Impact of Technological Progress and Industrial Structure Distortion on Energy Intensity in China

3.1 ISDs and Their Measurements

The above theoretical models suggest that, in a market economy with free entry and exit of labors and perfect competition, the quest to maximize the interests of economic entities will ensure the efficient allocation of productive resources such as labor for various purposes, resulting in equal labor productivity across sectors. However, when the free flow of factors is impeded, intersectoral differences in labor productivity emerge arise, resulting in distortions and imperfections in the economy. So, how should we measure the distortion in the economy? According to the studies of Nunn and Qian (2011) and Ando and Nassar (2017), equilibrium, under a free flow of factors, requires that labor productivity be equal across sectors, with the result that the shares of output and employment are necessarily equal in any sector. Based on the finding, Ando and Nassar (2017) proposed a method to measure ISDs using the Euclidean distance between employment and output shares.

In supposing a country has N sectors, we let VA_i and L_i represent the added value and employment of sector i, respectively. The number of employed persons L_i refers to the total one of employers, employees and self-employed persons in sector i. Ando and Nassar (2017) defined the Euclidean distance between added value and employment shares as:

d_i is the distance between shares of employment and output for sector i, and d is the Euclidean distance between shares of added value and employment in the economy as a whole. Obviously, if the distance d is equal to zero, it implies equalization of sectoral labor productivity:

P_i is the sector i’s labor productivity, while P is the labor productivity of the economy as a whole.

Under the condition that labor may freely enter and exit the labor market, people have an incentive to move from sectors with low labor productivity to those with high labor productivity, which can be expected as d→0, unless there is some barriers preventing the convergence of labor productivity. In this sense, d represents the overall distortion or ISD in an economy. The bigger d gets, the more distorted the economy turns. Also, d_i is used to measure the distortion of the sector i. d_i > 0 indicates that too many labor forces are involved in sector i, and vice versa.

In section 2, the findings of the theoretical analysis unveil {d,d_i}, an important property of the distortion index, meaning the index only reflects ISD due to market imperfections and policy distortions, and is not affected by technological progress. The results imply that the ISD index is exogenous to technological progress. It makes it possible for section 5 to identify the effects of ISD and technological progress (captured by R&D expenditure, FDI and foreign trade) on energy intensity.

3.2 Industrial Structure Distortion and Energy Intensity

Theoretically, any distortion leads to low-efficiency or zero-efficiency allocation of resources. In this sense, the distorted industrial structure necessarily increases energy intensity. The research of Nunn and Qian (2011) and Ando and Nassar (2017) demonstrated that the distorted industrial structure means the heterogeneity of sectoral labor productivity. In comparison to the equalization of labor productivity across sectors, labor is over-inputted relative to the size of its output in sectors with lower labor productivity, while it is under-inputted relative to the size of its output in sectors with higher labor productivity. This misallocation of labor resources at the industrial level essentially stems from that of resources at the micro level. The misallocation of micro-level resources often arises from distortions of product and factor markets. Product market distortions are commonly found in government regulations over energy prices. Currently, prices of energy products such as electricity and oil are strictly regulated by the Chinese government, while natural gas is priced under the guidance of the government. Energy prices do not adequately reflect the scarcity of energy and the externality of the environment. As a result, there is some price distortion of energy products in China (Hang and Tu, 2007; Ju et al., 2017). Evidently, the distortion must reduce the overall energy efficiency.

Distortions in factor markets affect the relative prices of labor and capital, which can lead to misallocation of resources. Take China’s restrictions on rural labor moving into cities and towns (such as hukou restrictions) as an example. This restriction keeps a large number of labor in the agricultural sector where labor productivity is already low. The surplus labor supply depresses the relative price of labor and capital. Farmers, governed by the cost minimization motive, would substitute labor for capital. In the agricultural sector, energy and capital (agricultural machinery) are usually complementary. The substitution of labor for capital also means the relative decline of energy utilization in agricultural production. The result is a relative decline in energy intensity of the agricultural sector. In the industrial sector, regulation of labor migration raises the relative price of labor. For a given output, a producer can substitute capital for labor in order to reduce the cost of production. In most cases, such practice would lead to an increase in energy use, thus enhancing energy intensity of the industrial sector. With a much higher share of output in the industrial sector than in the agricultural sector, the PTEI derived from a weighted average (the weight is sectoral output share) of energy intensity across sectors is bound to deteriorate. Such restrictions widen the gap in labor productivity between the agricultural sector and the industrial sector, compared to the situation where there is no regulation of rural labor migration. They reflect the fact that the share of labor in the agricultural sector is much higher than its share of output, while the opposite is true in the industrial sector, where the share of labor is lower than that of output. This is exactly what we mean by ISD. Thus, distortions in micro factor markets and ISDs are essentially two sides of the same coin, with the result that leads to resource misallocation and thus reduce the efficiency of resource allocation (including energy efficiency).

