Structural Change and Economic Dynamics 41 (2017) 64–77

Contents lists available at ScienceDirect

Structural Change and Economic Dynamics

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

Structural change and economic growth: Empirical evidence and

policy insights from Asian economies
K.M. Vu
National University of Singapore, 469C Bukit Timah Road, 259772, Singapore

a r t i c l e i n f o a b s t r a c t

Article history: This paper introduces a new measure of structural change labeled the effective structural change (ESC)
Received 29 February 2016 index, and applies it to study the effects of structural change on economic growth, using a sample of
Received in revised form 12 February 2017 19 Asian economies for the period from 1970 to 2012. This new approach provides an effective tool to
Accepted 14 April 2017
examine growth effects of structural change. The paper’s findings suggest the importance of reforms to
Available online 17 April 2017
foster productivity-enhancing structural change and need for strategies to overcome their short-term
costs. The paper also indicates that ESC can be used as a useful indicator for monitoring the impacts of
JEL classification:
structural reforms.
O40
© 2017 Elsevier B.V. All rights reserved.
O57

Keywords:
Structural change
Economic growth
Productivity
Asia
Policy

1. Introduction techniques, confirms a positive link between sectoral composition

Structural change defined as the reallocation of productive
resources among sectors in the economy is a prominent feature
of economic growth. The important role of structural change in
driving economic growth and productivity improvement has been
empirically supported by influential studies such as Lewis (1954),
Clark (1957), Kaldor (1966), Kuznets (1966), Kuznets (1979),
Denison (1967), Cheery et al. (1975), Syrquin (1988), Lin and Monga
(2010), and Lin (2009), Lin (2012a), Lin (2012b). The expected
nature of structural change dynamics is the continual shift of factor
inputs from lower to higher productivity sectors, which conse-
quently raises productivity at the aggregate level. Lewis (1954) uses
a classical framework of dual economy to provide insights into the
following dynamic: the shift of surplus labor from subsistence agri-
culture towards the modern sector increases worker productivity,
a country’s overall productivity, and output per capita.1 Echevarria
(1997), employing general equilibrium methods and simulation
E-mail address: sppkmv@nus.edu.sg
1
Lewis (1954) also points out that this pattern of growth will reach a turning
point as the surplus of labor is exhausted. Then, the modern sector would need to
raise wages to attract labor from agriculture for further expansion; and hence it may
be more challenging for the economy to sustain previous high rates of growth.
change and growth.
Empirical evidence on the effect of structural change on growth
has been found in a number of previous studies. Using growth
decomposition methods, Denison (1967: 319–322) shows that real-
location of productive inputs from agriculture to other sectors was
a significant factor explaining why the United States outperformed
the UK but was behind Germany in GDP growth between 1950 and
1962. Employing regression techniques to analyze OECD countries,
Kaniovski and Peneder (2002) and Dietrich (2012) provide further
evidence that structural change plays an important role in driv-
ing economic growth. Caselli and Coleman (2001), examining the
growth dynamics of the U.S. states, evidence that structural trans-
formation is a main factor driving the U.S. regional convergence.
With regard to Asia, van Ark and Timmer (2003) show that resource
reallocation from agriculture towards other sectors is a power-
ful source of growth for lower income countries, while for more
advanced economies the shift of labor towards services sectors such
as finance has a notable contribution to overall productivity growth.
Fan et al. (2003) find the essential role of sectoral composition
change in China’s economic growth.
Structural change, however, is not always found to be growth-
enhancing. For example, McMillan et al. (2014) show that, unlike in
Asia, the contribution of structural change to productivity growth

http://dx.doi.org/10.1016/j.strueco.2017.04.002
0954-349X/© 2017 Elsevier B.V. All rights reserved.
 K.M. Vu / Structural Change and Economic Dynamics 41 (2017) 64–77
was negative for Latin America during the period 1990–2005 and
for Africa during 1990–2000.
There are also studies contending that structural change may
not be conducive to productivity growth. Baumol (1967) shows that
labor may shift from a sector with higher and rapidly-growing pro-
ductivity to sector with a lower and stagnant productivity, which
causes a decline in the overall economy’s productivity growth
rate, ceteris paribus. The case of rapid expansion of the lower-
productivity service employment in the US can serve as a piece
of evidence Baumol et al. (1985). Furthermore, Ngai and Pissarides
(2007) introduce a model that suggests that the effect of structural
change may not show up in aggregate growth; while using a theo-
retical approach with restrictive assumptions, Meckl (2002) asserts
that sectoral composition change can be a byproduct of economic
growth and may have no feedback effect on the growth process. At
the sector level, Fagerberg (2000), examining manufacturing indus-
tries from a sample of 39 countries over period 1973–1990, finds
that structural change does not contribute to productivity growth.
Likewise, Timmer and Szirmai (2000) arrive at a similar conclusion
for Asian manufacturing.2
The discussions above calls for additional studies that provide
not only more conclusive evidence on the effect of structural change
on economic growth, but also a deeper understanding of the nature
of structural change and the mechanism through which structural
change influences growth. This paper aims to make contribution in
this direction by introducing a new approach to measure structural
change labeled as “effective structural change” (ESC) and use the
panel data of 19 Asian economies over the period from 1970 to
2012 to examine the effect of ESC on growth.
Among its main findings, the paper shows that ESC has a robust
positive effect on productivity, wage, and GDP growth; while its
effect on employment is negative and sizable in the short-term and
insignificant in the long-term. These findings reveal several impor-
tant policy insights. In particular, fostering productivity-enhancing
structural change is an effective way to promote economic growth.
However, the short-term cost of this process in term of rising unem-
ployment could be formidable, which policy makers may hesitate
to accept. This explains why structural reforms tend to be sluggish
in many countries.
The paper proceeds as follows. Section 2 provides an overview
of the dataset. Section 3 introduces the ESC measure and highlights
its patterns. Section 4 investigates the causal link between ESC and
economic performance, and discusses the empirical results. Section
5 makes concluding remarks with a brief policy discussion.
2. Overview of data and salient facts
This paper uses data from the Asian Productivity Organization
(APO) dataset,3 which covers 19 Asian economies over 43 years
from 1970 to 2012. The dataset, which is compiled based pri-
marily on the System of National Accounts (SNA),4 covers nine
sectors(under the international standard industrial classification
provided in Appendix A):(1) Agriculture; (2) Mining; (3) Manu-
2
Silva and Teixeira (2008), Krüger (2008), and Herrendorf et al. (2014) pro-
vide excellent reviews of major studies on the link between structural change and
growth.
3
This dataset is a joint research project between the APO and the Keio Economic
Observatory of Keio University, Tokyo (for details, see Nomura and Lau, 2014). The
salient advantage of this project is its close collaboration with the national statistical
agencies from the Asian economies. The online version of the dataset is available
athttp://www.apo-tokyo.org/wedo/measurement, accessed September 10, 2015.
4
The System of National Accounts 1993 was adopted by the United Nations Sta-
tistical Commission in 1993 as the international standard for compilation of national
accounts statistics and for the international reporting of comparable national
accounting data (source: http://unstats.un.org/unsd/nationalaccount/sna1993.asp).
65
facturing; (4) Utility; (5) Construction; (6) Trade and hotels; (7)
Transport and communications; (8) Finance, real estate, and busi-
ness services; and (9) Government and community services. Note
that the last four sectors ((6), (7), (8), and (9)) are usually combined
into the aggregate “services” sector in many datasets.
A snapshot of the 20 Asian economies is provided in Table 1
and Table 2.5 Table 1 highlights the key indicators of economic
performance,6 while Table 2 reports the simple measure of struc-
tural change in terms of changing employment share by sector
in the economy. From Table 1, three observations stand out.
First, labor productivity growth plays a major role in driving GDP
growth7 in most economies, especially in the economies with GDP
growth of as high as 5% or above. Labor productivity growth was
notably below employment growth only in five countries – Brunei,
the Philippines, Bangladesh, Nepal, and Pakistan, for all of which
GDP growth was well below 5%. This observation suggests that
robust average labor productivity (ALP) growth is an important
condition for an economy to achieve high GDP growth.
Second, for all countries, wage growth was generally equiva-
lent to ALP growth. However there is some distinction between
groups of countries. In the advanced economies, ALP grew faster
than wages (with the exception of South Korea, for which both
ALP and wage growth were both very strong). In most developing
countries, however, wages grew faster than ALP. The slower wage
growth compared to ALP in the Philippines, India, Bangladesh, and
Nepal is likely a result of high unemployment rates associated with
rapid population growth.
Third, total factor productivity (TFP) growth is positive for all
the countries in this dataset. The share of TFP growth in GDP
growth ranges from 15% to 35% for most countries. This share,
however, is notably low (below 10%) for Singapore (TFP growth
of 0.4% compared to GDP growth of 6.9%), Malaysia (0.5% vs. 6.3%),
the Philippines (0.3% vs. 4.1%), and Bangladesh (0.3% vs. 3.8%). This
implies that the share of TFP in GDP growth over a given period can
be low not only in low-performing economies but also in high-
performing economies. For low-performing economies, the low
share of TFP is due to stagnation in efficiency improvement and
slow technology progress. For high-income nation, the low share of
TFP can be explained by significant capital accumulation and rapid
employment expansion, which may have lowered the efficiency
use of these productive inputs.
In addition to employment share change by sector for each indi-
vidual economy over the period 1970–2012,8 Table 2 describes
employment share change by sector for the 19 countries under
investigation over period 1970–2012.9 The following observa-
tions are notable from the Table. First, the employment share
of the agriculture sector shrank in all economies. This contrac-
tion was largest for Korea (−44.1% points), China (−37.1), and
Thailand (−37.1). Developing economies with small contraction in
5
The dataset covers 20 Asian economies. However, the sector-level data needed
for structural change analysis is not available for Laos. Therefore, only 19 remaining
economies will be examined in this study.
6
In Table 1, the average labor productivity (ALP) and average wage are calculated
as GDP and total labor compensation, respectively, divided by the number of work-
ers. The ALP and average wage will be simply referred to as labor productivity and
wage, correspondingly.
7
Note GDP growth is equal to ALP growth plus employment growth.
8
Data on this measure is available only until 2012.
9
For a given economy, the Norm of Absolute Values (NAV) indexis calculated for

