Sunteți pe pagina 1din 19

China Economic Review 57 (2019) 101315

Contents lists available at ScienceDirect

China Economic Review


journal homepage: www.elsevier.com/locate/chieco

Agricultural productivity evolution in China: A generalized


T
decomposition of the Luenberger-Hicks-Moorsteen productivity
indicator

Zhiyang Shena,b, Tomas Baležentisc, Gary D. Ferrierd,
a
Anhui University of Finance and Economics, 962 Caoshan Road, 233030 Bengbu, China
b
China Eximbank, 30 FuXingMenNei Street, 100031 Beijing, China
c
Lithuanian Institute of Agrarian Economics, V. Kudirkos Str. 18-2, LT-03105 Vilnius, Lithuania
d
Department of Economics, University of Arkansas, Business Building 402, Fayetteville, AR 72701, USA

A R T IC LE I N F O ABS TRA CT

Keywords: China has undergone a series of agricultural policy reforms since 1978. The measurement of the
Total Factor Productivity productivity gains and identification of the underlying drivers thereof are important facets of
Luenberger-Hicks-Moorsteen productivity policy analysis. The commonly used Total Factor Productivity (TFP) measures often lack such
indicator desirable properties as completeness or independence of the direction of the optimization (or-
Productivity decomposition
ientation). In this paper, we take a top down approach by beginning with a TFP measure and then
Agricultural technology
Agricultural policy
decomposing it into three mutually exclusive, exhaustive elements. In particular, we begin with
the additively complete Luenberger-Hicks-Moorsteen (LHM) TFP indicator that takes into ac-
count both input and output changes when measuring productivity and then additively decom-
pose it into measures of technological progress, technical efficiency change, and scale efficiency
change. We develop a generalized decomposition of the LHM TFP indicator which encompasses
both input-oriented and output-oriented changes over time. We illustrate this additively com-
plete LHM TFP indicator using agricultural data from 31 Chinese provinces over the period
1997–2015. Our empirical results show that Chinese agricultural productivity growth (3.05% per
annum) was mainly driven by technological progress (2.35% p.a.), with relatively small con-
tributions from scale efficiency change (0.65% p.a.) and technical efficiency change (0.04% p.a.).
We also found that productivity change and the relative importance of its components varied
across both time and provinces.

1. Introduction

The growth of agricultural productivity affects the well-being and structural transformation of economies in various ways. First, a
more productive agricultural sector contributes to overall economic growth and can help to alleviate poverty, especially in rural
areas. Second, increased agricultural productivity may allow the release of labor from agriculture to manufacturing and other sectors.
Third, improved agricultural productivity may lead to lower food prices, improving the diets of consumers as the lower prices
increase real incomes. Fourth, enhanced agricultural productivity would increase the availability of food, the first step in lessening
food insecurity, and reduce hunger. Finally, the environmental pressures associated with agricultural activities can also be alleviated
if a sufficient amount of income is generated (Chen, Cheng, Nikic, & Song, 2018; Song, Fisher, Wang, & Cui, 2016; Zhao et al., 2017).


Corresponding author.
E-mail addresses: tomas@laei.lt (T. Baležentis), gferrier@walton.uark.edu (G.D. Ferrier).

https://doi.org/10.1016/j.chieco.2019.101315
Received 23 October 2017; Received in revised form 27 May 2019; Accepted 10 July 2019
Available online 11 July 2019
1043-951X/ © 2019 Elsevier Inc. All rights reserved.
Z. Shen, et al. China Economic Review 57 (2019) 101315

Given that China is one of the world's largest food producers as well as food consumers, its agricultural TFP growth is very
important. While China's agricultural productivity grew rapidly following the “reform and opening up” policies of 1978, the rate of
growth slowed in recent years. But with 18.4% of the world's population and only about 8.4% of its arable land (World Bank, 2019),
rising living standards and changing food consumption patterns, and an increasing reliance on agricultural imports, China's agri-
cultural productivity impacts global food markets and the influence of its policy on productivity is very important.
Both domestic (e.g., support policies) and external (e.g., climate change) factors shape the Chinese agricultural sector. For ex-
ample, to promote farmers' income and farming viability in general, the agricultural policy in China turned from taxation to sub-
sidization from 2006 onwards (Huang, Wang, & Rozelle, 2013; Huang & Yang, 2017). Input subsidies have allowed Chinese farmers
to expand their scales of operation and acquire higher quality inputs (Huang et al., 2013); however, the support programs did not
yield structural changes and small-scale farming still prevails in China. Wang, Yamauchi, Otsuka, and Huang (2016) found that
industrialization and rising wages have led to the substitution of machines for labor (supported by the development of markets for
machine rentals and machine services). Constraints on land sales, however, limit the expansion of firm size; in Wang et al.'s (2016)
sample, mean farm size was just 6.42 mu1 in 2008, which was little changed from 2004, but substantially smaller than in 1985.
Accurate measurement of TFP growth and its drivers, at both the aggregate and disaggregated levels, is essential for designing
policy and assessing its effectiveness. Wu (2011) concluded from a meta-analysis of the literature on TFP growth in China that
empirical findings on TFP vary widely by data sample as well as by choice of method used to measure TFP growth. Our paper
contributes to the literature on Chinese agricultural TFP growth by developing an additively complete Luenberger-Hicks-Moorsteen
(LHM) productivity indicator that simultaneously accounts for both input change and output change. Ratio-based productivity
measures such as the Malmquist index are undefined when one or more of the variables used to construct them are equal or close to
zero; this is not a problem for additive productivity measures such as the LHM productivity indicator. Furthermore, the LHM in-
dicator is additively complete because it is the difference between an output aggregator and an input aggregator. An incomplete
productivity measure may be a biased measure of TFP change since it does not simultaneously account for the effects of both input
and output changes. As a complete measure of productivity based on directional distance functions that simultaneously expand
outputs and contract inputs to move observations to the frontier of technology, the LHM indicator can be separated into output
growth and input growth effects. Our second contribution is to propose a decomposition of the additively complete LHM productivity
indicator that yields three mutually exclusive and exhaustive components—technological progress, technical efficiency change, and
scale efficiency change. Our third contribution is to present the non-parametric linear programming models (data envelopment
analysis [DEA] like models) needed to compute the directional distance functions used to construct the LHM TFP indicator and its
components. Previously, Ang and Kerstens (2017) applied the LHM TFP indicator to the analysis of agricultural productivity in the
US. However, they proposed particular cases of the LHM indicator (either input- or output-oriented), while, as noted above, this paper
proposes a comprehensive LHM TFP indicator that includes both the input- and output-orientations. In our illustration, we find that
each component of the TFP indicator has the same signs and similar magnitudes under the input- and output-orientations, but this
need not be the case. In cases in which the magnitudes and/or signs differ across orientations, using just one orientation could result
in biased information on economic performance, which could be detrimental to successful policy making. Finally, we illustrate the
proposed measures using provincial level panel data on China's agricultural sector over the period 1997–2015. The period covered
coincides with two recent cycles of economic reforms of the agricultural sector (Brümmer, Glauben, & Lu, 2006). Aggregate, regional,
and provincial level findings for TFP and its three components are discussed, as are the separate roles of input growth and output
growth on the components of TFP. We find that technological progress is the key driver of TFP growth in Chinese agriculture, but
results vary over time and space.
The remainder of the paper is organized as follows. Section 2 provides a literature review on the major approaches toward the
measurement of TFP and their manifestations for the Chinese agriculture. Section 3 begins with the building blocks of the LHM
indicator, and then presents the LHM TFP indicator along with its decomposition into technological change, technical efficiency
change, and scale efficiency change. Section 4 illustrates the calculation and decomposition of the LHM indicator using data from
China's agriculture sector over the period 1997–2015. Both the data and the insights gained when the LHM is used to investigate
productivity change in Chinese agriculture are discussed. The application demonstrates the value of using an additively complete TFP
measure to examines productivity change over time. Section 5 summarizes the paper's contributions and findings.

2. Literature review

The measurement of TFP involves a number of methodological challenges. In this section, we briefly review the major strands in
the TFP measurement literature, with a particular focus on DEA-based non-parametric approaches. We then discuss the literature on
TFP change in Chinese agriculture.

2.1. The measures and measurement of TFP

Following Chambers (1988) and Machek and Špička (2013) one can distinguish between the two types of TFP measures: price-
based measures and distance function-based measures. Price-based measures of TFP use price data to aggregate inputs and outputs.
The price-based approach has both merits and demerits. On the one hand, price-based aggregation allows just two observations to be

1
A mu is a unit of land measurement used in China. One mu is 1/15 ha or 0.1647 acres, so 6.42 mu is equal to 0.428 ha or 1.058 acres.

2
Z. Shen, et al. China Economic Review 57 (2019) 101315

compared and, thus, is less data-intensive. On the other hand, reliable price data may not be available and, therefore, the analysis
may be less certain. Examples of price-based TFP indices include the Fisher and Tornqvist indices. The distance function-based indices
utilize production frontiers and measures of productive efficiency (distance functions). These measures do not require price in-
formation, but relative to the price-based measures the distance function-based measures are more data intensive since they must
approximate production frontiers and estimate distance functions. The estimation of the production frontiers can be carried out either
parametrically (Cechura, Kroupova, & Rudinskaya, 2015) or non-parametrically (Coelli & Rao, 2005; O'Donnell, 2012b). Note that
the parametric approach often relies on stochastic frontier analysis (SFA). SFA easily allows for statistical inference but requires
assumptions on the functional form of the production technology and, in general, an assumption about the distribution of the
inefficiency. Furthermore, certain desirable axioms of technology may be violated unless parametric restrictions are imposed on the
empirical model. The non-parametric approach, which is widely applied, makes no assumptions regarding the functional form of the
production technology and only assumes that basic axioms of production theory, such as convexity, are satisfied. In this paper, we
focus on the non-parametric analysis of TFP.
The primary distance function-based approaches to TFP measurement are the Malmquist, Luenberger, Färe-Primont, and Hicks-
Moorsteen indices/indicators. While the Malmquist productivity index (Caves, Christensen, & Diewert, 1982) and the Luenberger
productivity indicator (Chambers, 2002) are probably the most widely applied measures, O'Donnell (2012a) showed that they were
not “multiplicatively complete” because they cannot be completely decomposed with respect to inputs and outputs. The Hicks-
Moorsteen productivity index (Bjurek, 1996) satisfies the conditions for completeness defined by O'Donnell (2012a). Briec and
Kerstens (2004) developed the additive Luenberger-Hicks-Moorsteen (LHM) TFP indicator. Additive measures of productivity—based
on differences, not ratios—allow for greater flexibility (e.g., they allow for zeros in the data and are translation invariant). Moreover,
the LHM indicator is potentially a “complete” TFP indicator since its decomposition can be carried out with respect to both inputs and
outputs, as we show below.

