Does Structural Change Contribute to Labor Productivity
Convergence? An Empirical Study on East and Southeast
Asian Countries
Master Thesis
Faculty of Economics and Business
International Economics and Business
Sun Yixiao y.sun.20@student.rug.nl
S3349721
Supervisor: Prof. Dr. Bart Los Co-assessor: Dr. Robbert K.J. Maseland
2018-6-19
Abstract
This paper studies the structural change and labor productivity growth in East and Southeast Asian countries from 1980 to 2010. Different from traditional structural change studies, this paper also attempts to find the linkages between structural change and labor productivity convergence. Our findings suggest that the primary source of labor productivity growth is the growth within sectors. However, structural change can provide considerable labor productivity growth at the early stage of development. Additionally, structural change did propel the tendency of labor productivity convergence, which worked better for countries that depend on the development of industrial sectors. Therefore, underdeveloped countries should pay more attention to structural change but keep continuous labor productivity growth within sectors at the same time.
Keywords: Structural change, labor productivity, convergence
Table of Contents
1. Introduction ... 1
2. Literature Review ... 3
2.1 Core Definitions ... 3
2.2 Various growth patterns ... 8
2.3 What made East and Southeast Asian countries stand out ... 10
3. Data and descriptive statistics ... 14
3.1 Labor productivity and GDP per capita data ... 14
3.2 control variables ... 15
4. Methodology ... 18
4.1 Regression method for hypothesis 1 ... 18
4.2Shift-share analysis and modification ... 20
4.3 Convergence regression ... 23
5.Results and analysis ... 25
5.1Results for hypothesis 1 ... 25
5.2 Results for shift-share analysis ... 29
5.3 Result for labor productivity convergence and hypothesis 2 ... 33
6. Robustness tests ... 34
7. Conclusion Remark ... 35
7. References: ... 38
1
1. Introduction
In 2015, governments, businesses and civil societies from all around the world met with the UN in order to address and adopt the seventeen sustainable development goals. The number one goal was to bring an end to poverty within the next fifteen years. Although government subsidies and foreign aid might be actions that help, the changes must come from within to be sustainable and long-lasting. One helpful way is to move workers from sectors characterized by low productivity to sectors that are more productive, thus increasing their income level. This process is called structural change. The majority of developed countries today have experienced structural change, which has been described as necessary to achieve sustainable economic development. In order to accomplish further development and fight against poverty in countries that are underdeveloped, a study focusing on structural change is needed.
Among the earliest and most central insights of the literature on economic development is the fact that development entails structural change (Lewis, 1954). Structural change is defined as the reallocation of labor from low- to high-productivity sectors. According to Timmer (2014), only countries in East and Southeast Asia managed to obtain productivity through structural change over the past three decades. This caused a series of studies, like de Vries (2013), Rodrik (2015) and Diao (2017), to analyze the existing problems in Africa and Latin America through decomposing labor productivity growth into growth within sectors and growth due to structural change. This paper analyzes structural change from a different perspective, focusing on structural change in East and Southeast Asian countries over the past three decades. By concentrating on successful cases, we expect to explore the impact of structural change on labor productivity growth and gain insights that can be useful for countries in other regions.
2
countries converge since the US has been the most productive country since 1970. This leads to the research question of this paper: To what extent did structural change contribute to labor productivity convergence of East and Southeast Asian countries towards the US from 1980 to 2010?
We use data from the GGDC 10-sector database, GGDC-Penn Worlds Table (PWT) and the World Bank database, ranging from 1980 to 2010. The main methods are modified shift-share analysis and regressions upon a panel data set. Based on traditional structural change studies, we combine two frequently discussed topics: structural change and convergence together and seek to find the link between them. Besides this, two other aspects are also included. First, the influence of country wealth (measured by real GDP per capita) on the inter-sector labor productivity gaps in developing country. Second, the different impacts of structural change on industrial-dependent countries and service-dependent countries (judged from where the most value-added is generated).
The main findings of this paper are as follows. On the one hand, countries in East and Southeast Asia have gone through fast labor productivity growth from 1980 to 2010. Although structural change functioned well at the early stage of development, its effect decreased as countries become more productive. Growth within sectors, instead, turned out to be the primary source of productivity growth. On the other hand, structural change did contribute to the productivity convergence towards the US over the past three decades. The contribution was more evident for countries that depend on the development of industrial sectors. These findings indicate that underdeveloped countries should pay more attention to structural change. Meanwhile, they need to keep continuous labor productivity growth within sectors. Additionally, as East and Southeast countries became richer, the labor productivity gap between industrial and agricultural sector increased. This correlation does not hold after we drop Japan from our sample.
3
2. Literature Review
This section reviews the main literatures concerning the topic structural change, labor productivity and convergence. This section is divided into two main parts. The first part defines the main concepts used in this article, and the second part illustrates the different patterns of economic growth as well as highlights the specific characteristics that made Asian countries stand out.
2.1 Core Definitions
2.1.1 Labor productivity
4
which is one of the main research objects in this article. Baumol (1986) summarizes that the standard of living of the country with lagging labor productivity will bear brunt of the burden as it is forced, increasingly, to compete by means of relatively low wages. One of the most important targets of policies formulation in East and Southeast Asia is to figure out a better way to improve living standard and alleviate poverty, which can be realized by referring to relevant studies. Also, when it comes to structural change, which focuses on reallocation of labors, more focus should be put on people. To sum up, this paper studies labor productivity change instead of TFP change.
5
Hypothesis 1: The inter-sector labor productivity gaps in East and Southeast Asian countries will decrease as countries grow richer (as measured by GDP per capita).
Now that the productivity gaps between sectors exist, countries are supposed to move labor to higher productivity sectors in order to promote productivity level and catch up with the more productive countries. This catching up process is defined as convergence. The next subsection will introduce convergence and explain some relevant questions.
2.1.2 Convergence
With the non-negligible importance of labor productivity on living standard as well as on economic development, developing countries that are lagging behind are supposed to enhance labor productivity promotion. Convergence happens when developing countries are able to increase their labor productivity growth at a faster rate than developed countries. This subchapter introduces two kinds of convergence: σ-convergence and β-convergence.
6
further analyze convergence, we need to know which type of convergence is studied, since convergence itself is a broad concept that includes different kinds of subclasses. Two most frequently discussed type of convergence are β-convergence and σ-convergence. A distinction is made by Sala-i-Martin (1996) in his work of demonstrating and distinguishing various approaches of convergence analysis: We say that there is β-convergence when the partial correlation between growth in labor productivity over time and its initial level is negative. In other words, there is β-convergence if poor economies tend to grow faster than wealthy ones. An example is given here:
γ𝑟,0,𝑡= α + βlog(y𝑟,0) + ε 𝑟,0
(1)
γ𝑟,0,𝑡 is the annualized labor productivity growth rate of economy r between 0 and t, and let the term
log(yr,0) be the logarithm of economy r’s labor productivity at time 0. ε 𝑟,0 is the error term.If we estimate the regression of equation 1 and find a β < 0, we can say that the data exhibits β-convergence. When β is negative, as mentioned by Young (2008), the partial correlation between growth in labor productivity over time and its initial level is negative, hence β-convergence.
The concept of σ-convergence can be defined as: a group of economies are converging in the sense of σ (known as standard deviation) if the dispersion of labor productivity level tends to decrease over time. That is, if
σ𝑡 < σ0
(2)
7
Hence, if we observe β-convergence, it implies σ-convergence, the potential problem of using β-convergence is solved. To sum up, this paper studies β-convergence instead of σ-convergence.
