STRUCTURAL CHANGE AND GROWTH IN CHINA UNDER ECONOMIC REFORMS: PATTERNS, CAUSES AND IMPLICATIONS

Authors


Yanqing Jiang: yjiang828@hotmail.com

Abstract

This paper investigates the patterns, causes, and implications of China's structural change and its contribution to China's regional growth. Among many other findings, our regression results show that conditional convergence exists across different regions in China. Regional structural change has a convergence effect and regional openness facilitates regional structural change. Structural shocks and structural transformation had the opposite effect on China's interregional convergence during the 1990s, though the combined effect of overall structural change is a convergence effect. We also find that Chinese regions rely more heavily on structural change for labor productivity growth as the economy evolves. In summary, the results of our empirical analysis support the hypothesis underlying the theoretical model of this paper.

1. Introduction

In the past 30 years, mainland China has achieved a continuous high rate of real GDP growth under various successful economic reforms. The country has experienced a drastic economic transition with fast economic growth and significant structural change. During 1978–2003, labor productivity grew at a remarkable 5.7% per year, and millions of workers left the agricultural sector to join the industrial sectors. The proportion of workers in non-agricultural sectors increased by more than 20 percent during that period. What is the relationship between China's fast growth and its structural change? What is the impact of structural change on China's interregional economic convergence? In this paper, we empirically study the patterns, causes, and implications of China's structural change, as well as its contributions to China's economic growth. We will pay particular attention to the potential convergence effect of structural change on the economy of the Chinese regions.

An analysis of the growth and convergence process across different economies often starts from the Solovian model. The starting point is the estimation of absolute β-convergence. The prediction of neoclassical growth theory is of convergence, with the growth rate of per capita income being negatively correlated with the initial level of per capita income during the adjustment process. However, the traditional one-sector neoclassical growth theory neglects growth due to structural change, which is a process involving the reallocation of labor and other resources from a lower productivity sector to a higher productivity sector. Structural change tends to have a convergence effect on regional productivity because poor regions with a relatively larger size of the low productivity sector are able to take advantage of structural change for larger gains over rich regions (See, for example, Abramovitz, 1986).

Although voluminous literature on growth and convergence since the 1980s has been dominated by the neoclassical framework, and inadequate attention has been paid to the role of structural change as a convergence mechanism (O’ Leary, 2006), yet there have emerged many recent studies that highlight the effects of structural factors on cross-country (-region) growth and convergence. These recent studies include Laitner (2000), Hansen and Prescott (2002), Gollin et al. (2002), O’Leary (2003, 2006), Dekle and Vandenbroucke (2006), Ngai and Pissarides (2007), Sassi (2007), Brandt et al. (2008), Lee and Malin (2009), Vollrath (2009), Dessy et al. (2010) and Jiang (2010), to name but a few, of which Dekle and Vandenbroucke (2006), Brandt et al. (2008), Lee and Malin (2009) and Jiang (2010) are related to China's growth and structural change under its economic reforms. Dekle and Vandenbroucke (2006) decompose China's economy into three sectors: the private agricultural sector, the private non-agricultural sector and the public non-agricultural sector. They use an accounting method to show that the main source of economic growth during 1978–2003 was total factor productivity (TFP) in the private non-agricultural sector. Out of the 5.7% annual growth of labor productivity, 1.9% can be accounted for by the reallocation of labor from the agricultural sector to the non-agricultural sectors. By calibrating a general equilibrium model where the driving forces of economic growth are public investment, employment and sectoral TFP, they show that the key driving forces behind China's structural change were TFP growth in the private non-agricultural sector and public capital accumulation. Brandt et al. (2008) find that there are four possible proximate forces behind the reallocation of labor: a deterioration of the agricultural sector's terms of trade, an increase in the capital intensity of the nonagricultural sector relative to the agricultural sector, a reduction in labor distortions in the nonagricultural sector, and an improvement in nonagricultural TFP relative to agricultural TFP. Lee and Malin (2009) explore education's role in improving the allocation of labor across China's agricultural and nonagricultural sectors and measure the portion of China's recent growth attributable to this channel. They find that the effect of education on labor reallocation between sectors accounts for about 9% of China's growth, whereas its impact on within-sector human capital growth explains only 2%. Jiang (2010) examines effects of different structural factors on regional growth and interregional convergence in China, and, among other findings, shows that in the 1990s structural shocks tended to widen the income gap between rich regions and poor regions in China and structural transformation worked to narrow the gap, while in contrast in the 1980s essentially both structural shocks and structural transformation worked to narrow the gap between rich regions and poor regions in China.