2.2. Chinese agriculture and TFP change

Noting the relative lack of consensus on TFP's contribution to China's economic performance, Wu (2011) performed a meta-
analysis of 74 studies from the literature on TFP change in the Chinese economy covering 1988–2008 (the studies analyzed in this
meta-analysis used data from within the period 1978–2007). For the agricultural sector, the meta-analysis found an average TFP
growth rate of 3.81% per annum (p.a.); technological progress was the main contribution to TFP growth. Wu (2011) noted that
results varied by region, methods used, and sample period studied. Tian and Yu (2012) also used meta-analysis to examine over 5000
measures of TFP growth at the aggregate and sector-specific levels for China. Their results suggested an annual rate of TFP growth of
2.03% for the Chinese agricultural sector over the period 1950–2008.
Li and Zhang (2013) analyzed TFP change in the Chinese agriculture by applying the DEA-based Malmquist index to provincial
level panel data. They obtained an average annual rate of TFP of 3.8% for 1985–2010. They found that technological progress
increased by 5.1% per year while technical efficiency declined by 1.2% per year. Li and Zhang (2013) also found that TFP growth
fluctuated over time, with the fluctuations directly related to the government's support of agriculture. Song, Han, and Deng (2016)
used a bootstrapped Malmquist TFP index to analyze of crop production across China's 31 provinces. Their results indicated an
average annual increase in TFP of 6.1% for the period of 1999–2008; technological progress was the main source of TFP growth.
Furthermore, Song, Han, and Deng (2016) found the rate of TFP growth changed over the period in response to changes in gov-
ernment policy. Chen, Yu, Chang, and Hsu (2008) applied the output-oriented Malmquist productivity index to panel data for 29
provinces to estimate TFP change in the Chinese agriculture. For the sub-periods of 1990–1996, 1996–1999, and 1999–2003, they
found average productivity growth rates of 1.4%, 0.3%, and 2.5%, respectively. Chen et al. (2008) concluded that technological
progress was the key driver of TFP change, while both technical efficiency and scale efficiency declined. The same approach was
followed by Ma and Feng (2013) for the period of 1994–2008. Again, technological progress was found to dominate efficiency change
as the driver of TFP growth; they found that efficiency was improving. Li, Jiang, Yu, and Shang (2017) applied an output-oriented
Malmquist productivity index to measure the energy efficiency of the Chinese agricultural sector. Based on provincial data, they
found an annual productivity growth rate of 3.3% for 1997–2014. They found that TFP growth varied geographically; it was greatest
in the east (3.7%) and lowest in the west (2.2%). Using a non-parametric Törnqvist-Theil TFP index, Wang, Tuan, Gale, Somwaru,
and Hansen (2013) found a 2.7% annual rate of TFP growth for Chinese agriculture in 25 provinces over the period 1985–2007. They,
too, found differences in productivity growth across regions, with coastal regions performing relatively better.
Zhang and Brümmer (2011) used stochastic frontier analysis with province-level data for 1979–2008 to appraise TFP change in
Chinese agriculture. They found average annual growth rates of 5.4% and 8.6% for 1998–2003 and 2004–2008, respectively. Gong
(2018) applied a semi-parametric smooth coefficient model to gauge the effects of the agricultural policies on TFP change in the
Chinese agricultural sector over the period of 1978–2015. Gong (2018) found that TFP varied by province and across time, un-
derscoring the need for a flexible model of TFP. While TFP growth was positive for much of the period covered, China's TFP growth in
agriculture fluctuated across the six different periods of reform since 1978.
As this brief review indicates, the literature on the productivity of China's agriculture sector has used a variety of methods and
data samples to assess productivity change and its sources. While findings are dependent on both methods and data (Wu, 2011), all of
the studies discussed above found positive rates of TPF growth; furthermore, those studies decomposing TFP growth found that with
technological change was the primary driver of TFP growth. Gong (2018) recently noted the importance of using flexible models.
Among the modeling choices, choosing just an input- or an output-oriented productivity index could affect results. Given the im-
portance of the productivity of China's agricultural productivity, there is a need for a generalized framework that combines the input-

3
Z. Shen, et al. China Economic Review 57 (2019) 101315

and output-oriented approaches. As regards the Chinese agricultural sector, the application of the generalized LHM TFP indicator is a
promising avenue as it may provide new insights into the sources of productivity growth, which could help with the development of
better policies. Consequently, the application of the generalized LHM TFP developed in this paper adds to the discussion of TFP
change in Chinese agriculture in both methodological and empirical terms.

3. Methodology

3.1. Building blocks—the production technology and distance function

The technology available to producers can be represented in various ways. Neo-Walrasian production theory was put forward by
Koopmans (1951), Debreu (1951), Shephard (1953), and Farrell (1957). Following these works, technology can be represented in
very general terms using the concept of a production possibility set—the set of all non-negative outputs that are feasible given
available inputs. These sets can change over time due to technological progress or firm-specific behavior. The neoWalrasian pro-
duction theory considers homogenous entities, decision making units (DMUs), that shape the production possibility sets.
Assume a multiple-input, multiple-output production technology under which DMUs consume N types of inputs (x) to produce M
types of outputs (y). The production possibility set at time t is given by:
T (t ) = { (xt, y t ) ∈ R+N + M : xt can produce y t } (1)
To assure that the technology satisfies standard economic assumptions, general axioms must hold for production sets. In parti-
cular, the following three basic axioms are usually imposed on the production possibility set (Shephard, 1953):

A1: (0, 0) ∈ T (t ) and if (0, y ) ∈ T (t ) then y = 0.A2: T (t ) is closed.A3: For each input x ∈ R+N , T (t ) is bounded. (2)
Assumption A1 implies inactivity is feasible and, conversely, that there is no free lunch (i.e., outputs cannot be generated without
inputs). Assumption A2 states that unlimited quantities of outputs cannot be produced from finite quantities of inputs, while A3
implies that production plans located on the efficient frontier belong to the technology. In addition to these three basic axioms, we
assume that technology can be characterized by variable returns to scale (VRS).
Following Chambers, Chung, and Färe (1996), the production technology can be represented by the directional distance function.
A generalized directional distance function (DDF) simultaneously expands outputs and contracts inputs for period a ∈ {t, t + 1} with
respect to a technology in period b ∈ {t, t + 1} to project observations onto the frontier. The generalized DDF is defined as:

(
D b x a, y a; g ax , g ay ) = max{δ ∈ R +: (x a
)
− δ g ax , y a + δ g ay ∈ T (b)} (3)
where (gxa, gya) ≥
0 are directional vectors along which inputs and outputs are projected onto the frontier of the technology, δ
measures the maximum possible increase in outputs and decrease in inputs, and (a, b) ∈ {t, t + 1} × {t, t + 1} allows for the mixed-
period directional distance functions needed to calculate productivity change over time.

3.2. The LHM indicator and its decomposition

3.2.1. The LHM indicator


The LHM productivity indicator is an additively complete TFP measure as defined by O'Donnell (2012a, p. 258). Similar to the
other distance function-based productivity indicators, the LHM indicator captures the change in TFP through the measurement of the
distances between observations and the frontiers for two successive periods t and t + 1. Two directions are utilized to measure the
components of the LHM productivity indicator. First, the output directional distance function is employed to measure the distances
from the frontier holding inputs constant. With the inputs fixed, this gives a measure of output change. Second, the input directional
distance function is applied to measure the distances from the frontier while holding outputs constant. With outputs fixed, this gives a
measure of input change. Note that the inputs or outputs are fixed at their base period levels, respectively, for the two measures; the
base period can be changed and the resulting measures averaged, providing a comprehensive measure of TFP change.
Start by treating period t as the base. Then the LHM TFP indicator is defined as the difference of changes in productivity due to
changes in outputs (keeping inputs fixed at period t levels) or inputs (keeping outputs fixed at period t levels):

⎡ ⎤
LHM t = ⎢Dt xtk , y tk ; 0, g ty − Dt ⎜⎛xtk , y tk+ 1; 0, g ty+ 1⎟⎞ ⎥
( )
⎣ ⎝ ⎠⎦
− [Dt (xtk+ 1, y tk ; g tx+ 1, 0) − Dt (xtk , y tk ; g tx , 0)] (4)
The first two bracketed terms in Eq. (4) measure the change in the directional distance function due a change in output quantities
from ykt to ykt+1 based on the technology for period t while keeping the input quantity fixed at xkt (i.e., gxt = 0 ). The last two
bracketed terms are also relative to the frontier of period t and gauge the change in the directional distance function resulting from
movement from input quantity xkt+1 to xkt with output quantity held constant at ykt (i.e., gyt = 0 ). Note that the directional distance
functions are defined for the directions given by the direction vectors gyt, gyt+1, gxt+1, and gxt. Values of the LHM productivity
indicator exceeding zero indicate gains in TFP, while values below zero indicate declines in TFP.
Analogously, using the technology in period t + 1 as the base, the LHM indicator is given by:

4
Z. Shen, et al. China Economic Review 57 (2019) 101315

⎡ ⎤
LHM t + 1 = ⎢Dt + 1 xtk+ 1, y tk ; 0, g ty − Dt + 1 ⎛⎜xtk+ 1, y tk+ 1; 0, g ty+ 1⎞⎟ ⎥
( )
⎣ ⎝ ⎠⎦
− [Dt + 1 (xtk+ 1, y tk+ 1; g tx+ 1, 0) − Dt + 1 (xtk , y tk+ 1; g tx , 0)] (5)
As noted above, averaging productivity indicators using different bases (t vs. t + 1) provides a more comprehensive measure that
obviates an arbitrary choice of the base period relative to which TFP is measured. In our case, since the LHM indicator is an additive
measure, we take the arithmetic mean of the indicators given in Eqs. (4) and (5):
1
LHM t , t + 1 = (LHM t + LHM t + 1)
2 (6)

3.2.2. A generalized decomposition of the LHM indicator


As shown by O'Donnell (2012a), a complete TFP index/indicator can be unambiguously decomposed into a technical change term
and inefficiency change terms. Following Diewert and Fox (2014, 2017) and Ang and Kerstens (2017), we decompose the LHM
indicator given in Eq. (6) into three terms:
LHM t , t + 1 = TP t , t + 1 + TEC t , t + 1 + SEC t , t + 1 (7)
t, t+1 t, t+1
where TP captures the effect of technological progress on productivity (i.e., a shift of the technological frontier), TEC
measures the contribution of technical efficiency change to productivity change, and SECt, t+1 indicates the effect of scale efficiency
change on productivity change.
The decomposition given in Eq. (7) can be carried out by projecting observations onto the frontier in either an output direction
(output-oriented) or an input direction (input-oriented). Our goal is to obtain a non-oriented decomposition based on the generalized
LHM indicator. Recall that a more general LHM indicator was obtained above by averaging the LHM indicators based on the
technologies of periods t and t + 1. Now we wish to generalize the decomposition by using two different orientations—the input-
orientation and the output-orientation—to measures the components of the LHM indicator. To do so, we present two oriented cases
along with the generalized decomposition of the LHM indicator.
We begin with the output-based decomposition which is based on the application of output directional distance functions as
shown in Fig. 1. The output-oriented technical efficiency change (TEC) component of the growth in TFP is the change in output-based
technical efficiency across two observed production plans relative to their corresponding contemporaneous frontiers:
t,t+1
TECoutput ( ) (
= Dt xtk , y tk ; 0, g ty − Dt + 1 xtk+ 1, y tk+ 1; 0, gty+ 1 ) (8)
t, t+1
where TECoutput > 0 (resp. TECoutputt, t+1
< 0) indicates growth (resp. regress) in TFP due to improved (resp. decreased)
technical efficiency.
As illustrated in Fig. 2, the technological progress (TP) term measures the growth in TFP that results from movements of the
production frontier over time (i.e., technological changes). In input-output space, the outward (resp. inward) movement of the
production frontier with respect to the point of origin implies that the frontier points are associated with higher (resp. lower)
productivity levels when compared to the preceding period. The output-oriented TP component compares the performance of two
input-output bundles across technologies. It's computed by considering the differences in distances of observations to frontiers across
time periods, by taking each period as the base in turn:

y
T (t+1)

Dt 1
xtk 1 , y tk 1 ; 0, g ty 1

Dt 1
xtk 1 , y tk 1 ; g tx 1 , 0 T (t)

D t xtk , y tk ; 0, g ty

D t xtk , y tk ; g tx , 0

0 x

Fig. 1. Technical efficiency change.

