The effect of Chinese import competition on
deindustrialization: a comparison between the North
and the South
Laurens van Dijk
s2482908
l.van.dijk.8@student.rug.nl
Supervisor: Tarek Harchaoui
Co-assessor: Abdul Erumban
1
Abstract: This thesis examines the impact of Chinese import competition on deindustrialization. We apply a unified model to a rich panel dataset of economies at various stages of economic development. We find that deindustrialization is accelerated as a result of Chinese import competition. This effect increased during the second phase of globalization due to a change in magnitude and type of imports from China and is having a bigger impact on developing economies, particularly those with a high relative share of real value added in manufacturing. The results remain robust to alternate specifications and source data.
Keywords: China, import competition, deindustrialization
2
1. Introduction
The increasingly important role of China in the world economy has been a cause of political debate, economic changes and even a changing world view. Many firms have started outsourcing some of their production activities to Chinese firms and imports from China have risen over the past few decades. Additionally, the exports from China started to rise exponentially and have only slowed down the past few years (Figure 1). This has not always been the case. China used to be a closed economy under the regime of Mao Zedong. According to Cheremukhin et al. (2015) three historic events have contributed to China becoming the open economy we know nowadays. In 1978, new economic reforms were introduced by Deng Xiaoping. In 1990 the Shanghai stock market reopened and in 2001 China joined the World Trade Organization (WTO). As a result, trading became easier and China started to get involved in the world economy, ultimately having a big impact.
Figure 1 – Total export of goods and services from China in current US$
3
together the literature on the dislocation of the labor in the North and the one on premature deindustrialization (Rodrik, 2015). Using a variety of estimation methods and alternate source data, our results highlight the following findings: 1) China's import competition increases the pace of deindustrialization. A higher initial share of manufacturing increases this negative effect. Higher levels of GDP per capita mitigate this negative effect; 2) The South is affected more than the North, as their type of manufactures competes the most with those of China; 3) The effect of Chinese import competition on deindustrialization has increased during the second phase of globalization for both the North and the South, due to a shift in magnitude and type of imports from China.
This thesis is structured as follows: the section 2 provides an overview of the literature related to similar topics. Thereafter, the datasets and methodology will be described. In the second part of that section the empirical results will be discussed along with several robustness tests. We conclude with some final remarks, limitations and policy implications.
2. Literature review
2.1 Deindustrialization
Industrialization has been a vital part in the development of countries. All developed countries have faced a period of industrialization leading up to their current state, which is often concentrated on the services sector. However, patterns of industrialization for the developing countries nowadays are changing. Often, developing countries find themselves deindustrializing prematurely, a term that Rodrik (2015) uses to address the phenomenon that "developing countries are turning into service economies without having gone through a proper experience of industrialization". Developing countries nowadays do not reach the same peak of industrialization as the developed countries did during that stage of development. On the contrary, the peaks in manufacturing employment and output nowadays only reaches about 40 percent of those before 1990. Different factors play a role in causing this shift.
One of which is the rise of China. The rise of China results in low-cost labor and thus lower prices for consumers but at the same time has many drawbacks. The effect that China has on other economies is not always the same and therefore it is useful to look at the relationship between the Chinese economy and structural change for different situations.
2.2 The Heckscher-Ohlin model
4
of similar countries will decrease, leading to competition and eventually a decrease in the size of this sector for the competing countries.
A similar point of view is used by Giovannetti et al. (2012), arguing that Italy was affected more than Germany, due to their specialization. These authors differentiate between the low cost manufactures and the more sophisticated categories of products. Therefore, Italy was affected more than Germany, as Italy had a comparative advantage in low cost manufactures, which decreased due to Chinese competition. The authors argue that it is necessary for these countries to follow a strategy of quality upgrading in order to keep its market power, leaving China to dominate the sector of low cost manufactures. Germany, however, was already exporting quality goods to China, creating new job opportunities (Marin, 2017). On a microeconomics level, Iacovone et al. (2013) mention that Mexican firms also increased their productivity, innovation and product quality as a result of foreign competition but that increased competition made more firms exit and produce fewer goods. From a Heckscher-Ohlin perspective, countries with a comparative advantage in manufacturing goods will suffer the most from Chinese import competition. However, Hanson (2016) mentions that the reality is much more complicated as there are many factors influencing deindustrialization.
2.3 Other causes of deindustrialization
The reason that reality is more complicated is because there are many factors affecting both structural change directly, but also the effect China has on structural change.