The current paper empirically investigates the patterns, causes, and implications of China's structural change, as well as its contributions to China's regional growth and interregional income inequality.1 The central hypothesis underlying our analysis is that knowledge accumulation is one of the key driving forces of structural change and economic growth. We take the Chinese regions basically as follower economies whose TFP growth is ultimately driven by technology diffusion from the exogenously expanding world technology frontier. Knowledge accumulation of Chinese workers depends on technology spillovers from the world technology frontier. The actual effectiveness of technology spillovers from the frontier in promoting knowledge accumulation of Chinese workers is in turn dependent on three factors: 1) the degree of exposure of the region to the world technology frontier, which is affected by the degree of economic openness of the region to the outside world, 2) the workers’ absorptive capacity in the region, which should depend on their level of knowledge preparedness as determined by their level of education, as well as 3) other influencing forces within a framework of the general comprehensive socio-economic environment of the region that includes aspects such as the infrastructure, institution and economic policy. We will see that the major results of our empirical analysis in this paper, to a large extent, support our hypothesis above.

The rest of the paper is structured as follows. In Section 2, we present a model to illustrate a mechanism through which structural change is determined and how it is related to economic growth. In Section 3, we decompose total per worker output into three mutually exclusive components with meaningful interpretations to get prepared for the empirical analysis in the following section. In Section 4, we carry out our empirical analysis and present our major estimation results. Finally, Section 5 concludes.

2. Structural change in terms of labor shares

In this section, we present a simple model of Lucas (2009) (with minor adaptation) to illustrate how structural change (in terms of sectoral labor shares) occurs and how it is related to output growth. The model in this section will motivate our empirical analysis in Subsection 4.3. We first consider a world of one-sector economies in which an economy's per capita GDP is proportional to its stock of knowledge. There is a leading economy (or group of economies) where its stock of knowledge, denoted by H, grows exogenously at a constant rate inline image:

image

Knowledge accumulation of any follower economy (a Chinese region in the current case) is determined by

image(1)

where the stock of knowledge in a follower economy is denoted by h. This equation implies that while per capita GDP of the leader grows at the constant rate inline image, per capita GDP of a follower grows at the rate

image

where inline image is greater than inline image since inline image. That is, the follower grows faster than the leader, at a rate that is positively related to the size of the GDP gap, inline image, and the size of the spillover parameter, inline image. We assume that the magnitude of the spillover parameter inline image is jointly determined by the three major factors mentioned earlier: 1) the degree of exposure of the follower economy (i.e. a Chinese region) to the world technology frontier, which can be empirically proxied for by the degree of trade openness of the region to foreign countries, 2) the workers’ absorptive capacity in the region, which, for example, can be empirically proxied for by their (average) level of education, as well as 3) other influencing forces related to the region's socio-economic aspects such as the infrastructure, institution and economic policy. The effect of this third factor can be (partly) captured by including a region heterogeneity variable in the fixed-effects panel data regressions of our empirical analysis in later sections.

The solution to the differential equation for inline image with the initial value inline image is

image

In order to study structural change, we have to modify the model to accommodate multiple sectors. Suppose in a dual economy we have two sectors. Call the two sectors “farm” and “city”. A fraction inline image of each unit of labor in the economy is allocated to the city sector, where it produces

image

The remaining fraction inline image is allocated to the farm sector, where it produces inline image units of the same, single output good:

image

Here land per person is taken as fixed and incorporated into inline image. The parameter inline image, according to Lucas (2009), can be interpreted as reflecting a spillover effect of city knowledge on agricultural productivity. Labor is assumed to be mobile between the two sectors so that the equilibrium and optimal allocation of labor coincide. In this case, equilibrium output is given by

image

If inline image, then inline image. Otherwise x is given by

image(2)

The two functions, inline image and inline image, defined by this model show the way the employment share in agriculture should vary with per capita GDP. When inline image, inline image: the traditional agricultural sector eventually empties out. Equation (1) is now modified to the form

image(3)

where another parameter inline image has been added. The new term inline image can be thought of as capturing an agglomeration effect, according to which the rate of knowledge inflow to any individual is an increasing function of the fraction of city labor. Combining (2) and (3) yields

image(4)

If the knowledge level inline image is high enough, then equation (4) implies that inline image will grow indefinitely. The added agglomeration term will approach one and the follower economy will eventually behave exactly as in the previous one-sector model, with both the level of GDP and its growth rate approaching the values of the leading economy. For a low value of inline image, however, the growth rate of inline image approaches zero. The model is so designed as to eliminate the growth advantage of extreme poverty, and Lucas (2009) believes that the value of the parameter inline image can be carefully chosen to give good quantitative agreement to the growth performance of predominantly agricultural economies.

In Subsection 4.3 below, we will carry out the part of our empirical analysis that is motivated by the basic idea of the theoretical model above. However, as will become clearer later, the empirical analysis coming in Subsection 4.3 will be solely based on the theoretical model we have just presented (which is a model describing structural change in terms of the sectoral labor shares), and will not take account of structural factors that are not captured by the changing sectoral labor shares. Therefore, in the next section, in order to look closer into the patterns and causes of structural change and growth in China, we will distinguish two types of structural factors, i.e. structural shocks and structural transformation. The next section, Section 3, will form the basis of our empirical analysis in Subsections 4.4 and 4.5.