2.1.3 Structural change
8
labor productivity is to reallocate labors from low-productivity sectors to higher ones. Additionally, this paper focuses on East and Southeast Asian countries, where a large number of people suffering from poverty are located. Therefore, labor movements can be closely linked to people’s living status as well as welfare conditions. Because of the reasons mentioned above, the main focus in this paper is put on labor reallocation aspect. Following McMillan (2011) and de Vries (2013), this paper narrowly defines structural change as the reallocation of labor from low- to high-productivity levels. It’s worthwhile to mention that this definition also reflects a limit of structural change. That is, the sector with high productivity might have limited capacity. For example, mining does operate at high level of labor productivity, but we cannot put everyone in this high-productivity sector. This case explains the decreasing effect of structural change, which we will discuss in the analysis section.
2.2 Various growth patterns
9
following the path depicted by Kuznets and other relevant authors. Also, some developing countries have shown the momentum of growing this way. China, for example, started reallocating labor from agriculturial to industrial sectors few decades ago. However, it remains unknown whether these countries will continue to develop service sector and increase the employment share in service.
10
potential productivity loss, which could have been produced by the industrial sector. Therefore, it can be concluded that the first pattern is more beneficial for sustainable growth.
Due to different growth patterns and different stages of development, some countries depend on the development of the industrial sector while other countries focus on the development of the service sector. However, as already mentioned, the industrial sector provides faster productivity growth. When it comes to labor productivity convergence brought by structural change, it might be the case that countries dependent on industrial sector converge faster towards the US than service-dependent countries? This leads to our second hypothesis.
Hypothesis 2: Industrial-dependent countries1 converge at a faster rate towards the US than
service-dependent countries in term of labor productivity brought by structural change from 1980 to 2010.
However, the growth pattern is decided by a set of factors, such as trade environment, policy guidance and factor endowment. We discuss some characteristics that made East and Southeast Asian countries grow faster than other developing countries in the next subsection.
2.3 What made East and Southeast Asian countries stand out
This subsection introduces the reasons why East and Southeast Asia countries stood out while other developing countries stayed relatively stagnated in the recent period. Three main aspects are concluded: openness to trade, technology upgrade and comparative advantage. The economic structure of Asia will also be briefly discussed.
Remarkable growth in East and Southeast Asian countries has been witnessed in the past few decades. Collins and Bosworth (1996) recorded the fast growth rate in term of GDP per capita, approximately at an average level of 4% in 1980s. Recently published national account data by the OECD (2018) also recorded the high GDP per capita growth rate. The data shows East and
1 Industrial-dependent means that most value-added are generated in industrial sectors, same for service-dependent. the classification of
11
12
opening trade policy, which is the main reason why these countries kept shining during and after the postwar period.
The second aspect refers to technology upgrade, which is composed of both technology access and technology generation. Thanks to trade openness, they got the opportunity to access technology, especially cutting-edge technology. Starting from executing basic manufacture works, these countries gradually improved their technology level. Also, the abilities to generate new technology in Asian countries were massively improved, which is important for overall economic development and labor productivity growth. Fagerberg (1994) believes, except for the factors mentioned in neoclassical models (such as capital accumulation and income distribution), current growth differences across countries is influenced by the technology gap. He points out the potential problems in neoclassical models: the growth brought by R&D and innovation activities is missing. Therefore, the ability to generate new technology can decrease the technology gap across countries thus induce economic growth. Scherer (1982) explores the relationship of technology, R&D and labor productivity, drawing the conclusion that labor productivity gains substantial return from the generation of better technology. With trade openness and technology availability, both of which were missing in both Africa and Latin America, Asian countries were possible to propose structural change and realize substantial growth.
13
to transfer labor from agriculture into manufacturing, which contributes to the structural change. Although Asian countries mentioned above have made impressive achievement in the past few decades, some developed through different patterns while others are at the different stage of development, resulting in diverse economic structures. Figure 1 depicts the employment share change for different East and Southeast Asian countries from 1980 to 2010. The country on the left has least GDP per capita (PPP, 2005 USD) in 1980 while the country on the right side has the most.
Figure 1, Economic structure of East and Southeast Asian countries, 1980-2010.
Data Source: GGDC-10 sector database.
Note: Countries are ordered by GDP per capita (PPP, 2005 USD)in year 1980 from low to high. This figure depicts the economic structure of East and Southeast Asian countries, taking year 1980 and 2010 as comparison. Y-axis denotes employment share. Abbreviations are as follows CHN: China, IND: India, IDN: Indonesia, THA: Thailand, PHL: Philippines, MYS: Malaysia, KOR: Korea (rep. of), TWN: Taiwan, HKG: Hong Kong, SGP: Singapore, JPN: Japan.
It can be concluded from Figure 1 that East and Southeast Asian countries have gone through obvious structural change process. Poorer countries tend to have larger agriculture employment share, and labor moved out of agriculture into industrial and service sectors as countries develop. Richer countries, on the opposite, have relatively large service employment share. Labor move from both agriculturial and industrial sectors into service.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
CHN IND IDN THA PHL MYS KOR TWN HKG SGP JPN
Economic structure of East and Southeast Asian countries
14
3. Data and descriptive statistics
This paper uses data from GGDC 10-sector database, GGDC Penn World Table (PWT), and the World Bank database. The following countries and regions are included in the sample: China, Hong Kong, India, Indonesia, Japan, Korea (Rep. of), Malaysia, Philippines, Singapore Taiwan and Thailand. Other countries such as Mongolia, Laos, Cambodia and Myanmar are excluded due to lack of reliable data source. The time range is set from 1980 to 2010 mainly due to data availability, which imposes some limits on our study. The rapid labor productivity growth in Korea, Taiwan, Singapore and Hong Kong in the 1960s and 1970s would not be included using this time period; thus there is a downward bias on the impact of structural change. Also, the scale of our sample is limited by this time selection. It’s worthwhile to mention that two countries are special in the sample: India and Japan. The former does not belong to the region we’re studying (India belongs to South Asia). But India has shown a greater impact on the world economy in the recent decades. It is hard to ignore such a huge economy. As for Japan, it belongs to the region we study, making it hard to ignore such a huge economy. However, Japan is considered as the only developed country in East and Southeast Asia, which means it might influence our result concerning developing countries. To deal with the potential bias brought by these two countries, we implement robustness test by dropping them from the sample, which will be shown in section 6. Additionally, according to OECD data, the US has been the most productive country since 1970; thus the target country is set as the US.
3.1 Labor productivity and GDP per capita data
15
and constant prices (2005 price level) is taken from National Accounts of various countries. As these have all been complied according to the UN system of National Accounts, international comparability is high, in principle. Besides, for sectoral GDP, GGDC 10-sector database starts with GDP level for the most recent available benchmark year, while historical national accounts were linked to the bench year, through which the growth rate of individual series are retained. Employment is defined as “all persons employed”, thus including all paid employees. Population census is used to ensure the full coverage of working population and a reliable sectoral breakdown. For the purpose of comparability, following McMillan (2011), we combine two of the original sectors (Government Services and Community, Social and Personal Services) into one sector; thus the number of sectors is reduced to 9. Detailed sector coverage, as well as classification, will be shown in Table A1 in appendix.
As for GDP data, we use GDP per capita from the GGDC Penn World Table (PWT), version 7.1. This version provides real GDP per capita data in US dollar at 2005 price level, with purchasing power parity (PPP) index converted, which is consistent with the data in GGDC 10-sector database (Also using PPP 2005, USD). Also, PWT has been regarded as a reliable database for more than 4 decades. Another accessible database is the World Bank. However, they do not provide real GDP per capita at 2005 constant price level.