n
period [0,T] as follows: NAV = 0.5 ∗ |SkT − Sk0 | Where n is the number of sectors
k=1
in the economy; Sk0 and SkT represent the employment share of sector k in time 0
and T, respectively. The 0.5 factor is to correct the double count of employment share
changes. This measure is also called. More details can be found in Dietrich (2012).
66 K.M. Vu / Structural Change and Economic Dynamics 41 (2017) 64–77

Table 1
Selected Statistics on the 20 Asian Economies.

Country Average Annual Growth, 1970–2012* Level in 2012 Population in

2012
(Million)

GDP GDP per capita ALP TFP Average Wage Employment GDP per capita ALP

China-India
China 8.6% 7.5% 6.8% 3.0% 6.8% 1.9% 11,700 20,682 1,360.7
India 5.3% 3.4% 3.0% 1.4% 2.8% 2.3% 5194 13,388 1,240.9

Advanced Economies
Japan 2.6% 2.1% 2.2% 0.6% 1.8% 0.4% 36,340 71,454 127.3
Hong Kong 5.5% 4.2% 3.5% 1.6% 2.8% 2.0% 51,400 99,897 7.2
Singapore 6.9% 4.7% 3.1% 0.4% 2.2% 3.8% 77,589 124,942 5.4
Korea, Rep. 6.8% 5.7% 4.5% 1.6% 5.9% 2.2% 30,852 61,812 50.2
Taiwan 6.3% 5.3% 4.3% 1.5% 3.9% 2.0% 41,214 87,838 23.4

ASEAN
Brunei 2.3% -0.3% -1.4% ... ... 3.7% 73,812 157,830 0.4
Cambodia 7.0% 4.7% 3.8% ... 5.8% 3.2% 3,126 5164 14.7
Indonesia 5.9% 4.1% 3.2% 0.9% 5.1% 2.6% 9,679 20,943 246.7
Lao PDR 6.3% 4.2% 3.9% ... ... 2.4% 4,772 9,185 6.6
Malaysia 6.3% 3.9% 3.1% 0.5% 3.4% 3.2% 21,884 49,612 29.9
Myanmar 6.0% 4.6% 4.1% ... ... 1.9% 4,739 7,677 49.2
Philippines 4.1% 1.8% 1.3% 0.3% 1.2% 2.8% 6,503 16,648 97.2
Thailand 5.6% 4.0% 3.5% 1.9% 4.3% 2.0% 15,160 25,787 66.5
Vietnam 5.7% 4.0% 2.9% 1.7% 6.8% 2.8% 5,265 9,047 89.7

South Asia
Bangladesh 3.8% 2.0% 1.4% 0.3% -0.4% 2.4% 2,581 6,643 153.8
Nepal 4.2% 2.2% 1.8% ... 1.3% 2.5% 2,587 5,212 27.1
Pakistan 4.7% 2.1% 2.0% 1.4% 5.3% 2.6% 4,501 14,902 183.1
Sri Lanka 5.0% 3.8% 3.2% 1.9% 4.4% 1.7% 9,646 25,980 20.5

Data Source: APO. Notes: * The period may vary slightly for some countries and variables due to availability of data.

Table 2
Change in Employment Share by Sector (in % points), 1970–2012.

Country Total Change* Agricul-ture Mining Manufac-turing Construc-tion Utility Trade Transport Finance Gov.& social Total services
services

China-India
China 37.4 −37.1 -0.3 5.7 7.7 0.3 8.4 1.8 0.4 13.0 23.7
India 20.3 −20.3 0.0 1.9 7.8 0.0 5.5 2.4 1.6 1.2 10.6

Advanced economies
Japan 25.0 −14.8 −0.4 −9.8 0.0 0.2 2.0 1.2 1.5 20.1 24.8
Hong Kong 42.1 −1.3 ... −40.8 1.9 0.0 11.8 4.7 14.5 9.1 40.1
Singapore 9.1 −2.1 ... −4.2 0.7 −0.4 1.3 0.2 6.9 -2.3 6.1
Korea, Rep. 45.3 −44.1 −1.2 4.0 4.3 0.0 6.0 3.5 11.8 15.8 37.0
Taiwan 24.9 −19.9 −0.8 −3.1 0.2 −0.1 8.7 −0.9 7.5 8.4 23.7

ASEAN
Brunei 14.9 −9.2 −3.8 1.8 −0.6 −0.8 10.2 −0.5 2.9 0.0 12.6
Cambodia 23.4 −23.4 0.1 8.5 2.5 0.2 5.5 1.0 0.4 5.2 12.1
Indonesia 30.7 −30.7 1.4 7.2 4.2 0.1 9.2 2.1 2.2 4.3 17.8
Lao PDR ... ... ... ... ... ... ... ... ... ... ...
Malaysia 20.2 −18.6 −0.4 2.0 2.0 0.5 7.8 2.3 5.8 -1.3 14.6
Myanmar 7.4 −5.1 0.1 1.3 1.0 0.0 −0.2 −0.7 −1.4 4.9 2.6
Philippines 25.5 −21.8 0.2 −3.7 1.9 0.1 14.9 3.6 2.4 2.4 23.2
Thailand 37.2 −37.1 0.0 8.7 5.0 0.1 13.7 1.0 2.4 6.2 23.3
Vietnam 23.6 −23.3 −0.4 5.9 3.7 0.2 8.8 1.0 0.6 3.5 13.9

South Asia
Bangladesh 18.4 −16.4 0.0 4.0 3.6 0.0 5.2 3.9 1.6 −1.9 8.9
Nepal 21.6 −21.6 0.2 5.5 3.0 0.9 6.7 1.7 0.8 2.9 12.0
Pakistan 15.2 −13.1 0.0 −2.1 2.9 0.1 6.1 0.3 0.5 5.3 12.1
Sri Lanka 25.0 −24.9 0.7 9.2 3.9 −0.1 4.9 0.8 2.9 2.6 11.2

Data Source: APO. Note: ∗ the total magnitude change in employment share by sector in the economy, correcting for double counting. This measure is known as Norm of
Absolute Values (NAV), which is defined in Section 3.

the agriculture sector’s employment share10 were Myanmar (−5.1),
Pakistan (−13.1), and Bangladesh (−16.4), which were among
low-performing countries. The slow reallocation of labor from
10
Note that the agriculture sector’s employment share is insignificant for two city-
states, Hong Kong and Singapore.
agriculture to other sectors observed for these three economies
is likely an important factor contributing to their lagging GDP
growth. Second, while the employment share of the manufactur-
ing sector expanded in developing countries (with the exception
of the Philippines (−3.7% points) and Pakistan (−2.1)), it shrank
in advanced economies (with the exception of South Korea (+4%
points)). This reflects the typical inverted U-shape pattern of the
 K.M. Vu / Structural Change and Economic Dynamics 41 (2017) 64–77
manufacturing sector’s growth: expanding in lower-income coun-
tries (industrialization) and shrinking in higher-income nations
(de-industrialization). Third, the employment share of the total
services sector, which consists of four services sectors (Trade,
Transport, Finance, and Government/Social Services), expanded in
all economies. For all the economies, the magnitude of employment
share change was much larger in the aggregate services sector than
in the manufacturing sector.11 This indicates the importance of the
services sectors in job creation and facilitation of structural change.
Lastly, the simple structural change measure is not sophisticated
enough to capture the degree to which this change contributed
to growth. For example, the Philippines, on average, outperformed
Vietnam, Malaysia, and India on this measure for 1970–2012; how-
ever, it was well behind these three countries on economic growth.
3. Effective structural change (ESC) and patterns
This section introduces a new measure of structural change,
hereafter referred to as effective structural change (ESC). This mea-
sure is built on the two existing approaches – the shift-share
method and the Norm of Absolute Values (NAV) index, both of
which are widely used for investigating the pace of structural
change and its effect on economic growth. The paragraphs below
briefly summarize these two methods, in order to establish the
context for introducing the ESC measure.
3.1. The shift-share method
This analysis is often used for capturing the contribution
of structural change to labor productivity growth in a given
economy.12 The method decomposes aggregate labor productivity
growth in a multi-sector economy into the contribution of techno-
logical progress and structural change effects. A typical shift-share
decomposition framework of labor productivity growth over period
[0, T ] is derived from the following equation (for example, see
Fagerberg, 2000):
P  S P  P S  S P
n n n
i0 i i0 i i i
= + + (1)
P0 P0 P0 P0
i=1 i=1 i=1
Where subscripts i and 0 indicate sector i and the initial year; P
denotes the level of labor productivity computed as the value-
added13 divided by the number of workers; S represents the
employment share by sector in the economy;  in front of a variable
indicates its change over period [0,T].
The three terms on the right-hand side of Eq. (1) can be inter-
preted as the three distinctive sources of ALP growth as follows14 :