3. Decomposition of output growth

In order to carry out our empirical analysis in Subsections 4.4 and 4.5 below, we have to first decompose total per worker output growth into its different components. We assume any region inline image has inline image sectors. The output per worker of region inline image at time inline image, inline image, can be written as

image(5)

where inline image is output per worker in sector inline image, inline image is the share of labor in sector inline image, inline image, and inline image is the growth rate of output per labor in sector inline image, inline image. We denote total per worker output growth during inline image, inline image by inline image:

image(6)

We next perform a decomposition of inline image.2 In order to do so, we follow Barro and Sala-i-Martin (1995) and define a hypothetical level of output per worker inline image as

image(7)

where inline image. That is, inline image is the average value of all the inline image over the sample of all inline image regions. Based on inline image, we can follow Barro and Sala-i-Martin (1995) and formulate a hypothetical growth rate as

image(8)

The superscript SS stands for “structural shocks”, suggesting inline image captures the part of inline image resulting from “structural shocks”. Structural shocks can largely be understood as the uneven effects of the same changing factor on different regions. These regions receive uneven effects (shocks) owing to two major facts: 1) growth rates of labor productivity vary across different sectors, and 2) different regions have different sectoral compositions, i.e. the same inline image sectors take up different labor shares across different regions (Jiang, 2010). Comparing equations (6) and (8), we can see that inline image would equal inline image, if under the assumption of constant sector shares of labor over time, each sector in that region were to develop at the average growth rate of the whole sample, i.e. for region inline image, inline image for all inline image's.

We now follow Paci and Pigliaru (1997) in constructing another hypothetical growth rate:

image(9)

in which inline image is defined as

image(10)

We can see that in equation (7) the construction of inline image uses the shares of labor across the sectors at the initial time inline image while in equation (10) the construction of inline image uses the share of labor across the sectors at the current time inline image. Compared with inline image, inline image (where the superscript simply stands for “structural”) takes account of varying shares of labor across the sectors within a region over time. The difference between inline image and inline image is denoted by

image(11)

Therefore, inline image, where the superscript stands for “structural transformation, is a constructed hypothetical growth rate capturing the magnitude of the part of inline image that results from structural transformation in terms of the changes in labor shares of the sectors.

What is left in inline image after inline image is extracted is now denoted by

image(12)

inline image, where the superscript stands for “residual”, captures regional per worker output growth resulting from increases in region-specific sectoral labor productivities, due to, for instance, increases in regional technology or investment. inline image should be conceptually closer than inline image to growth driven by capital deepening as described by the Solow model (Jiang, 2010). So far, we have broken inline image up into three components:

image(13)

Thus, the overall regional per worker output growth inline image can be divided into growth due to structural shocks as measured by inline image, growth due to structural transformation as measured by inline image, and a “residual” growth due to region-specific labor productivity increases as measured by inline image. Based on this decomposition, we carry out our empirical analysis in the next section.

4. Empirical analysis and results

4.1. The data

The sample uses 31 province-level regions (provinces, province-level municipalities and ethnic minority autonomous regions) in mainland China over the period 1980–2004. Our sample period does not cover the most recent years since 2005 because the National Bureau of Statistics of China changed some of its statistical categories in 2005 and thus data on the sectoral variables since 2005 are not directly comparable to corresponding data before 2005. We use a four-year-span panel data structure in which the entire period 1980–2004 is divided into five equal-length sub-periods: 1980–1984, 1985–1989, 1990–1994, 1995–1999, and 2000–2004. The data are obtained from (or calculated based on) Chinese Statistical Yearbooks (1980–2006), which include, for each region in each year of the sample period, regional overall GDP, regional GDP of each of the three big sectors—the agricultural sector, the manufacturing sector and the service sector, regional population of all working people, regional labor shares of the three sectors, as well as regional total exports and total imports. Hypothetical variables inline image, inline image, inline image, inline image, inline image and inline image are calculated according to their definitions, with the number of sectors being three (the three sectors mentioned above).

4.2. Variance decomposition of output growth

In this subsection, we perform a variance decomposition exercise which is important in itself and which will also serve as a stepping stone on our way to the empirical analysis in Subsections 4.4 and 4.5.

Equations (11) and (12) directly lead to the following result:

image(14)

By using equation (14) we can gain a general idea of how much of the variation in inline image can be accounted for by variations in inline image, inline image and inline image.3 The decomposition in equation (14) is equivalent to looking at the OLS coefficients from separate regressions of inline image (or inline image and inline image separately) and inline image on inline image, respectively. Therefore, this decomposition shows how much higher the conditional expectations of inline image (or of inline image and inline image separately) and inline image would be if inline image is one unit higher.