3.2 Control variables
Except for GDP per capita, there are some other variables that can be influential to labor productivity gaps between sector. This subsection introduces these variables and discusses why they are relevant to the productivity gaps between sectors.
16
GDFP) from World Bank database, the time coverage is 1980 to 2010. This indicator can reflect the openness to trade since export and import are the main activity in international trade, measuring the sun as a share of GDP can show the extent of involvement in the trade.
Secondly, labor abundance, which can be measured as population. Because large population provides abundant labor supply, especially in industrial sectors. As we discussed already, numerous labor supply attracts firms to offshore their activities, mainly manufacturing activities to countries with low unit labor cost. As a consequence, the labor productivity gap between both industrial and service sector and industrial and agricultural sector would increase as population keep growing. Population data is also taken from PWT, version 7.1, which is in line with the population data from the World Bank, which is considered to be reliable.
17
investments, the labor productivity level should also increase faster than other sectors. Therefore, the labor productivity gap between both industrial and service sector and industrial and agricultural sector would increase with more FDIs. We use FDI net inflow from the World Bank database. One potential problem of FDI data is that it uses current price. In order to minimize the bias, we use GDP deflator to process it. Moreover, in order to scale the data to be in line with GDP per capita, we make further calculation to transfer it into FDI per capita. To provide more details about the data used, Table 1 provides data for all countries in year 2010 while Table 2 provides detailed data descriptions including number of observations, mean, standard deviation, maximum and minimum.
Table 1 Descriptive statistics, part 1.
Population GDP per capita FDI per capita Agricultural labor productivity Industrial labor productivity Service labor productivity Openness to trade Education SGP 5076732 55838.63 10848.68 11294.17 497963.99 384143.73 373.44 28.55 HKG 7024200 38688.13 3900.36 14398.09 629756.28 383381.48 404.77 55.90 TWN 23024956 32117.70 - 20514.89 1833527.61 351942.05 - - MYS 28112289 11961.50 387.22 24799.81 1092085.96 205439.84 157.94 - KOR 49554112 26613.77 191.66 21572.61 634878.21 141688.48 95.65 104.21 THA 67208808 8065.57 219.42 3444.60 600690.55 113767.22 127.25 49.39 PHL 93726624 3193.63 11.42 4310.61 207399.89 384143.73 71.42 28.55 JPN 128070000 31453.08 58.10 17809.60 425830 266128.17 28.61 57.32 IDN 242524123 3965.80 63.05 4857.02 303724.66 109970.49 46.70 22.38 IND 1230980691 3476.78 12.42 2704.56 144956.54 111268.50 49.69 16.11 CHN 1337705000 7129.74 182.18 3521.02 180464.52 151527.90 48.89 22.40
Data Source: GGDC 10-sector database, PWT and World Bank database
18 Table 2. Descriptive statistics, part 2
Variable Obs Mean Std. Dev. Min Max
Population 341 245000000 403000000 2413945 1340000000
GDP per capita (PPP 2005, $) 341 12658.63 11919.58 563.57 55838.63 FDI per capita (Constant price, $) 306 0.0621 0.1648 -0.0033 1.1249
Openness to trade (%) 310 113.69 109.88 12.35 441.60
Education (%) 244 25.06 20.23 1.14 104.21
Agricultural labor productivity 341 10073.04 8416.97 849.66 38732.50 Industrial labor productivity 341 368591.40 361190.90 23238.20 1977470.00 Service labor productivity 341 184867.60 112810.90 15453.02 447890.60 Gap 1: Industrial/agricultural 341 46.82 32.15 3.49 174.39
Gap 2: Industrial/Service 341 2.38 2.51 0.30 12.94
Data Source: GGDC 10-sector database, PWT and World Bank database
Note: This table shows the descriptive data from 1980 to 2010. Obs means the number of observations. Std.Dev denotes standard deviation. Measurement units are as follows. GDP per capita, dollar, 2005 constant price, PPP; FDI per capita, dollar, constant price; productivity: dollar/person employed, 2005 constant price, PPP; Openness to trade, %, the ratio of the sum of import and export and GDP; Education, %, tertiary school enrollment rate.
4. Methodology
To answer our research question and examine our above-mentioned hypotheses, this paper uses both modified shift-share analysis and regressions upon a panel dataset. This section introduces the methodologies used in this paper following the following order. First, the regression model for hypothesis 1. Second, modifying Shift-share analysis. Third and last, β-convergence and hypothesis2.
4.1 Regression method for hypothesis 1
19
calculated as the ratio of labor productivity in industrial sector and labor productivity in service sector (Industrial/Service). Through this way, the possibility of productivity gap being minus is solved.
After defining the labor productivity gap between sectors, models are constructed by adding control variables mentioned. Equation 3 and Equation 4 are the main regression models for hypothesis 1:
𝑙𝑛 (gap1)𝑟,𝑡=𝑎0+𝑎1ln(GDP)𝑟,𝑡+𝑎2ln(N𝑟,𝑡)+𝑎3𝑙𝑛(FDI𝑟,𝑡)+𝑎4EDU𝑟,𝑡+𝑎5EDU2𝑟,𝑡+𝑎6OTT𝑟,𝑡+λr+µt+ ϵr,t (3)
𝑙𝑛 (gap2)𝑟,𝑡 =𝑏0 +𝑏1 ln(GDP)𝑟,𝑡 +𝑏2 ln(N𝑟,𝑡) +𝑏3𝑙𝑛 (FDI𝑟,𝑡) +𝑏4EDU𝑟,𝑡 +𝑏5EDU2𝑟,𝑡 +𝑏6OTT𝑟,𝑡 +λr +µt +
ήr,t (4)
ln(GDP)𝑟,𝑡 denotes the logarithm of GDP per capita and ln(N𝑟,𝑡)is log population. 𝑙𝑛(FDI𝑟,𝑡) represents the FDI per capita. r and t are economy r and time r, separately. There are three other variables that are not logged. First, EDU is the tertiary school enrollment rate Second, EDU2, as mentioned, is a term to test the decaying effect of education. Third and last, OTT means the openness to trade, which is measured as the sum of exports and imports divided by GDP. At the same time, these two equations include economy (country) and time fixed effects, with λi being
a country-specific effect and µt a time-specific effect. By combining country fixed effect and time fixed effect together, the impact of each variable is less distorted by a possible omitted variable bias. ϵr,t and ήr,t are error terms.
20
multicollinearity. If the value is far lower than 10, we can say that the problem of multicollinearity does not matter. Table A2 reports the information concerning multi-collinearity. It can be concluded from table A2 that multicollinearity is not a problem here. Second, the problem of heteroskedasticity. A white test is undertaken to test the existence of heteroskedasticity. It turns out that the problem of heteroskedasticity does exist; thus we need to use command "robust" to deal with it. Third, the problem of endogeneity. The main potential endogeneity in the model is that GDP per capita level might in turn influence the labor productivity gap, which we call reverse causality. To deal with this problem, we use the lagged variables. The result of reverse causality test will be given in the robustness tests part.