5
Z. Shen, et al. China Economic Review 57 (2019) 101315

y
T (t+1)

t 1
TPoutput
t 1
TPinput
t
TPoutput T (t)

t
TPinput

0 x

Fig. 2. Technological progress.

1 ⎡ ⎤
t,t+1
TPoutput = ( ⎢Dt + 1 xtk , y tk ; 0, g ty − Dt ⎛⎜xtk , y tk ; 0, g ty⎞⎟ ⎥
( )
2 ⎝ ⎠⎦

⎡ ⎤
+ ⎢Dt + 1 xtk+ 1, y tk+ 1; 0, g ty+ 1 − Dt ⎛⎜xtk+ 1, y tk+ 1; 0, gty+ 1⎞⎟ ⎥ )
( )
⎣ ⎝ ⎠⎦ (9)

where the first two terms capture the distance between the frontier for the two consecutive periods for an observation in period t and
the last two terms capture an analogous distance for the observation in period t + 1. Therefore, TPt, t+1 > 0 (resp. TPt, t+1 < 0)
indicates an increase (resp. a decrease) in TFP due to technological progress.
Since the LHM indicator is a complete TFP indicator, it can be exhaustively decomposed. Therefore, the scale efficiency change
term, SEC can be defined residually once LHM, TEC, and TP are known. The SEC component reflects changes in TFP that accrue from
changes in the scale of operation. In the output-oriented case, the movement toward or away from the most productive scale size is
captured by considering the change in the gradient of the frontier at the points of observations for periods t and t + 1:
t,t+1 t,t+1 t,t+1
SECoutput = LHM t , t + 1 − TECoutput − TPoutput

⎡ ⎤
= 1 2 {( ⎢Dt xtk+ 1, y tk+ 1; 0, g ty+ 1 − Dt ⎛⎜xtk , y tk+ 1; 0, g ty+ 1⎞⎟ ⎥
( )
⎣ ⎝ ⎠⎦
− [Dt (xtk+ 1, y tk ; g tx+ 1, 0) − Dt (xtk , y tk ; g tx , 0)])

⎡ ⎤
+ ( ⎢Dt + 1 xtk+ 1, y tk ; 0, g ty − Dt + 1 ⎜⎛xtk , y tk ; 0, g ty⎟⎞ ⎥
( )
⎣ ⎝ ⎠⎦
− [Dt + 1 (xtk+ 1, y tk+ 1; g tx+ 1, 0) − Dt + 1 (xtk , y tk+ 1; g tx , 0)]) } (10)

where the first four terms following the second equal sign allow computing the gradient of the frontier for period t by considering
changes in the efficient output level due to changes in the observed input use given by vectors xt and xt+1, and the last four terms
measure the gradient of the frontier for period t + 1 by considering changes in the efficient output level on the corresponding frontier
due to the same change in the input vector as it is defined for period t frontier.
Following Diewert and Fox (2017) and Ang and Kerstens (2017), the SEC component in Eq. (10) can be re-written by considering
efficient points on the frontiers shown in Fig. 3a. By virtue of the translation property of the directional distance function, one can
rewrite (10) as follows:

1 ⎡ t t t,∗ ⎤
t,t+1
SECoutput = {( D x k , y k ; 0, g ty − Dt ⎛⎜xtk , y tk+ 1, ∗∗; 0, g ty+ 1⎞⎟ ⎥
( )
2 ⎢ ⎣ ⎝ ⎠⎦
− [Dt (xtk+ 1, y tk ; g tx+ 1, 0) − Dt (xtk , y tk ; g tx , 0)])

⎡ ⎤
+ ( ⎢Dt + 1 xtk+ 1, y tk, ∗∗; 0, g ty − Dt + 1 ⎛⎜xtk+ 1, y tk+ 1, ∗; 0, g ty+ 1⎞⎟ ⎥
( )
⎣ ⎝ ⎠⎦
− [Dt + 1 (xtk+ 1, y tk+ 1; g tx+ 1, 0) − Dt + 1 (xtk , y tk+ 1; g tx , 0)])} (11)

6
Z. Shen, et al.

y y
D t 1 (xtk 1 , y tk,** ; 0,g ty ) D t 1 (xtk 1 , y tk 1,* ; 0,g ty 1 ) T (t 1) D t (xtk,* , y tk ; g tx ,0) D t (xtk 1,** , y tk ; g tx 1 ,0) T (t 1)

D t 1 (xtk,** , y tk 1; g tx ,0) D t 1 (xtk 1,* , y tk 1; g tx 1 ,0)


D t (xtk , y tk,* ; 0,g ty ) D t (xtk , y tk 1,** ; 0,g ty 1 )
T (t ) T (t )

(xtk 1 , y tk 1 ) (xtk 1 , y tk 1 )

7
(xtk , y tk ) (xtk , y tk )
D t 1 (xtk 1 , y tk 1; 0,g ty 1 ) D t 1 (xtk 1 , y tk
D t 1 (xtk 1 , y tk 1; g tx 1 ,0) D t 1 (xtk , y tk 1; g tx ,0)
D t (xtk , y tk 1; 0,g ty 1 ) D t (xtk , y tk ; 0,g ty )
D t (xtk 1 , y tk ; g tx 1 ,0) D t (xtk , y tk ; g tx ,0)

0 x 0 x

a – Output-oriented Approach b – Input-oriented Approach


Fig. 3. Scale efficiency change.
China Economic Review 57 (2019) 101315
Z. Shen, et al. China Economic Review 57 (2019) 101315

where

(
y tk, ∗ = y tk + Dt xtk , y tk ; 0, g ty g ty and )
(
y tk+ 1, ∗∗ = y tk+ 1 + Dt xtk+ 1, y tk+ 1; 0, g ty+ 1 g ty+ 1 ) (12)

are the projections of ykt and ykt+1 onto the frontier in period t, and

(
y tk, ∗∗ = y tk + Dt + 1 xtk , y tk ; 0, g ty gty and )
y tk+ 1, ∗ = y tk+ 1 + Dt + 1 ( xtk+ 1, y tk+ 1; 0, gty+ 1 )g t+1
y (13)
t t+1
define the projections of yk and yk onto the frontier in period t + 1.
This completes the output-oriented decomposition of the LHM productivity indicator.
To continue toward the generalized LHM productivity indicator, we now present the input-oriented decomposition of the LHM
productivity indicator. Based on the input directional distance function, the input-oriented LHM indictor can also be decomposed into
the three terms of technological progress, technical efficiency change, and scale efficiency change.
The input-oriented TEC component relies on the input directional distance functions as shown in Fig. 1. Holding output at its
observed level, the input direction distance function measures how far an input vector lies from the frontier of technology. The input-
oriented TEC component measures the difference in excess input use across the two consecutive time periods:
t,t+1
TECinput = Dt (xtk , y tk ; g tx , 0) − Dt + 1 (xtk+ 1, y tk+ 1; g tx+ 1, 0), (14)

where TECinputt, t+1 > 0 (resp. TECinputt, t+1 < 0) indicates gains (resp. declines) in productivity due to movement toward (resp.
away from) the contemporaneous frontier (i.e., changes in technical efficiency).
As illustrated in Fig. 2, the input-oriented TP component of the LHM indicator is obtained by considering the distances of two
input-output bundles, (xt, yt) and (xt+1, yt+1), between the frontiers in periods t and t + 1 in the input direction:
1
t,t+1
TPinput = ([Dt + 1 (xtk , y tk ; g tx , 0) − Dt (xtk , y tk ; g tx , 0)]
2
+ [Dt + 1 (xtk+ 1, y tk+ 1; g tx+ 1, 0) − Dt (xtk+ 1, y tk+ 1; g tx+ 1, 0)]), (15)

where the difference between the first two terms measures the change in distance of the base period inputs and outputs from the
frontiers in periods t and t + 1, while the difference between the last two terms measure the difference of the period t + 1 inputs and
outputs between the frontiers in period t and t + 1. That is, the performance of each input-output combination is compared across
two technologies, allowing the effect of technical change to be captured. Note that TPinputt, t+1 < 0 denotes an inward shift of the
technological frontier which corresponds to a reduction in TFP, whereas TPinputt, t+1 > 0 is associated with an increase in TFP
associated with technological progress and an outward shift of the frontier.
The input-oriented SEC term is also computed residually as:
t,t+1 t,t+1 t,t+1
SECinput = LHM t , t + 1 − TECinput − TPinput
1
= {([Dt (xtk+ 1, y tk+ 1; g tx+ 1, 0) − Dt (xtk+ 1, y tk ; g tx+ 1, 0)]
2
⎡ ⎤
− ⎢Dt xtk , y tk+ 1; 0, g ty+ 1 − Dt ⎛⎜xtk , y tk ; 0, g ty⎞⎟ ⎥ )
( )
⎣ ⎝ ⎠⎦
+([Dt + 1 (xtk , y tk+ 1; g tx , 0) − Dt + 1 (xtk , y tk ; g tx , 0)]

⎡ ⎤
−⎢Dt + 1 xtk+ 1, y tk+ 1; 0, g ty+ 1 − Dt + 1 ⎛⎜xtk+ 1, y tk ; 0, g ty⎞⎟ ⎥ ),
( )
⎣ ⎝ ⎠⎦ (16)

where the first four terms following the second equality capture the gradient of the frontier for period t and the last four terms capture
it for the frontier of period t + 1. As shown in Fig. 3b, the input-oriented measure can be rewritten by considering the efficient points
on the frontier. Specifically, the changes in the efficient levels of input use are compared to changes in the output levels:
1
t,t+1
SECinput = { ([Dt (xtk, ∗, y tk ; g tx , 0) − Dt (xtk+ 1, ∗∗, y tk ; g tx+ 1, 0)]
2
⎡ ⎤
− ⎢Dt xtk , y tk+ 1; 0, g ty+ 1 − Dt ⎛⎜xtk , y tk ; 0, g ty⎞⎟ ⎥ )
( )
⎣ ⎝ ⎠⎦
+ ([Dt + 1 (xtk, ∗∗, y tk+ 1; gtx , 0) − Dt + 1 (xtk+ 1, ∗, y tk+ 1; gtx+ 1, 0)]

