Total Factor Productivity Growth in China’s Agricultural Sector[1]

Po-Chi Chen

Dept. of International Trade, ChungHuaUniversity, Hsinchu, Taiwan

Ming-Miin Yu

Dept. of Transportation Technology, NationalTaiwanOceanUniversity, Keelung, Taiwan

Ching-Cheng Chang

Institute of Economics, Academia Sinica, Taipei, Taiwan.

Shih-Hsun Hsu

Dept. of Agricultural Economics, NationalTaiwanUniversity, Taipei, Taiwan,

Abstract

A panel dataset of 29 provinces in China is used to analyze the productivity growth in China’s agricultural sector over the period 1990-2003. We first compute the output-oriented Malmquist productivity indexes and their decomposition using a sequential data envelopment analysis approach. The results indicate that the major source of TFP growth is technical progress and that the regional disparities in TFP growth worsen over time. The second-stage regression results show that rural tax reform, increased spending on rural education, and an expansion in R&D and infrastructure investment should be the policy priorities for improving efficiency.

Keywords: Total factor productivity, agriculture, China, Malmquist productivity indexes

JEL Classification: O13, O33, O47

Total Factor Productivity Growth in China’s Agricultural Sector

1.Introduction

Since the implementation of China’s economic reform in 1979, Chinese agricultural productivity has become a popular topic among researchers (McMillan et al., 1989; Fan, 1991; Lin, 1992; Wen, 1993; Kalirajan et al., 1996; Mao and Koo, 1997; Xu, 1999; Wu et al., 2001; Fan and Zhang, 2002; Mead, 2003; Fan et al., 2004). The measurement of Chinese agricultural productivity is important in two major respects. First, the reform in 1979 gave Chinaa chance to change the basis of its rural economic institution from the commune system to the household responsibility system (HRS), which was referred to as the individual household-based farming system. It was a dramatic institutional change. After the rural reforms were implemented over the 1979-1984 period, Chinese agriculture as a whole grew at a rate of 7.8 percentper annum. In contrast to the average growth rate of 2.8 percent per annumover the period from 1952-1978, this was a remarkable achievement. Thus, many economists were interested in engaging in comparative studies to evaluate the impact of this institutional reform on Chinese agricultural productivity. Second, Chinaneeds to feed almost one quarter of the global population with only 7 percent of the world’s arable land (Liu and Zhuang, 2000). If China cannot provide enough food for its people and has to rely on imports, thiscould have a serious impact on the global grain market. Therefore, the analysis of whether China has the ability to supply enough food by itself is an important issue that warrants careful research.

At the present time, as we review these two reasons again, the importance of the former has gradually disappearedsince the impact of the rural reforms has diminished in recent years. For example, Lin(1992) estimated that the implementation of HRS constituted about 47 percent of the agricultural output growth during 1978-1984. However, its contribution went down to 0percent during the 1984-1987 period. Fan, et al. (2004) further indicated that while the rural reforms accounted for more than 60 percent of the total growth of Chinese agricultural production over the 1978-1984 period, its impact on the growth of agricultural productivity was not significant during 1985-2000.