n
S
i0 Pi
(i) The first term ( P0
) captures the contribution of within-
i=1
sector productivity improvement. This term is also referred
to as within-effect or the “technological progress” effect. This
11
Jorgenson and Timmer (2011) show that deeper analysis of individual service
sectors is important to better understand the process of economic growth and struc-
tural change.
12
For example, see Syrquin (1984), Peneder (2002), Van Ark and Timmer (2003),
Felipe et al. (2009), McMillan and Rodrik (2011), McMillan al. (2014). This analy-
sis can also be conducted for a multi-industry sector such as manufacturing. For
example, see Fagerberg (2000), Timmer and Szirmai (2000).
13
The value-added is measured in constant price. The shift-share analysis implic-
itly assumes that value-added generated by each sector has the same price deflator
as the entire economy’s GDP.
14
Most studies in this stream employ this approach; for example, see Fagerberg
(2000), Timmer and Szirmai (2000), Felipe et al. (2009), and Timmer et al. (2015).
67
term is usually positive because labor productivity in most
sectors tends to grow over time.

n
P i0 Si
(ii) The second term ( P0
) captures the contribution of real-
i=1
location of employment among sectors, keeping the initial
level of labor productivity of each sector constant. This term
is can be referred to as the “static structural effect.” This term
is positive if workers move to sectors with above-average pro-
ductivity.

n
S P i i
(iii) The third term ( P0
) represents the joint effect of
i=1
employment reallocation and productivity growth. This term
can be called as “dynamic structural effect”. For sector i, Si Pi
is positive if its productivity growth is positive and its labor
share is expanding or its productivity growth is negative and
its labor share is shrinking. The third term therefore is positive
if workers tend to shift from productivity declining sectors to
productivity growing sectors.
The shift share method can be modified to make the analy-
sis more straightforward, decomposing labor productivity growth
into only two sources: within-sector productivity improvement
(within-effect) and between-sector labor reallocation (between-
effect). A typical framework for this approach is to average the
levels of employment share and labor productivity over the period
[0,T] to yield the identity below15 :
P  S P  P S
n n
i i i i
= + (2)
P0 P0 P0
i=1 i=1
Where the bar on the top of a variable denotes its average value
over the period [0,T]; while other notations are similar to those in
Eq. (1).
Although the shift-share method provides a convenient way
to quantify the contribution of labor reallocation to productivity
growth, its results may be problematic due to its assumption that
productivity growth within each sector is independent of structural
change (Timmer and de Vries, 2009). Likely due to this unjustified
assumption, a number of studies using the shift-share method do
not find significant evidence confirming the positive contribution of
structural change to growth. For example, Timmer and Vries (2009)
find that growth accelerations in 19 countries in Asia and Latin
America over the period from 1950 to 2005 are explained by within-
sector productivity growth, not by reallocation of employment
to more productive sectors; while Fagerberg (2000), examining a
sample of 39 countries between 1973 and 1990, finds that the con-
tribution of labor reallocation to productivity growth was negative
for most countries.
It is important to overcome the limitation of the shift-share
approach in its unjustified assumption about the independence
between the “within-effect” and “between-effect” to productivity
growth. One way to address this issue is to combine the two effects
for each sector and to consider this combined effect as its structural-
15
This approach has been employed by several previous studies. For example, Tim-
mer and Vries (2009) use this framework to analyze the contribution of structural
change to productivity growth in Asian and Latin American economies.
68 K.M. Vu / Structural Change and Economic Dynamics 41 (2017) 64–77

related contribution to productivity growth. The combined effect 3.3. The Effective Structural Change (ESC) index
can be quantified from rewriting Eq. (2) as follows16 :

 n   
n
The Effective Structural Change (ESC) Index proposed by this
P S̄i Pi P̄ Si paper is built on the strengths of the shift-share method and the
= + i = Ci (3)
P0 P0 P0 NAV index presented above. The ESC index is computed as follows:
i=1 i=1  
S̄i Pi P̄i Si
ESC = 0.5 ∗ |SiT − Si0 |X = i such that Ci > 0 (5)
Where, Ci = + P0
is the total contribution of sector i to
P0 i∈X
the overall economy’s productivity growth. This total contribu-
S̄i Pi Where n, Si0 , SiT , and the subscripts are the same as in the NAV
tion consists of components: within-effect ( P0
) and between (or
index defined above. X is the set of sectors i such that Ci > 0; That is,
P̄ S
“structural change”) effect ( iP i ).
For a given sector, the condition only the sectors with a positive contribution to labor productivity
0
Ci > 0 is satisfied in the following scenarios: growth are accounted for in the computation of ESC.
By definition, the ESC index has two main advantages. First, by
(i) The sector’s productivity is growing and its employment share examining the overall contribution instead of its two components
is expanding. In this scenario, the sector is booming, which may (within-effect and between-effect) separately, the ESC measure
be driven by rapid technological progress, economic reforms, overcomes the limitation of the shift-share method’s unjusti-
and/or a surge in foreign direct investment. fied assumption that these two effects are independent of each
(ii) The sector’s productivity is growing and its employment share other. Second, the ESC accounts only for productivity-enhancing
is shrinking, while the effect of the former outweighs that of the structural change, eliminating productivity-decreasing structural
latter. This scenario can be observed for sectors experiencing adjustment from the NAV.
significant restructuring; The paragraphs below provide some observations about the
(iii) The sector’s productivity is declining and its employment share salient patterns of ESC and its related measures. Fig. 1 graphs the
is expanding, while the effect of the latter is greater than that ESC and NAV indexes18 by individual country along the time dimen-
of the former. This scenario can be observed for the sectors that sion. It shows that ESC is significantly below NAV in most years for
enjoy a rapid increase in market demand for its products and most countries, which means that not all structural change cap-
services, while its technological improvement is limited. tured by the NAV measure would boost productivity growth. This
observation supports the findings by McMillan et al. (2014) that
structural change may not enhance productivity growth as seen
At the same time, the condition Ci > 0 is not satisfied if:
for the case of Latin America (during 1990–2005) and Africa (during
1990–1999).To gain more insights into the trend of ESC and the ESC-
(i) The sector’s productivity is declining and its employment share NAV gap, the 19 economies are divided into three groups by their
is contracting. respective patterns of ESC. Group 1 includes economies with declin-
(ii) The sector’s productivity is growing and its employment share ing ESC trends. This group comprises Japan, Korea, Taiwan, and
is contracting, while the effect of the former is dwarfed by that Thailand19 (Fig. 1, Panel A). That the ESC-NAV gap for these coun-
of the latter. tries, with the exception of Japan, has recently narrowed implies
(iii) The sector’s productivity is declining and its employment share that structural change in these economies has become more effec-
is expanding, while the effect of the former is greater than that tive in enhancing productivity growth.
of the latter. Group 2 includes countries with increasing ESC trends. This
group consists of China, India, Sri Lanka, Vietnam, Bangladesh, and
3.2. The Norm of Absolute Values (NAV) index Myanmar (Fig. 1, Panel B). While all six economies belong to the
low- or middle-income group, their ESC patterns exhibit some
For a given economy, the Norm of Absolute Values (NAV) index differences. The upward trend of ESC is more solid for India and
is calculated for period [0,T] as follows: Vietnam, while that of China and Sri Lanka is more fluctuating.
For China, ESC and NAV surged in periods associated with exter-

n

NAV = 0.5 ∗ |SiT − Si0 |
k=1
Where n is the number of sectors in the economy; Si0 and SiT
represent the employment share of sector i in time 0 and T,
respectively.17
This method provides a quick measure of the overall magnitude
of structural change. However, this does not make a distinction
as to whether structural experienced by a sector is productivity-
enhancing or decreasing. This limitation, therefore, makes the
measure less meaningful in providing insights into the growth
effects of structural change.
The Effective Structural Change (ESC) Index introduced below
is constructed to combine the strengths of the shift-share method
and the NAV index to overcome their limitations.
(4) nal shocks (1986–1987: global economic recession; 1997–1998:
Asian financial crisis; and 2008–2009: global financial crisis). One
may interpret from this pattern that China accelerated its efforts to
foster structural change during slow-growth years. It is also worth
noting that the ESC-NAV was quite narrow for Vietnam and China
(since the early 1990s) compared to other countries. This may be an
encouraging sign of the economic reform approaches adopted by
both countries in embracing international trade and foreign direct
investment, promoting manufacturing development and exports,
and fostering reallocation of labor from agriculture to other sectors,
all of which are expected to enhance the effectiveness of structural
change.
Group 3 comprises countries with mixed ESC trends. This group
is further divided into two subgroups by income level: the higher-
income subgroup includes Hong Kong, Singapore, Malaysia, and

16
Note that the value of Ci is the same regardless of whether one chooses the
decomposition framework provided by Eq. (1), Eq. (2), or Eq. (3).
17
The 0.5 factor is to correct the double count of employment share changes. This
measure is also called Michaely Index or Stoikov Index; more details can be found
in Dietrich (2012).
18
The value is five-year moving average to eliminate short-term fluctuations.
19
Note that all countries in this group are of high income, except for
Thailand, which is an upper-middle-income nation. The income classification is
provided by World Bank (http://data.worldbank.org/about/country-and-lending-
groups, accessed September 2, 2015).
 K.M. Vu / Structural Change and Economic Dynamics 41 (2017) 64–77 69

Japan Korea, Rep.