We perform the variance decomposition across the Chinese regions for each four-year period, and the results are contained in Table 1. We can see that inline image, i.e. growth due to structural transformation, accounts for about 10% of the total cross-region variation in inline image in each of the five periods. The contribution of inline image, i.e. growth due to structural shocks, has remained less stable. The combined effect of inline image and inline image accounts for about 7%∼25% of the total cross-region variation in inline image in different periods. The contribution of inline image, i.e. the “residual” growth due to region-specific labor productivity improvement, takes the lion's share (75%∼93%) in accounting for total cross-region variation in inline image in each of the periods. In addition to the within-period cross-region decomposition, the last row of Table 1 lists the results of our variance decomposition exercise when the five different periods are pooled together. We see in this case inline image accounts for as much as nearly 40% of the total variation in inline image. Since when the periods are pooled together inline image varies both across regions and across periods, we can conclude that inline image plays a more important role in accounting for variation in inline image over time than across regions.

Table 1.  Variance decomposition of output growth
Periodinline imageNo. of Obs
inline imageinline imageinline imageinline image
1980–19840.75340.24660.14750.099127
1985–19890.93100.0690−0.0036 0.072631
1990–19940.90830.09170.01330.078431
1995–19990.83700.16300.05870.104331
2000–20040.84370.15630.01610.140231
Pooled0.53810.46190.37860.0833151 

Figures 1 and 2 plot the values of inline image, inline image, inline image and inline image for each of the 31 regions for the periods 1990–1994 and 1995–1999. One prominent feature of this region-by-region decomposition, as we can see, is that inline image has exhibited more cross-region variation than inline image in both periods.

Figure 1.

The regions are arranged on the X-axis in ascending order in terms of initial per worker output in 1989. They are Guizhou, Gansu, Tibet, Anhui, Sichuan, Chongqing, Yunnan, Jiangxi, Hunan, Shaanxi, Henan, Guangxi, Hubei, Qinghai, Ningxia, Shanxi, Inner Mongolia, Hebei, Jilin, Shandong, Jiangsu, Hainan, Xinjiang, Zhejiang, Fujian, Heilongjiang, Liaoning, Guangdong, Tianjin, Beijing, and Shanghai.

Figure 2.

The regions are arranged on the X-axis in ascending order in terms of initial per worker output in 1994. They are Guizhou, Gansu, Guangxi, Sichuan, Anhui, Yunnan, Tibet, Qinghai, Henan, Shaanxi, Chongqing, Jiangxi, Hunan, Hubei, Ningxia, Shanxi, Inner Mongolia, Hebei, Shandong, Hainan, Jilin, Xinjiang, Heilongjiang, Jiangsu, Zhejiang, Liaoning, Fujian, Guangdong, Tianjin, Beijing, and Shanghai.

4.3. Sectoral labor shares, openness, and human capital

In line with the spirit of the model in Section 2, we now examine the impacts of regional openness to foreign trade and regional human capital accumulation on the changes in regional sectoral labor shares across the Chinese regions. Our regression results are summarized in Table 2.4 The explained variables are the changes in sectoral labor shares for each of the three sectors “inline image”, which is defined as inline image for inline image each denoting the agricultural, the manufacturing, and the service sector, respectively. The explanatory variables are 1) the log of the initial level of regional per worker GDP “inline image”, 2) the regional openness index “inline image”, which is constructed as the ratio of total value of foreign trade (exports plus imports, converted to RMB yuan) to regional GDP of the same year, averaged over the corresponding time span, and 3) the schooling rate “inline image”, which is meant to be a flow measure of human capital accumulation and is constructed as the number of students enrolled in secondary education divided by the working population in the same year, averaged over the corresponding time span.5

Table 2.  Regressions: sectoral labor shares, openness, and human capital accumulation
Regressions
Explained variableExplanatory variablesPeriod
1980–1984 Obs: 261985–1989 Obs: 311990–1994 Obs: 311995–1999 Obs: 312000–2004 Obs: 31Pooled Obs: 150
  1. Standard errors are in parentheses. *denotes significant at the 5% level while **denotes significant at the 10% level. For the sake of brevity, we do not report the estimated intercept term in this table. The pooled regression in the last column includes period dummy variables (four of them altogether) besides the common intercept.

inline imageln(yi,t–1)−0.0060.0060.040*0.043*0.0050.013**
 (0.024)(0.011)(0.013)(0.018)(0.017)(0.007)
Fit−0.137−0.094**−0.064*−0.037−0.006−0.026*
 (0.147)(0.047)(0.017)(0.028)(0.028)(0.013)
shit−0.279−0.272−0.250−0.136−0.005−0.063
 (0.608)(0.337)(0.509)(0.602)(0.607)(0.239)
<R2>0.1420.1620.4270.2380.0030.250
inline imageln(yi,t–1)0.010−0.008−0.054*−0.050*−0.033**−0.024*
 (0.021)(0.008)(0.011)(0.014)(0.019)(0.007)
Fit0.0730.090*0.073*0.046*0.0490.041*
 (0.126)(0.031)(0.014)(0.022)(0.031)(0.012)
shit−0.0960.1570.924*0.2210.1670.149
 (0.522)(0.226)(0.437)(0.463)(0.672)(0.219)
<R2>0.0910.2460.5630.4020.1030.312
inline imageln(yi,t–1)−0.0040.0030.0140.0070.028**0.011*
 (0.008)(0.007)(0.012)(0.010)(0.015)(0.005)
Fit0.0640.004−0.009−0.010−0.042**−0.015**
 (0.051)(0.029)(0.015)(0.016)(0.024)(0.008)
shit0.375**0.115−0.673−0.086−0.162−0.086
 (0.210)(0.211)(0.475)(0.330)(0.514)(0.155)
<R2>0.2190.0360.0740.0220.1260.170