Nevertheless, the latter problem as to whether China’s agricultural production is sustainable remains an issue of concern. On the supply side, the annual growth rate of agricultural output decreased from 8 percent in 1991-1998 to 4.3 percent in 1998-2003. On the demand side, with the growing population and rapid economic development, China’s demand for food is expected to increase continuously (Zhu, 2004). It has been estimated that China’s population will increase from 1.34 billion in 2005 to 1.38 billion in 2010, which reflects an increase of about 10 million people over a 5-year period (Tian, 2004). Some of the agricultural production resources, such as cultivated land, irrigated water and the rural agricultural labor force, are decreasing as the economy further develops (Mao, 2000; Tian, 2004). It can be seen that,between 1990 and 2003, China’s rural labor force declined by 0.5 percent annually. The total cultivated area also decreased from 1.95 billion acresin 1996 to 1.84 billion acres in 2004, and the average cultivated area per farm was only 1.41 acres, which was less than one third of the world average. Besides these trends, the marginal productivity of fertilizer, which was the input with the most important contribution to the growth of output in the 1980s[2], appears to have been diminishing. On average, instead of producing 11.56 kg of crops as it did over the period 1981-85, one more kilogram of fertilizer used could only produce an additional 2.03 kgof crops in 1991-95. It also goes without saying that, along with the rapid development of the industrial sector, the added stress on the environment (e.g., air pollution, acid rain and irrigated water pollution)has significantly threatened agricultural performance (Mao, 2000; Tian, 2004). Following China’s accession to the WTO, the traditional yield-enhancing measures such as price support and input subsidies have conflicted with the WTO principles and will thus have to be abolished or significantly curtailed (Zhu, 2004). As a result, the trade status of agricultural commodities in China switched from a surplus to a deficit in 2004, something that had nothappened since 1984 (Chen, 2005).

In general, there are two major factors that contribute to the growth of output, namely, input growth and the change in total factor productivity (TFP). Faced with declining resources and the other problems mentioned above, it is apparent that the only way for China to raise its agricultural output is to increase TFP. Therefore, it is very important to identify the factors that can promote its agricultural productivity.Recently, the Malmquist index has gained considerable popularity in the measurement of TFP since Färe, et al. (1994) applied the data envelopment analysis (DEA) approach to calculate the distance functions that make up the Malmquist index. The reason for the index’s increasing popularity is that the Malmquist productivity-change index relies exclusively on the quantity of information, requiring neither price information nor a behavioral assumption in its construction[3]. Most importantly, it allows for the further decomposition of TFP growth into changes in efficiency and changes in technology. This decomposition can help us measure the sources of changes in productivity and is important for facilitating a multilateral comparison that may help explain and characterize the differences and similarities in growth patterns for different regions (Chang and Luh, 2000). Furthermore, from the policy point of view, it is important to know whether technological progress has stagnated over time or whether the given technology has been used in such a way as to realize its full potential (Kalirajan et al., 1996). Because technical advances and efficiency change constitute different sources of TFP growth, different policies may be required to address them. Therefore, the decomposition will be useful for policy-makers.

While there have been manystudies that have analyzed Chinese agricultural TFP growth, we have not yet found any study which has both calculated Chinese agricultural TFP and its components and at the same time quantified the major determinants of TFP growth. For example, McMillan, et al. (1989), Wen (1993), Wu (1995), Kalirajan et al. (1996), Mao and Koo (1997)[4], Xu (1999), Wu et al. (2001), Fan and Zhang (2002), and Carter et al. (2003) engaged in measuring the changes in TFP using either a parametric approach, nonparametric approach or the Törnqvist index. However, they did not quantify the determinants of TFP change and theperiods they studied weremostly before 1997. On the other hand, Lin(1992)[5], Fan and Pardey (1997)[6], Jin et al. (2002)[7], Fan et al. (2004)[8] and Zhu (2004)[9] concentrated on explaining the sources of output growth, TFP change or labor productivity growth in Chinese agriculture. However, they did notdifferentiate between changes in technology and changes in efficiency.

The main purpose of this study is to measure Chinese agricultural TFPover the last decade using the Malmquist index and to investigate the factors that might explain the shift in technology and relative efficiency. The data used in this study consist of panel data for 29 provinces in mainland China over the period 1990-2003. The sequential data envelopment analysis (sequential DEA) approach proposed by Tulkens and Eeckaut (1995) is also used to rule out the possibility of technical regress. Then, maximum likelihood estimation methodis applied to identify the major determinants of TFP growth and its components. To be specific, the role of government policies, investment in infrastructure, and education in the process of TFP growth are investigated.

The remainder of the paper is organized as follows. The next section briefly describes the methodology used to measure efficiency and productivity under sequential technology. Section 3 describes the dataset, and Section 4 measures TFP growth using the Malmquist (ML) index approach. Regression results on the major determinants are presented in Section 5 followed by concluding remarks in the final section.