4.1 Model Setup

The above theoretical analysis suggests that ISD is an important factor affecting the PTEI. However, the ISD index in a region is measured based on the methodology in the previous section, with the premise that factors move only across industries within a region. In practice, however, factors flow between regions, in addition to between various industries within a region. Especially when there are significant differences in labor productivity and the resulted differences in wage between regions, the movement of labor factors across provinces and regions is more common and may even be more important than that between industries within regions. In this scenario, the ISDs in different regions may be intrinsically linked. In other words, the ISD in a region may affect those in other regions. The spatial correlation of distortions in industrial structure also means that economic output and energy use in different regions are also spatially correlated. As such, the energy intensity in a region is influenced not only by distortions in its own industrial structure, but also by those in industrial structures of other regions. To analyze the spatial correlation of ISDs and their impact on energy intensity, we set up several spatial econometric models as follows.

ln E Y i t denotes the logarithm of energy intensity in the region i during the period t, E represents the region i’s energy consumption, and Y represents the real gross regional product in the region i. lnd_it represents the logarithm of the ISD index in the region i during the period t. Z_it is a vector consisting of the logarithms of a set of control variables. μ_i is the individual effect, which is assumed to be μ i ∼ N 0 , σ μ 2 in the case of random effects and the parameter to be estimated in the case of fixed effects. The standard assumption of ε i t  is  ε i t ∼ N 0 , σ ϵ 2 ,  and when  i ≠ j  and  t ≠ s , E(ε_itε_js) = 0 . In addition, the symbol W represents the spatial weight matrix. Limited by the availability of data, the paper constructs a spatial weight matrix based on the geographical distance between two regions. We calculated the Euclidean distance between any two provincial capital cities based on their geographical coordinates (latitude and longitude) and constructed the spatial weight matrix of inverse distance using the spmat command based on Stata software provided by Drukker et al. (2013).

The equation (15), commonly referred to as a spatial autoregressive model (SAR), was used to only consider the spatial correlation of explained variables. The equation (16), representing a spatial autocorrelation model (SAC), was utilized to consider the spatial correlation of explained variables and error terms. The equation (17), a spatial Durbin model (SDM), was applied to consider the spatial dependency of both the explained and explanatory variables.

4.2 Selection and Definition of Variables

The explained variable in this paper is the regional energy intensity, while the core explanatory variable the ISD index. The first control variable we choose is energy price. Energy price is one of the key driving forces for improving energy intensity. Many studies revealed significantly negative correlation coefficients between PTEI and energy price index (Hang and Tu, 2007; Lin and Du, 2017). The second control variable is R&D expenditure intensity. Technological change is considered to be the most effective way to improve energy efficiency. The empirical literature suggests a significant inverse relationship between technological change and energy intensity (Golder, 2011; Lin and Du, 2017). In the R&D literature, technological progress was considered to be the result of R&D investment (Young, 1998; Howitt, 1999). The increase in R&D expenditure intensity is likely to promote technological progress, which in turn could contribute to the decline in PTEI.

An important source of technological progress, in addition to independent innovation, is from technology spillovers of FDI and foreign trade. One of the most appealing ideas in modern economic growth theory is that there is technological catchup between countries: a country or region lagging behind the world’s technological frontier can close the technological gap by imitating technologies invented by countries at the technology frontier. FDI and foreign trade are the two channels through which lagged countries receive R&D spillovers from technologically advanced countries. According to the research, FDI and foreign trade can help reduce energy intensity by facilitating technology spillovers (Elliott et al., 2013). Of course, studies show that FDI and foreign trade have a negative impact on energy intensity of a country (Zheng et al., 2011; Yu, 2012). Based on this literature, we include FDI and foreign trade as two additional control variables.

