiali_Junya7101a_std Tianjin G5 1 M 15.8 5 0.8 39 3 3.7 138 3.92
12 AudiA6_2.8L_std Changchun, Jinlin G1 2.8 A 9.8 7.2 1.5 140 6 4.9 228 50.41
13 Jetta_CIF16L_mt_std Changchun, Jinlin G4 1.6 M 13 6.1 1.1 68 4 4.4 175 10.26
14 Bora_1.8L_mt Changchun, Jinlin G3 1.8 M 11.1 6.4 1.3 92 4 4.4 206 16.99
15 Chery_Qiyun16GL_mt_std Wuhu, Anhui G4 1.6 M 10 5.9 1.1 65 4 4.4 186 7.93
16 BJ_Jeep2024 Beijing G3 2.2 M 32 11.5 1.6 76 4 4.3 115 13.49
17 Changan_SC6350 Chongqing G5 1 M 21 6.4 1 39 4 3.5 105 3.92
18 Changcheng_SaifuCC6470BY Shijiazhuang G4 2.2 M 18 10 1.7 74.5 4 4.5 120 8.98
19 Changfeng_CFA6470 Changsha, Hunan G2 2.4 M 16 10.8 1.8 92 4 4.8 140 19.96
20 Hafei_MinyiHF6370 Harbin, Heilongjiang G5 1.1 M 21 5.8 1.1 51 4 3.7 110 4.11 Notes: Disp=Displacement; Accel.=Accelerate(Seconds, 0-100km/hrs); Fuel Cons=Fuel consumption (L/100km); Horse PW=Horse power; Cylin.=Cylinder; Spd=Speed
26
Appendix II. Geographic Distribution of 36 cities in China