4.2 Shift-share analysis and modification
For the purpose of analyzing structural change and labor productivity growth, a decomposition on labor productivity growth using shift-share analysis is needed. Based on the conventional shift-share analysis used by McMillan (2011) and Rodrik (2015), this paper modifies some details and put forward the modified shift-share analysis based on the work by de Vries (2013). Conventional shift-share analysis divides labor productivity growth into two parts: First, productivity can grow within sectors, through capital accumulation, technological change, etc. Second, labor can move across sectors, from low-productivity sectors to high-productivity sectors, increasing overall labor productivity in the economy. This can be written in equations using the following decomposition:
𝑝𝑡− 𝑝0 = ∑ (𝑝
𝑖𝑡− 𝑝𝑖0) 𝑠𝑖0 𝑛
𝑖=1 + ∑𝑛𝑖=1(𝑠𝑖𝑡-𝑠𝑖0) 𝑝𝑖𝑡 (5) Where 𝑝𝑡 and 𝑝0 are aggregate labor productivity of time t and 0, 𝑠
𝑖0 and 𝑠𝑖𝑡 are the employment share of sector i in time 0 and t, respectively. Meanwhile, 𝑝𝑖𝑡 and 𝑝𝑖0 represent the
21
between effect depends on whether labors are reallocated to higher-productivity sector or lower ones.
However, this conventional method is not always undisputed. The within effect term in equation 5 is only a static measure of the structural change effect since it depends on the difference in productivity levels across sectors, not growth rates. This means that we do not know whether workers move into sectors that have faster growth rate. To deal with this “static” problem, an alternative method accounts for the possibility that growth and levels across sectors are negatively correlated. This decomposition method uses base period for both employment share and productivity. Additionally, a third term is introduced into the equation:
𝑝𝑡− 𝑝0 = ∑ (𝑝
𝑖𝑡− 𝑝𝑖0) 𝑠𝑖0 𝑛
𝑖=1 + ∑𝑛𝑖=1(𝑠𝑖𝑡-𝑠𝑖0) 𝑝𝑖0 + ∑𝑛𝑖=1(𝑝𝑖𝑡− 𝑝𝑖0) * (𝑠𝑖𝑡− 𝑠𝑖0) (6) The first term is the within effect, similar to equation 6. It is positive when the change in labor productivity levels in sectors is positive. The second term now uses the productivity in base period t=0, measuring the contribution of labor reallocation across sector. This term is positive when labor moves from less to more productive sectors. The third term is known as the interaction term, which represents the joint effect of changes in employment and productivity. The interaction term is positive if workers are moving to sectors that are experiencing positive productivity growth.
22 𝑝𝑡− 𝑝0= ∑ (𝑝 𝑖𝑡− 𝑝𝑖0)𝑠𝑖0 𝑛 𝑖=1 +∑ (𝑠𝐽𝑗 𝑗𝑡-𝑠𝑗0)(𝑝𝑗0− 𝑝0∗)+∑ [(𝑝𝐽𝑗 𝑗𝑡− 𝑝𝑗0)− (𝑝𝑡∗− 𝑝0∗)]*(𝑠𝑗𝑡− 𝑠𝑗0) (7) 𝑝0∗= ∑ (𝑆 𝑘𝑡− 𝑆𝑘0) 𝐾 𝑘 𝑝𝑘0/∑ (𝑆𝐾𝑘 𝑘𝑡− 𝑆𝑘0) (8) 𝑝𝑡∗= ∑ (𝑆 𝑘𝑡 − 𝑆𝑘0) 𝐾 𝑘 𝑝𝑘𝑡/∑ (𝑆𝐾𝑘 𝑘𝑡 − 𝑆𝑘0) (9) Where J denotes the set of expanding sectors and K is the set of shrinking sectors. 𝑝0∗and 𝑝𝑡∗are the average labor productivity of shrinking sectors at time 0 and t. With the application of J and K, the problem that expanding sectors might have below-average productivity level is solved. Therefore, equation 7 is our main model for shift-share analysis. As for the three terms in equation 7, while the first term measures the growth within sectors, the second term measures whether workers move to above-average productivity sectors (static structural change). The third term, still works as an interaction term, measuring whether productivity is higher in sectors that expand (dynamic structural change).
Recall our research question: we want to study the contribution of structural change on
productivity convergence. Therefore, based on our decomposition results, only the second and
the third term are needed for our convergence study. We add up these two terms and generate
labor productivity growth due to structural change, which can be written as equation 10. Psc,r,0,t
represents the labor productivity growth due to structural change in economy r from time 0 to
t.
Psc,r,0,t = ∑𝐽𝑗(𝑠𝑗𝑡-𝑠𝑗0) (𝑝𝑗0− 𝑝0∗) + ∑ [(𝑝𝐽𝑗 𝑗𝑡− 𝑝𝑗0) − (𝑝𝑡∗− 𝑝0∗)] * (𝑠𝑗𝑡− 𝑠𝑗0) (10)
With the decomposition results of modified shift-share analysis, we can decompose the
labor productivity growth rate and get the growth rate due to structural change. Also,
regressions on labor productivity convergence due to structural change can be
23
4.3 Convergence regression
According to the definition of β-convergence, there is β-convergence if the partial correlation between labor productivity growth rate and the initial productivity level is negative. However, in order to see the specific convergence brought by structural change, we make a change to equation 1. The new convergence equation is given by equation 11.
𝛾𝑟,0,𝑡 𝑆𝐶= α + blog(𝑦𝑟,0) + 𝜀 𝑟,0 , b= -β (11)
Where 𝛾𝑟0,𝑡 𝑆𝐶 denotes the labor productivity growth rate due to structural change. 𝑦𝑟,0is the initial productivity level of country r and 𝜀 𝑟,0 is the error term. Due to the default setting of STATA (STATA runs the linear regressions with plus only), we use b instead of β to keep the
sign of estimated value of β in line with the definition.
According to Sala-i-Martin (1995), we can test the existence of β-convergence through
regressing equation 11. However, a limit of equation 11 is that the time lengths have to be
identical. Since the model restricts the time period from 0 to t for all countries. In order to
compare the speed of convergence using data that have different time lengths, a modification
on equation 11 is needed. The following non-linear equation is used by Sala-i-Martin and
subsequent studies:
γ𝑟,0,𝑡 𝑠𝑐= 𝑎 − [(1 − 𝑒−𝛽𝑇)/𝑇]log (y𝑟,0)+ εr,0,t (12)
T is the length of period. The regression result can be seen as the speed of β -convergence since
the comparability is better. Note that both equation 11 and equation 12 can prove the existence of β -convergence. Because we have data with different time length, we use equation 12 as our main regression model for β -convergence. If we find a β >0, we can say that the data exhibits β -convergence. The larger β is, the faster countries converge towards the US in term of labor productivity.
With plausible convergence equation, we can carry out the test for hypothesis 2. Since countries
are divided into industrial-dependent and service-dependent, we need value-added data for
24
Figure 2 the share of Value added in different sectors, from 1980 to 2010
Data source: GGDC 10-sector database.
Note: countries are ordered from least to most GDP per capita in 1980 (PPP, 2005 $). Y-axis is the share of labor productivity, and X-axis shows countries and time. Abbreviations are as follows CHN: China, IND: India, IDN: Indonesia, THA: Thailand, PHL: Philippines, MYS: Malaysia, KOR: Korea (rep. of), TWN: Taiwan, HKG: Hong Kong, SGP: Singapore, JPN: Japan.