2. The Sequential TechnologiesandMalmquist index

The underlying assumption when applying DEA to calculate the ML productivity index is that one can construct a best practice frontier in each time period as a reference technology. As mentioned by Nin et al. (2003), there are two ways of constructing the frontier of the production set. One is called the contemporaneous production set that is commonly used in many studies, and the other is the sequential production set.

According to Tulkens and Eeckaut (1995), both contemporaneous and sequential technologies could be specified as follows. For the contemporaneous production set,it is assumed that for each time period t= 1, 2, …, T, and denote, respectively, an input vector and an output vector for period t (t=1,2,…, T). The set of production possibilities is given by the closed set,

,(1)

where the technology is assumed to have the standard properties, such as convexity and strong disposability, as described in Färe, et al (1994). The output sets are defined in terms of as:

.(2)

With this definition, successive production sets are essentially unrelated to each other.

The sequential production set, on the other hand, assumes that there is a certain form of dependence between the production sets across time. This dependence stems from the assumption that “production units can always do what they did before in the production process.” Thus, the construction of the latest set would require information on inputs and outputs before any time period. Assuming that, for each time period t= 1, 2, …, T, the are transformed into . The set of production possibilities is given by the closed set,

with s = 0, 1, 2, …, t-1. (3)

The superscript in denotes sequential technology and the output sets defined in terms of are:

. (4)

These production sets state that the input–output mix used in previous years is always available and is part of the technology in period t.

In this study, we adopt the output-oriented ML productivity index specified by Färe et al. (1994). In order to calculate the ML usingthe sequential DEA approach, thedefinition of the output distance function for each time period thas to be modified as follows:

. (5)

This distance function still represents the smallest factor,, by which an output vector is deflated so that it can be produced with a given input vector under period t’stechnology. However, this reference technology is sequential instead of contemporaneous.

According toFäre et al. (1994), the output-oriented ML index takes the following form:

. (6)

The four output distance functions in eq. (6) are defined in eq. (5). In order to calculate this productivity index, we need to compute these four distance functions using sequential DEA. For each firm, the four distance functions can be calculated as solutions to the following linear programming (LP) problem under the assumption of constant returns to scale (CRS):

subject to

m = 1, …, M

n = 1, …, N

k = 1, …, K, and s = 1, …, t+I, (7)

where is calculated with (i, j)=(0, 0),

is calculated with (i, j)=(1, 1),

is calculated with (i, j)=(0, 1),

is calculated with (i, j)=(1, 0).

The subscript c denotes the CRS benchmark technology, while K, N, M,T represent, respectively, the total number of firms, inputs, outputs, and time periods in the sample, denotes a scalar of the proportional expansion in output for a given input vector, and is an intensity variable indicating at what intensity production unit k may be employed in production. Instead of usinga one-year observation to construct its own frontier, the LP problem presented above for the sequential DEA approach shows that the frontier for each year is constructed on the basis of all observations generated up to that year, i.e. the first frontier consists of the sample for the year 1990, while the last frontier covers all observations for the period 1990-2003 (Hjalmarsson et al., 1996).

The ML formula in (6) can be decomposed into two components: technical change and efficiency change (Färe et al., 1994). The efficiency change can further be decomposed into pure efficiency change and scale efficiency change. The “technical change” component measures the shift in the frontier over time and can be interpreted as providing evidence of innovation for the province considered. The “pure efficiency change” component measures the extent to which observed production is moving toward (or away from) the frontier, which is constructed by the best practice provinces based on the variable returns to scales (VRS) technology. The pure efficiency change component, therefore, captures the performance relative to the best practice in the sample and can be interpreted as the catching-up effect. The “scale efficiency” in a given period captures the deviations between the VRS technology and the CRS technology at observed input levels. So eq. (6) can be rewritten as follows:

, (8)

whereTECH = ,

EFFI = ,

PUREFF = , and

SCAL = .

