2.1 Iron and steel industry in China

In the 1980s, China's iron and steel was produced in inefficient facilities, and its industry lagged behind that in developed countries. Until 1980, because of internal and external political factors, limited technology transfer from developed countries contributed to this lag, and China was restricted to introduce aging technologies and equipment from the former Soviet Union and Eastern Europe. China's iron and steel industry was eventually modernized following economic reform and liberalization policies.

In 1980, by which time Japan had ceased production in open-hearth furnaces, 35 per cent of production in China was still from open-hearth furnaces (see figure 1). In the early 1990s, they still accounted for approximately 20 per cent of total iron and steel production. China's slow process in phasing out its open-hearth furnaces in the 1980s and early 1990s occurred for two reasons. First, converter furnace technology requires a significant monetary investment, and second, the iron and steel industry was running at full capacity to meet China's rapidly growing demand, making the rapid phasing out of open-hearth furnace production very difficult.

Figure 1.

Share of steel production by operation method.

Source: Steel Statistical Yearbook 2005 (International Iron and Steel Institute, 2005a).

Casting processes are also important in improving energy resource efficiency. The continuous casting share of the steel production in China increased rapidly during the 1990s, rising from 20 per cent in 1990 to almost 90 per cent in 1999 (see figure 2). Adapting modern production processes led to productivity and energy efficiency improvements. Energy consumption per unit of crude steel production dramatically improved when open-hearth furnaces were phased out and the continuous casting method came into general use during the 1990s.

Figure 2.

Crude steel production and continuous casting ratio.

Source: World Steel in Figures 2005 (International Iron and Steel Institute, 2005b).

An increase in the water-recycling rate during the 1990s was another important achievement resulting from technological modernization of China's iron and steel industry. Three major factors improved water usage efficiency (China Industrial Water Conservation Report (CIWCR), 2004). First, technological improvements in blast furnaces reduced water usage. Second, two new drying methods for removing furnace dust required less water usage than conventional methods and facilitated water conservation. The third factor was simplifying production processes by installing electric arc furnaces and using the continuous casting method. These improvements reduced the amount of water required, and the water-recycling rate also increased because of wastewater quality improvement. The water-recycling rate of China's iron and steel industry rose from only 57 per cent in 1980 to 75 per cent in 1990, and to 85 per cent in 1999 (CIWCR, 2004). From 1990 to 1999, the volume of fresh water inputs and industrial wastewater per ton of crude steel dropped by more than half.

Changes in policy and regulations affecting the iron and steel industry played an important role in improving the business environment underlying the technological modernization. China imposed price controls on steel product prices until 1992. Price deregulation affected the performance in China's iron and steel industry. The price of steel rapidly increased after price deregulation. Wire rod, for example, was 1,600 Yuan per ton in February 1992 before the policy change, but immediately after the change in April 1992, the price increased to 1,750 Yuan per ton. Wire rod reached 3,500 Yuan per ton in December 1992, meaning the price more than doubled within a year. Although the price of steel continued to increase during 1993, it started declining in late 1993 and bottomed out at 2,000 Yuan per ton in December 1993 (Ministry of Metallurgical Industry: Yearbook of Iron and Steel Industry of China, 1994, 1995).

These rising prices enabled firms to make large profits in a short period. However, another market reform also provided stable foundations for the iron and steel industry to ensure larger profits during the same period. Under the contract management system, which was the sales system introduced in the 1980s, the enterprise was permitted to sell steel products only to customers designated by the government (Sugimoto, Reference Sugimoto1993). However, after the price and market reforms in 1992, firms were allowed to have direct contracts with their customers. In subsequent years, this new market system gained popularity throughout China. As a result, the proportion of steel products sold in the new market system rapidly increased from 27 per cent in 1988 to 91.8 per cent in 1995 (Ministry of Metallurgical Industry: Yearbook of Iron and Steel Industry of China, 1989, 1996).

Iron and steel firms greatly increased their profits because of increasing prices and market reforms. These high returns encouraged technological advancements in iron and steel production. Higher profits enabled firms to invest in new technologies such as continuous casting equipment and basic oxygen furnaces and to build new manufacturing factories. Additionally, the demand for high valued-added steel products, such as seamless pipe and thin steel plate, also increased because of the economic development of China and export expansion, which required advanced technologies (Sugimoto, Reference Sugimoto1993). The demand push factors further enhanced the acceleration of investments into advanced technologies.