NAV ESC NAV ESC
3.0% 7.0%

2.5% 6.0%

5.0%
2.0%
4.0%
1.5%
3.0%
1.0%
2.0%
0.5% 1.0%

0.0% 0.0%
1975
1978
1981
1984
1987
1990
1993
1996
1999
2002
2005
2008
2011

1975
1978
1981
1984
1987
1990
1993
1996
1999
2002
2005
2008
2011
Taiwan Thailand
NAV ESC NAV ESC
4.5% 14.0%
4.0% 12.0%
3.5%
10.0%
3.0%
2.5% 8.0%
2.0% 6.0%
1.5%
4.0%
1.0%
2.0%
0.5%
0.0% 0.0%
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
2010
2012
1983
1985
1987
1989
1991
1993
1995
1997
1999
2001
2003
2005
2007
2009
2011

Fig. 1. Effective Structural Change (ESC) and Norm of Absolute Values (NAV) Indices (five-year moving average values).

Brunei (Fig. 1, Panel C), while the lower-income subgroup consists
of Cambodia, Nepal, Indonesia, Pakistan, and the Philippines (Fig. 1,
Panel D). The ESC of the economies in the higher-income subgroup
follows a sine-wave shape, while their ESC-NAV gap tends to widen
in recent years (Hong Kong since 1997, Singapore since 2001, and
Malaysia since 2009). At the same time, the ESC-NAV gap is large
for Brunei in most years.20 These observations tend to suggest that
reducing the ESC-NAV gap, which is associated with enhancing the
effectiveness of structural change, is a challenging task facing all the
four economies in this subgroup. In the lower-income subgroup
(Panel D), including Cambodia, Nepal and Pakistan, has seen ESC
surge to a peak then decline. On the other hand, the ESC of Indonesia
and the Philippines follows a declining trend before entering into
a slightly increasing pattern.
nation’s success through a revealing comparison of the economic
performance of the Philippines and South Korea between 1960
and 1988. In 1960, when the two countries had similar levels of
income, the conditions for rapid development were more favorable
in the Philippines than in South Korea, particularly in education. The
remarkable transformation of South Korea in contrast to the stag-
nation of the Philippines in the past five decades can be regarded as
an economic miracle Lucas (1993). Can ESC help explain this mir-
acle? Fig. 2 shows that the disparity in effective structural change
(captured by the ESC metric) between Korea and the Philippines
has a strong link with the gap between the two countries in income
growth until mid–2000s (Panels A and B).
4.2. Regression model and estimation methods

4. Effective structural change and economic growth
This section examines the causal link of the ESC index, which
captures the magnitude of effective structural change, with key
growth indicators. Before entering into rigorous analyses with
econometric techniques, it is helpful to motivate this investigation
with some illustrative evidence.
Hall and Jones (1999), Acemoglu et al. (2001), Acemoglu et al.
(2008), and Eicher and Schreiber (2010) provide a useful approach
to dynamic panel data analysis with a parsimonious model to
detect the causal link between variables of interest. Following this
approach, the investigation of the effect of ESC on economic growth
is based on the dynamic model below:
gi,t = ˇ0 + ˇ1 gi,t−1 + ˇ2 yi,t−1 + ˇ3 ESCi,t−1 + ˇ4 ESCi,t + i

4.1. Motivation +ωt + εi,t (6)

Nobel Laureate Economist Robert E. Lucas, in his 1993 article

“Making a Miracle,” aimed to better understand factors driving a
20
It should also be noted that, compared to other economies, the ESC and NAV
of Brunei experienced large fluctuations. A possible reason is the country’s heavy
dependence on oil production (on average, the mining sector accounts for approx-
imately 70% of GDP) and its employment concentration in services (66%) and
construction (16%).
where gi,t is growth rate of country i in year t in an economic
performance indicator of interest, includinglabor productivity, TFP,
GDP per capita, GDP, wage,and employment; thelagged variable
gi,t−1 is included to capture the effects of unobservable factors
underlying the persistent pattern of growth variable gi,t . The lagged
value of the income level yi,t−1 captures the convergence effect
(Barro, 1991; Barro and Sala-i-Martin, 1995), which means, in this
context, that a country’s growth tends to slow as its level of income
70 K.M. Vu / Structural Change and Economic Dynamics 41 (2017) 64–77

China India
NAV ESC NAV ESC
4.5% 3.5%
4.0% 3.0%
3.5%
2.5%
3.0%
2.5% 2.0%
2.0% 1.5%
1.5%
1.0%
1.0%
0.5%
0.5%
0.0% 0.0%

1975
1978
1981
1984
1987
1990
1993
1996
1999
2002
2005
2008
2011
1983
1985
1987
1989
1991
1993
1995
1997
1999
2001
2003
2005
2007
2009
2011
Sri Lanka Vietnam
NAV ESC NAV ESC
8.0% 4.0%
7.0% 3.5%
6.0% 3.0%
5.0% 2.5%
4.0% 2.0%
3.0% 1.5%
2.0% 1.0%
1.0% 0.5%
0.0% 0.0%
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
1975
1978
1981
1984
1987
1990
1993
1996
1999
2002
2005
2008
2011

Bangladesh Myanmar
NAV ESC NAV ESC
4.0% 4.0%
3.5% 3.5%
3.0% 3.0%
2.5% 2.5%
2.0% 2.0%
1.5% 1.5%
1.0% 1.0%
0.5% 0.5%
0.0% 0.0%
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
2010
2012

Fig. 1. (Continued)

increases, and that lower-income economies tend to grow faster
than the richer ones. The structural change variable ESC is broken
down to enter into the model in two terms21 : ESCi,t−1 and ESCi,t .
The lagged value ESCi,t−1 captures the delayed effect of effective
structural change (ESC) taking place in the previous year, which is
expected to have a positive sign. ESCi,t is the change of ESC from
year t − 1 to year t. A positive value of ESCi,t indicates an acceler-
ation in effective structural change, while its negative value means
a deceleration. ESCi,t is expected to have a positive effect on
economic growth. Finally, i and ωt capture country-specific and
time-fixed effects, respectively, while εi,t represents the random
error term. Note that country-fixed effects and the lag of depen-
dent variable capture most unobserved factors that can influence a
country’s growth. These factors range from geographic conditions,
21
Note that ESCi,t = ESCi,t−1 + ESCi,t . Therefore, if the model includes ESCi,t instead
of ESCi,t−1 and ESCi,t , it will unreasonably force the effects of ESCi,t−1 and ESCi,t
to be equal.
openness, and institutional quality (Barro, 1991) to income distri-
butions and environmental performance (Antoci et al., 2009; López,
2007).
The panel data used for this exercise consists of 19 countries
across 42 years (from 1971 to 2012).22 The Im–Pesaran–Shin (Im
et al., 2003) and Fisher-type Choi (Choi, 2001) tests for all the key
variables in Eq. (6) both reject the null hypothesis that they contain
panel unit roots. This means that it is proper to use these time series
variables for regression analyses.23 Table 3 provides a summary of
statistics for the variables used in this regression exercise.
22
Data for Laos and for 2013 is not available for this analysis. Due to the growth
calculation, year 1970 is eliminated. This panel data is not balanced, which means
data is not available for some countries in some years.
23
Because the panel data is not strongly balanced (data for some early or recent
years is not available for some countries and industries), it is not possible to conduct
the panel unit root tests (as done by Harris and Tzavalis (1999) and Levin et al. (2002))
as such tests require strongly balanced data.
 K.M. Vu / Structural Change and Economic Dynamics 41 (2017) 64–77 71

Hong Kong SAR, China Singapore

NAV ESC NAV ESC
10.0% 7.0%
9.0%
6.0%
8.0%
7.0% 5.0%
6.0% 4.0%
5.0%
4.0% 3.0%
3.0% 2.0%
2.0%
1.0%
1.0%
0.0% 0.0%
1983
1985
1987
1989
1991
1993
1995
1997
1999
2001
2003
2005
2007
2009
2011

1975
1978
1981
1984
1987
1990
1993
1996
1999
2002
2005
2008
2011
Malaysia Brunei
NAV ESC NAV ESC
6.0% 14.0%

5.0% 12.0%

10.0%
4.0%
8.0%
3.0%
6.0%
2.0% 4.0%

1.0% 2.0%

0.0% 0.0%
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
2010
2012
1987
1989
1991
1993
1995
1997
1999
2001
2003
2005
2007
2009
2011

Fig. 1. (Continued)

Table 3
Effective Structural Change and Economic Growth: Regression results.