One of the drivers of structural change which is mentioned by Nickel et al. (2008) is technological progress. If technological progress takes place in one sector, it means that the total factor productivity (TFP) is likely to increase as well. Workers then tend to reallocate to the sectors where productivity is highest. They describe how not only the technological progress within a country matters for structural change, but also that of other countries and industries, due to competing uses for these factors. Acemoglu et al. (2016) estimate both the direct and indirect effects of Chinese import penetration. They conclude that countries that are more exposed face higher job losses. Besides technological progress, prices also affect structural change directly (in the country itself) and indirectly (via competition). Related to the indirect effects, Hanson and Robertson (2008) mention that some developing countries are affected more when their trade partners also have low trade costs with China. Those trade partners become more likely to import from China, resulting in decreased manufacturing exports and thus deindustrialization in other developing countries. However, they find that even developing countries with a comparative advantage in manufacturing are only facing a modest negative shock as a result of China's expansion.
5
employment of low-skilled workers. Rodrik then continues to argue that the effect is larger for low income countries that do not specialize in manufacturing. Countries that do specialize in manufacturing would face an increase in this sector as a result of technological change. Seemingly contradictory, Autor et al. (2013) argue that countries with a higher initial share of employment in manufacturing face a stronger decline in this sector as a result of Chinese import. This indicates that having multiple factors into play makes it much more difficult to predict what part of deindustrialization is due to Chinese import competition. Autor et al. estimate that about 44 percent of the decline in manufacturing employment is caused by Chinese import competition between 1990 and 2007. This effect was larger between 1990 and 2000, showing that the time period also matters. Estimating the effect on output might provide different results than for employment. For that reason, this thesis looks at the effect of Chinese import competition on output, measured in real value added, and compares different time periods.
Another driver of deindustrialization is the demand structure. This relates to the aforementioned, as the demand structure might cause differences between the effect on employment and output. In some cases, for example, the indirect effects of China's competition cause a decrease in exports of a country but not in output, due to increased domestic demand. Wood and Mayer (2010) show that the effect of China's opening was larger for countries' export market than it was for sales in the home market. However, there will be a shift in the demand of the home market as well. As countries develop, the demand for goods declines and the demand for services increases, amplifying the effect of deindustrialization itself. The same theory holds for an increase in Chinese import competition, as workers are expected to reallocate as well. Autor et al. (2016) show that in the short run, this might cause stagnation in GDP growth, as labor is not perfectly mobile and will thus not immediately shift towards the services sector or non-traded jobs. Additionally, most workers will remain working in trade-exposed industries, even though they change employer. This leads to higher rates of unemployment, lower wages and thus a lower demand in general. As a result of a lower demand, unemployment rates rise even further and economic growth is hindered.
6
2.4 Benefits of China's opening
China's opening does not necessarily lead to lower GDP growth rates and deindustrialization, as long as countries react efficiently. Turning back to the example from Giovannetti et al., Italy was able to deal with Chinese competition due to its quality upgrading. For example, countries might impose reforms in order to increase the level of education. If a country fails to do so properly, it will not be able to compete with China and thus face deindustrialization. Harchaoui et al. (2017) show this with the example of Tunisia, for which its education strategy backfired. Countries at an early stage of development do not always provide job opportunities for high-skilled workers. For Tunisia, this led to more self-employed and decreasing productivity levels. On the other hand, if a country successfully upgrades, it is able to niche in more sophisticated goods and benefit from the rising demand from China for this type of products. Thus, there is a tradeoff between the competition from China and the rising demand from China and the specialization of a country determines the outcome of this tradeoff. Hanson and Robertson (2008) look at several developing countries and find that this tradeoff turned out positively for Sri Lanka, which experienced a large increase in exports of tea to China, offsetting the negative effects from Chinese competition. The effect China has on the structural change of a country therefore highly depends on the level of education within a country, the ability to reallocate properly and its specialization.
Another benefit from exposure to China's import is the decrease in prices that cause an increase in demand. This increase in demand could offset the decreasing demand due to the direct and indirect effects of Chinese competition. Additionally, Autor et al. (2016) mention that countries also benefit from the extended range of products and the increased levels of R&D. Even if this is the case, and a country is able to increase its productivity and GDP as a result of China's opening, it is still likely that this larger pie will not be distributed equally across the country (Hanson, 2016). Each country is heterogeneous and consists of multiple groups. This thesis will show that Chinese import competition leads to shrinkage of the manufacturing sector. If workers are not able to reallocate to another sector and get unemployed, they lose from Chinese import competition. Nickell et al. (2008) mention that there are also differences in the effect of Chinese import competition, depending on gender and skill level. Thus, even though net welfare might increase if countries are able to innovate, there will always be losers.