For the agricultural and manufacturing sectors, the estimated coefficients on both inline image and inline image have the expected signs in all regressions (except one). For the agricultural sector, the estimated coefficients on both inline image and inline image are all negative, while for the manufacturing sector, they are all positive except the one on inline image for the period 1980–1984 (which is insignificantly negative). Owing to the small sample sizes, the single-period regressions do not produce very precise estimates: many of the estimated coefficients are insignificant. However, for the periods 1985–1989, 1990–1994 and 1995–1999, inline image is shown to have a significant positive effect on the expansion of the manufacturing sector.6 Specifically, for the period 1990–1994, inline image is also shown to have a significant positive effect on the shrinking of the agricultural sector. Also for this period, inline image is shown to have a significant positive effect on the expansion of the manufacturing sector. The estimated coefficients on inline image also have the expected signs. For the manufacturing sector the estimated coefficients on inline image are negative while for the agricultural sector they are positive (except for the period 1980–1984). Specifically, for the periods 1990–1994 and 1995–1999 the estimated coefficients on inline image are significant. The signs of these coefficients show that, at least in the 1990s, (initially) poorer regions tend to experience faster structural transformation (with labor moving from the agricultural to the manufacturing sector), which in turn suggests that structural transformation had a convergence effect—it worked to narrow the income gap between rich and poor regions in China during the 1990s. The last regression in Table 2 pools the five different periods together. Results of this pooled regression reinforce two major findings above: 1) regional openness facilitates structural transformation in terms of labor moving from the agricultural to the manufacturing sector, and 2) poorer regions tend to experience a faster process of such structural transformation, which in turn contributes to convergence across different regions in China.

4.4. Structural change and its effects on convergence

The study of convergence across different Chinese regions has important policy implications as it is directly related to the issue of interregional income inequality in China. We now move on to examine the convergence effects of structural shocks and structural transformation. To do this, we run regressions of inline image as well as its different components on the explanatory variables inline image, inline image and inline image. Regression results are summarized in Table 3. By construction (equations (11) and (12)), the estimated coefficients on the three explanatory variables inline image, inline image and inline image in the regressions of inline image are the sums of the corresponding estimated coefficients in the regressions of inline image and inline image, and the estimated coefficients on the three explanatory variables in the regressions of inline image are the sums of the corresponding estimated coefficients in the regressions of inline image and inline image.

Table 3.  Regressions: structural change, openness, and human capital accumulation
Regressions
Explained variableExplanatory variablesPeriod
1980–1984 Obs: 261985–1989 Obs: 311990–1994 Obs: 311995–1999 Obs: 312000–2004 Obs: 31Pooled Obs: 150
  1. Standard errors are in parentheses. *denotes significant at the 5% level while **denotes significant at the 10% level. For the sake of brevity, we do not report the estimated intercept term in this table. The pooled regression in the last column includes period dummy variables (four of them altogether) besides the common intercept.

inline imageln(yi,t–1)−0.154*−0.185*0.0380.078−0.039−0.054*
 (0.068)(0.046)(0.082)(0.057)(0.046)(0.027)
Fit0.3080.861*0.289*−0.0370.0290.156*
 (0.420)(0.193)(0.106)(0.088)(0.075)(0.048)
shit1.3284.053*2.590−1.2341.5251.481**
 (1.734)(1.384)(3.267)(1.853)(1.608)(0.892)
< R2 >0.2640.4760.3480.0810.0470.641
inline imageln(yi,t–1)−0.092−0.162*0.1060.108*−0.015−0.018
 (0.068)(0.050)(0.072)(0.051)(0.049)(0.026)
Fit0.1520.730*0.212*−0.093−0.0020.099*
 (0.419)(0.208)(0.094)(0.079)(0.080)(0.047)
shit1.4573.505*2.228−1.4002.0581.528**
 (1.728)(1.494)(2.874)(1.673)(1.727)(0.865)
< R2 >0.1070.365)0.4250.1550.0510.211
inline imageln(yi,t–1)−0.061−0.023−0.068*−0.030−0.023−0.036*
 (0.040)(0.019)(0.019)(0.026)(0.032)(0.011)
Fit0.1560.1310.077*0.0560.0310.057*
 (0.249)(0.079)(0.024)(0.041)(0.051)(0.020)
shit−0.1290.5480.3610.165−0.533−0.046
 (1.026)(0.569)(0.747)(0.858)(1.109)(0.373)
< R2 >0.1920.1020.4350.0750.0350.823
inline imageln(yi,t–1)−0.038*0.0020.015*0.018*−0.002−0.004
 (0.012)(0.003)(0.005)(0.009)(0.004)(0.004)
Fit−0.0990.002−0.0080.0020.0070.012
 (0.072)(0.011)(0.006)(0.013)(0.006)(0.007)
shit−0.642*−0.0230.381**0.3510.306*0.059
 (0.298)(0.076)(0.199)(0.276)(0.132)(0.139)
< R2 >0.7760.0450.5520.4970.2100.962
inline imageln(yi,t–1)−0.024−0.025−0.083*−0.048**−0.021−0.032*
 (0.043)(0.018)(0.018)(0.025)(0.032)(0.011)
Fit0.2550.129**0.085*0.0540.0250.045*
 (0.263)(0.075)(0.023)(0.038)(0.052)(0.021)
shit0.5120.571−0.020−0.185−0.839−0.106
 (1.086)(0.541)(0.713)(0.810)(1.119)(0.382)
<R2>0.0440.1110.5850.1510.0470.259