2.2 Productivity measurement

Many productivity evaluation techniques are based on the frontier efficiency to evaluate inefficiency by specifying the production frontier with the best performing observations, and measuring the distance of inefficient samples from the frontier. The empirical specification methods of frontier efficiency divide into two types: parametric and nonparametric. The parametric method uses Stochastic Frontier Analysis (SFA) with production frontiers like those of the Cobb–Douglas and Translog functions (see Schmidt (Reference Schmidt1986) for survey). The nonparametric method uses the DEA developed by Charnes et al. (Reference Charnes, Cooper and Rhodes1978), where nonparametric linear programming techniques are applied. Previous studies empirically analyzing productivity in the Chinese iron and steel industry have used both SFA and DEA. (For review of previous studies, see online Appendix available at http://journals.cambridge.org/EDE.)

Even though many articles have already used DEA and SFA, it is not simple and easy to expand these to consider pollution issues, partly because of the methodological difficulty when incorporating undesirable outputs into production measurements. To address this issue, Färe et al. (Reference Färe, Grosskopf, Lovell and Pasurka1989) developed the Hyperbolic Distance Function (HDF), which can consistently identify undesirable output as an extension of the nonparametric approaches. However, the HDF imposes the strong assumption of the same multiplier and divisor for increases in desirable output and decreases in undesirable output, while maintaining inputs constant. Chambers and Pope (Reference Chambers and Pope1996) and Chung et al. (Reference Chung, Färe and Grosskopf1997) developed the directional distance function (DDF) approach to ease the restriction. There is a growing literature on methodologies for measuring productivity efficiency with DDF incorporating environmental pollution (Hailu and Veeman, Reference Hailu and Veeman2000; Managi and Kaneko, Reference Managi and Kaneko2009; Managi et al., Reference Managi, Opaluch, Jin and Grigalunas2005; Zhou et al., Reference Zhou, Ang and Poh2008). We examine explicitly the relative importance of multiple environmental pollution such as both global and local pollution.

3.1 Directional distance function

Let x ∈ ℜ^L₊, b ∈ ℜ^R₊, y ∈ ℜ^M₊ be vectors of inputs, environmental output (or undesirable output), and market outputs (or desirable output), respectively, and then define the production technology as

(1)

We assume that the good and bad outputs are null joint; a company cannot produce desirable output without producing undesirable outputs (Shephard and Färe, Reference Shephard and Färe1974):

(2)

Before defining the directional vectors in a productivity analysis that considers undesirable outputs, either strong disposability or weak disposability needs to be assumed. The difference between the strong and weak disposability of undesirable outputs is attributed to the opportunity cost of pollution abatement (see Färe et al., Reference Färe, Grosskopf and Pasurka1986, Reference Färe, Grosskopf, Lovell and Pasurka1989). The strong disposability assumes that the undesirable output is not regulated and a firm can freely dispose it. Otherwise, regulation on the undesirable output should potentially incur the cost for the firm to dispose. This conceptual clarification was translated into modeling framework of frontier productivity analysis in such a way that weak disposability applies when an undesirable output is disposed of in proportion to the reduction of desirable outputs, holding the input vector constant (Färe and Primont Reference Färe and Primont1995). In this case, the reduction of good output is regarded as a cost of disposing the undesirable output.

Considering that the pollutants in this study are CO₂ emission and wastewater discharge, CO₂ emission is assumed under strong disposability and wastewater discharge is assumed under weak disposability, respectively, since wastewater is regulated in China. Here, the possibility that CO₂ emission reduction achieved through energy and resource saving with technological modernization, while maintaining the level of desirable outputs, is specifically treated by setting directional vectors. If resource inputs potentially causing pollution are not considered in the model, there is the possibility of bias in the productivity measurements under the strong disposability assumption. Therefore, we include energy inputs in the models and assume CO₂ emission can be reduced without changing the production level of the desirable output, but reducing energy inputs.

Weak disposability can be mathematically expressed as below (Färe et al., Reference Färe, Grosskopf, Lovell and Pasurka1989):

(3)

Under the null-joint hypothesis and weak disposability, this directional distance function can be computed for firm k by solving the following optimization problem:

(4)

(5)

(6)

(7)

(8)

where l, m, r represent types of input, desirable output, and undesirable output, respectively; x is an input matrix with dimensions L × N; y is a desirable-output matrix with dimensions M × N; and b is an undesirable-output matrix with dimensions R × N. Furthermore, g_x is the directional vector of the input matrix, g_y is the directional vector of the desirable-output matrix, g_b is the directional vector of the undesirable-output matrix, β_k is the inefficiency score of the firm k, and λ_i is the weight variable. To estimate the inefficiency score of all firms, the model needs to be applied independently to each of the N firms.