7
3. Quantitative Analysis
3.1. Modeling and Estimation Strategies
We use a polynomial function to quantify how to determine how and to what extent China has altered the patterns of structural transformation in developed and developing economies. This sort of specification has proved to be successful in the study of patterns of structural transformation performed by Bah (2007) and Dabla-Norris et al. (2013). For each sector 𝑖 the following equation is estimated:
𝑠𝑖𝑐𝑡 = 𝛼 + 𝛽1ℓ𝑛(𝑙𝑒𝑣𝑑𝑒𝑣𝑐𝑡) + 𝛽2(ℓ𝑛(𝑙𝑒𝑣𝑑𝑒𝑣𝑐𝑡)) 2
+ 𝜓1ℓ𝑛(𝐶ℎ𝑖𝑛𝑎𝑡) + 𝜓2(ℓ𝑛(𝐶ℎ𝑖𝑛𝑎𝑡) ×
ℓ𝑛(𝑙𝑒𝑣𝑑𝑒𝑣𝑐𝑡)) + ∑ 𝜙𝑗 𝑗𝑍𝑗𝑐𝑡+ 𝜖𝑖𝑡 (1)
where 𝑠𝑖𝑡 is the share of sector 𝑖 in the total economy of country 𝑐 in year 𝑡 and 𝑙𝑒𝑣𝑑𝑒𝑣𝑖𝑐𝑡 is the level of development attained by the country 𝑐 in year 𝑡, 𝐶ℎ𝑖𝑛𝑎𝑡 is the impact of China in year 𝑡
and 𝑍𝑗𝑖𝑐𝑡 is a set of country-specific control variables observed in year 𝑡.
Equation (1) has the structure of a pooled model as no provision is made for differences across countries that may be captured by differences in the coefficients. Under this model specification, the errors 𝜖𝑖𝑡 have zero mean and constant variance, are uncorrelated over time and countries, and are uncorrelated with the set of explanatory variables. This model is clearly restrictive and is entertained here as a benchmark. One way to relax some of these restrictions is to estimate the quantile regression model which allow the parameter estimates to differ across the distribution of sectoral output shares, namely the 10th, 25th, 50th, 75th, and 90th quantiles. A battery of alternate estimates is performed to assess the robustness of the results. These alternate estimates range from applying the pooled model and the quantile regression technique to different sets of economies grouped by level of development, different time-periods including the use of the fixed-effect model. We also contrast the results with the use with alternate source data.
3.2. Source Data
We use two alternate source data. The first source data, considered as the benchmark, builds on the one assembled by Dabla-Norris et al. (2013) from which have eliminated all sorts of missing values.1 Originally, the data comprises 168 economies at different stages of development the majority of which are observed over the 1970-2010 period but after the suppression of the missing values we were left with a complete panel of 87 countries tracked over the 1980-2009
1 Dropped first 10 years due to missing values for va_agr, va_man and va_ser. Dropped 48 countries due to missing
8
period (see Table A1 in the appendix).2 Despite dropping 2731 observations as a result of this clean-up exercise, we are still left with a reasonably large panel comprising 2,610 observations. To this new dataset, we appended data on our set of economies’ imports from China. The primary source of these imports is CEPII’s bilateral trade data converted to $US, rather than the current British Pound as in the original database. The imports measure flows of merchandise which includes manufacturing and resource. Table 1 provides the summary statistics of the variables retained in this first source data. As is clear from a cursory reading of the table, this panel is heterogeneous in many respects: in terms of level of development (gdppc_ppp), size (population), economic structures (va_agr, va_man and va_ser), geography (ieco), etc.)