Again, owing to small sample sizes, the single-period regressions in Table 3 do not produce very precise estimates. Nevertheless, we can still draw useful conclusions from those estimates that are significant. For the regressions of inline image, those estimated coefficients that are significant all have the expected signs. Of all the significant estimates, the coefficients on inline image and inline image are positive, suggesting positive partial effects of the two variables on inline image once inline image is controlled for—a result in conformity with the central idea of the model in Section 2. The coefficients on inline image are negative, suggesting (conditional) convergence in income levels across different regions, at least during the 1980s. The pooled regression produces results that reinforce the two findings above: 1) both regional openness and regional human capital accumulation facilitates regional labor productivity growth, and 2) there exists conditional convergence in income levels across different regions in China (at least during the 1980s).

For the single-period regressions of inline image, the estimated coefficients on inline image are all negative, though most of them are not significant. This negative sign suggests that structural change (structural shocks plus structural transformation) contributes to interregional convergence in income levels, that is, structural change works to narrow the income gap between rich and poor regions in China, at least during 1990–1994, where the estimate is significant. The estimated coefficients on inline image are all positive, but only the one for the period 1990–1994 is significant. This positive sign suggests that regional openness facilitates regional structural change, at least during the period 1990–1994. The pooled regression produces results that reinforce the two findings above: 1) regional structural change has a convergence effect, and 2) regional openness facilitates regional structural change.

For the regressions of inline image, we get results similar to those for the regression of inline image: the estimated coefficients on inline image are all negative while those on inline image are all positive. Accordingly, we have two similar conclusions: 1) regional structural transformation has a convergence effect, and 2) regional openness facilitates regional structural transformation. As to the regressions of inline image, most estimates are insignificant. However, one result stands out: the estimated coefficients on inline image are significantly positive for the periods 1990–1994 and 1995–1999. This result shows that during the 1990s, structural shocks have a divergence effect, i.e. it works to widen the income gap between rich and poor regions in China. In other words, the two types of structural change—structural shocks and structural transformation—have opposite effects on China's interregional convergence during the 1990s, though the combined effect (i.e. that of overall structural change) is shown to be a convergence effect.

4.5. Explaining the relative sizes of growth components

One issue of interest is how the relative sizes of the different growth components change in response to a certain set of factors. In order to examine this, we use the shares of the growth components as the dependent variables and regress each of them on a set of explanatory variables. The pooled OLS and the fixed-effects (FE) panel data methods are applied to our four-year-span panel data.7 The regression results are summarized in Table 4. By construction again, the estimated coefficients on the explanatory variables in the regressions of (inline image) are the sums of the corresponding estimated coefficients in the regressions of (inline image) and (inline image). In order to take account of secular differences over time, we have included four period dummy variables D89, D94, D99 and D04 respectively for the four periods 1985–1989, 1990–1994, 1995–1999 and 2000–2004. Obviously, the major difference between the OLS regressions and the FE regressions in Table 4 is that in the latter regressions we have controlled for not only the set of explanatory variables listed in the table, but also the unobserved time-constant region-specific effects.