⎡ ⎤
− ⎢Dt + 1 xtk+ 1, y tk+ 1; 0, gty+ 1 − Dt + 1 ⎛⎜xtk+ 1, y tk ; 0, gty⎞⎟ ⎥ )},
( )
⎣ ⎝ ⎠⎦ (17)

where

8
Z. Shen, et al. China Economic Review 57 (2019) 101315

xtk, ∗ = xtk + Dt (xtk , y tk ; g tx , 0) g tx and

xtk+ 1, ∗∗ = xtk+ 1 + Dt (xtk+ 1, y tk+ 1; g tx+ 1, 0) g tx+ 1 (18)

are the projections of xkt and xkt+1onto the frontier in period t and

xtk, ∗∗ = xtk + Dt + 1 (xtk , y tk ; g tx , 0) g tx and

xtk+ 1, ∗ = xtk+ 1 + Dt + 1 (xtk+ 1, y tk+ 1; g tx+ 1, 0) g tx+ 1. (19)

are the projections of xkt and xkt+1 onto the frontier in period t + 1.
Having developed the input- and output-oriented measures of productivity and their decompositions, we can now decompose the
change in TFP into three generalized terms that consider both input change and output change. These three generalized TFP com-
ponents are averages of their corresponding oriented terms:

1 t,t+1 t,t+1
TEC t , t + 1 = (TECoutput + TECinput ),
2
1 t,t+1 t,t+1
TP t , t + 1 = (TPoutput + TPinput ),
2
1 t,t+1 t,t+1
SEC t , t + 1 = (SECoutput + SECinput ).
2 (20)

This completes the derivation of the additively complete Luenberger-Hicks-Moorsteen productivity indicator and its decom-
position into technological progress, technical efficiency change, and scale efficiency change. Both input-oriented and output-or-
iented measures were presented, as well as general measures that combine the two orientations. We now turn to the specification of
non-parametric models that can be used to empirically implement the measures we have developed above.

3.3. Model specification

We adopt a non-parametric approach to estimate the directional distance functions needed to compute the LHM productivity
indicator and its components presented above. This approach allows us to impose a set of economic assumptions (i.e., monotonicity,
convexity, and variable returns to scale) on the underlying technology without specifying a particular functional for the technology.
While we follow the DEA-approach to implement the LHM indicators, one could opt for other estimators (e.g., SFA); however, this
may come at the cost of violating of certain economic axioms, possibly jeopardizing the interpretability of the model and results.
We specify and compute the directional distance functions as DEA-like linear programs. For the sake of brevity, we limit our
exposition of the models to two general cases. Specifically, we provide the linear programming problems for the input and output
distance functions for input-output vectors from period a ∈ {t, t + 1} and the technologies defined by observations from period
b ∈ {t, t + 1}. Assume there are K production units (provinces in the illustration below) indexed by k = 1, 2, …, K. The input-output
vectors of these production units (provinces in our illustration below) are used to construct the empirical frontiers of technology.
Using input and output data from period a and the production technology available in period b, the output directional distance
function Db(xa, ya; 0, gya) is obtained by solving the following linear programming problem:

( )
D b x a, y a; 0, g ay = max δ
δ , λk
K
s. t . ∑ λk ykm,b ≥ y m,a + δgym,a, ∀ m = 1, ⋯, M ;
k=1
K
∑ λk xkn,b ≤ x n,a , ∀ n = 1, ⋯, N ;
k=1
K
∑ λk = 1;
k=1
λk ≥ 0, ∀ k = 1, …, K , (LP1)

where the λks are intensity variables that indicate the weights placed on each observation k when constructing the “best-practice”
frontier, δ is the value of the output directional distance function giving the maximum expansion of outputs in the direction
(0, gya),and the summation constraint on the intensity variables imposes variable returns to scale on the technology. In our appli-
cation below, we apply the proportional directional distance function by setting gya = ya.
The input directional distance function Db(xa, ya; gxa, 0) is obtained by solving the following linear programming problem:

9
Z. Shen, et al. China Economic Review 57 (2019) 101315

D b (x a, y a; g ax , 0) = max ϕ
ϕ, λk
K
s. t . ∑ λk ykm,b ≥ y m,a , ∀ m = 1, …, M ;
k=1
K
∑ λk xkn,b ≤ x n,a − ϕgxn,a, ∀ n = 1, …, N ;
k=1
K
∑ λk = 1;
k=1
λk ≥ 0, ∀ k = 1, …, K ; (LP2)

where the λks are the intensity variables, ϕ is the value of the input directional distance function which measures the maximum
contraction of inputs in the direction specified by (gxa, 0),and the summation constraint on the intensity variables imposes variable
returns to scale on the technology. Analogous to the output-orientation, for the input-orientation we use the proportional directional
distance function as gxa = xa in the illustration given in the next section.
In the two linear programming models above, (LP1) and (LP2), full output and input efficiency, respectively, are represented by
zero values of δ and ϕ, whereas positive values of δ and ϕ indicate inefficiency. Note that the distances are expressed as a percentage
of the direction vectors. All required distance functions presented above can be obtained by generalizing the linear programming
models (LP1) and (LP2).

4. Illustration using the Chinese agricultural sector

In 1978 China instituted important agricultural reforms, which led to a rapid increase in agricultural productivity (for information
on China's rural reforms, see Zhang and Brümmer (2011), who divided the reforms into six different periods). However, the gains
from reform were not felt uniformly across time or place. For example, some reforms liberalized the production of cash crops, but this
most benefited regions where land was better suited to cash crops. Over time, China's central government has implemented additional
agricultural policies at the national level. For example, in 2004 the central government announced agricultural tax reform which led
to the abolishment of the agricultural tax in 2006. Other reforms included direct payments to grain producers, general inputs
subsidies, improved-seed subsidies, machinery and equipment subsidies, property rights reform that allow farmers to transfer their
land-use rights, and a land retirement program. Accurately measuring the overall change in productivity and the drivers of the change
are critical to understanding the effects of these policies and to gain insight into the areas that future policies might target. We use the
generalized LHM productivity indicator developed above to measure productivity change in China's agricultural sector and determine
the underlying drivers of change.

4.1. Data

To assess the productivity of the Chinese agricultural sector, we apply the measures derived above to a balanced panel of data for
China's 31 mainland provinces, municipalities, and autonomous regions (hereafter, “provinces”) over the years 1997–2015. For ease
of analysis, we group the 31 provinces into three large economic zones based on their average volume of agricultural output over the
sample period; this allows us to compare performance by “size.”2 The zone comprised of the major agricultural producers includes the
following 10 provinces: Hebei, Heilongjiang, Jiangsu, Anhui, Shandong, Henan, Hubei, Hunan, Guangdong, and Sichuan. The second
zone consists of 10 provinces of medium agricultural importance: Inner Mongolia, Liaoning, Jilin, Zhejiang, Fujian, Jiangxi, Guangxi,
Yunnan, Shaanxi, and Xinjiang. The third zone consists of 11 provinces of minor agricultural importance: Beijing, Tianjin, Shanxi,
Shanghai, Hainan, Chongqing, Guizhou, Tibet, Gansu, Qinghai, and Ningxia.
The various directional distance functions used to assess the productivity change in China's agricultural sector are estimated using
one output and five inputs. The specification of inputs and outputs parallels previous research (e.g. Chen et al., 2008; Ito, 2010;
Kerstens, Shen, & Van de Woestyne, 2018 and Shen, Baležentis, Chen, & Valdmanis, 2018). Output is measured as the gross output
value of agriculture (100 million yuan in constant 2010 prices).3 The five inputs in the analysis are labor, machinery (“capital”), land,

2
China's provinces are often divided into groups for analysis. For example, Chen et al. (2008) compared high income provinces and low income
provinces, while Gong (2018) divided provinces by geography (eastern, central, and western provinces). The difference between these two ap-
proaches may be slight—Chen et al. (2008) noted that most of the high income provinces were on the coast and most low income provinces were in
the central and western regions of China, so grouping by income and by geographic location would be very similar.
3
China follows the common practice of classifying economic activity into three broad sectors—the primary sector, the secondary sector, and the
tertiary sector (commonly referred to as “agriculture,” “manufacturing,” and “services”). The primary sector (“agriculture”) includes economic
activity related to farming, forestry, animal husbandry and fishery, as well as services in support of farming, forestry, animal husbandry and fishery.
Our output measure, the gross output value of agriculture, measures the output of agriculture as a subsector of the primary sector (i.e., farming; it
does not include the outputs of the forestry, animal husbandry, and fishery subsectors, nor any of the support services of the primary sector). That is,
our output measure is based on the narrower definition of agriculture given, for example, in the introduction to Chapter 12 of the China Statistical
Yearbook (2015), “Agriculture: refers to cultivation of farm crops, including cereals, beans, tuber crops, cotton, oil-bearing crops, sugar crops, hemp,
tobacco leaves, vegetables, gardening plants, fruits, nuts, crops for beverages and spices, medicinal herbs and other farm crops.” Given changes in