Table 1 – Descriptive statistics of the variables included in the benchmark source data
variable N min max mean p50 skewness kurtosis
va_agr 2610 0.266892 93.90939 13.43943 7.885994 1.531312 5.575878
va_man 2610 0.582508 26.41718 12.16129 12.83938 -0.19127 2.556914
va_ser 2610 4.072005 73.29621 47.18027 47.64846 -0.36854 2.759929
CHNimportUS 2610 19813 3.59e+11 2.95e+09 1.01e+08 13.6427 228.3677
gdppc_ppp 2610 82.07873 68649.49 9091.14 4331.743 1.95657 7.372398
teco 2610 0 1 0.011494 0 9.165786 85.01163
ieco 2610 0 1 0.045977 0 4.335688 19.79819
va_mu 2610 0.76918 75.94196 22.10949 18.55241 1.701393 6.002747
landarea 2610 320 9161920 876738.1 276840 3.582112 15.49371
population 2610 115641 1.21E+09 3.83E+07 9982905 7.354194 64.87739
arable 2610 0.06 73.39 14.92298 10.05 1.490307 4.917195
agedep_young 2610 20.78 106.43 59.32433 60.35 -0.02275 1.632806
agedep_old 2610 0.54 34.17 10.58894 7.335 1.086618 2.850233
Notes: va_agr = Share of agriculture in terms of relative real value added; va_man = Share of manufacturing in terms of relative real value added; va_ser = Share of services in terms of relative real value added; CHNimportUS = Chinese imports in current U.S. dollar; gdppc_ppp = GDP per capita, in constant PPP; teco = Transition economy dummy; ieco = Island economy dummy; va_mu = Share of mining in terms of relative real value added; landarea = Land area in squared kilometers; population = total population; arable = percentage of arable land; agedep_young = age dependency ratio, young, as percentage of working-age population; agedep_old = age dependency ratio, old, as percentage of working-age population
The second source data possesses a narrower coverage of imports from China, represented by manufacturing.3 The accuracy gained with this variable comes at the cost of its availability for only a shorter time frame, namely 1992-2009, albeit with 100 countries, much larger than the sample included in the benchmark source data (see Table A2). The time availability of the other
2 From now on, the prefix “A” for each table and figure refers to the appendix.
3 Chemicals and related products, n.e.s.; Leather, leather manufactures, n.e.s., and dressed furskins; Rubber
9
variables reported in Table 1 had to be adjusted accordingly. Table 2 displays the summary statistics of this second source data and by and large the heterogeneity that features the earlier dataset is preserved under this one.
Table 2 – Descriptive statistics of the variables compiled from the World Bank Data
variable N min max mean p50 skewness kurtosis
va_agr 1800 0.266892 93.90939 13.79986 8.069409 1.578589 5.658421 va_man 1800 0.582508 26.41718 11.71253 12.31397 -0.03081 2.475651 va_ser 1800 4.072005 82.81077 48.40809 48.48861 -0.27232 2.922627 CHNman 1800 21.752 2.43E+08 2710690 160934.7 12.01456 177.9499 gdppc_ppp 1800 82.07873 68649.49 10611.34 5307.501 1.565618 5.394753 teco 1800 0 1 0.01 0 9.849371 98.0101 ieco 1800 0 1 0.09 0 2.865312 9.210012 va_mu 1800 0.76918 75.94196 21.16749 18.23535 1.731833 6.381021 landarea 1800 320 9161920 768190.7 212195 3.911728 18.16972
population 1800 97113 1.21E+09 3.71E+07 9190151 7.717696 70.05881
arable 1800 0.06 66.14 14.71561 10.31 1.37497 4.426678
agedep_young 1800 20.78 100.91 56.50539 55.775 0.050862 1.733402
agedep_old 1800 0.54 34.17 10.70334 7.575 1.135874 3.020861
Notes: CHNman = Chinese manufactures import in thousands of U.S. dollar; all other variables are the same as described in Table 1.
Table A12 displays the way each of these variables has been measured along with its underlying source.
3.3. Econometric Implementation
3.3.1 Preamble
Prior to the estimation, we performed several diagnostic checks, including normality of the dependent variable, collinearity and heteroskedasticity. The values for skewness and kurtosis between -2 and +2 are considered acceptable in order to prove normal univariate distribution (see Table A3). The values reported in Tables 1 and 2 for var_man and va_ser comply somewhat with this criterion but not va_agr.4 However, given the large sample size, one would expect that the distribution of these variables can reasonably be approximated by a normal distribution. There is also evidence of heteroskedasticity, as indicated by White's test, a robust standard errors has been performed as a result. As for collinearity, the correlation matrix (Table A4) reveals that only the variables age dependency ratio – old and age dependency ratio – young somewhat correlated. However, the Variance Inflation Factor (VIF) in Table A5 still lower than four, indicating that the effect of none of the variables is inflated drastically as a result of collinearity.
4 A Jarque-Bera test was also performed in order to test for normality. The results also showed that not all variables
10 3.3.2. Econometric Results
We provide a vast array of results and perform along the way robustness checks. Our baseline results consist of the pooled model and the quantile regression for the entire sample of economies over the 1980-2010 period.5 We provide two variants of these results. The first retains the same period but slices the sample of economies according to their level of development. The second not only slices the sample of economies according to their level of development but also the time period into 1980-1991 and 1992-2009 sub-periods. This split is justified on the ground that the latter period covers the second phase of globalization that reshaped considerably the articulation between the North and the South. Additionally, the role of China in the world market increased this period as a result of the re-establishment of the Shanghai stock exchange in November 1990.