Explanatory g [t]=ALP growth g [t]=TFP growth g [t]=GDP per capita growth

Variable
LSDV SYS-GMM LSDV SYS-GMM LSDV SYS-GMM

(1a) (1b) (1c) (2a) (2b) (2c) (3a) (3b) (3c)

** *** * *** *** *** ***
g [t-1] 0.040 0.098 0.299 0.079 0.138 0.204 0.283 0.280 0.466***
(0.039) (0.042) (0.066) (0.042) (0.044) (0.061) (0.038) (0.039) (0.058)
Income [t-1] −0.001 −0.002 −0.005 −0.013** −0.012** −0.003** −0.015*** −0.017*** −0.003
(0.006) (0.006) (0.003) (0.006) (0.006) (0.002) (0.005) (0.005) (0.002)
ESC [t] 0.594*** – – 0.439*** – – 0.171*** – –
(0.066) (0.056) (0.056)
ESC [t-1] – 0.393*** 0.295*** – 0.264*** 0.221*** – 0.154** 0.179***
(0.086) (0.114) (0.075) (0.085) (0.075) (0.059)
ESC [t] – 0.643*** 0.683*** – 0.482*** 0.481*** – 0.184*** 0.235***
(0.067) (0.160) (0.058) (0.101) (0.058) (0.058)
Country FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
Adj-R2 0.321 0.335 – 0.311 0.326 – 0.386 0.390 –
N 633 633 633 557 549 549 675 664 664

P-values of hypothesis tests

AR(2) test of serial correlation – – 0.862 – – 0.727 – – 0.108
Sargan test of overid. restrictions – – 0.637 – – 0.991 – – 0.345

Note: Figures in parentheses under coefficients are robust standard errors. * p < 10%; ** p < 5%; *** p < 1%.

Least Squared Dummy Variable (LSDV) estimation is a good start eralized Method of Moments (GMM) estimation strategy should
for analysis. The advantage of this method is its control for country- be used, as it allows endogenous variables to be instrumented
specific and time-fixed effects, which eliminates the bias caused with their own lags (Roodman, 2006). Two commonly-used GMM
by unobservable factors related to time-invariant characteristics of estimators are first-difference GMM (DIF-GMM) and system GMM
countries and by time-related effects. The LSDV estimates as shown (SYS-GMM).24 For this analysis exercise, the SYS-GMM estimator is
by Nickell (1981) can serve as useful benchmarks to examine the
magnitude of coefficients provided by other estimation methods.
The LSDV estimates, however, are subject to potential bias due to 24
The DIF-GMM estimator developed by Holtz-Eakin et al. (1988) and Arellano and
the endogeneity associated with possible reverse effects of the left- Bond (1991) uses first-differences to remove entity-fixed effects, and instruments
hand side variable (economic growth) on ESC variables (effective these first-differences with the earlier values of explanatory variables. The GMM
structural change). To address the endogeneity problem, the Gen- estimator proposed by Arellano and Bover (1995) and Blundell and Bond (1998)
augments the DIF-GMM to obtain a system of two equations: one in differences and
72 K.M. Vu / Structural Change and Economic Dynamics 41 (2017) 64–77

Cambodia Nepal
NAV ESC NAV ESC
7.0% 5.0%
4.5%
6.0%
4.0%
5.0% 3.5%
4.0% 3.0%
2.5%
3.0% 2.0%
2.0% 1.5%
1.0%
1.0%
0.5%
0.0% 0.0%
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012

1975
1978
1981
1984
1987
1990
1993
1996
1999
2002
2005
2008
2011
Indonesia Pakistan
NAV ESC NAV ESC
12.0% 6.0%

10.0% 5.0%

8.0% 4.0%

6.0% 3.0%

4.0% 2.0%

2.0% 1.0%

0.0% 0.0%
1981
1983
1985
1987
1989
1991
1993
1995
1997
1999
2001
2003
2005
2007
2009
2011

1975
1978
1981
1984
1987
1990
1993
1996
1999
2002
2005
2008
2011
Philippines
NAV ESC
7.0%

6.0%

5.0%

4.0%

3.0%

2.0%

1.0%

0.0%
1975
1978
1981
1984
1987
1990
1993
1996
1999
2002
2005
2008
2011

Fig. 1. Effective Structural Change (ESC) and Norm of Absolute Values (NAV) Indices (five-year moving average values).

chosen due to its superior performance in the presence of a highly
persistent dependent variable (Blundell and Bond, 1998).25 The
greater asymptotic efficiency of the SYS-GMM estimator is also evi-
dent in other studies such as Bond et al. (2001) and Soto (2009). In
particular, Soto (2009) demonstrates that SYS-GMM is more valid
than DIF-GMM when the number N of entities (such as countries)
is small.
4.3. Empirical results and discussions
Results from regressions based on a model represented by Eq.
(6) are reported in Table 3 (for three dependent variables: labor
productivity growth, TFP growth, and GDP per capita growth) and
Table 4 (average wage growth, employment growth, and GDP
growth). For each individual dependent variable, the regression
results are reported in three columns, of which columns [a] and
[b] are for LSDV estimations, while column [c] is for the SYS-GMM
estimation method, which will be referred to hereafter as GMM.
For GMM regressions, it is necessary to test the null hypotheses
of second order autocorrelation and of the non-validity of over-
identifying restrictions of instrumental variables. The p-values
of these tests (AR(2) for serial correlation and Sargan for over-
identifying restrictions)confirm that all the GMM regressions are
properly specified.26

one in levels. In the levels equation the variables are instrumented with the lagged
values of their own first differences.
25
The GMM estimator, however, requires a mild assumption about the orthogo-
nality between the lagged first difference in the dependent variable (yt−1 ) and the
composite error terms in the levels equation. This assumption allows yt−1 to be
used as valid instruments for the levels equation.
26
The p-value = 1.0 of the Sargan test for g [t] = Employment growth (Table 4,
column [c]), however, should be read with caution. This may indicate an overfit-
ting problem, which usually occurs for GMM regressions with a small N and large
T (Roodman, 2009). This problem can be found in some previous studies such as
Giuliano and Ruiz-Arranz (2009, Table 3: 148).
 K.M. Vu / Structural Change and Economic Dynamics 41 (2017) 64–77 73

Fig. 2. GDP per Capita Growth, ESC, and NAV: Korea and the Philippines in Comparisons.
Note: The graphs are based on 5-year moving average (5-YMA) values to eliminate short-term fluctuations.

In the discussions of empirical results below, the GMM coeffi- hereafter be referred to as g[t], g[t-1], y[t-1], ESC[t-1], and ESC[t],
cients are considered the most reliable estimates, while the LSDV respectively. Summary statistics of the key variables are provided
coefficients serve as useful benchmarks. For ease of exposition, in Appendix B.
variables gi,t , gi,t−1 , yi,t−1 , ESCi,t−1 , and ESCi,t , from Eq. (6) will
74 K.M. Vu / Structural Change and Economic Dynamics 41 (2017) 64–77

Table 4
Effective Structural Change and Economic Growth: Regression results.

Explanatory g [t]=Average wage growth g [t]=Employment growth g [t]=GDP growth

Variable

LSDV SYS-GMM LSDV SYS-GMM LSDV SYS-GMM

(4a) (4b) (4c) (5a) (5b) (5c) (6a) (6b) (6c)

g [t-1] 0.148*** 0.174*** 0.242*** -0.099*** -0.031 0.076 0.290*** 0.286*** 0.442***
(0.039) (0.043) (0.062) (0.038) (0.042) (0.121) (0.038) (0.039) (0.055)
Income [t-1] −0.008 -0.008 −0.005* -0.009* −0.008** 0.001 −0.014*** -0.015*** -0.004*
(0.010) (0.011) (0.003) (0.004) (0.004) (0.002) (0.005) (0.005) (0.002)
ESC [t] 0.796*** – – -0.345*** – – 0.176*** – –
(0.098) (0.041) (0.056)
ESC [t-1] – 0.605*** 0.732*** – -0.181*** -0.080 – 0.161** 0.190***
(0.133) (0.137) (0.053) (0.064) (0.074) (0.068)
ESC [t] – 0.846*** 0.964*** – −0.392*** −0.417*** – 0.188*** 0.216***
(0.103) (0.116) (0.043) (0.072) (0.057) (0.058)
Country FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
Adj-R2 0.250 0.243 – 0.229 0.256 – 0.376 0.381 –
N 586 578 578 674 665 665 675 664 664

P-values of hypothesis tests

AR(2) test of serial correlation – – 0.149 – – 0.366 – – 0.148
Sargan test of overid. restrictions – – 0.521 – – 1.000 – – 0.456

Note: Figures in parentheses under coefficients are robust standard errors. * p < 10%; ** p < 5%; *** p < 1%.