Using value-added from figure 2, we divide countries into two groups judging from where the majority of value-added are generated. The first group is industrial-dependent countries. Four countries are included in this group. They are China, Indonesia, Malaysia and Thailand. The second group is service-dependent countries. India, Japan, Korea, the Philippines, Singapore, Hong Kong and Taiwan are included. The speed of productivity convergence will be measured for both of the groups. Some might argue that for Malaysia, the share of value-added in service sector is more than industrial sectors in 2010. Since the disparity of value-added share between industrial and service sector is larger in 1980 than 2010. We take Malaysia as an industrial-dependent country from 1980 to 2010.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.91
Value added from 1980 to 2010 (2005 constant
prices, PPP $)
25
5. Results and analysis
This section discusses main results of this paper. As mentioned in the methodology section, the approaches are based on modified shift-share analysis as well as fixed effect regressions. We organize the order of results as follows: First, results for hypothesis 1. Second, results for modified shift-share analysis. Third and last, results for convergence regression as well as hypothesis 2. Corresponding analysis and explanations of these results are also given.
5.1 Results for hypothesis 1
The results for hypothesis 1 are shown in table 3. Panel (a) shows the correlation between gap1(labor productivity gap between industrial sector and agriculture) and GDP per capita while panel (b) shows the correlation between gap2 (labor productivity gap between industrial sector and service) and GDP per capita. As we already mentioned, each panel shows 6 different regression results. Column 6 is the result with all control variables included. Also, as it can be seen from the tables, we account for both country and time fixed effects.
26
that not all countries were focusing on the development of industrial sector. We have already mentioned that the growth patterns as well as the development phases of East and Southeast Asian countries are diverse. This fact indicates that some countries would focus on the development of service sector, even at the expense of industrial sector.
As for the result being opposite to the findings on the EU and the US, we conclude that the development of industrial sector in those developed countries has reached a high level, which leaves little space for labor productivity growth in industrial sector. However, industrial sectors for countries in East and Southeast Asia remain to be developed. Therefore, the results on productivity gap between industrial and service sector turn out to be different from empirical studies concerning developed countries in western-world.
The results for gap2 (gap between industrial sector and service sector) turns out to be insignificant. Similar to the small coefficient of results for gap1, different growth patterns and phases of development result in the insignificance. Countries that depend on the development of industrial sector have relatively weak productivity growth in service sector. Therefore, as countries become richer, some realized more productivity in service sector while other countries made industrial sector more productive. With different growth patterns, it is hard to find a significant correlation between GDP per capita and gap 2. However, the positive sign still to some extent indicates that industrial sector can generate more labor productivity than service sectors.
27
Table 3 (a)．Regression result for hypothesis 1. Effect of variables on the labor productivity gap between industrial and agricultural sectors (gap1)
*** p<0.01, ** p<0.05, * p<0.1 Robust standard errors in parentheses Data source: GGDC 10-sector database & PWT, World Bank database.
Note: This table shows the impact of country’s wealth on its inter-sector productivity gap. Gap1 denotes: labor productivity in industrial sector/ labor productivity in agricultural sector. Term EDU2 is calculated as Education*Education.
(1) (2) (3) (4) (5) (6)
VARIABLES Log gap1 Log gap1 Log gap1 Log gap1 Log gap1 Log gap1 Log GDP per capita 0.437 0.426 0.387 0.560** 0.439 0.574*** (0.246) (0.311) (0.338) (0.237) (0.282) (0.131)
Log Population -0.156 0.306 0.701 1.168 0.267
(1.055) (0.842) (0.956) (0.891) (0.651)
Log FDI per capita 0.0932 0.0681 0.0541 -0.0017
(0.0879) (0.0967) (0.0796) (0.0347) Education -0.0003 0.0262 0.0090 (0.0062) (0.0219) (0.0098) EDU2 -0.0002 -0.0007 (0.0002) (0.0007) Trade Openness 0.0078** (0.0030) Observations 341 341 291 227 227 227 R-squared 0.346 0.346 0.401 0.459 0.498 0.696
Country FE Yes Yes Yes Yes Yes Yes
28
Table 3 (b)．Regression result for hypothesis 1. Effect of variables on the labor productivity gap between industrial and service sectors (gap2)
*** p<0.01, ** p<0.05, * p<0.1 Robust standard errors in parentheses Data source: GGDC 10-sector database & PWT, World Bank database.
Note: This table shows the impact of country’s wealth on its inter-sector productivity gap. Gap2 denotes: labor productivity in industrial sector/ labor productivity in service sector. Term EDU2 is calculated as Education*Education.
(1) (2) (3) (4) (5) (6)
VARIABLES Log gap2 Log gap2 Log gap2 Log gap2 Log gap2 Log gap2 Log GDP per capita 0.286 0.197 0.258 0.328 0.254 0.327
(0.429) (0.446) (0.466) (0.326) (0.324) (0.281)
Log Population -1.248 -1.284 0.337 0.623 0.140
(0.999) (1.161) (0.952) (0.823) (0.750)
Log FDI per capita 0.0288 -0.0149 -0.0235 -0.0534
(0.0438) (0.0502) (0.0446) (0.0339) Education 0.0161** 0.0324** 0.0232* (0.0050) (0.0106) (0.0115) EDU2 -0.0002 -0.00005 (0.00008) (0.00008) Trade Openness 0.0042** (0.0018) Observations 341 341 291 227 227 227 R-squared 0.300 0.353 0.452 0.611 0.624 0.677
Country FE Yes Yes Yes Yes Yes Yes
29
5.2 Results for shift-share analysis
Using the modified shift-share analysis in equation 10, the decomposition results are listed in this subsector. Table 4 shows the decomposition of annual labor productivity growth rate in East and Southeast Asia from year 1980 to year 2010.
Table 4. Decomposition result using modified shift-share analysis.
Data Source: GGDC 10-SECTOR database. This table records the decomposed annual labor productivity growth rate from 1980 to 2010. Within means that labor productivity growth due to within-sector growth, which is the first term on the right-hand side of equation 10. Static and dynamic corresponds to the second and the third term, denoting the labor productivity growth due to structural change. And the term total is the sum of within, static and dynamic.
Table 4 is consisted of 4 columns Column 1 to 3 represent the three terms in equation 7. First, “within” means the growth within sectors. Second, “static” is the static structural change term, which measures whether workers move to above-average productivity sectors. Third, “dynamic” represents the dynamic structural change term, measuring whether productivity is higher in
(1) (2) (3) (4)
Within Static Dynamic Total
30
sectors that expand. Colum 5 is the aggregate labor productivity growth rate, which equals to the sum of column 1 to 3.
It can be seen from Table 4 that all countries in the sample have gone through labor productivity growth in the past three decades. China realized the fastest productivity growth, with an annual growth rate of 7.61% while Philippines has relatively lower annual growth rate, only 0.53% per year. Besides, other countries and regions such as Hong Kong, India and Korea also grew at a fast speed. Another conclusion is that within effect is the main factor that contributes to productivity growth, taking the majority of the total productivity growth. Negative dynamic structural change effects are observed for Hong Kong, Japan, Korea, Philippines, Malaysia, Singapore and Taiwan. This indicates that the sectors that expand their employment shares had productivity growth rate below those of shrinking sectors, which might due to labors shifting away from industrial sector, mostly into service sector. As labor keep moving out of industrial sector into service sector, the contribution of dynamic structural change becomes negative. Therefore, structural change works better for industrial-dependent countries, like China and Thailand. Additionally, the only country that has negative static structural change growth is Malaysia, which indicates that workers in Malaysia have moved to below-average productivity sectors. Other than the points mentioned above, attention should be paid to the growth rate of America in the past thirty years. The labor productivity growth rate of the US is lower than most countries in the sample, only exceeding the Philippines. Also, the sum of static term and dynamic term is negative, indicating that structural change has been dragging the productivity growth in the US during the past three decades. This implies that those countries that grasped the chance to implement structural change, especially moving labor from agriculturial to industrial sector, should be able to catch up with the US in term of labor productivity growth. However, we cannot conclude the tendency of productivity convergence from table 4, nor can we know which groups of countries caught up faster, which is why we need to regress the convergence equation to further study this process of catching up.