A. The 1980-2010 Period
We begin with the pooled OLS regression and the results are reported in Table 3. Except for manufacturing, the models explain a reasonably high level of the variance of the dependent variables, with an R-squared of 0.850 for agriculture and 0.785 for services. At the level of the parameter estimates, the results confirm the presence of a significant quadratic relationship between GDP per capita and the relative size of each sector. By and large, the results conform to the predictions of the textbook case of structural transformation (see Figures A1). The relative size of agriculture declines precipitously with the level of development represented by real GDP per capita (lngdppc_ppp), with a tendency of this decline to level-off at higher level of development. The hump-shaped pattern of manufacturing is also corroborated by the results, compared to a more steady increase for services. The “made in China” effect significantly affects only the manufacturing sector as one would expect. The coefficient is significantly negative but this effect tends to get smaller for countries at a higher level of development which is indicated by the significantly positive interaction term of Chinese import and GDP per capita of 0.12.
From this regression, it appears that Chinese import competition is significantly and negatively affecting the real value added share of the manufacturing sector. The sign of the parameter estimates of the control variables are also in accordance with our expectations. For example, land endowment (lnlandarea) and arable land (arable) lead to a higher agricultural share, contrasting the manufacturing and services sectors where these effects are negative. The variable population constitutes a proxy for scale economies, an indicator of efficiency. These efficiencies contribute to lower the relative size of agriculture and to increase those of services and, to a lesser extent, manufacturing. Age dependency ratios are strongly negatively related with the share of manufacturing (and less so agriculture), but are positive and statistically significant for services,
5 Another way to control for heterogeneity of the sample is to use the fixed effects model. Note that fixed effects
11
likely reflecting high demand for services resulting from the need to care for young and old populations. There is also compelling evidence that a high share of mining activity has an adverse effect on the relative importance of sectors, an indication if any of a Dutch disease, resulting in a lower demand for other goods and services. Except for services, there is no evidence that island economies’ experience of structural transformation differs from other economies while transition economies display behaviours consistent with the typical pattern of structural transformation6.
Table 3 – Pooled OLS Model: All countries
VARIABLES Agriculture Manufacturing Services
lnCHNimUS 0.350 -1.159*** 0.361 (0.571) (0.213) (0.541) lngdppc_ppp -40.53*** 9.802*** 28.71*** (1.939) (0.625) (1.818) c.lnCHNimUS#c.lngdppc_ppp -0.00696 0.118*** -0.0408 (0.0633) (0.0254) (0.0621) lngdppc_ppp2 1.904*** -0.758*** -1.155*** (0.140) (0.0518) (0.133) teco -2.922*** 1.905*** 2.317*** (0.520) (0.502) (0.678) ieco -0.428 -0.0967 2.013*** (0.410) (0.429) (0.651) va_mu -0.108*** -0.0771*** -0.771*** (0.0115) (0.00863) (0.0143) lnlandarea 1.039*** -0.679*** -0.279*** (0.0787) (0.0736) (0.101) lnpopulation -2.084*** 1.359*** 0.543*** (0.133) (0.106) (0.138) arable 0.0742*** -0.0399*** -0.0241** (0.0128) (0.00634) (0.0119) agedep_young -0.0202 -0.141*** 0.172*** (0.0129) (0.00795) (0.0122) agedep_old -0.0562*** -0.0835*** 0.184*** (0.0191) (0.0171) (0.0246) Constant 229.5*** -13.80*** -107.9*** (10.21) (3.551) (9.705) Observations 2,610 2,610 2,610 R-squared 0.850 0.421 0.785 Number of countries 87 87 87
Time dummies YES YES YES
Notes: Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
6 Results for teco should be interpreted with some caution, as Hungary is the only transition economy in these
12
Based on the results from Table 3, we can estimate the effect China's import competition has on deindustrialization. Simplifying Equation (1) and implementing the coefficients from Table 3, we find the following equation for the effect of GDP per capita on the real value added share of manufacturing:
𝑠𝑐𝑡 = −13.8 + 9.802ℓ𝑛(𝑙𝑒𝑣𝑑𝑒𝑣𝑐𝑡) − 0.758(ℓ𝑛(𝑙𝑒𝑣𝑑𝑒𝑣𝑐𝑡)) 2
(2) In this equation, the effect of GDP is singled out. Equation (3) shows the same equation after adding the effect of Chinese import competition and its interaction with GDP per capita.