The definitions of relevant variables are specified below.

(1) Energy intensity (E/Y): energy intensity is defined as the ratio of primary energy consumption to real GDP.

(2) ISD index (d): the ISD index is defined as the square root of the sum of the squared deviations of the employment share from the output share for all sectors in a region. The specific formula is equation (13) in section 2.

(3) Energy price index (P): we measure the energy price index using the fuel price index in the retail price indexes of goods by region.

(4) R&D expenditure intensity (RD): R&D expenditure intensity is defined as the ratio of a region’s internal expenditure on R&D to its GDP.

(5) FDI intensity (FDI): FDI intensity is defined as the ratio of FDI flows absorbed by a region to its GDP.

(6) Foreign trade intensity (Trade): foreign trade intensity is defined as the ratio of a region’s total exports and imports to its GDP.

4.3 Data

Data on primary energy consumption, regional GDP, FDI, total imports and exports, and fuel price index by region are taken from the CEIC China Economic Database, with regional GDP calculated in constant 1978 values. Data on the shares of employment and output in the primary, secondary (industry plus construction) and tertiary industries by region are taken from the statistical yearbook or development yearbook of each province. Data on internal expenditure on R&D by region are extracted from the China Science and Technology Statistical Yearbook in relevant years. The time window for the employment share and output share data by region ranges from 1978 to 2016, which gives us a longer time series to examine changes in ISDs. The time window for primary energy consumption, fuel price index (1986 is the base period), R&D expenditure, FDI, import and export data ranges from 1998 to 2014. Table 1 reports the statistical description of related variables.

5.1 Industrial Structure Distortion Index

We calculated the ISD index by region based on equation (13) using data on employment share and output share of the three industries in 30 provincial-level administrative regions of mainland China (excluding Xizang) from 1978 to 2016. According to equation (13), the number N of sectors equals to 3, and the ISD index d takes a value ranging from 0 to 3 The higher the value d, the more severe the structural distortion. Due to the limitation of space, we cannot give the value of the ISD index for each province (city or district) each year, but only for each province and district in 1978 and 2016 (see Figure 2). Figure 2 shows that the ISD index decreased in most provinces and regions in 2016 compared with 1978, but a few regions (e.g., Inner Mongolia, Jilin, Guizhou, and Shaanxi) experienced an increase in their ISD index. Figure 2 also indicates that, in 1978 and 2016, the regions with a lower ISD index were basically the eastern coastal provinces and regions, while those with higher ISD indexes were mostly the provinces and cities in the western region.

Figure 2

Industrial Structure Distortion Index by Region

5.2 Estimating the Impact of Industrial Structure Distortion Index on Energy Intensity

Common estimation methods for spatial econometric models are maximum likelihood (ML), quasi-maximum likelihood (QML), and instrumental variable (IV) methods (Anselin, 2008). We estimate the three spatial econometric models in the paper using the QML method. In the estimation, we consider space-specific effects to control the effects of unobservables that vary with space but not with time.

Because spatial econometric models use a complex dependency structure between sample units, changes in explanatory variables for a given unit potentially affect all other units in addition to the unit itself. In the case of this paper, the energy intensity of a province or region is affected not only by changes in its own ISD index but may also be affected by the index of ISDs in other provinces and regions. The first type of change is called the direct effect, while the second type of change is called the indirect effect. The total effect is the sum of direct and indirect effects. Table 2 gives the direct, indirect and total effects based on the estimations of models (15) ~ (17). ^[1] The first and second columns provide fixed and random effects estimates for the SAR model, the third column provides fixed effects estimates for the SAC model, and the fourth and fifth columns provide fixed and random effects estimates for the SDM model.