31
32
Figure 3 labor productivity growth rate due to structural change in Asian countries, from 1980 to 2010. Dividing time into 6 periods.
Note: Data source: GGDC 10-sector database. Y-axis is the percentage of productivity growth while X-axis shows different countries and different time periods.
Figure 4 aggregate labor productivity growth rate in Asian countries, from 1980 to 2010. Dividing time into 6 periods.
Note: Data source: GGDC 10-sector database.Y axis is the percentage of productivity growth while X-axis shows different countries and different time periods.
-0.02 -0.015 -0.01 -0.005 0 0.005 0.01 0.015 0.02 0.025 0.03 0.035
HKG KOR CHN IDN JPN IND PHL SGP THA TWN MYS
Annual Labor productivity growth due to structrual change (1980-2010), divided into 6 time periods
1980-1984 1985-1989 1990-1994 1995-1999 2000-2004 2005-2010 -0.06 -0.04 -0.02 0 0.02 0.04 0.06 0.08 0.1 0.12
HKG CHN KOR MYS IND TWN THA IDN JPN SGP PHL
Annual Labor productivity growth (1980-2010) divided into 6 time periods
33
5.3 Result for labor productivity convergence and hypothesis 2
34 Table 5. result for convergence regression
(1) (2) (3) (4) β 0.0023*** 0.0025*** 0.0030** 0.0029*** (0.0008) (0.0007) (0.0012) (0.0009) α 0.0285*** 0.0299*** 0.0334*** 0.0349*** (0.0079) (0.0074) (0.0116) (0.0091) Observations 72 72 30 48
Data source: GGDC 10-sector database. Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 Note: column 1 is the result for equation 11 while column 2-4 are results of regression using equation 12.
6. Robustness tests
35
result to be significant.
As for labor productivity convergence, we also study the impact of sample change on β. The results are shown in Table A32 in the appendix. We can see that the results are still in line with the initial findings, which confirms the robustness. This means that countries in our sample have been converging towards the US in term of labor productivity. It is not driven by a single or a few countries.
Another potential problem in the model is endogeneity, or reverse causality, which means that the independent variable is influenced by the dependent variable thus the independent variable is not exogenous. The labor productivity gap between sectors might in turn influence the GDP per capita level. Since the magnitude of productivity gap might influence the economic structure, it is worthwhile to test whether this reverse causality exists. As mentioned in the result analysis section, we will use the lagged variables, through which the dependent variable does not have direct impacts on the independent variable. Table A6 shows the results using lagged variables. It can be included that the results still hold. Therefore, we exclude the potential threat of reverse causality.
7. Conclusion Remark
Through modified shift-share analysis and regressions using panel data, the following conclusions are drawn. First, East and Southeast Asian countries have gone through successful labor productivity growth in the past three decades, due to both structural change and growth within sectors. Although the tendency of growth is the same for those countries, the developing patterns they went through show various characteristics. While some countries like China and Thailand relying on industrial sectors, other countries like Tai Wan, Korea and Japan focus more on encouraging labor to move into service sector. Despite different structural change patterns, these countries show some common features that distinguish them from developing countries in other regions. Three main aspects are summarized: openness to trade, technology upgrade, and labor abundance.
2
36
Second, different from the EU and the US, countries in East and Southeast Asia show different results concerning the relationship of country wealth and inter-sector gap. As countries get richer, the productivity gap between industrial and agricultural sector increases. But this result does not hold for developing countries after we drop Japan from our sample. Additionally, no correlation between country wealth and productivity gap between industrial and service sectors is found.
Third, East and Southeast Asian countries have realized stable labor productivity growth in the past three decades. China realized the most remarkable growth, at an annual growth rate of 7.6% while the Philippines had the lowest growth rate in the sample, with only 0.53% per year. While growth within sectors takes the most proportion of labor productivity growth, structural change seems to work better for less developed countries. This impact decreases as the process of structural change goes further. Sometimes structural change might even bring negative productivity growth. Therefore, structural change is not the master key for growth. It can be rather helpful in the early stage of growth, but this benefit decreases as countries develop. Fourth and last, Structural change did contribute to the labor productivity convergence of East and Southeast Asian countries towards the US. This finding indicates that underdeveloped countries should pay more attention to structural change since it provides a chance of catching up. However, the growth within sectors should also be emphasized. Because growth within sectors is the main source of labor productivity growth and it is sustainable in long-term. To sum up, although structural change is not the “master key” for labor productivity growth, it does provide appreciable productivity growth for underdeveloped countries. In order to further reinforce labor productivity growth, both the growth within sectors and structural change are needed.
37
deflator so that the bias is reduced as much as possible, it still remains to be a problem of our study. Besides, some education data is missing. Moreover, due to data limitation, many countries in Southeast Asia, such as Laos, Cambodia, Vietnam are excluded. The development of these countries is also of great impact. Excluding them will influence our result on East and Southeast Asia. Second, our model concerning convergence regression still exposes some problems. The most important problem is the lack of control variables. Since β-convergence is derived from neoclassical model concerning economic growth, it is related to some aspects such as steady state, which is not consistent with labor productivity study. Therefore, finding proper control variables is a big challenge. Moreover, the convergence speed we have now is a speed for the whole sample, which means that we cannot calculate the speed for a specific country. Although the current result can reflect the existence of labor productivity convergence as well as the speed, this method remains to be optimized. Third, there is a potential bias in our sample selection. Since we want to study the successful cases and gain inspirations that can be helpful, countries in our sample are mostly developing countries with relatively good growth in recent times. This might bias the impact of structural change upward.
38
7. References:
Baldwin, R. E. (2004). Openness and growth: What's the empirical relationship? In Challenges to Globalization: Analyzing the economics 499-526. University of Chicago Press.
Barro, R.J. & Sala-i-Martin, X. (1992). Convergence, Journal of Political Economy, Vol.100. No. 2. 223-251.
Baumol, W. J. (1986). Productivity Growth, Convergence, and Welfare: what the long run data show. American Economic Review. Vol.76. Issue 5, 1072-1085
Baumol, W. J. (2001). Paradox of the services: exploding costs, persistent demand. Raa, Schettkat, 3-28.
Berman, E., Bound, J., & Griliches, Z. (1994). Changes in the demand for skilled labor within US manufacturing: evidence from the annual survey of manufactures. The Quarterly Journal of Economics, 109(2), 367-397.
Bosworth, B., & Collins, S. M. (2008). Accounting for growth: comparing China and India. Journal of Economic Perspectives, 22(1), 45-66.
Collins, S. M., Bosworth, B. P., & Rodrik, D. (1996). Economic growth in East Asia: accumulation versus assimilation. Brookings papers on economic activity, 1996(2), 135-203.
Dasgupta, S., & Singh, A. (2006). Manufacturing, services and premature deindustrialization in developing countries: A Kaldorian analysis (No. 2006/49). Research Paper, UNU-WIDER, United Nations University (UNU).
De Long, J. (1988). Productivity Growth, Convergence, and Welfare: Comment. The American Economic Review, 78(5), 1138-1154.
De Vries, G. J., Timmer, M. P., & de Vries, K. (2013). Structural transformation in Africa: Static gains, dynamic losses. (GGDC Working Papers; Vol. GD-136). Groningen: GGDC.