10
Z. Shen, et al. China Economic Review 57 (2019) 101315

pesticides, and fertilizer. Labor is measured by the total number of employed persons in the primary sector (in 10,000 s of people).4
Machinery is measured by the sum of all agricultural machinery power (in 10,000 s of kw) and provides a measure of the mechanical
power available in agriculture. Land is measured by the total sown area of farm crops (in 1000 s of hectares). Pesticides are measured
by the annual use of agricultural pesticide (in 10,000 s of tons). Finally, chemical fertilizer is measured as the applied quantity of
effective fertilizer components (in 10,000 s of tons)—i.e., gross weight is converted to the weight of fertilizer's 100% effective
components (e.g., the 100% potassium oxide content in potash fertilizer).5 The input and output data were collected from the China
official statistical database (National Bureau of Statistics of China, 2017).6,7 Because Chongqing became a municipality in 1997, we
select 1997–2015 as the sample period to enhance data availability and comparability. This also corresponds to the most recent
policy cycles (Brümmer et al., 2006; Zhang & Brümmer, 2011). Of course, altering the research period may render different estimates
of TFP change and its components. Table 1 reports the annual growth rates for the input and output variables used in our illustration
of the generalized LHM TFP indicator.
The data in Table 1 provide several insights into the dynamics of TFP for the Chinese agricultural sector. At the national level, the
average annual rate of growth in gross agricultural output was 4.16%, which exceeded the growth rates for labor (−1.44% p.a.), land
(0.41% p.a.), pesticides (2.63% p.a.), and fertilizer (2.58% p.a.).8 Notably, the labor input declined by −1.44% p.a. (the sole case of
an observed decline among the variables employed), due in part to the rise in off-farm employment associated with industrialization
and urbanization in China. The growth rate for machinery, however, 5.53%, surpassed the growth in agricultural output by 1.37
percentage points. The rates of change in the output and input data give rise to two observations. First, the Chinese agriculture has
seen an expansion in scale during 1997–2015 as evidenced by the positive rates of growth for output and all but one input (i.e., labor)
and output. Second, machinery has grown much faster than labor.
The rapid mechanization of Chinese agriculture poses a challenge for growth in TFP as the growth in the input machinery exceeds
that in agricultural output. Ito (2010), for example, found that machinery's contribution to TFP growth was negative for most of
China's provinces over the period 1991–2004 because of declining land-labor ratios. Machinery is a quasi-fixed (indivisible) in-
vestment that may not be fully utilized, especially given the relatively small size of farms in China. However, as Zhang, Yang, and
Reardon (2017) note, there are two types of division of labor—one within a producer and one across producers. While most farms in
China are too small to acquire machinery (allowing labor and capital to specialize), specialization across producers provides farmers
the opportunity to outsource power intensive activities such as harvesting to specialized firms that offers mechanization services.
These firms can travel around the country providing their services as needed. Thus, the development of mechanization services as
well as cooperatives can facilitate the use of machinery. Agricultural machinery's importance extends beyond its direct role in
agriculture—machines free labor, which can flow to rural enterprises or move to cities, fueling economic growth in other sectors. In
addition, machinery could support larger farms, enhancing scale efficiency. Like Ito (2010), we include machinery as an input in our
empirical illustration to capture the effects of mechanization on TFP growth.
The growth rate for agricultural output was highest in the medium agricultural producing provinces (4.59% p.a.), followed by the
minor (4.08% p.a.) and major provinces (3.98% p.a.). This suggests that the provinces belonging to the regions of medium and minor
agricultural importance managed to increase their output levels and tended to converge with the region of major importance.
The most obvious differences in input use dynamics are seen in machinery and the use of pesticides. Specifically, the medium and
minor agricultural producing zones show the highest rates of growth in machinery use (6.13% and 5.45%, respectively), whereas the
major zone provinces show a relatively low growth rate of 5.31%. Conversely, the minor and medium agricultural producing pro-
vinces exhibit higher rates of growth in the pesticide use (5.94% and 3.49%, respectively), compared to the provinces of major
importance (1.81%). Fertilizer growth, however, does not vary to the same degree as pesticide growth, but the growth was greater for

(footnote continued)
coverage and changes in the classification system over time, using official Chinese data can be challenging; guidance can be found in Holz (2013).
4
Data on employment in the agricultural sector are not available at the provincial level for the period 1997–2015. Instead, we use data on
employment in the primary sector; Gong (2018), Zhan et al. (2017), and others have also used this measure of labor. As explained above, the
primary sector is commonly referred to as “agriculture,” which is its main component, but it is broader than agriculture—it includes agriculture,
forestry, animal husbandry, and fishery activities Given this measure of labor overstates the actual amount of labor used in agriculture per se, our
measures of TFP growth are likely understated.
5
For more details on variable definitions, see the appropriate appendices of the China Statistical Yearbook (2015).
6
It should be noted that some variable definitions change slightly over time. For example, after 2003 the gross value of agricultural output
includes total value of products of agriculture, forestry, animal husbandry, and fishery as well as the value of services in support of agriculture,
forestry, animal husbandry and fishery.
7
Concerns exist regarding the accuracy and reliability of China's economic statistics. For agricultural data, the concern is fueled in part by gaps
between official production and consumption data. Plekhanov (2017) reviews the academic literature on the accuracy and reliability of China's
official statistics and identifies the key criticisms of the official statistics. Changes in definitions and coverage, underfunding and understaffing of
data agencies, incentives to over-report output, and the transition from a state dominated economy to one of numerous small-scale enterprises are
some of the reasons behind the qualms about data validity. See Rawski and Xiao (2001) for an introduction to a series of papers on Chinese economic
statistics that appeared in the China Economic Review in 2001. Zheng (2001) provides an overview of China's data collection and reporting systems
and discusses the “remarkable progress” that has been made and challenges that remain. It should noted that the concern about Chinese data is not
universal; Chow (2006; p. 396) noted that while “some statistics are not reliable,” in his view “the statistics are by and large reliable and useful for
drawing conclusions about the Chinese economy.”
8
Note that the average annual rates of growth reported here are based on the province-level data and stochastic trends. Therefore, they cannot be
directly compared to the national-level data from official statistics.

11
Z. Shen, et al. China Economic Review 57 (2019) 101315

Table 1
Annual growth rates of inputs and outputs.
Variable China Major Medium Minor

Inputs Labor −1.44% −1.79% −0.91% −1.06%


Machinery 5.53% 5.31% 6.13% 5.45%
Land 0.41% 0.36% 0.58% 0.21%
Pesticides 2.63% 1.81% 3.49% 5.94%
Fertilizers 2.58% 2.10% 3.67% 2.42%
Output Gross Output 4.16% 3.98% 4.59% 4.08%

Note: Stochastic growth rates are given in the table. Major, Medium, and Minor denote the regions of major, medium, and minor agricultural
producing provinces, respectively.

the medium (3.67% p.a.) and minor (2.42% p.a.) province than for the major provinces (2.58% p.a.). Overall the input growth rates
imply a certain divergence in terms of machinery-based and biochemical-based growth among Chinese provinces. This is consistent
with Ito (2010), who found inter-regional differences in the use of biochemical technologies and mechanical technologies. Ito (2010)
attributed the divergence in biochemical technological attainment to differences in public spending on R&D—with more public
spending, the use of biochemical increases. On the other hand, Ito (2010) found that mechanization increased most in provinces with
high land-to-labor ratios. Indeed, Wang et al. (2016) showed that as off-farm employment increased, the accompanying wage growth
led to a sharp increase in mechanization.
This relationship between labor and machinery can be seen in the growth rates of these two inputs. As shown in Table 1, there's an
inverse relationship between the growth rates for these two variables at the regional level. Specifically, the steepest (resp. mildest)
decline in labor input is observed for the major (resp. minor) agricultural producing provinces, whereas the lowest (resp. highest)
growth in machinery use is observed there. This suggests that both scale and substitution effects determine the input levels in Chinese
agriculture. However, more detailed analysis of this issue falls beyond the scope of our paper.

4.2. Results

We now discuss results obtained from estimating the models developed above using the data on China's agricultural sector. We
believe this application provides useful insights into the patterns and sources of TFP change in China's agricultural sector. However,
given concerns about the accuracy and consistency of official Chinese statistics, the results of our analysis should be interpreted with
caution.
The average annual national and regional growth rates of TFP and their components are presented in Table 2. Table 2 also reports
the complete decomposition of the LHM TFP indicator based on both the input and output directional distance functions as well as the
generalized measures based on the means of the two oriented measures. We found that, on average, China's agricultural TFP grew by
3.05% p.a. over the period 1997–2015. Comparing our results to those of other non-parametric studies, our rate of TFP growth is
lower than the estimate of 3.8% reported in both the meta-analysis by Wu (2011) as well as by the DEA-Malmquist application by Li
and Zhang (2013), the 6.1% growth by Song, Han, and Deng (2016) using bootstrapped DEA, and the energy efficiency growth rate of
3.3% reported by Li et al. (2017) based on an output-oriented Malmquist index. On the other hand, our estimate is higher than the
rate of 2.03% p.a. found in the meta-analysis of Tian and Yu (2012), the growth rate of 1.5% found by Chen et al. (2008) using an
output-oriented Malmquist index, and the growth rate of 2.7% obtained by Wang et al. (2013) by using the Törnqvist-Theil pro-
ductivity indicator. Given that these studies use a wide variety of non-parametric methods and data sets that cover different units of
observation and time periods, a direct comparison of results is not possible. Nonetheless, our estimates seem reasonable. Given that
the LHM productivity indicator is, in principle, free of bias it should on average perform better than other measures. Of course, the
proper specification of inputs and outputs and the quality/accuracy of the data remain possible issues.

Table 2
Annual average regional growth rates of TFP indicator and its decomposition.
Variable China Major Medium Minor

TFP 3.05% 2.33% 3.90% 2.91%


TEC 0.04% −0.16% −0.21% 0.45%
TEC-output 0.04% −0.22% −0.29% 0.56%
TEC-input 0.04% −0.11% −0.12% 0.33%
TP 2.35% 3.20% 3.06% 0.94%
TP-output 2.31% 2.85% 3.30% 0.92%
TP-input 2.40% 3.55% 2.83% 0.96%
SEC 0.65% −0.70% 1.05% 1.52%
SEC-output 0.70% −0.30% 0.90% 1.43%
SEC-input 0.60% −1.11% 1.20% 1.62%

Note: We report TFP and its components for China overall and for the three agricultural zones (Major, Medium, and Minor) discussed above; the
components of TFP are further decomposed by output- and input-directional distance functions.