In the case of lnd, the core explanatory variable in this paper, the results in Table 2 indicate that its direct effect is greater than 0 in the first three regressions but not statistically significant, while that is less than 0 in the last two regressions and also not statistically significant. This implies that the ISD of a region does not have a significant direct effect on its energy intensity. The reasons for the statistically insignificant direct effects would be analyzed in the following part where we study the effects of sectoral distortions on energy intensity. In terms of indirect effects, the coefficient of the variable lnd is statistically insignificant in the SAR and SAC models, but it is 0.923 and 0.942 in the two estimates of the SDM model, respectively, which is highly statistically significant. Therefore, according to the SDM model estimation, the ISD has a significant negative impact on energy intensity through indirect effects. The total effect of the variable lnd is also insignificant in the SAR and SAC models but is significantly greater than zero in the SDM model. By considering the total effect of the SDM model, a 1% increase in the index of industrial structural distortion would ultimately increase energy intensity by 0.89%~0.92% overall.

The direct effect of the variable lnRD is highly statistically significant positive in the first three regressions, but highly statistically significant negative in the last two regressions. Its indirect effect is highly significant positive in the SAR and SAC models. While the indirect effect is negative in the SDM model, it is statistically significant in the SDM_FE estimate. The total effect is statistically significant positive in the estimates of SAR and SAC models. It is statistically insignificant in the two estimates of the SDM model, despite being negative. This result for the total effect suggests that R&D expenditure may not have reduced regional energy intensity in China. One possible reason is that, although R&D expenditure in China is growing rapidly, it is relatively inefficient. There is misallocation of public financial investment in innovation in particular (Wei et al., 2017). Also, the rapid growth of R&D expenditure reduces the resources (e.g., material capital and human capital) that would otherwise be available for producing goods and services in the context of budget constraints faced by enterprises. R&D investment has a crowding-out effect on investment in producing goods and offering services. In addition, when there is information asymmetry and government rent-seeking, government R&D incentives can lead to manipulative behavior of enterprises in R&D expenditure. Enterprises may strategically meet policy thresholds to obtain policy preferences, which ultimately reduces their R&D performance (Yang et al., 2017).

The direct effect of the variable lnFDI is negative in all five regressions, but statistically insignificant in the SAR_FE and SAC_FE estimates and statistically significant in the remaining three estimates. This unveils that, in most cases, FDI in a region helps to reduce its energy intensity. Its indirect effect is negative but statistically insignificant in the SAR and SAC models, while it is 0.309 and 0.294 in estimating fixed and random effects of the SDM model, respectively, and is highly statistically significant. The results from the SDM model show that, for a region, FDI attracted by other regions would have a significant negative impact on the energy intensity of that region. The reason may be that the growth of FDI inflows from other regions has a magnetic effect on the talent and capital of neighboring regions, thus negatively affecting the economic growth of neighboring regions. If the decline in neighboring output exceeds that in energy consumption, the energy intensity would rise. Finally, the total effect is also negative but statistically insignificant in the SAR and SAC models. The highly statistically significant estimates in the SDM model are 0.266 and 0.244, respectively, indicating that a 1% increase in FDI ultimately enhances energy intensity by 0.24%~0.27%. Therefore, although the inflow of FDI to a region may reduce its energy intensity through direct effects, the introduction of FDI does not have a positive effect on reducing its energy intensity through its technology spillover by considering the adverse effects of indirect effects.

The three effects of the foreign trade intensity variable lnTrade are highly significant negative in all regressions, suggesting that the development of foreign trade in a region can both contribute to its own reduction in energy intensity and have a positive effect on reducing energy intensity in other regions through spillover effects. Specifically, the direct effect ranges from −0.530 to −0.479 in the five regressions, indicating that a 1% increase in the share of a region’s foreign trade in GDP can reduce its energy intensity by 0.48%~0.53%. The indirect effect is also highly significant negative in the five regressions, ranging from −2.099 to −1.310. This indicates that, for a region, a 1% increase in the share of foreign trade in GDP in other regions can reduce the energy intensity of that region by 1.31%~2.10%. In the end, regarding the total effect, a 1% increase in the share of foreign trade in GDP can reduce energy intensity by 1.79%~2.63%, when other conditions are constant.