The subscript v denotes the VRS benchmark technology. In addition, the LP problems for VRS technology can be solved by adding the following constraint to the LP model specified in (6):

(9)

3. Dataand Variable Specification

This section presents the definitions of inputs and outputs and the dataset used in this study. The sample consists of 29 provinces, autonomous regions and municipalities in mainland China for fourteen consecutive years, 1990-2003. Since Chongqing was once part of Sichuan province and its data was unavailable until 1997, the data forChongqing from 1997 to 2003 are added to those of Sichuan province in this study. Besides, data for Hainan province are not included in this sample. All the data are obtained from various issues of the China Statistical Yearbook, China’s Agricultural Yearbook and China’s Rural Statistical Yearbook.

The output variable used in our empirical analysis is the value-added of production in farming, forestry, animal husbandry and fisheries at 1990 constant prices. Please refer to Appendix I for the procedures to deflate these value-added statistics. Most previous studies on China’s agricultural productivity have adopted the gross value of agricultural output (GVAO) as the output variable, except for Mao and Koo (1997). However, the GVAO not only includes the value created by primary inputs, but also the value created by intermediate inputs. To avoid the error of double counting, the values of intermediate inputs, e.g., seed, breeding animals, fertilizer, and other materials, are subtracted from the GVAO.

Fourprimary inputs arespecified: labor, land, machinery, and draft animals and they are all measured in real terms. Labor is measured as the number of rural workers at the year-end. Land is defined as the sown area because the provincial-level data for the cultivated area has not been available since 1998[10]. The sown areacan reflect the actual utilization condition of the cultivated land in China. Machinery refers to the total power of farm machinery. Draft animals for cropping activities and rural transportations are measured in the middle of the year in terms of head counts. More detailed explanations are also reported in Appendix I.[11]

Table 1 lists the sample means, standard deviations and annual growth rates of the input and output variables for our sample period 1990-2003. Following Mao and Koo (1997)[12], we divide the provinces into twocategories:those characterized by advanced technology and those by low technology. The advanced-technology provinces are mostly located in the coastal region, while the low technology provinces are mostly located in the central or western regions.[13]

It is shown that Chinese agriculture in general has the tendency to use more machinery and less labor to generate output. On the other hand, if we compare the geometric meansof the growth rates for these two categories, it can be found that the output growth was higher and that input growth was lower in the advanced-technology provinces than in the low technology ones. This implies that the advanced-technology provinces should perform better in terms of TFP growth than the low technology provinces. Moreover, from the means of the input and output variables, it is very obvious that the advanced-technology provinces generate a larger amount of output with more machinery and less labor, land and draft animals than the low technology provinces do. In other words, the production structures for these two categories are different.

Table 2 presents the average input ratios for each province/region over the 1990-2003 period. It is also shown that the ratio of machinery to labor has the highest growth of all of them. By contrast, the ratios of draft animals per person decline in nearly half of the 29 provinces/regions. These findings indicate that the input usage pattern has been gradually changing from being draft-animal-intensive to machinery-intensive.

4. Decomposition Results

Therural reforms carried out in Chinaproceeded in two phases. The first one consisted of the reforms initiated in the late 1970’sthat focused primarily on decentralizing the system of agricultural production. The second phase liberalizedboth the factor and output markets beginning in 1985 (Fan and Pardey, 1997). Regarding the degree of agricultural market liberalization, Dai (2004) calculated a comprehensive index and hefurther divided his study period into four sub-periods, namely, 1979-1986, 1986-1996, 1996-1999, and 1999-present. During the1979-1986 period, the initial liberalization phase, the procurement system was transformed from one that was mandatory to one that was voluntary or contract-based. From 1986-1996, the liberalization continued. However, the liberalization was impeded in 1995. According to Lin (1997), the Chinese government resorted again to administrative intervention in grain production and marketing due to a sharp increase in grain prices led by a large shortage in supply. From 1999 onwards, these administrative restrictions were lifted and the free market mechanism restored. Thus the degree of market liberalization increased significantly during the final period.