12
Z. Shen, et al. China Economic Review 57 (2019) 101315

Our estimate of TFP is less than those in two studies using stochastic frontier analysis (SFA) or its variants. Zhang and Brümmer
(2011) estimated TFP growth of 5.4% and 8.6% over the periods 1998–2003 and 2004–2008, respectively, while Gong (2018) found
an average annual rate of TFP growth of about 5%.9 As with the non-parametric studies, the SFA-based research suggests that TPF
growth varies by time and location.
Consistent with the previous literature (e.g., Li & Zhang, 2013, Song, Han, & Deng, 2016), the major source of TFP change for
China was technological progress (i.e., an outward shift of the production frontier), which averaged of 2.35% p.a. Off-farm em-
ployment driven by industrialization and urbanization may have led to more widespread use of advanced agricultural technology as
farm size expands, enhancing technological progress (Song et al., 2016; Wang et al., 2016). Changes in technical efficiency and scale
efficiency also contributed to TFP growth, though by much lesser amounts—by 0.04% and 0.65% p.a. on average, respectively, for
China overall. A comparison of the results of the input- and output-oriented measures and their decompositions reveals little dif-
ference across the two orientations at the national level. Overall, China's agricultural sector appears to have experienced important
improvements in TFP, driven primarily by technological progress.
Regional differences are observed in the dynamics of TFP growth, its sources, and differences across the input- and output-
oriented measures. The major agricultural producing provinces showed the lowest rate TFP of growth, 2.33% p.a. The relatively low
growth rate may be due in part to the fact that this region has a larger base level of output (and, hence, TFP). The minor agricultural
producing provinces come next, with an average rate of TFP growth of 2.91% p.a. The strongest growth in TFP, 3.9% p.a., was
observed for the medium importance provinces.
The size and relative contributions of the components of agricultural TFP growth also vary by zone. Technological progress
contributed positively, though not equally, to TFP for all three zones, with gains for the major, medium, and minor regions averaging
3.2%, 3.06%, and 0.94% p.a., respectively. Technical efficiency change had relatively small negative impacts on TFP change in the
major and medium agricultural producing provinces (−0.16% and − 0.21% p.a., respectively). The minor provinces had a modest
gain in technical efficiency change of 0.45%, suggesting the minor agricultural provinces were catching up to the frontier. It is
noteworthy that the major agricultural producing provinces had a negative rate of growth in scale efficiency component of TFP
(−0.7% p.a.), suggesting they strayed from the most productive scale size. On the other hand, both the medium and minor regions
experienced positive contributions from scale efficiency change (1.05% and 1.52% p.a., respectively); in fact, scale efficiency gains
(1.52% p.a.) accounted for more than that half of the minor provinces' TFP gains (2.91% p.a.). This might be mitigated by introducing
corrections in agricultural support schemes and further land reforms such as those introduced in 2016 (i.e., three rights separation;
see Wang & Zhang, 2017). On the other hand, the trend might be reversed once investments into other factors render sufficient
growth in the output.
As noted above, the average size of a farm in China is quite small, which suggests possible scale inefficiency. However, there are a
number forces affecting scale efficiency in Chinese agriculture. First, increased urbanization (and off-farm employment) could lead to
more land rental market transactions, which may, in turn, lead to larger sized agricultural operations that would enhance economies
of scale (Song, Han, & Deng, 2016). Huang, Gao, and Rozelle (2012) noted that the government has long been supportive of cul-
tivated land rental transactions because they allow operational land holdings to grow larger, which should enhance productivity
through scale efficiency. Second, as noted above, the emergence of nation-wide, migratory mechanization services has helped farms
overcome the limitations of smallholdings through outsourcing (Zhang et al., 2017).
Summarizing the patterns of the components of TFP change by zone, the major agricultural provinces had a negligible effect from
technical efficiency change (−0.16% p.a.), a strong benefit from technological progress (3.2% p.a.), and modest losses due to scale
efficiency (−0.7% p.a.); the medium agricultural provinces had minimal average technical efficiency change (−0.21% p.a.), large
gains from technological progress (3.06% p.a.), and modest benefits from scale efficiency change (1.05% p.a.); and the minor
agricultural provinces experienced relatively balanced gains from technical efficiency change (0.45% p.a.), technological progress
(0.94% p.a.), and scale efficiency change (1.52% p.a.). These results broadly agree with Chen et al. (2008), who, using an output-
oriented Malmquist productivity index, found that, on average, technological progress was the main driver of TFP, while technical
efficiency and scale efficiency declined over the period 1990–2003. Thus, major and medium agricultural provinces have enjoyed
larger TFP increases primarily due to technical progress, whereas minor agricultural provinces have benefited primarily from im-
proved scale efficiency for TFP growth.
The minor agricultural provinces faced a subdued shift of the production frontier, but there is evidence of “catching up” as
individual provinces within this group reduced their performance gaps relative to the frontier (the minor agricultural region was the
only region with a positive average rate of technical efficiency change). However, looking at the land and labor productivity in-
dicators based on our dataset we see that the minor agricultural provinces have remained less productive when compared to the other
regions. Thus, the relative lack of technological progress poses a clear problem for further development of agriculture in the minor
provinces.
The patterns of TFP change across the three zones can be related to growth in agricultural output. The difference between output
growth and TFP growth represents output gains due to input increases. For China as a whole, output growth of 4.16% p.a. compared
to TFP growth of 3.05% p.a. suggests that much of the output growth was due to TFP gains. However, the results vary across the
regions. The largest gap between the growth rates of TFP and output is observed for the major agricultural zone (2.33% and 3.98%
p.a., respectively). This suggests the major agricultural producing provinces based their output growth about equally on increasing
the use of inputs and TFP gains. Therefore, expansion in agricultural output might be reduced in this zone in the future due to

9
Gong (2018) presented results graphically rather than in numerical tables so it isn't possible to report his exact finding.

13
Z. Shen, et al. China Economic Review 57 (2019) 101315

constraints on the agricultural inputs, particularly land and labor. The smallest agricultural producing provinces show the second
largest gap between the growth rates in TFP and output (2.91% and 4.08% p.a., respectively). In this case, the contributions of the
inputs use and TFP gains are more balanced. Finally, the medium agricultural producing provinces largely relied on productivity
gains—their annual average output growth of 4.59% p.a. is largely explained by TFP growth of 3.9% p.a. Thus, the provinces of
medium agricultural importance seem to have enjoyed the highest growth rates and increasing application of machinery and
agrochemicals raised land and labor productivity there.
Xu (2017) argued that there have been regional disparities in agricultural productivity that have persisted since the era of
collectivization. To quantify the legacy effects of collectivization, he constructed a collective legacy index (CLI) based on rural
infrastructure, education, and health under collectivization. Using data for 28 provinces from 1980 to 2008, he found that the CLI has
long-term effects on agricultural development following decollectivization. We used Xu's (2017) CLI at the provincial level to ex-
amine its relationship with our measures of TFP. We found a significant positive correlation between the CLI and technological
progress and a significant negative correlation between the CLI and scale efficiency. Following Xu's (2017) line of argument, pro-
vinces that performed well under collectivization have experienced experience lasting benefits in the form of higher rates of tech-
nological change because of superior infrastructure and human capital (as measured by the CLI) already in place under socialism.10
Because the proposed LHM is a generalized TFP indicator, we can also analyze the differences in the components of the LHM
indicator across the input- and output-orientations of the directional distance functions underlying the LHM. As can be seen from the
results shown in Table 2, at the national level the components of the LHM indicator differ little across orientations. However,
differences can be observed within and across zones, though neither orientation produces uniformly higher or lower outcomes. In the
case of technological progress, the mean annual growth rates are higher under the output-oriented than under the input-oriented case
for the medium agricultural producing provinces; however, the opposite holds for the major producing provinces. For the minor
provinces, the two orientations indicate essentially the same rates of growth. The difference across orientations is negligible for the
minor producing provinces. While the gains in technical efficiency are modest at the national level, there is considerable variation
across the three regions (the contribution of the technical efficiency change ranges between −0.21% and 0.45% at the regional
level), and across orientation, for this contributor to TFP growth.
Province-specific results, reported by zone, of the change in LHM productivity indicator and its components are presented in
Table 3. The average annual rate of change in TFP ranges from 6.17% p.a. for Yunnan to −1.88% p.a. for Tibet. The two provinces
with the highest average annual increases in TFP, Yunnan and Jilin, benefited primarily from improvements in scale efficiency, with
technological progress also contributing to TFP growth. Many of the provinces show productivity gains from both technical efficiency
change and technological progress, but the contributions of the scale efficiency change component are more uneven across the
provinces. Specifically, we found that the scale efficiency change component of TFP was negative for 11 of the 31 provinces, meaning
these 11 provinces moved away from the optimal scale size. In addition, 8 provinces had negative contributions to TFP from technical
efficiency change, while technological progress negatively impacted 3 provinces. The findings with respect to technical efficiency
change and scale efficiency change suggest that better management of input use and scale would be fruitful areas for improving the
agricultural TFP in China. The continued development of machinery rental and mechanization services, land rental, and land reform
hold promise for improving scale efficiencies.
Three provinces show negative rates of growth for the agricultural TFP over the sample period—Tibet (−1.88% p.a.), Hubei
(−1.71% p.a.), and Sichuan (−0.24% p.a.).11 These provinces have previously been identified in the literature (e.g., Chen et al.,
2008) as having TFP regress. However, each of these provinces displays a different pattern of the sources of their negative rates of
changes in TFP growth. Sichuan's slightly negative rate of TFP growth was driven by the scale efficiency change term (−1.59%)
which more that fully offset gains from technological progress (1.35% p.a.). For Hubei, the scale efficiency component (−1.81%) was
reinforced by a decline in technical efficiency change (−1.08%). For the worst performing province, Tibet, technological change of
−2.1% accounted for almost all of its TFP regress (the two efficiency change components were essentially zero for Tibet). Looking at
the dynamics in inputs and outputs may provide insights into the underlying causes of the decline in TFP in these three provinces. All
three provinces showed relatively high rates of growth in machinery use (as measured by machinery power available) that exceeded
their rates of growth in the agricultural output. As regards machinery use, Sichuan and Hubei had growth rates of 6.62% p.a. and
7.87% p.a., respectively, whereas the value for Tibet was 11.74% p.a. Obviously, these values exceed the region-specific averages
given in Table 1. In Tibet, the growth rates for the use of fertilizers and pesticides (3.90% p.a. and 4.22% p.a., respectively) also

10
Xu (2017) provided the following example to illustrate the relevance of the CLI to agricultural productivity. Hebei has a high CLI; 40% of its
sown land was irrigated as of 1980 while 50% was irrigated as of 2013. On the other hand, Guizhou has a low CLI; from 1980 to 2013 the percent of
its sown land under irrigation increased only slightly, from 16% to 17%.
11
It may seem odd that three provinces averaged negative productivity growth over a sample period covering 1997–2015. However, Zhang
(2011) noted that China's agricultural sector has experienced significant shortfalls from expected output that can't be explained by natural con-
ditions. He argued that while weather might account for short term deviations, it can't account for decades long shortfalls. He argued that ab-
normally large fluctuations were the result of structural policy changes. In the case of Tibet, both natural conditions and policy may be hindering its
agricultural productivity. On a high plateau, Tibet has limited arable land and temperatures that are too low for warm season crops. Fischer (2011)
noted that while Tibet's overall GDP grew rapidly from 2000 to 2008, agriculture grew much slower and its share of Tibet's GDP fell dramatically.
Fischer (2011, p. 63) argued that output growth in Tibet was the result of massive subsidies which turned Tibet into a “quintessential aid economy
par excellence” rife with inefficiencies. In addition, policies directed Tibet's labor out of agriculture. The combination of poor natural conditions and
policy initiatives may explain Tibet's negative average annual TFP growth over the sample period. More detailed analysis of the negative rates of TFP
growth, while beyond the scope of this study, would help to address provinces' lagging performances.

14
Z. Shen, et al. China Economic Review 57 (2019) 101315

Table 3
Annual average provincial growth rates of TFP indicator and its decomposition.
Region Province LHM TEC TP SEC

Major agricultural producers Anhui 3.84% −1.09% 2.87% 2.06%


Guangdong 2.96% 0.04% 3.97% −1.06%
Hebei 3.90% 0.31% 4.82% −1.23%
Heilongjiang 3.52% 0.01% 2.39% 1.12%
Henan 2.40% 0.00% 6.16% −3.77%
Hubei −1.71% −1.08% 1.18% −1.81%
Hunan 2.00% 0.07% 2.58% −0.65%
Jiangsu 3.93% 0.12% 4.91% −1.09%
Shandong 2.74% 0.00% 1.75% 0.99%
Sichuan −0.24% 0.00% 1.35% −1.59%
Medium agricultural producers Fujian 4.25% 0.08% 4.38% −0.21%
Guangxi 5.26% 0.87% 2.48% 1.91%
Inner Mongolia 0.50% −3.12% 1.50% 2.12%
Jiangxi 4.04% −0.24% 1.36% 2.92%
Jilin 5.68% −0.51% 2.71% 3.49%
Liaoning 2.94% 0.18% 2.69% 0.08%
Shaanxi 4.32% 0.87% 4.09% −0.65%
Xinjiang 1.08% 0.00% 5.21% −4.13%
Yunnan 6.17% −0.26% 1.45% 4.98%
Zhejiang 4.80% 0.05% 4.76% −0.01%
Minor agricultural producers Beijing 5.22% 0.00% 2.60% 2.62%
Chongqing 3.56% 0.85% −0.30% 3.01%
Gansu 3.95% 1.07% 0.91% 1.97%
Guizhou 1.24% −0.39% −0.43% 2.06%
Hainan 5.14% 1.25% 3.06% 0.84%
Ningxia 2.78% 0.28% 1.24% 1.27%
Qinghai 2.56% 1.13% 0.86% 0.57%
Shanghai 3.76% 0.00% 2.14% 1.62%
Shanxi 3.75% 0.80% 0.55% 2.39%
Tianjin 1.96% −0.05% 1.81% 0.20%
Tibet −1.88% 0.00% −2.10% 0.22%

Note: Provincial TFP and its components are calculated as per Eqs. (7) and (20).

exceeded the rate of growth in agricultural output (3.65% p.a.).