Finally, the direct effect of the energy price index variable lnP is statistically significant negative in the first three regressions but statistically significant positive in the last two regressions. However, the indirect effect is statistically significant negative in all five regressions. The total effect is also negative in all five regressions, but statistically insignificant in the last two regressions. Estimates of the total effect of the variable range from −0.696 to −0.136. When other conditions are constant, a 1% increase in energy prices can lead to an eventual reduction in energy intensity by 0.14%~0.70%.

5.3 Model Selection Test

Of the SAR, SAC, and SDM models, which would be more compatible with the sample data? First, according to the Hausman test, it is estimated that spatial fixed effects are preferable to random effects. Therefore, we perform model selection tests for three spatial fixed effect models: spatial autoregressive model for fixed effect estimation (SAR_FE), spatial Durbin model for fixed effect estimation (SDM_FE) and spatial autocorrelation model for fixed effect estimation (SAC_FE). Table 3 provides relevant test results. Since the SDM model is imbedded in the SAR model, the former can be turned into the latter when the coefficients of the spatially lagged explanatory variables are all zero. The LR test in the first row of Table 3 rejects the original hypothesis that the coefficients of the spatially lagged explanatory variables are all equal to zero. Therefore, the SDM model should be chosen through its comparison with the SAR model. Since the SDM model is not a nested form of the SAC model, we test the choice of the two models using AIC and BIC information criteria. According to the results in Table 3, the SDM_FE model has significantly smaller values of AIC and BIC than the SAC_FE model, indicating that the SDM_FE model is more compatible with sample data. In this way, section 6 summarizes the findings based on the SDM_ FE estimation.

5.4 Impact of Sectoral Distortion Index on Energy Intensity

In this part, we estimate the QML for the SDM_FE model using the sectoral distortion index instead of the overall ISD index. The sectoral distortion index is calculated using the first equation in formula (13). The sectoral distortion indexes for the primary, secondary and tertiary industries are denoted as d₁, d₂ and d₃. Since the sectoral distortion indexes for the secondary and tertiary industries are negative, this paper takes absolute and logarithmic values. Table 4 presents the direct, indirect and total effects of the QML estimations for the SDM_FE model. ^[1]

Regression (6) in Table 4 reports the effect of the sectoral distortion index of the primary industry on energy intensity. It can be seen that the direct, indirect and total effects of the sectoral distortion index of the primary industry are highly significant positive in statistics. The total effect suggests that a 1% increase in the sectoral distortion index of the primary industry ultimately leads to a 0.84% increase in energy intensity, when other conditions are constant. Although the primary industry is a small part in output and energy consumption shares, the gap between the employment and output shares of the industry significantly worsens PTEI.

Regression (7) gives the effect of the sectoral distortion index of the secondary industry on PTEI. The results revealed that the direct effect of the sectoral distortion index on energy intensity is negative and highly statistically significant, but its indirect and total effects are highly significant positive in statistics. The value of the total effect reveals that a 1% increase in the sectoral distortion index of the secondary industry can ultimately increase the PTEI by 0.28%. Regression (8) reports the direct, indirect and total effects of sectoral distortions in the tertiary industry. With respect to the total effect, the sectoral distortions of the tertiary industry ultimately have no significant effect on PTEI.

Finally, we return to the question in Table 2 that the ISD index has no direct effect on energy intensity. Since the overall ISD index is an integrated result of sectoral distortions in the primary, secondary and tertiary industries, the direct effect of overall ISDs on energy intensity should also be an integrated result of the direct effect of distortions in the three industries. As can be seen in Table 4, the direct effects of the sectoral distortion indexes of the three industries almost cancel each other out and their combined effect is close to zero. This explains why the ISD index in Table 2 has no direct effect on energy intensity.

5.5 Robustness Test

The spatial weight matrix is changed in this paper to test the robustness of the estimated results of the spatial Durbin model for fixed effect (SDM_FE) in Table 2. We replace the spatial weight matrix of inverse Euclidean distance with the Queen adjacency space weight matrix and the 5th order nearest neighbor space weight matrix. Then, we re-estimate the SDM_FE. The estimated results are reported in Table 5. In both estimates, the direct, indirect and total effects of the variable lnd are positive and highly statistically significant in most cases, indicating the robust results of SDM_FE in Table 2.