3.2 The Luenberger productivity indicator

In order to analyze changes in efficiency over time, aggregated indices such as the Malmquist Index and Luenberger Productivity Indicator have been developed (Luenberger, Reference Luenberger1992; Chambers and Pope, Reference Chambers and Pope1996; Chambers, Reference Chambers2002). They are derived from the efficiency scores of production frontier models. These productivity indices are measures of total factor productivity (TFP), when the efficiency score comes from economic production frontier models. TFP includes all categories of productivity changes and can be decomposed further to provide a better understanding of the relative importance of various components, including “Technical change” (TECHCH) and “Efficiency change” (EFFCH) (Färe et al., Reference Färe, Grosskopf, Norris and Zhang1994). “Technical change” measures shifts in the production frontier, the so-called frontier shift. “Efficiency change” measures changes in the position of a production unit relative to the frontier, the so-called catching-up factor.

We employ the Luenberger Productivity Indicator (Chambers et al., Reference Chambers, Chung and Färe1998) as a TFP measure because the LPI is believed to be more robust than the widely used Malmquist Indicator (Luenberger, Reference Luenberger1992; Chambers et al., Reference Chambers, Chung and Färe1998). Change in the LPI is further decomposed into technical change and efficiency change.

The LPI is computed with the results of the DDF model and derived as follows (Luenberger, Reference Luenberger1992; Chambers et al., Reference Chambers, Chung and Färe1998):

(9)

(10)

(11)

where, x_t represents input for year t, x_{t +1} is input for year t+1, y_t is desirable output for year t, and y_{t +1} is desirable output for year t+1. b_t represents undesirable output for year t, and b_{t +1} is undesirable output for year t+1. The LPI score is computed based on inefficiency scores derived from the different combinations of observations and frontiers between t and t+1. denotes the inefficiency score of year t based on the frontier in year t, while is the inefficiency of year t based on the frontier in year t + 1.

The LPI can be calculated as the cumulative score over time as follows:

(12)

The DDF model commonly assumes either constant return to scale (CRS) or variable return to scale (VRS). In this study, we apply CRS to avoid an infeasible calculation in time series analysis. We note that imposing CRS does not eliminate the possibility of infeasible linear programming (LP) problems, when weak disposability is imposed on bad outputs.Footnote ⁴ When we model bad outputs in LP, the potential for infeasible LP problems results from imposing weak disposability on the undesirable outputs and specifying an LP problem that uses a period t reference technology with observations from period t+1.Footnote ⁵ Based on Färe et al. (Reference Färe, Grosskopf, Norris and Zhang1994) and Färe and Grosskopf (Reference Färe and Grosskopf1996), the VRS model tends to have infeasible results more often than the CRS model when computing productivity change because the VRS has stronger restrictions for solving a linear program.Footnote ⁶

We confirmed with our models that the calculation of productivity changes under the VRS is infeasible.Footnote ⁷ Therefore, we apply the CRS model in this study.

3.3 Data

Most of our dataset comes from the China Iron and Steel Industry Fifty-Year Summary (CISIFS) and CIWCR. Data for the 27 firms covers the decade from 1990 to 1999. The selected firms account for 68 per cent of the crude steel production in China, based on fiscal year 2000. Mostly they are the larger firms in the industry (see table A1 in online Appendix available at http://journals.cambridge.org/EDE). These 27 firms account for approximately 65 per cent of the fresh water input and industrial wastewater discharge and 45 per cent of total energy consumption of the iron and steel industry in 1999.

Profits (in real value added) and crude steel production (tonnes) are both used as the desirable outputs, whereas fixed capital stock and gross wages are input data for productivity estimation. Because the existence of high volatility and extreme hikes in prices of steel products cause disturbance in, and instability of, the computation of productivity measurements, we employ both physical outputs and value of outputs as our desirable outputs. We admit that these two measures correlate when the price is stable. However, the correlation does not cause problems in computation of DDF models when the price and/or value of products are significantly changing over time in our study. However, during the 1990s, as China's industry strove to improve energy efficiency in crude steel production, it was also expanding into processing steel products and increasing the production of high-value-added products, such as steel plates and steel pipes. This manufacturing process necessitates the input of additional energy to generate the same output. Firms that followed the strategy of developing and manufacturing high-value-added products during the 1990s may display lower efficiency if the efficiency is evaluated as the ratio of total energy consumption to crude steel production. Therefore, developing new and high-value-added products may actually increase energy use.