These trends indicate that Sichuan and Hubei adjusted their input structure but did not manage to substantially improve the
agricultural output as the input factors became less productive (if compared to the region of the production possibility set for their old
input-mixes). Natural disasters also contributed to the negative technical efficiency change (Chen et al., 2008). Therefore, the
agricultural practices need to be improved to ensure that the inputs are used efficiently; otherwise, the excessive amounts of inputs
need to be identified and reallocated (in this case, the machinery input appears to be the most problematic). The case of Tibet is more
complex as more inputs are involved in the undesirable pattern of the rates of growth. Therefore, the changes in the input-mix for
Tibet put it into a region of a production possibility set where its boundary moved inward. This suggests a need for improved
agricultural practices (e.g., better varieties and breeds allowing for higher yields, extension services, etc.). The heterogeneity of the
underlying technologies may also be a factor in this case; however, this topic falls beyond the scope of the present research.
As shown in Table 2, at the national level China experienced positive rates of growth for LHM TFP indicator and each of its three
components. Fig. 4 presents the trends of TFP growth and its components for China over the sample period 1997–2015. The cu-
mulative contribution of technological progress rose steadily; Fig. 4 reveals this component's importance to overall productivity
growth. The cumulative effect of scale efficiency change also increased steadily, but the gains it delivered paled in comparison to
technological progress. On the other hand, the cumulative contribution of the technical efficiency change was negative throughout
the period 1997–2015, although by 2015 its cumulative contribution was close to nil. For China overall we found an average annual
TFP growth rate of 3.05%; technological progress was 2.35% indicating that an outward movement of the technological frontier
provided much of the productivity gain in Chinese agriculture over the period 1997–2015. Our estimates of productivity change are
closer to those in previous studies that used non-parametric techniques (e.g., Li et al., 2017; Wang et al., 2013) as opposed to those
based on parametric methods (e.g., Zhang and Brümmer, 2011). However, when comparing our results with those of earlier studies,
there are three considerations to keep in mind. First, studies vary widely in terms of their specifications of production technology
making direct comparison difficult. Second, we applied the very general additively complete LHM TFP indicator, whereas earlier
studies utilized a variety of other (total factor) productivity indexes and indicators. Third, we provide disaggregated measures of TFP
with regard to different orientations (i.e., input contraction and output augmentation), which provides additional insights into
agricultural performance at the regional and provincial levels.
As seen in Table 3, there are differences in the rates of TFP growth, technological progress, technical efficiency change, and scale
efficiency change across China's 31 provinces. An important policy issue is whether the dispersion of agricultural TFP levels across
China's provinces has declined over time. To address this issue, we test for sigma-convergence—a narrowing of the dispersion of the

15
Z. Shen, et al. China Economic Review 57 (2019) 101315

60%
TFP
50%
TP

Cumulative TFP change


40%

30%

20%

10% SEC

0%
TEC

-10%
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
Fig. 4. Cumulative TFP and its components.

agricultural LHM TFP indicator across provinces over time. First, we compute the coefficients of variation (CVs) for the provinces
within each of the three zones for the years 2000–2015 (see Fig. 5). Second, we use ordinary least squares to the find the linear trends
of the CVs over time. The estimated coefficients and t-statistics of the annual growth rates are shown for each agricultural zone in
Table 4. The results are very similar for all zones, with decreases in the values of CVs indicating an increasing convergence across the
provinces within all three zones over the sample period—i.e., there is evidence of catching-up by the slower growing provinces. As
can be seen in Fig. 5, much of the convergence had occurred by 2004; after 2004, the major agricultural producing provinces did not
experience much further convergence, though the other two zones witnessed some further reductions in their respective CVs. Our
finding agrees with Zhan et al. (2017) who estimated a distance function-based stochastic frontier for agriculture using data for 29
Chinese provinces over the period 1986–2011; they found evidence of convergence in TFP growth. Zhan et al. (2017) found that R&D
as well as human capital had positive effects on convergence. Our results also comport with Li and Zhang (2013) who found evidence
of TFP convergence across regions based on a DEA-Malmquist approach applied to province level data covering the period
1985–2010. Tong, Fulginiti, and Sesmero (2012) calculated TFP using both an output-oriented DEA-Malmquist index and stochastic
frontier analysis. For provincial data covering 1993–2005, Tong et al. (2012) found evidence of TFP convergence over time with their
non-parametric Malmquist index, but they did not find evidence of TFP convergence with their SFA model. In contrast, using an
output-oriented Malmquist productivity index with provincial data from 1990 to 2003, Chen et al. (2008) found that regional
differences in productivity widened over time.

Fig. 5. Dynamics in coefficients of variation for cumulative TFP within regions.

16
Z. Shen, et al. China Economic Review 57 (2019) 101315

Table 4
Test of sigma convergence for trends of coefficients of variation displayed in Fig. 5.
Zone Coefficient t-value

Major agricultural producing provinces −6.05% −3.81


Medium agricultural producing provinces −6.20% −9.46
Minor agricultural producing provinces −11.47% −9.82

5. Conclusion

We have proposed a generalized, additively complete LHM productivity indicator based on both input- and output-oriented
directional distance functions. The proposed indicator is completely and exhaustively decomposable and thus can be applied to study
the dynamics of TFP growth across time and regions.
Annual data on China's agriculture sector covering 1997–2015 were used to illustrate the proposed LHM productivity indicator
and its decomposition using non-parametric models, which are presented. Employing a complete LHM productivity indicator, we
found differences in the results rendered by the input- and output-oriented measures, especially at the disaggregate (provincial) level.
The proposed additively complete LHM productivity indicator appears to be useful in delivering robust analysis of dynamics in TFP
growth and the underlying sources.
In pursuit of agricultural productivity growth, China has experimented with agricultural policy reform for decades (see, e.g.,
Brümmer et al., 2006). The importance of agricultural reform is evidenced by a series of “Number 1 central documents,” the first
policy statements made each year which establish government priorities, that have been devoted to agricultural reform. Over the
years, collectivism has given way to the household responsibility system; more recently, land management rights reform allows
villages to collectively transfer land to corporations in exchange for annual payments. The first change was intended to incentivize
farmers to be more productive; the latter change should enhance scale efficiency. The latter change should also enhance moder-
nization, since larger entities likely have better access to capital than small landholders. In addition, government investments have
been made in irrigation, seed development, land management, etc. Ito (2010) found that biochemical technological advances (e.g.,
more intensive use of intermediate inputs such as fertilizer and the adoption of new seed varieties) were a major contributor to TFP
growth12; the contribution of mechanization was smaller. Pricing mechanism reforms were undertaken as well, with the intent of
giving farmers a greater incentive to be more innovative and more efficient.
Some of China's policy most directly impact technological progress (e.g., new seed varieties), others will have a more direct
impact on technical efficiency (e.g., pricing reforms) or scale efficiency (e.g., land reform and the development of mechanization
services). We observe that technological progress has been the major driving force behind an increase in China's agricultural TFP by
3.05% p.a. year over the sample period 1997–2015. Continued technological innovations are needed to further push the production
frontier and enhance productivity. While the contributions of efficiency change to TFP growth were found to be relatively small,
enhancing scale efficiency with greater mechanization (perhaps through greater use of mechanization services and rental) and land
reform, and improving the efficiency of fertilizers, pesticides, and irrigation would provide a welcome boost to productivity growth.
Looking at the results for different groups of provinces, we note that the major and minor agricultural producing provinces saw lower
rates of TFP growth relative to the provinces in the medium agricultural zone. In addition, the major agricultural producing provinces
were hampered by a negative component of TFP growth—its average scale efficiency change component of TFP growth was −0.7%
p.a. Therefore, the issues of input usage, mechanization, and land reform are of particular importance in this zone. The negative
contributions of technical efficiency change in the major (−0.16% p.a.) and medium (−0.21% p.a.) agricultural producing provinces
are also a concern. Finally, the analysis of coefficients of variation suggests the presence of convergence in the cumulative growth in
TFP within each of the three regions considered in the analysis.
It should be noted that given concerns about the accuracy and consistency of the data contained in the China Statistical Yearbooks
results should be viewed with caution. In particular, our study used labor force in the primary sector as an input variable, whereas the
output variable refers specifically to agricultural output (the chief subsector of the primary sector). Thus, the results obtained may be
downward biased as the primary sector includes some additional activities (i.e., animal husbandry, fisheries, and forestry) that are
small relative to agriculture. In addition, our model does not allow for statistical noise, so any noise in the data will be reflected in the
estimates of TFP and its components. Finally, because our decomposition is exhaustive, an imperfect estimate of one component
necessarily results in an imperfect estimate of at least one other element of the decomposition. Future research could examine the
implication of these issues, perhaps using stochastic frontier analysis (which, as noted above, has its own shortcomings).

References

Ang, F., & Kerstens, P. J. (2017). Decomposing the Luenberger–Hicks–Moorsteen total factor productivity indicator: An application to U.S. agriculture. European

12
Recently Cui et al. (2018) reported that field experiments on enhanced management practices in China's major agricultural zones led to
increased grain output and the simultaneous reduction in the application of nitrogen fertilizers. The project, involving 21 million farmers, de-
monstrated that reducing an input can lead to an increases in output (i.e., an increase in technical efficiency), not to mention the environmental
benefits that would also result.

17
Z. Shen, et al. China Economic Review 57 (2019) 101315

Journal of Operational Research, 260, 359–375. https://doi.org/10.1016/j.ejor.2016.12.015.