𝑠𝑐𝑡 = −13.8 + 9.802ℓ𝑛(𝑙𝑒𝑣𝑑𝑒𝑣𝑐𝑡) − 0.758(ℓ𝑛(𝑙𝑒𝑣𝑑𝑒𝑣𝑐𝑡))2− 1.159ℓ𝑛(𝐶ℎ𝑖𝑛𝑎𝑡) +
0.118(ℓ𝑛(𝐶ℎ𝑖𝑛𝑎𝑡) × ℓ𝑛(𝑙𝑒𝑣𝑑𝑒𝑣𝑐𝑡)) (3)
Plotting Equation (2) and Equation (3) in the same graph (Figure 2) visualizes the negative effect Chinese import competition has on the share of value added in the manufacturing sector7.
Figure 2 – Log of GDP per capita against real value added share in manufacturing
Note: va_man_noCHN plots Equation (2). va_man_CHN plots Equation (2) and shows the quadratic fitted values. Results are based on the coefficients from Table 3
Figure 2 shows that countries with the same level of GDP per capita have smaller shares of manufacturing as a result of Chinese import competition. The slope changes as well, which is a
7 Note that this only shows the predicted effect of the log of GDP per capita on the real value added share of
13
result of the interaction term in Equation (3). Equation (4) shows the derivative in order to mathematically describe the change in the slope.
𝑑𝑠𝑐𝑡
𝑑ℓ𝑛(𝑙𝑒𝑣𝑑𝑒𝑣𝑐𝑡)= 9.802 − 1.516ℓ𝑛(𝑙𝑒𝑣𝑑𝑒𝑣𝑐𝑡) + 0.118ℓ𝑛(𝐶ℎ𝑖𝑛𝑎𝑡) (4)
As Equation (4) shows, the Chinese import competition has a positive effect on the slope of the relation between GDP per capita and share of manufacturing, implying that the positive slope at lower levels of GDP steepens and the decreasing slope at higher levels of GDP flattens. By following this reasoning and by looking at Figure 2, we conclude that at high levels of GDP per capita, Chinese import competition complements the manufacturing sector and thus does not accelerate deindustrialization. For other countries, however, the share of manufacturing is smaller than without the China effect.
A variant of the results presented in Table 3 sliced along the level of development of economies is displayed in Table A6. It appears that much of the results reported in Table 3 are driven by those reported by emerging economies. The results related to services in advanced economies remain generally robust and consistent with those outlined above on the basis of Table 3, while only the control variables of low income economies remain significant, except for arable. The effects on the share of real value added of agriculture also show no large differences across the levels of development.
Given that manufacturing is at the center of the debate on deindustrialization, we now place the focus on this sector by providing results based on a quantile regression across different parts of the distribution. The results reported in Table 4 indicate that Chinese import competition is significant and negatively affects the manufacturing sector at all levels of the relative size of manufacturing. These results are in line with the pooled OLS model, albeit with a higher level of resolution. It now becomes clear that countries at both ends of the distribution are affected the most. Countries in the lowest quantile (such as Liberia, Syria and Gabon) and in the highest quantile (such as Thailand, Malaysia and the Philippines) face higher levels of deindustrialization as a result of Chinese import competition compared to other countries. Overall, once GDP per capita increases, the effect of Chinese import competition on the manufacturing sector diminishes.
14
Table 4 – Quantile regression for the manufacturing sector: All countries
VARIABLES Q10 Q25 Q50 Q75 Q90 lnCHNimUS -1.489*** -0.361** -0.612*** -1.258*** -1.974*** (0.214) (0.182) (0.171) (0.182) (0.375) lngdppc_ppp 4.419*** 6.786*** 10.70*** 13.76*** 10.39*** (0.562) (0.487) (0.427) (0.470) (1.158) c.lnCHNimUS#c.lngdppc_ppp 0.175*** 0.0297 0.0407** 0.124*** 0.252*** (0.0261) (0.0231) (0.0203) (0.0224) (0.0484) lngdppc_ppp2 -0.508*** -0.482*** -0.692*** -1.003*** -0.950*** (0.0477) (0.0463) (0.0323) (0.0492) (0.103) teco 2.265*** 3.789*** 1.981*** 1.215*** 1.077** (0.280) (0.912) (0.290) (0.247) (0.540) ieco -2.233*** 0.872 -0.259 0.990*** 0.679 (0.364) (1.318) (0.396) (0.379) (0.643) va_mu -0.0716*** -0.0844*** -0.118*** -0.0395** 0.135*** (0.00514) (0.00621) (0.00531) (0.0154) (0.0338) lnlandarea -0.298*** -0.413*** -0.517*** -0.933*** -1.010*** (0.0727) (0.0674) (0.0580) (0.0545) (0.0902) lnpopulation 0.635*** 1.254*** 1.279*** 1.329*** 1.013*** (0.0922) (0.0980) (0.0874) (0.0817) (0.135) arable -0.0235*** -0.0579*** -0.0249*** -0.0457*** -0.0550*** (0.00592) (0.00490) (0.00677) (0.00424) (0.00855) agedep_young -0.143*** -0.148*** -0.134*** -0.108*** -0.0778*** (0.00635) (0.00745) (0.00664) (0.00845) (0.00893) agedep_old -0.0541*** -0.0695*** -0.131*** -0.0255 -0.00589 (0.0205) (0.0159) (0.0123) (0.0196) (0.0426) Constant 12.34*** -13.12*** -24.29*** -25.77*** -4.802 (3.376) (2.990) (2.811) (3.126) (4.889) Observations 2,610 2,610 2,610 2,610 2,610 Number of countries 87 87 87 87 87
Time dummies YES YES YES YES YES
Pseudo R2 0.314 0.324 0.273 0.193 0.190
Notes: Robust standard errors in parentheses. The quantiles represent the relative share of manufacturing *** p<0.01, ** p<0.05, * p<0.1
15
manufacturing. This increase is smaller than the decrease in less manufacturing abundant countries and only significant at a 10% level.