We use gross wages to reflect the quality of engineers and managers because the number of employees or working hours does not reflect the quality of the work. We assume that the salaries of the employees managing the modern equipment would change during the technological modernization period.

Other variables used in the models are CO₂ emissions and wastewater discharge as undesirable outputs with energy consumption and fresh water consumption as inputs. Energy consumption is aggregated by inputs in physical units such as electricity, coal, coke, oil, and natural gas (CISIFS) and net calorific value coefficient (IPCC, 2006). We specially consider double counting of the energy consumption between coal consumption and the coke-making process (CISIFS) when the firm operates the coke-making process, and the induced energy consumption for purchased coke otherwise. Similarly, energy for power generation is properly treated by the national average conversion factor for electricity (China Energy Statistics). CO₂ emission data is calculated using energy consumption data and IPCC CO₂ coefficients (IPCC, 2006). All monetary values of fixed capital stocks, gross wages, and value added are deflated to fiscal year 1990 levels using the ex price indicator for the iron and steel industry in the China Statistical Yearbook 2000 (State Statistical Bureau: China Statistical Yearbook 2000).

3.4 Application of models

This study applies the DDF to four models; namely, the Market model, the Water model, the Energy model and the Joint model. Only the Market model has combinations of input and output variables different from those in the other three. In the Market model, profits and crude steel production are the desirable outputs, and capital stock and wage are the inputs. On the other hand, in the other three models, profits and crude steel production are the desirable outputs, and CO₂ emission and wastewater discharge are the undesirable outputs. Furthermore, energy and fresh water consumption are used as inputs in addition to capital stock and wages. The difference between the three models is found in the directional vector combinations as summarized in table 1.Footnote ⁸ The Water, Energy and Joint models seek to capture ESP relating to water, energy and joint effects on water and energy, whereas the Market model is set for reference.

Table 1.

Combinations of directional vector in each model for calculating inefficiency of firm k

Note: All of these variables are used as data in three ESP models. The nonzero value indicates the variable is also used as a directional vector. The value of zero indicates that the variable is not used as a directional vector but used as data. Fresh water, energy, and two undesirable output variables are not used as data in the Market model.

The directional vector specifies for inefficient firms the way to improve productivity toward the frontier production line. In the Joint, Water and Energy models, we set the directional vector as (g_y, g_x, g_b) = (y, −x, −b), while in the Market model, we set it as (g_y, g_x) = (y, −x). This type of directional vector assumes that an inefficient firm can improve productivity while increasing desirable outputs and decreasing undesirable outputs and/or inputs in proportion to the initial combination of actual inputs and outputs.

For further analysis of the results, we separate the firms by company scale and regional characteristics. First, we divided the sample into three groups by size of production: large, medium, and small. Companies that produced more than 5 million tons of crude steel annually during the 1990s are categorized as large, firms with production ranging between 1 and 5 million tons as medium, and firms with production of less than 1 million tons as small. The sample sizes for large, medium, and small are 4, 14, and 9 firms, respectively. Second, for regional characteristics, we divided the sample into north and south groups by location of the firms.Footnote ⁹ Particular attention is paid to the northern part of China, which experienced water scarcity in the 1990s, and hence we hypothesize that firms in the north use water resources more efficiently than those in the south. The northern group consists of 11 firms (2 large, 7 medium, and 2 small) located in Beijing, Gansu, Hebei, Henan, Inner Mongolia, Jilin, Liaoning, Shangdong, and Shanxi provinces. The southern group consists of 16 firms (2 large, 7 medium, and 7 small) located in Anhui, Chongqing, Guangdong, Guizhou, Hunan, Hubei, Jiangsu, Jiangxi, Shanghai, Sichuan, Yunnan, and Zhejiang provinces.

Large and medium firms are equally located in the north and the south, but more small firms are located in the south. Large firms started to set up continuous casting earlier than small and medium firms in the early 1990s, but steel production in open-hearth furnaces was still widespread even in the large firms. This can be explained by two major reasons: the rapid increase in China's domestic steel demand and the extremely high costs of replacing and updating equipment in large firms.