Diao, X. S., Harttgen, k & McMillan, M (2017). The Changing Structure of Africa’s Economies, The World Bank Economic Review, Volume 31, Issue 2, 1 June 2017, 412–433
Enright, M. J., Scott, E. E., & Chang, K. M. (2005). Regional powerhouse: the greater Pearl River Delta and the rise of China (p. 68). Chichester: Wiley.
Fabricant, S. (1942). Employment in Manufacturing, 1899-1939. National Bureau of Economic Research Inc. Number 41.
39
Fei, C.H. & Ranis, G. (1961). A theory of Economic development. The American Economic Review. Vol.51, No.4. 533-565
Heston, A., Summers, B. & Aten, b. (2012) Penn World Table Version 7.1 Center for International Comparisons of Production, Income and Prices at the University of Pennsylvania, November 2012.
Islam, N. (1995). Growth empirics: a panel data approach. The Quarterly Journal of Economics, 110(4), 1127-1170.
Jorgenson, D. W., & Timmer, M. P. (2011). Structural change in advanced nations: a new set of stylised facts. The Scandinavian Journal of Economics, 113(1), 1-29.
Kaldor, N. (1967). Strategic Factors in Economic Development. Economic Development and Cultural Change. Vol.17, No.2. 277-280
Kinoshita, Y., & Campos, N. F. (2003). Why does FDI go where it goes? New evidence from the transition economies (No. 3-228). International Monetary Fund.
Krueger, A. O. (1997). Trade policy and economic development: how we learn (No. w5896). National Bureau of Economic Research.
Kuznets.S (1966) Modern economic growth: Rate, structure, and spread. Journal of Political Economy, 1966, 37, 475-486
Kuznets.S (1971) Economic Growth of Nations. Irvington, 1971, 35 (1) :275-326
Kuznets.S (1973) Modern economic growth: findings and reflections. The American Economic Review Vol. 63, No. 3 pp. 247-258.
Lewis, W. A. (1954). Economic Development with Unlimited Supplies of Labour. The Manchester School of Economic and Social Studies, 22, 139–191.
Los, B., Timmer, M. P., & Vries, G. J. (2015). How global are global value chains? A new approach to measure international fragmentation. Journal of Regional Science, 55(1), 66-92.
Maddison, A. (1980). Economic growth and structural change in the advanced countries. Western Economies in Transition.
40
McMillan, M., Rodrik, D., & Verduzco-Gallo, Í. (2014). Globalization, Structural Change, and Productivity Growth, with an Update on Africa. World Development, 63, 11–32.
Miketa, A., & Schrattenholzer, L. (2004). Experiments with a methodology to model the role of R&D expenditures in energy technology learning processes; first results. Energy Policy, 32(15), 1679-1692.
Nelson, R. R. & Pack, H. (1999). The Asian Miracle and Modern Growth Theory. The Economic Journal, Vol.109, No.456, 416-436
O'Brien, R. M. (2007). A caution regarding rules of thumb for variance inflation factors. Quality & Quantity, 41(5), 673-690.
OECD (2018), Economic Outlook for Southeast Asia, China and India 2018. OECD Publishing, Paris. http://dx.doi.org/10.1787/9789264286184-en
Prescott, C. (1998). Lawrence R. Klein Lecture 1997: Needed: A Theory of Total Factor Productivity International Economic Review. vol 39, No. 3. 525-551
Ramasamy, B., & Yeung, M. (2007). Malaysia–Trade policy review 2006. The World Economy, 30(8), 1193-1208.
Rodriguez, F., & Rodrik, D. (2000). Trade policy and economic growth: a skeptic's guide to the cross-national evidence. NBER macroeconomics annual, 15, 261-325.
Rodrik, D. (2015). Premature Deindustrialization. Journals of Economic Growth, March 2016, Volume 21, Issue 1, 1–33
Sala-i-Martin, X. (1996). The Classical Approach of Convergence Analysis. The Economic Journal. Vol.106, No.436. 1019-1036
Sala-i-Martin, X. (1994). Cross-sectional regressions and the empirics of economic growth. European Economic Review. Vol.38, Issue 3-4. 739-747
Sally, R. (2007). Thai Trade Policy: From Non‐discriminatory Liberalisation to FTAs. The world economy, 30(10), 1594-1620.
Scherer, F. M. (1982). Inter-industry technology flows and productivity growth. The review of economics and statistics, 627-634.
Solow, R.M. (1956), A Contribution to the Theory of Economic Growth, The Quarterly Journal of Economics, Volume 70, Issue 1, 1 February 1956, 65–94
The World Bank database. Net FDI Inflow.
41
The World Bank database. Tertiary school enrollment.
https://data.worldbank.org/indicator/SE.TER.ENRR
The World Bank database. Trade (% of GDP).
https://data.worldbank.org/indicator/NE.TRD.GNFS.ZS
Timmer, M. P., Erumban, A. A., Los, B., Stehrer, R., & de Vries, G. J. (2014). Slicing up global value chains. Journal of Economic Perspectives, 28(2), 99-118.
Timmer, M. P., de Vries, G. J., & de Vries, K. (2015). Patterns of Structural Change in Developing Countries. Routledge Handbook of Industry and Development. 65-83.
Timmer, M. P., & de Vries, G. J. (2009). Structural change and growth accelerations in Asia and Latin America: a new sectoral data set. Cliometrica, 3(2), 165-190.
42
Appendix
Table A1. GGDC 10-Sector database, Sector coverage and classification
Sector Name Classification
1. Agriculture, hunting, forestry and fishing (AtB); Agricultural sector
2. Mining and quarrying (C); industrial sector
3. Manufacturing (D); industrial sector
4. Electricity, gas and water supply (E); industrial sector
5. Construction (F); industrial sector
6. Wholesale and retail trade, hotels and restaurants (GtH); Service sector 7. Transport, storage, and communication (I); Service sector 8. Finance, insurance, real estate and business services (JtK); Service sector 9. Government services,Community, social and personal services (OtP, LtN); Service sector
Note: It is mentioned in the data part that Government services and Community, social and personal services are merged as one sector in our study.
Table A2. Variation Inflator Factor value
Variable VIF OTT 5.85 lfdipc 3.89 lpop 3.77 lgdp 3.57 EDU 2.28 Mean VIF 3.87
Note: OTT, Openness to trade; lfdipc, logarithm of FDI per capita; lpop logarithm of population; lgdp, logarithm of GDP per capita EDU, education
Table A3, result of convergence robustness test
Drop India Drop Japan
β 0.002265*** 0.002302***
(0.0008) (0.0008)
α 0.0278*** 0.0285 ***
(0 .0084) (0.0079)
43
Table A4 (a)．Result of dropping hypothesis 1, India from the sample. Effect of variables on the labor productivity gap between industrial and agricultural sectors (gap1)
*** p<0.01, ** p<0.05, * p<0.1 Robust standard errors in parentheses Data source: GGDC 10-sector database & PWT, World Bank database.
Note: Gap1 denotes: labor productivity in industrial sector/ labor productivity in agricultural sector. Term EDU2 is calculated as Education*Education.