Bjurek, H. (1996). The Malmquist total factor productivity index. The Scandinavian Journal of Economics, 98, 303–313. https://doi.org/10.2307/3440861.
Briec, W., & Kerstens, K. (2004). A Luenberger-Hicks-Moorsteen productivity indicator: Its relation to the Hicks-Moorsteen productivity index and the Luenberger
productivity indicator. Economic Theory, 23, 925–939. https://doi.org/10.1007/s00199-003-0403-2.
Brümmer, B., Glauben, T., & Lu, W. (2006). Policy reform and productivity change in Chinese agriculture: A distance function approach. Journal of Development
Economics, 81, 61–79. https://doi.org/10.1016/j.jdeveco.2005.04.009.
Caves, D. W., Christensen, L. R., & Diewert, W. E. (1982). The economic theory of index numbers and the measurement of input, output, and productivity.
Econometrica, 50, 1393–1414. https://doi.org/10.2307/1913388.
Cechura, L., Kroupova, Z., & Rudinskaya, T. (2015). Factors determining TFP changes in Czech agriculture. Agricultural Economics, 61, 543–551. https://doi.org/10.
17221/14/2015-AGRICECON.
Chambers, R. (1988). Production economics. Cambridge, UK: Cambridge University Press.
Chambers, R. G. (2002). Exact nonradial input, output, and productivity measurement. Economic Theory, 20, 751–765. https://doi.org/10.1007/s001990100231.
Chambers, R. G., Chung, Y., & Färe, R. (1996). Benefit and distance functions. Journal of Economic Theory, 70, 407–419. https://doi.org/10.1006/jeth.1996.0096.
Chen, J. D., Cheng, S. L., Nikic, V., & Song, M. L. (2018). Quo vadis? Major players in global coal consumption and emissions reduction. Transformations in Business &
Economics. Vol. 17. Transformations in Business Economics (pp. 112–132).
Chen, P. C., Yu, M. M., Chang, C. C., & Hsu, S. H. (2008). Total factor productivity growth in China's agricultural sector. China Economic Review, 19, 580–593. https://
doi.org/10.1016/j.chieco.2008.07.001.
China Statistical Yearbook. http://www.stats.gov.cn/tjsj/ndsj/2015/indexeh.htm (Accessed 20 May 2019).
Chow, G. (2006). Are Chinese official statistics reliable? CESifo Economic Studies, 52, 396–414. https://doi.org/10.1093/cesifo/ifl003.
Coelli, T. J., & Rao, D. S. (2005). Total factor productivity growth in agriculture: a Malmquist index analysis of 93 countries. 1980–2000. Agricultural Economics, 32,
115–134. https://doi.org/10.1111/j.0169-5150.2004.00018.x.
Cui, Z., et al. (2018). Pursuing sustainable productivity with millions of smallholder farmers. Nature, 555, 363–366. https://doi.org/10.1038/nature25785.
Debreu, G. (1951). The coefficient of resource utilization. Econometrica, 19, 273–292. https://doi.org/10.2307/1906814.
Diewert, W. E., & Fox, K. J. (2014). Reference technology sets, free disposal hulls and productivity decompositions. Economics Letters, 122, 238–242. https://doi.org/
10.1016/j.econlet.2013.11.026.
Diewert, W. E., & Fox, K. J. (2017). Decomposing productivity indexes into explanatory factors. European Journal of Operational Research, 256, 275–291. https://doi.
org/10.1016/j.ejor.2016.05.043.
Farrell, M. J. (1957). The measurement of productive efficiency. Journal of the Royal Statistical Society, Series A (General), 120, 253–290. https://doi.org/10.2307/
2343100.
Fischer, A. M. (2011). The great transformation of Tibet? Rapid labor transitions in times of rapid growth in the Tibet autonomous region. HIMALAYA, The Journal of
the Association for Nepal and Himalayan Studies, 30, 14.
Gong, B. L. (2018). Agricultural reforms and production in China: Changes in provincial production function and productivity in 1978–2015. Journal of Development
Economics, 132, 18–31. https://doi.org/10.1016/j.jdeveco.2017.12.005.
Holz, C. A. (2013). Chinese statistics: Classification systems and data sources. Eurasian Geography and Economics, 54, 532–571. https://doi.org/10.1080/15387216.
2014.908312.
Huang, J., Gao, L., & Rozelle, S. (2012). The effect of off-farm employment on the decisions of households to rent out and rent in cultivated land in China. China
Agricultural Economic Review, 4, 5–17. https://doi.org/10.1108/17561371211196748.
Huang, J., Wang, X., & Rozelle, S. (2013). The subsidization of farming households in China's agriculture. Food Policy, 41, 124–132. https://doi.org/10.1016/j.foodpol.
2013.04.011.
Huang, J., & Yang, G. (2017). Understanding recent challenges and new food policy in China. Global Food Security, 12, 119–126. https://doi.org/10.1016/j.gfs.2016.
10.002.
Ito, J. (2010). Inter-regional difference of agricultural productivity in China: Distinction between biochemical and machinery technology. China Economic Review, 21,
394–410. https://doi.org/10.1016/j.chieco.2010.03.002.
Kerstens, K., Shen, Z., & Van de Woestyne, I. (2018). Comparing Luenberger and Luenberger-Hicks-Moorsteen productivity indicators: How well is total factor
productivity approximated? International Journal of Production Economics, 195, 311–318. https://doi.org/10.1016/j.ijpe.2017.10.010.
Koopmans, T. C. (1951). An analysis of production as an efficient combination of activities. In T. C. Koopmans (Ed.). Activity analysis of production and allocation,
proceeding of a conference (pp. 33–97). London: John Wiley and Sons, Inc.
Li, N., Jiang, Y., Yu, Z., & Shang, L. (2017). Analysis of agriculture total-factor energy efficiency in China based on DEA and Malmquist indices. Energy Procedia, 142,
2397–2402.
Li, Z., & Zhang, H. (2013). Productivity growth in China's agriculture during 1985–2010. Journal of Integrative Agriculture, 12, 1896–1904. https://doi.org/10.1016/
S2095-3119(13)60598-5.
Ma, S. Z., & Feng, H. (2013). Will the decline of efficiency in China's agriculture come to an end? An analysis based on opening and convergence. China Economic
Review, 27, 179–190. https://doi.org/10.1016/j.chieco.2013.04.003.
Machek, O., & Špička, J. (2013). Measuring performance growth of agricultural sector: A total factor productivity approach. International Journal of Economics and
Statistics, 1, 200–208.
National Bureau of Statistics of China (2017). China's official statistical database is available from. http://data.stats.gov.cn/english/index.htm.
O'Donnell, C. J. (2012a). An aggregate quantity framework for measuring and decomposing productivity change. Journal of Productivity Analysis, 38, 255–272. https://
doi.org/10.1007/s11123-012-0275-1.
O'Donnell, C. J. (2012b). Nonparametric estimates of the components of productivity and profitability change in US agriculture. American Journal of Agricultural
Economics, 94, 873–890. https://doi.org/10.1093/ajae/aas023.
Plekhanov, D. (2017). Quality of China's official statistics: A brief review of academic perspectives. Copenhagen Journal of Asian Studies, 35, 76–101. https://doi.org/10.
22439/cjas.v35i1.5400.
Rawski, T. G., & Xiao, W. (2001). Roundtable on Chinese economic statistics introduction. China Economic Review, 12, 298–302. https://doi.org/10.1016/S1043-951X
(01)00074-8.
Shen, Z., Baležentis, T., Chen, X., & Valdmanis, V. (2018). Green growth and structural change in Chinese agricultural sector during 1997–2014. China Economic
Review. https://doi.org/10.1016/j.chieco.2018.04.014.
Shephard, R. W. (1953). Cost and production functions. Princeton, NJ: Princeton University Press.
Song, M.-L., Fisher, R., Wang, J.-L., & Cui, L.-B. (2016). Environmental performance evaluation with big data: Theories and methods. Ann. Oper. Res.. https://doi.org/
10.1007/s10479-016-2158-8.
Song, W., Han, Z., & Deng, X. (2016). Changes in productivity, efficiency and technology of China's crop production under rural restructuring. Journal of Rural Studies,
47, 563–576. https://doi.org/10.1016/j.jrurstud.2016.07.023.
Tian, X., & Yu, X. (2012). The enigmas of TFP in China: A meta-analysis. China Economic Review, 23, 396–414. https://doi.org/10.1016/j.chieco.2012.02.007.
Tong, H., Fulginiti, L., & Sesmero, J. (2012). Agricultural productivity in China: National and regional growth patterns, 1993-2005. In K. O. Fuglie, V. E. Ball, & S. L.
Wang (Eds.). Productivity growth in agriculture: An international perspective (pp. 163–177). Cambridge, MA: CABI.
Wang, Q., & Zhang, X. (2017). Three rights separation: China's proposed rural land rights reform and four types of local trials. Land Use Policy, 63, 111–121.
Wang, S. L., Tuan, F., Gale, F., Somwaru, A., & Hansen, J. (2013). China's regional agricultural productivity growth in 1985–2007: A multilateral comparison.
Agricultural Economics, 44, 241–251. https://doi.org/10.1111/agec.12008.
Wang, X., Yamauchi, F., Otsuka, K., & Huang, J. (2016). Wage growth, landholding, and mechanization in Chinese agriculture. World Development, 86, 30–45. https://
doi.org/10.1016/j.worlddev.2016.05.002.

18
Z. Shen, et al. China Economic Review 57 (2019) 101315

World Bank. World bank open data. (2019). https://data.worldbank.org/ (Accessed 15 April 2019).
Wu, Y. R. (2011). Total factor productivity growth in China: A review. Journal of Chinese Economic and Business Studies, 9, 111–126. https://doi.org/10.1080/
14765284.2011.568682.
Xu, Z. (2017). Decollectivization, collective legacy, and uneven agricultural development in China. World Development, 98, 290–299. https://doi.org/10.1016/j.
worlddev.2017.04.037.
Zhan, J., Xu, T., Zhang, Y., Yang, X., Qu, Z., & Tan, T. (2017). The effects of agricultural R&D on Chinese agricultural productivity growth: New evidence of
convergence and implications for agricultural R&D policy. Canadian Journal of Agricultural Economics, 65, 453–475. https://doi.org/10.1111/cjag.12137.
Zhang, J. (2011). China's success in increasing per capita food production. Journal of Experimental Botany, 62, 3707–3711. https://doi.org/10.1093/jxb/err132.
Zhang, X., Yang, J., & Reardon, T. (2017). Mechanization outsourcing clusters and division of labor in Chinese agriculture. China Economic Review, 43, 184–195.
https://doi.org/10.1016/j.chieco.2017.01.012.
Zhang, Y., & Brümmer, B. (2011). Productivity change and the effects of policy reform in China's agriculture since 1979. Asian-Pacific Economic Literature, 25, 131–150.
https://doi.org/10.1111/j.1467-8411.2011.01307.x.
Zhao, Q. R., Chen, Q. H., Xiao, Y. T., Tian, G. Q., Chu, X. L., & Liu, Q. M. (2017). Saving forests through development? Fuelwood consumption and the energy-ladder
hypothesis in rural southern China. Transformations in Business & Economics. Vol. 16. Transformations in Business Economics (pp. 199–219).
Zheng, J. (2001). China's official statistics: Growing with full vitality. China Economic Review, 12, 333–337. https://doi.org/10.1016/S1043-951X(01)00068-2.

19

S-ar putea să vă placă și