From a quick glance, three general observations stand out. First,
the coefficient ˇ1 on g[t-1] is positive and significant at the 1% level
in all the GMM regressions, with the exception of the regression
on employment growth (Table 4, column [5c]). This suggests that
growth patterns of labor productivity, TFP, GDP per capita, wage,
and GDP are highly persistent.
Second, the coefficient ˇ2 on the lagged value of income level
y[t-1] has its expected negative sign in all regressions. However,
the GMM estimate of this coefficient is not statistically significant
for ALP growth, GDP per capita growth, and employment growth
(Table 3, columns [1c], [3c], and Table 4, column [5c]), which means
that the income level tends not to be a significant drag on these
growth variables. On the other hand, the GMM estimate of this
coefficient is statistically significant at the 5% level for the TFP
growth equation (Table 3, column [2c]), which indicates that TFP
growth tends to be lower for the economies with higher income
levels. This finding is plausible because as a country gets richer, it
begins to approach the global technology frontier. As a result, it will
have less “backwardness advantage” (Gerschenkron, 1962), which
means less room for efficiency gains from importing foreign tech-
nology and improving management skills. Similar implications are
also found for average wage and GDP growth, for which the GMM
estimate of ˇ2 is significant at the 10% level (Table 4, columns [4c]
and [6c]).
Third, for the dependent variable, the coefficient on
ESC[t](column [a]) lies between the coefficients on ESC[t-1]
and ESC[t] (column [b]for LSDV and column [c] for GMM esti-
mates). More specifically, the coefficient on ESC[t], in the absolute
value, is larger than the coefficient on ESC[t-1] and smaller than
the coefficient on ESC[t]. This means that breaking down27
ESC[t] into ESC[t-1] and ESC[t] to include in Eq. (6) is a proper
approach. This helps to show that the delayed effect (ˇ3 ) and
the contemporaneous effect (ˇ4 ) of structural change are not the
same, and the magnitude of the latter is significantly larger than
that of the former (|ˇ4 | > |ˇ3 |).
Lastly, for each individual dependent variable, the LSDV coef-
ficients from column [b] are generally comparable to the GMM
coefficients reported in column [c], with the exception of the coef-
ficient on g[t-1], for which the LSDV estimate is notably below the
GMM estimate. This observation is consistent with the claim by
Nickell (1981) that for a dynamic panel model, the LSDV coefficient
on the lagged dependent variable is biased downwards, while the
coefficients on other variables are less affected.28
The paragraphs below discuss the results in more detail, with a
focus on the magnitude of effects of ESC on growth. The discussion
concentrates on the GMM estimates of the coefficients, which are
believed to be least affected by the bias caused by the endogeneity
problem.
The coefficient on ESC[t-1] is statistically significant in all regres-
sions, with the exception of the GMM coefficients on ESC[t-1] for
employment growth (Table 4, column [5c]).The GMM coefficient on
ESC[t-1] indicates that increasing the magnitude of effective struc-
tural change by one percentage point would raise growth rate in
the following year by 0.3% points for ALP, 0.22 for TFP, 0.18 for GDP
per capita, 0.73 for average wage, and 0.19 for GDP growth. The
effect of ESC[t-1] has a negative sign but it is not significant.
The coefficient on ESC[t] is robustly significant at the 1% level
across regressions. The GMM coefficient on ESC[t] implies that
accelerating effective structural change by one percentage point
increases growth rates by 0.68% points for ALP, 0.48 for TFP, 0.24
for GDP per capita, 0.96 for average wage, and 0.22 for GDP. At
the same time, because this coefficient is negative for employment
growth, it means that one percentage point increase in effective
structural change accelerationwould reduce employment growth
by 0.42 percentage points.
Combining the GMM estimates for ESC[t-1] and ESC[t] reveals
several important policy insights. First, the delayed effects and con-
temporaneous effect (ˇ4 ) of effective structural change are much
larger for wage (ˇ3 = 0.732 ; ˇ4 = 0.964) and ALP growth (ˇ3 = 0.295
; ˇ4 = 0.683), compared to those for other growth variables. This
means that fostering effective structural change produces more
palpable growth effects on workers’ wage and productivity, which
are deemed most essential to the competitiveness of the business
sector.
The second policy insight is about the channel through which
effective structural change fosters labor productivity growth. Note

28
This point is also raised by Judson and Owen (1999), Gaduh (2002), Hauk and
27
Note that ESC[t] = ESC[t − 1] + ESC[t]. Wacziarg (2004), and Eicher and Schreiber(2010), among others.
 K.M. Vu / Structural Change and Economic Dynamics 41 (2017) 64–77
that labor productivity growth comes from two sources, capi-
tal deepening, which means increase in capital stock per worker,
and TFP growth, which is associated with efficiency improvement
in value creation. The GMM coefficients on ESC[t-1] for TFP and
ALP growth suggests that approximately 75% (0.221/0.295) of the
delayed effect of effective structural change on labor productiv-
ity growth is through fostering TFP growth. At the same time, the
estimates of GMM coefficients on ESC[t] for TFP and ALP growth
indicate that TFP accounts for 70% (0.481/0.683) of the contempo-
raneous effect of effective structural change on labor productivity
growth. These findings show that the contribution of effective
structural change to labor productivity growth is mainly through
fostering efficiency improvement, which is far more effective than
promoting capital deepening.
Third, while the growth benefits from fostering effective
structural change are sizeable, the cost in terms of short-term
unemployment is high. The GMM coefficient on ESC[t] for
employment growth indicates that accelerating effective structural
change by one percentage point lowers employment growth by
0.42 percentage points (Table 4, column [5c]). This negative effect
is high but appears to become insignificant in the long-term. In
fact, the GMM coefficient on ESC[t-1] for employment growth is
negative but not statistically significant.
Finally, effective structural change strongly boosts GDP growth
through both its lagged and contemporaneous effects (Table 4,
column [6c]). This implies that the positive effects of effective
structural change on productivity growth far outweigh its nega-
tive effects on employment growth. The policy challenge, therefore,
is how to overcome the sensitive cost of rising unemployment
when launching reforms to boost growth through fostering effec-
tive structural change.
4.4. Is ESC an improved approach to measuring structure change?
As presented in Section 3, ESC is built on the advantages of the
existing two approaches in capturing the effect of structural change
– the NAV and Shift-Share Analysis (SSA). This subsection provides
some evidence that ESC is an improved approach compared to the
NAV and SSA.
While the computation of NAV is straightforward as indicated
in Eq. (4), the SSA structural change slightly varies by researchers.
McMillan et al., (2014:19) define “structural change” as the second
term in the equation below:

n

n
P = Si0 Pi + PiT Si (7)
i=1 i=1
Dividing both sides of Eq. (7) by P0 yields
P  S P  P S
n n
i0 i iT i between 1970 and 2012. Among the main findings of the paper
= + (7 )
P0 P0 P0
i=1 i=1
That is labor productivity growth can be expressed as a sum-
mation of two terms, the “within” effect (the first term) and the

n
P
iT Si
“structural change”. The second term, MRV 1 = P0
can be
i=1
considered as a raw measure of SSA structural change. An improve-
ment of this measure is its log-transformation, MRV 2 = ln (MRV 1),
which improves the variable’s normality.
Using NAV, MRV1, and MRV2 as a structural change measure
(SCM) to replace ESC in Eq. (6) above, one can examine the effect of
these measures on economic growth. Table 5 reports the SYS-GMM
regression results for these measures of structural change. To save
space, the results are reported only for two productivity dependent
variables: ALP growth and TFP growth.
75
The following observations stand out. First, as shown in Columns
(R1) and (R2), NAV has no significant effect on ALP growth and TFP
growth. The inclusion of the results for the non-effective structural
change (NESC),29 which is a residual of NAV after extraction of ESC,
can help shed light on this observation. The statistically significant
negative effect of NESC on growth as shown in Columns (R3) and
(R4) cancel out the statistically significant positive effect of ESC on
growth presented in Subsection 4.3; and hence NAV does not have
its significant effect on growth.
Second, while MRV1 shows a strong significant effect on growth
(Columns (R5) and (R6)); MRV2 does not (Columns (R7) and (R8)).
Note that MRV1 is an additive component in the growth Eq. (7) its
strong association with growth is not meaningful enough as MRV2,
a log-transformation of MRV1, does not have a significant effect on
growth. This somehow implies the limitation of the SSA structural
change measure used by McMillan et al., (2014).
The discussion, therefore, suggests that ESC is a significantly
improved approach to capturing the effect of structure change on
growth. In addition, its simple calculation can make it a meaning-
ful indicator for monitoring and promoting structural change that
enhance productivity growth.
5. Conclusion
A wealth of studies on structural change has laid a solid foun-
dation for the hypothesis that reallocation of productive inputs,
particularly labor, among sectors plays an important role in driving
economic growth. However, this hypothesis is still challenged by
a number of studies that find no conclusive evidence of a positive
contribution by structural change to growth. Moreover, the exist-
ing methods used to capture the effectiveness of structural change
and quantify its contribution to growth have significant limitations
that may compromise the accuracy of their empirical results.
This paper introduces a new measure of structural change called
“effective structural change” (ESC), which takes into account the
effectiveness of the shift of labor among sectors in contribution
to overall labor productivity growth. This novel approach is built
on two existing methods: the shift-share analysis and the norm of
absolute values (NAV) index, both of which are widely used for eval-
uating the effects of structural change. The ESC measure is designed
to overcome the main shortcomings of the shift-share analysis
and the NAV index. In the ESC measure, the unjustified assump-
tion by the shift-share analysis about the independence between
the within-effects and between-effects contributed by structural
change is no longer necessary. At the same time, the ESC measure
is more effective than the NAV index in capturing the structural
change that contributes to growth.
This study focuses on Asian economies, taking advantage of the
richness of the APO dataset, which is available for 19 economies
are the following. First, effective structural change has a strong
impact on growth of labor productivity, TFP, GDP per capita, GDP,
and wages. This impact is transmitted through two channels, the
delayed effect of effective structural change in the previous year
and the contemporaneous effect of effective structural accelera-
tion. In addition, these two effects are found notably larger for
labor productivity and wage growth compared to other growth
variables, implying that the impact of effective structural change
is particularly strong for enhancing competitiveness and raising