Secondly, and most importantly, the emerging economies are affected the most, with a significantly negative relationship between Chinese import competition and the share of manufacturing for all levels of manufacturing. The effect becomes larger as the relative share of manufacturing grows, providing evidence that countries with a comparative advantage in manufacturing deindustrialize the most as a result of Chinese competition. Notable in this case is how the effect becomes larger as the share of manufacturing increases, similar to the relation observed for the full sample, showing that the full sample reflects much of the effects in emerging economies.
Thirdly, there are no clear effects for the manufacturing sector in low income countries. Only the countries with the highest level of manufacturing face a decline in this sector. The lack of significant results might be due to the fact that relative to other countries, low income counties have a small share of manufacturing on average. This explains why those in the highest quantile (Ghana during the 1980s, Honduras the last decade) do significantly deindustrialize.
Overall, the conclusion from these quantile regressions is that the manufacturing sector of emerging countries decreases the most as a result of Chinese import competition. This effect becomes even larger if a country initially has a high share of manufacturing and if a country has a lower level of GDP per capita.
B. Pre- vs. Post-Second Phase of Globalization
We now split the 1980-2009 period between the pre- and post-second phase of globalization (1980-1991 and 1992-2009, respectively). During the latter period, China emerged as the manufacturing world powerhouse. In terms of the results, we expect to see that the order-of-magnitude and significance of the “made in China” effect to be quantitatively larger. The results are presented in Table 5 for the pooled OLS regression.
16
Table 5 – Pooled OLS regression for different sub-periods: All countries
1980-1991 1992-2009
VARIABLES Agriculture Manufacturing Services Agriculture Manufacturing Services
lnCHNimUS -1.838** -0.168 1.989** -0.597 -0.488* 0.452 (0.759) (0.408) (0.906) (0.829) (0.261) (0.739) lngdppc_ppp -43.45*** 12.66*** 28.09*** -44.80*** 11.28*** 32.90*** (3.291) (1.030) (3.181) (2.705) (0.828) (2.426) c.lnCHNimUS#c.lngdppc_ppp 0.257*** -0.0256 -0.220** 0.135 0.0657** -0.0968 (0.0881) (0.0478) (0.109) (0.0860) (0.0328) (0.0809) lngdppc_ppp2 1.873*** -0.804*** -0.984*** 1.941*** -0.785*** -1.280*** (0.195) (0.0740) (0.205) (0.192) (0.0690) (0.172) teco -2.242*** 0.907* 2.239** -2.986*** 2.299*** 2.278*** (0.719) (0.523) (0.899) (0.586) (0.732) (0.832) ieco 0.499 0.570 0.743 -0.168 -1.010** 2.354*** (0.607) (0.763) (1.239) (0.523) (0.488) (0.688) va_mu -0.0990*** -0.0791*** -0.789*** -0.0913*** -0.0798*** -0.778*** (0.0167) (0.0133) (0.0209) (0.0166) (0.0114) (0.0197) lnlandarea 0.690*** -0.329*** -0.154 1.094*** -0.825*** -0.287** (0.128) (0.122) (0.180) (0.0970) (0.0878) (0.120) lnpopulation -1.667*** 1.154*** 0.380* -2.398*** 1.284*** 0.826*** (0.187) (0.164) (0.200) (0.213) (0.147) (0.230) arable 0.0731*** -0.0134 -0.0431** 0.0683*** -0.0547*** -0.00877 (0.0185) (0.00952) (0.0189) (0.0160) (0.00819) (0.0143) agedep_young -0.0212 -0.163*** 0.183*** -0.00960 -0.131*** 0.171*** (0.0185) (0.0104) (0.0190) (0.0186) (0.0119) (0.0165) agedep_old -0.0970*** -0.157*** 0.289*** -0.0346 -0.0497** 0.144*** (0.0315) (0.0284) (0.0425) (0.0253) (0.0212) (0.0304) Constant 254.1*** -30.15*** -117.5*** 265.6*** -26.47*** -133.4*** (17.96) (6.104) (17.82) (14.57) (4.894) (13.53) Observations 1,044 1,044 1,044 1,566 1,566 1,566 R-squared 0.854 0.473 0.748 0.861 0.420 0.818 Number of countries 87 87 87 87 87 87