Table 4.  Regressions: explaining the relative sizes of output growth components
Explanatory VariablesDependent variable: (inline image/inline image), (inline image/inline image), (inline image/inline image)
OLS (1) Obs: 151OLS (2) Obs: 150FE (1) Obs: 151FE (2) Obs: 150
  1. Standard errors are not reported in this table for the sake of brevity. The first, second, and third row for each explanatory variable corresponds to the dependent variable inline image/inline image, inline image/inline image, and inline image/inline image, respectively. * denotes significant at the 5% level while ** denotes significant at the 10% level. The estimated common intercept term is not reported in this table. The goodness-of-fit reported for the FE regressions is the within R2.

ln(yi,t–1)−0.0074 0.1072* 0.6812* 1.0125*
0.0122  0.1328*0.6845*0.9675*
0.0196 −0.0255 0.0033  0.0451
Fit −0.1638**−0.4793*
  −0.2089* 0.3861*
   0.0451 0.0932 
shit−5.5928*−5.8296*
 5.0546* 5.8049*
 0.5382  0.0248 
D89−0.1085**−0.1598*−0.4655*−0.6184*
0.00350.0566 0.3486*0.4820*
0.1050*0.1032*0.1169*0.1364*
D94−0.1720*−0.2933*−0.9509*−1.2708*
0.1135**0.2333*0.8706*1.1520*
0.05840.0601*0.0804 0.1188 
D99−0.1197−0.2972*−1.5687*−2.1525*
0.0049  −0.1820**1.4064*1.9052*
0.1246*0.1153  0.1623 0.2474 
D04−0.0762  −0.2241**−1.8693*−2.5132*
0.07620.2418*1.8235*2.3630*
0.00000.0177*0.0459 0.1502 
< R2 > 0.0657  0.1311   0.2474  0.3431 
0.05210.1343 0.2738 0.3630 
0.2016 0.20490.2665 0.2703

For the regressions of (inline image), we get one insignificantly negative and three significantly positive estimated coefficients on inline image. This result basically indicates that controlling for the explanatory variables, a higher initial level of regional labor productivity is associated with a larger share of contribution of structural change to regional labor productivity growth. In other words, a Chinese region will generally rely more and more on structural change for its overall labor productivity growth as the regional economy progresses. The estimated coefficients on inline image and inline image are all (significantly) negative. This result shows that the “residual” growth inline image increases by a larger fraction than inline image does in response to an increase in inline image or in inline image. Such a result is understandable because regional openness and regional human capital accumulation are more directly associated with “region-specific” productivity, whose growth effect is captured in inline image, but not in inline image. For the regressions of (inline image), the estimated coefficients on inline image, inline image and inline image are quite close to their counterparts in the regressions of (inline image). This result is also seen from the insignificant estimated coefficients on inline image, inline image and inline image in the regressions of (inline image), where the insignificance of these coefficients suggests that each of the explanatory variables inline image, inline image and inline image, ceteris paribus, basically does not affect the ratio of inline image to inline image, though both inline image and inline image are affected by those explanatory variables. In other words, the share of contribution of structural transformation to regional labor productivity growth is not shown to be diminishing with ceteris paribus increases in regional initial labor productivity, regional openness, or regional human capital accumulation (whereas the share of contribution of structural shocks to regional labor productivity growth has been shown to be decreasing in regional openness and regional human capital accumulation and increasing in regional initial labor productivity, ceteris paribus).

The estimated coefficients on the period dummy variables D89, D94, D99 and D04 are telling too. They show how the relative sizes of the different growth components vary across different time periods—an issue that should be related to the changing policy environment across different time periods. From our regression results, there is some evidence that regional structural change tends to have a decreasing share of contribution to regional labor productivity growth over time. In other words, over time, it seems that regional labor productivity growth tends to rely more on region-specific factors such as regional technology improvement or regional capital accumulation, and it becomes harder (ceteris paribus) to take advantage of structural change in achieving regional labor productivity growth. This is the reason why regional openness and regional human capital accumulation are important in fueling continued growth in regional labor productivity.

4.6. Robustness of our empirical results

All of our regressions above are based on our division of the entire sample period 1980–2004 into short five-calendar-year spans 1980–1984, 1985–1989, 1990–1994, 1995–1999, and 2000–2004. The application of this five-calendar-year partition in this paper imitates that of Jiang (2010). An important issue is whether different temporal partitions of the sample period will significantly affect our major empirical results above. Compared with the analyses of Jiang (2010) and this paper, another related regional growth study of China, that of Li and Huang (2006), has used longer time spans of 12 years in addition to the five-year spans. To check the robustness of our regression results, we re-run the regressions in Tables 2, 3 and 4 but with the temporal partition changed to 1980–1992 and 1992–2004. That is, we now re-run the regressions in the tables concerning two longer time spans 1980–1992 and 1992–2004.

It turns out that although this change in the time partition alters the estimated magnitudes of the partial effects of the explanatory variables in varying degrees, it does not alter the major results we have obtained above. It is shown that, at least during 1992–2004, regional openness facilitates structural transformation with labor moving from the agricultural to the manufacturing sector, and poorer regions tend to experience a faster process of such structural transformation, which contributes to convergence across the different Chinese regions. In addition, structural shocks and structural transformation are shown to have opposite effects on China's interregional convergence during 1992–2004, though the combined effect is a convergence effect. Moreover, regional structural change is shown to have a smaller share of contribution to regional labor productivity growth during 1992–2004 than during 1980–1992. All of these results are completely compatible with our prior results obtained within the five-year-span setup.