29
NESC consists only the structural change that has a negative contribution to ALP
growth. Similar to ESC, NESC can be defined as NESC = 0.5 ∗ |SiT − Si0 | where
 i∈Y
Y= i such that C i < 0.
76 K.M. Vu / Structural Change and Economic Dynamics 41 (2017) 64–77

Table 5
Results from GMM Regressions for different structural change measures.

Explanatory Dependent Variable: Growth Rate g[t]

Variable
ALP growth TFP growth ALP growth TFP growth ALP growth TFP growth ALP growth TFP growth
SCM=NAV SCM=NESC SCM=MRV1 SCM=MRV2

(1) (2) (3) (4) (5) (6) (7) (8)

g [t-1] 0.202** 0.159* 0.266*** 0.249*** 0.220*** 0.188*** 0.307** 0.258***

(0.101) (0.092) (0.071) (0.047) (0.081) (0.061) (0.132) (0.081)
Income [t-1] −0.004 −0.003 −0.001 −0.001 −0.004 −0.002 -0.001 −0.001
(0.003) (0.002) (0.002) (0.001) (0.003) (0.002) (0.002) (0.002)
SCM [t-1] −0.049 0.006 −0.861*** −0.604*** 0.343*** 0.215*** −0.001 −0.002
(0.092) (0.056) (0.162) (0.172) (0.089) (0.060) (0.001) (0.002)
SCM [t] 0.036 0.042 −1.281*** −0.923*** 0.390*** 0.252*** 0.001 −0.001
(0.078) (0.059) (0.209) (0.208) (0.099) (0.055) (0.001) (0.001)
Country FE Yes Yes Yes Yes Yes Yes Yes Yes
Time FE Yes Yes Yes Yes Yes Yes Yes Yes
N 633 549 633 549 633 531 341 303

P-values of hypothesis tests

AR(2) test of serial correlation 0.448 0.293 0.865 0.325 0.675 0.400 0.056** 0.224
Sargan test of overid. restrictions 0.115 1.000 0.073 1.000 0.344 1.000 0.452 0.611

Note: Figures in parentheses under coefficients are robust standard errors. * p < 10%; ** p < 5%; *** p < 1%.

living standards. Furthermore, for all growth variables the con-
temporaneous effect of effective structural change is larger than
its delayed effect, which means decisive efforts to foster effective
structural change can quickly produce palpable results. The cost
of pushing for effective structural change, however, should not be
underestimated. The paper also finds that accelerating effective
structural change has a significant negative impact on employ-
ment growth. In fact, increasing ESC by one percentage point would
reduce employment growth by 0.42% points, which means a sizable
rise in unemployment rates should be expected.
Several important policy insights can be drawn from the find-
ings of this study. First, fostering effective structural change is a
powerful way to promote economic growth.30 Because its effects
are quickly palpable – especially the variables vital to national level
competitiveness and living standards – this effort, once decisively
launched, would create important momentum for next steps.
Second, the cost of fostering effective structural change in terms
of rising unemployment should be foreseen and prepared for. This
requires an effective strategy to raise awareness and understanding
about the benefits of a given structural change, and an action plan to
support laid-off workers with training for new skills and assistance
finding new jobs (likely in new sectors). Finally, while this study
examines the effect of effective structural change on economic
growth only at the overall level, the results imply that this effect
can be expected to occur at deeper layers. That is, robust produc-
tivity effects can be expected as production factors, namely labor,
capital, and land, shift from industry to industry within a given sec-
tor (the sector level). These effects can also be seen at increasingly
lower scales: from firm to firm within a given industry31 (the indus-
try level), from product to product within a given firm32 (the firm
level), and even from activity to activity within a given product of
a given firm33 (the product level).
It should be noted that fostering effective structural change can-
not be done only through a one-time decisive reform effort, but
must be pursued through a series of systematic and continuous
policy initiatives. Promoting knowledge diffusion and innovation
strengthens within-effects, while deepening global integration cre-
ates opportunities for new structural change. Finally, improving
the business environment and upgrading training systems lowers
costs and enlarges benefits of new investments as well as shifting
resources to more profitable sectors and more productive users.
This paper has not only endeavored to shed additional light on
the impacts of structural change, but has also introduced a novel
analytical and methodological approach that complements a rich
body of existing growth literature. Future research can test this
methodology in other temporal and geographical contexts, includ-
ing the post-industrialization epoch of Western Europe, North
America, and other developed economies. It can also be applied to
studies of developing economies and regions experiencing rapid
structural transformation. One example is Africa, where in the
coming decades a shift from agriculture to low-level manufactur-
ing is likely to have significant impacts on productivity and labor
force dynamics. A similar transformation is imminent in China and
Southeast Asia, as countries in the region seek to move up the value-
added production chain with better education and up-skilling of
workers. The ESC method is flexible enough to accommodate these
and other contexts, and will likely be an important tool for under-
standing the rapidly evolving global economy.
Acknowledgements
The author is grateful to the anonymous referees for their
valuable comments and suggestions and acknowledges the staff
research support grant provided by the Lee Kuan Yew School of
Public Policy for this study.
Appendix A. The Industry Classification of the Nine
Economic Sectors

Sector Sector’s Complete Title ISIC Rev.3

1. Agriculture Agriculture, hunting, forestry, and 01,02,05

30
This insight is also consistent with the finding by Hausmann et al. (2005) that fishing
sustained acceleration in economic growth is significantly driven by a major eco- 2. Mining Mining and quarrying 10–14
nomic policy reform, which is often associated with substantial structural change. 3. Manufacturing Manufacturing 15–35
31
This suggests the importance of the business environment and legal system, 4. Utility Electricity, gas and water supply 40–41
which effectively foster competition, facilitate merging and acquisition (M&A) trans- 5. Construction Construction 45
actions, and resolve insolvency. 6. Wholesale and Wholesale and retail trade, repair of 50–52,55
32
Entrepreneurs’ self-discovery spirits (Hausmann and Rodrik, 2003). retail vehicles and household goods, hotels
33
For example, a shirt-making company reallocates its resources from production and restaurants
to design and marketing as it moves up the value chain.
 K.M. Vu / Structural Change and Economic Dynamics 41 (2017) 64–77 77