(1) (2) (3) (4) (5) (6)
VARIABLES Log gap1 Log gap1 Log gap1 Log gap1 Log gap1 Log gap1 Log GDP per capita 0.425 0.420 0.378 0.574* 0.457 0.590*** (0.266) (0.326) (0.357) (0.256) (0.293) (0.139)
Log Population -0.0730 0.386 0.786 1.242 0.300
(1.068) (0.864) (0.970) (0.907) (0.671)
Log FDI per capita 0.0974 0.0688 0.0532 -0.00519
(0.0938) (0.105) (0.0870) (0.0387) Education -0.000931 0.0256 0.00904 (0.00626) (0.0215) (0.0103) EDU2 -0.000193 -7.49e-05 (0.000152) (6.82e-05) Trade Openness 0.00776** (0.00308) Observations 310 310 266 206 206 206 R-squared 0.349 0.349 0.402 0.466 0.504 0.698
Country FE Yes Yes Yes Yes Yes Yes
44
Table A4 (b)．Result of hypothesis 1, dropping India from the sample. Effect of variables on the labor productivity gap between industrial and service sectors (gap2)
*** p<0.01, ** p<0.05, * p<0.1 Robust standard errors in parentheses Data source: GGDC 10-sector database & PWT, World Bank database.
Note: Gap2 denotes: labor productivity in industrial sector/ labor productivity in service sector. Term EDU2 is calculated as Education*Education.
(1) (2) (3) (4) (5) (6)
VARIABLES Log gap1 Log gap1 Log gap1 Log gap1 Log gap1 Log gap1 Log GDP per capita 0.305 0.220 0.290 0.370 0.291 0.365
(0.462) (0.472) (0.500) (0.360) (0.356) (0.306)
Log Population -1.192 -1.252 0.358 0.666 0.147
(0.998) (1.160) (0.935) (0.791) (0.738)
Log FDI per capita 0.0243 -0.0263 -0.0368 -0.0691*
(0.0477) (0.0544) (0.0472) (0.0337) Education 0.0160** 0.0340*** 0.0248* (0.00552) (0.00970) (0.0117) EDU2 -0.000131 -6.54e-05 (7.67e-05) (8.17e-05) Trade Openness 0.00428* (0.00188) Observations 310 310 266 206 206 206 R-squared 0.310 0.357 0.460 0.624 0.640 0.694
Country FE Yes Yes Yes Yes Yes Yes
45
Table A5 (a)．Result of hypothesis 1, dropping Japan from the sample. Effect of variables on the labor productivity gap between industrial and agricultural sectors (gap1)
*** p<0.01, ** p<0.05, * p<0.1 Robust standard errors in parentheses Data source: GGDC 10-sector database & PWT, World Bank database.
Note: Gap1 denotes: labor productivity in industrial sector/ labor productivity in agricultural sector. Term EDU2 is calculated as Education*Education.
(1) (2) (3) (4) (5) (6)
VARIABLES Log gap1 Log gap1 Log gap1 Log gap1 Log gap1 Log gap1 Log GDP per capita 0.275 0.00481 -0.00508 0.156 0.121 0.402
(0.293) (0.457) (0.448) (0.436) (0.464) (0.231)
Log Population -1.698 -1.216 -1.977 -1.420 -1.241
(1.602) (1.402) (2.109) (1.765) (0.905)
Log FDI per capita 0.0947 0.0735 0.0641 0.00754
(0.0799) (0.0988) (0.0914) (0.0474) Education -0.00881 0.00771 -0.00274 (0.00733) (0.0141) (0.00906) EDU2 -0.000113 -2.15e-05 (0.000111) (6.94e-05) Trade Openness 0.00722** (0.00251) Observations 310 310 264 204 204 204 R-squared 0.392 0.445 0.509 0.556 0.567 0.726
Country FE Yes Yes Yes Yes Yes Yes
46
Table A5 (b)．Result of hypothesis 1, dropping Japan from the sample. Effect of variables on the labor productivity gap between industrial and service sectors (gap2)
*** p<0.01, ** p<0.05, * p<0.1 Robust standard errors in parentheses Data source: GGDC 10-sector database & PWT, World Bank database.
Note: Gap2 denotes: labor productivity in industrial sector/ labor productivity in service sector. Term EDU2 is calculated as Education*Education.
(1) (2) (3) (4) (5) (6)
VARIABLES Log gap1 Log gap1 Log gap1 Log gap1 Log gap1 Log gap1 Log GDP per capita 0.223 -0.213 -0.203 0.00613 -0.0178 0.124
(0.488) (0.453) (0.440) (0.404) (0.409) (0.339)
Log Population -2.742* -3.372** -1.846 -1.467 -1.377
(1.258) (1.313) (1.918) (1.735) (1.392)
Log FDI per capita -0.00975 -0.0420 -0.0484 -0.0768**
(0.0406) (0.0479) (0.0472) (0.0330)
Education 0.0105 0.0217* 0.0165
(0.00643) (0.0110) (0.0152)
EDU2 -7.67e-05 -3.09e-05
(7.80e-05) (0.000102)
Trade Openness 0.00363**
(0.00155)
Observations 310 310 264 204 204 204
R-squared 0.311 0.457 0.600 0.669 0.673 0.710
Country FE Yes Yes Yes Yes Yes Yes
47
Table A6 (a)．Result of reverse causality test.
*** p<0.01, ** p<0.05, * p<0.1 Robust standard errors in parentheses Data source: GGDC 10-sector database & PWT, World Bank database.
Note: Gap1 denotes: labor productivity in industrial sector/ labor productivity in agricultural sector. L. represents the lagged data.
(1) (2) (3) (4) (5) (6)
VARIABLES Log gap1 Log gap1 Log gap1 Log gap1 Log gap1 Log gap1
L.Log GDP 0.441* 0.426 0.384 0.565** 0.447 0.578*** (0.236) (0.310) (0.336) (0.237) (0.280) (0.125) L.Log Population -0.197 0.260 0.634 1.091 0.223 (1.093) (0.888) (0.968) (0.908) (0.691) L.Log FDI 0.0890 0.0665 0.0528 -0.000885 (0.0818) (0.0946) (0.0776) (0.0329) L.Education -0.00141 0.0245 0.00794 (0.00598) (0.0217) (0.00941) L.EDU2 -0.000189 -7.28e-05 (0.000154) (6.20e-05) L. Trade Openness 0.00749** (0.00314) Observations 300 300 257 206 206 206 R-squared 0.351 0.351 0.410 0.466 0.504 0.698
Country FE Yes Yes Yes Yes Yes Yes
48
Table A6 (a)．Result of reverse causality test.
*** p<0.01, ** p<0.05, * p<0.1 Robust standard errors in parentheses Data source: GGDC 10-sector database & PWT, World Bank database.
Note: Gap2 denotes: labor productivity in industrial sector/ labor productivity in service sector. L. represents the lagged data.
(1) (2) (3) (4) (5) (6)
VARIABLES Log gap1 Log gap1 Log gap1 Log gap1 Log gap1 Log gap1
L.Log GDP 0.289 0.191 0.245 0.332 0.278 0.343 (0.396) (0.418) (0.442) (0.305) (0.299) (0.263) L.Log Population -1.277 -1.328 0.185 0.393 -0.0342 (1.022) (1.182) (0.943) (0.860) (0.764) L.Log FDI 0.0265 -0.0179 -0.0241 -0.0506 (0.0346) (0.0398) (0.0369) (0.0294) L.Education 0.0135** 0.0253** 0.0171 (0.00471) (0.00992) (0.0111)
L.EDU2 -8.62e-05 -2.88e-05
(7.03e-05) (7.22e-05)
L. Trade Openness 0.00369*
(0.00169)
Constant -2.090 21.30 21.38 -6.285 -9.678 -2.767
(3.316) (19.52) (22.26) (16.66) (15.21) (13.38)
Country FE Yes Yes Yes Yes Yes Yes