Another aspect of our robustness checks is concerned with the possibility of interregional labor flows across the Chinese regions. For example, agricultural labor in backward hinterland regions can move to urban sectors in rich coastal regions. Thus our empirical analysis above may have mingled the contribution of cross-region spatial change with that of within-region structural change in the inland and coastal regions of China. In order to partly address this problem, we re-run the regressions in Tables 2, 3 and 4 but with the four regions Shanghai, Beijing, Tianjin, and Guangdong excluded from our sample. By dropping from our sample these four regions, which are supposed to be the four main recipient regions of cross-region labor inflows, we hope to be able to mitigate the problem caused by interregional labor mobility. It turns out that the exclusion of the four regions from our sample alters most of our prior results in Tables 2, 3 and 4 only insignificantly, and therefore does not have an important bearing on the major conclusions of this paper.

5. Concluding remarks

This paper has empirically investigated the patterns, causes, and implications of China's structural change (structural shocks and structural transformation) and its contributions to China's regional growth and interregional income convergence. Our empirical results show that at least during the 1980s, regional openness and regional human capital accumulation promotes regional labor productivity growth, and there exists conditional convergence in income levels across different regions in China. Regional structural change has a convergence effect and regional openness facilitates regional structural change. During the 1990s, the two components of overall structural change have opposite effects on China's interregional convergence: structural shocks works to widen the income gap between rich and poor regions in China while structural transformation works to narrow the gap, but the combined effect of overall structural change is a convergence effect. We also find that a Chinese region will generally rely more and more on structural change for its overall labor productivity growth as the regional economy progresses. However, over time, ceteris paribus, it tends to become harder for the region to take advantage of structural change in achieving regional labor productivity growth. Therefore, ever-increasing regional openness and/or continued regional human capital accumulation should be needed for a Chinese region to fuel its long-term labor productivity growth. In sum, the results of our empirical analysis in this paper, to a large extent, support the hypothesis underlying the model we presented in Section 2. These results have significant policy implications: if structural change promotes regional labor productivity and at the same time promotes interregional income convergence, then it is important that we properly measure this dual effect and make full use of it.

Before we end this paper, we have to point out one shortcoming that we have not been able to fully overcome in the empirical analysis of this paper. Unlike cross-country studies, the present paper is a cross-region study of the Chinese provinces, where inter-provincial labor flows are a common phenomenon. Specifically, agricultural labor in backward regions tends to move inter-regionally to urban sectors in wealthier regions. In other words, processes of regional structural change and interregional spatial change tend to take place simultaneously. If this is the case, then our empirical analysis in this paper may have been prone to misattribute the contribution of cross-region spatial change to within-region structural change in the inland and coastal regions. Unfortunately, given the scope of the analysis in this paper and owing to lack of data, we are currently not able to address this issue very satisfactorily within the framework of the present paper. However, a further study toward this direction is indeed on our agenda—we hope to be able to seriously tackle issues of interregional labor mobility and interregional spatial change in China in our next study on China's regional growth and development.

Footnotes

  • 1

    In this paper, as will become clear in later sections, we take “structural change” as including both structural shocks and structural transformation, whose formal definitions will be given in Section 3.

  • 2

    The process of such a decomposition exercise is also found in Li and Huang (2006) and Jiang (2010), albeit in different notations.

  • 3

    As will become clearer in the next paragraph (and in Table 1), in this analysis we perform the decomposition only along the cross-section dimension (except for the last row of Table 1).

  • 4

    In the regressions in this and the following tables, the usual standard errors are calculated and used for drawing statistical inferences. For all the regressions in this paper, it can be shown that the alternative use of the heteroskedasticity-robust standard errors (not reported in the tables) does not alter any of our important results.

  • 5

    Needless to say, this schooling rate is a rather coarse measure of human capital formation. The general idea behind this measure is that the variation in the fraction of the population devoted to formal education reflects the variation in investment in human capital. Our choice of a feasible flow or stock measure of human capital formation is severely restricted by data unavailability.

  • 6

    All at the 5% significance level if not otherwise stated.

  • 7

    Generally, a dynamic panel data regression would render the OLS estimator (and the random effects estimator) biased and inconsistent. However, we argue that, for the analysis in the current paper, it is sufficient to include the fixed effects (FE) regressions alongside with the OLS regressions in Table 4. This is because we believe that the FE estimator is permissible in the present case. The presence of a lagged dependent variable makes the FE estimator inconsistent when asymptotics are considered in the direction of N→∞. However, Amemiya (1967) shows that when the asymptotic properties are considered in the direction of T→∞, the FE estimator proves to be consistent and asymptotically equivalent to the Maximum Likelihood Estimator (MLE). In a cross-country growth study, Islam (1995) used both this FE estimator and the Minimum Distance (MD) estimator proposed by Chamberlain (1982), but found that there were no significant differences between results of the two estimators. This further justifies the use of the FE estimator in such cases.

Ancillary