7. Transport and Transport, storage and 60–64 Hauk, W., Wacziarg, R., 2004. A Monte Carlo Study of Growth Regressions.
communications communications NBERTechnical Working Paper 296.
8. Finance and real Financial intermediation, real estate, 65–67,70-74 Hausmann, R., Rodrik, D., 2003. Economic development as self-discovery. J. Dev.
estate renting and business activities Econ. 72 (2), 603–633.
9. Government and Community, social and personal 75,80,85,90–93,95,99 Hausmann, R., Pritchett, L., Rodrik, D., 2005. Growth accelerations. J. Econ. Growth
community services 10 (4), 303–329.
services Holtz-Eakin, D., Newey, W., Rosen, H., 1988. Estimating vector autoregressions
with panel data. Econometrica 56 (8), 1371–1395.
Source: APO Im, K.S., Pesaran, M.H., Shin, Y., 2003. Testing for unit roots in heterogeneous
panels. J. Econo. 115 (1), 53–74.
Jorgenson, D., Timmer, M., 2011. Structural change in advanced nations: a new set
Appendix B. Summary statistics of key variables
of stylised facts. Scand. J. Econ. 113 (1), 1–29.
Judson, R., Owen, A., 1999. Estimating dynamic panel data models: a guide for
Variable N Mean SE 95% Conf. Interval macroeconomists. Econ. Lett. 65 (1), 9–15.
Kaldor, N., 1966. Causes Of The Slow Rate Of Economic Growth Of The United
g[t]=ALP growth 652 2.6% 0.2% 2.2% 2.9% Kingdom. Cambridge University Press, London.
g[t]=TFP growth 645 1.3% 0.1% 1.0% 1.5% Kaniovski, S., Peneder, M., 2002. On the structural dimension of competitive
g[t]=Employment growth 831 2.4% 0.1% 2.2% 2.6% strategy. Ind. Corp. Change 11 (3), 257–279.
g[t]=Wage growth 666 3.7% 0.2% 3.3% 4.2% Kuznets, S., 1966. Modern Economic Growth: Rate Structure and Spread. Yale
g[t]=GDP per capita growth 825 3.7% 0.1% 3.4% 4.0% University Press, London.
g[t]=GDP growth 825 5.4% 0.1% 5.1% 5.7% Kuznets, S., 1979. Growth and structural shifts. In: Galenson, W. (Ed.), Economic
Level of per capita income (y[t]) 845 8.619 0.046 8.529 8.710 Growth and Structural Change in Taiwan. The Postwar Experience of the
ESC[t] 687 2.42% 0.09% 2.25% 2.60% Republic of China. Cornel University Press, London.
López, R.E., 2007. Structural change, poverty and natural resource degradation. In:
ESC[t] 668 −0.03% 0.11% −0.24% 0.18%
Atkinson, G., Dietz, S., Neumayer, E. (Eds.), Handbook of Sustainable
NAV[t] 687 3.4% 0.11% 3.2% 3.6%
Development. Edwards Elgar, Cheltenham, UK, Northhampton, MA, USA.
Levin, A., Lin, C.F., Chu, C.S.J., 2002. Unit root tests in panel data: asymptotic and
References finite-sample properties. J. Econ. 108 (1), 1–24.
Lewis, W.A., 1954. Economic development with unlimited supplies of labour.
Manch. Sch. Econ. Soc. Stud. 22, 139–191.
Acemoglu, D., Johnson, S., Robinson, J., 2001. The colonial origins of comparative
Lin, J.Y., Monga, C., 2010. The Growth Report and New Structural Economics. Policy
development: an empirical investigation. Am. Econ. Rev. 91 (5), 1369–1401.
Research Working Paper 5336. The World Bank.
Acemoglu, D., Johnson, S., Robinson, J.A., Yared, P., 2008. Income and democracy.
Lin, J.Y., 2009. Economic Development and Transition. Cambridge University Press,
Am. Econ. Rev. 98 (3), 808–842.
Cambridge.
Antoci, A., Russu, P., Ticci, E., 2009. Distributive impact of structural change: does
Lin, J.Y., 2012a. New Structural Economics: A Framework for Rethinking
environmental degradation matter? Struct. Change Econ. Dyn. 20 (4), 266–278.
Development and Policy. The World Bank, Washington, DC.
Arellano, M., Bond, S.R., 1991. Some tests of specification for panel data: Monte
Lin, J.Y., 2012b. The Quest For Prosperity: How Developing Economies Can Take
Carlo evidence and an application to employment equations. Revi. Econ. Stud.
Off. Princeton University Press, Princeton.
58 (2), 277–297.
Lucas, R.E., 1993. Making a miracle. Econometrica 61 (2), 251–272.
Arellano, M., Bover, O., 1995. Another look at the instrumental variable estimation
McMillan, M.S., Rodrik, D., 2011. Globalization, Structural Change and Productivity
of error-components models. J. Econ. 68 (1), 29–51.
Growth. NBER Working Paper No. 17143. National Bureau of Economic
Barro, R., Sala-i-Martin, X., 1995. Economic Growth. McGraw-Hill, New York.
Research, Cambridge M.A.
Barro, R.J., 1991. Economic growth in a cross section of countries. Q. J. Econ. 106
McMillan, M., Rodrik, D., Verduzco-Gallo, Í., 2014. Globalization, structural change,
(2), 407–443.
and productivity growth: with an update on Africa. World Dev. 63, 11–32.
Baumol, W.J., Blackman, S.A.B., Wolff, E.N., 1985. Unbalanced growth revisited:
Meckl, J., 2002. Structural change and generalized balanced growth. J. Econ. 77 (3),
asymptotic stagnancy and new evidence. Am. Econ. Rev. 75 (4), 806–817.
241–266.
Baumol, W.J., 1967. Macroeconomics of unbalanced growth: the anatomy of urban
Ngai, R., Pissarides, C., 2007. Structural change in a multisector model of growth.
crisis. Am. Econ. Rev. 57 (3), 415–426.
Am. Econ. Rev. 97 (1), 429–443.
Blundell, R., Bond, S., 1998. Initial conditions and moment restrictions in dynamic
Nickell, S., 1981. Biases in dynamic models with fixed effects. Econometrica 49 (6),
panel data models. J. Econ. 87 (1), 115–143.
1417–1426.
Bond, S., Hoeffler, A.E., Temple, J., 2001. GMM Estimation of Empirical Growth
Nomura, K., Lau, E.Y., 2014. APO Productivity Databook 2014. Asian Productivity
Models. CEPR Discussion Paper No. 3048. Centre for Economic Policy Research,
Organization, Tokyo.
London.
Peneder, M., 2002. Industrial Structure and Aggregate Growth. WIFO Working
Caselli, F., Coleman, W., 2001. The U. S. structural transformation and regional
Papers No. 182. Austrian Institute of Economic Research.
convergence: a reinterpretation. J. Polit. Econ. 109 (3), 584–616.
Roodman, D., 2006. How to Do Xtabond2: an Introduction to “Difference” and
Choi, I., 2001. Unit root tests for panel data. J. Int. Money Finance 20 (2), 249–272.
“System” GMM in Stata. Working Paper Number 103. Center for Global
Clark, C., 1957. The Conditions of Economic Progress, 3rd ed. Macmillan, London.
Development, Washington.
Denison, E.F., 1967. Why Growth Rates Differ. Brookings, Washington DC.
Roodman, D., 2009. A note on the theme of too many instruments. Oxf. Bull. Econ.
Dietrich, A., 2012. Does growth cause structural change, or is it the other way
Stat. 71 (1), 135–158.
around? A dynamic panel data analysis for seven OECD countries. Empir. Econ.
Silva, E.G., Teixeira, A.A.C., 2008. Surveying structural change: seminal
43 (3), 915–944.
contributions and a bibliometric account. Struct. Change Econ. Dyn. 19 (4),
Echevarria, C., 1997. Changes in sectoral composition associated with economic
273–300.
growth. Int. Econ. Rev. 38 (2), 431–452.
Soto, M., 2009. System GMM Estimation with a Small Sample. Barcelona
Eicher, T.S., Schreiber, T., 2010. Structural policies and growth: time series
Economics Working Paper No. 395. Institut d’Anàlisi Econòmica.
evidence from a natural experiment. J. Dev. Econ. 91 (1), 169–179.
Syrquin, M., 1984. Resource allocation and productivity growth. In: Syrquin, M.,
Fagerberg, J., 2000. Technological progress, structural change and productivity
Taylor, L., Westphal, L.E. (Eds.), Economic Structure and Performance. Essays in
growth: a comparative study. Struct. Change Econ. Dyn. 11 (4), 393–411.
Honor of Hollis B. Chenery. Academic Press Inc, Orlando.
Fan, S., Zhang, X., Robinson, S., 2003. Structural change and economic growth in
Syrquin, M., 1988. Patterns of structural change. In: Chenery, H., Srinavasan, T.N.
China. Rev. Dev. Econ. 7, 360–377.
(Eds.), Handbook of Development Economics. North Holland, Amsterdam.
Felipe, J., Leon-Ledesma, M., Lanzafame, M., Estrada, G., 2009. Sectoral engines of
Timmer, M.P., Szirmai, A., 2000. Productivity growth in Asian manufacturing: the
growth in developing Asia: stylized facts and implications. Malays. J. Econ.
structural bonus hypothesis re-examined. Struct. Change Econ. Dyn. 11 (4),
Stud. 46 (2), 107–133.
371–391.
Gaduh, A., 2002. Properties of Fixed Effects Dynamic Panel Data Estimators for a
Timmer, M.P., de Vries, G.J., 2009. Structural change and growth accelerations in
Typical Growth Dataset. CSIS Working Paper Series WPE 062.
Asia and Latin America: a new sectoral data set. Cliometrica 3, 165–190.
Gerschenkron, A., 1962. Economic Backwardness in Historical Perspective: a Book
Timmer, M., de Vries, K., de Vries, G., 2015. Structural transformation in Africa:
of Essays. Belknap Press of Harvard University Press, Cambridge.
static gains, dynamic losses. J. Dev. Stud. 51 (6), 674–688.
Giuliano, P., Ruiz-Arranz, M., 2009. Remittances, financial development, and
van Ark, B., Timmer, M., 2003. Asia’s productivity performance and potential: the
growth. J. Dev. Econ. 90 (1), 144–152.
contribution of sectors and structural change. In: Paper Presented at the
Hall, R., Jones, C., 1999. Why do some countries produce so much more output per
RIETI-KEIO Conference on Japanese Economy: Leading East Asia in the 21 st
worker than others? Q. J. Econ. 114 (1), 83–116.
Century, Tokyo.
Harris, R.D.F., Tzavalis, E., 1999. Inference for unit roots in dynamic panels where
the time dimension is fixed. J. Econ. 91 (2), 201–226.