Time dummies YES YES YES YES YES YES
Notes: Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
17
Moving now to the quantile regression of which the results are displayed in Table 6.1 and 6.2. Over the 1980-1991 period, China’s import competition has no effect at all on any of the manufacturing sectors. Yet, this result is misleading as it masks a great deal of heterogeneity between countries at different stages of development. As shown by Tables A9, for both the advanced economies and low income countries, the effect depends on the size of the manufacturing sector. For advanced economies, with a slightly above average level of manufacturing, Chinese imports actually had a complementary effect. For low income countries, however, it negatively affected the manufacturing sector. Note that the size of the manufacturing sector in low income countries in that time period was generally lower than that of advanced economies. Chinese import competition had a significantly negative effect on most levels for emerging economies during this time period.
Table 6.1 – Quantile regression for the manufacturing sector, all countries: 1980-1991
VARIABLES Q10 Q25 Q50 Q75 Q90 lnCHNimUS -0.555 0.0117 0.296 -0.383 -0.328 (0.403) (0.441) (0.493) (0.296) (0.544) lngdppc_ppp 10.66*** 7.854*** 10.92*** 15.73*** 16.42*** (0.815) (1.107) (1.058) (1.012) (1.490) c.lnCHNimUS#c.lngdppc_ppp 0.0519 -0.0363 -0.0937 -0.000941 -0.0176 (0.0512) (0.0569) (0.0587) (0.0340) (0.0646) lngdppc_ppp2 -0.814*** -0.498*** -0.594*** -0.997*** -1.079*** (0.0669) (0.0909) (0.0796) (0.0681) (0.101) teco 0.222 2.670*** 1.178** 0.170 -0.737 (1.459) (0.658) (0.496) (0.193) (6.496) ieco -3.672*** 0.0567 0.152 2.186*** 1.467** (0.450) (1.434) (1.623) (0.416) (0.577) va_mu -0.0737*** -0.0862*** -0.0912*** -0.101*** -0.00823 (0.0107) (0.0141) (0.0162) (0.0154) (0.0341) lnlandarea 0.561*** -0.0375 -0.326*** -0.464*** -0.670*** (0.120) (0.120) (0.1000) (0.0576) (0.136) lnpopulation -0.427*** 0.941*** 1.325*** 1.007*** 1.207*** (0.155) (0.166) (0.151) (0.0832) (0.174) arable 0.0623*** -0.0103 -0.00215 -0.0150*** -0.0458*** (0.0139) (0.0141) (0.0106) (0.00462) (0.0143) agedep_young -0.226*** -0.195*** -0.162*** -0.158*** -0.142*** (0.00831) (0.0129) (0.0133) (0.0119) (0.0186) agedep_old -0.158*** -0.239*** -0.178*** -0.250*** -0.0814 (0.0392) (0.0341) (0.0324) (0.0288) (0.0578) Constant -5.320 -13.58** -31.37*** -34.13*** -36.05*** (5.344) (6.662) (7.606) (6.242) (9.005) Observations 1,044 1,044 1,044 1,044 1,044 Number of countries 87 87 87 87 87
Time dummies YES YES YES YES YES
Pseudo R2 0.392 0.371 0.314 0.238 0.213
18
Table 6.1 – Quantile regression for the manufacturing sector, all countries: 1992-2009
VARIABLES Q10 Q25 Q50 Q75 Q90 lnCHNimUS -1.837*** -0.146 0.490** 0.746*** -0.253 (0.287) (0.240) (0.249) (0.237) (0.544) lngdppc_ppp 4.801*** 8.424*** 13.45*** 15.02*** 10.40*** (1.013) (0.661) (0.684) (0.901) (1.027) c.lnCHNimUS#c.lngdppc_ppp 0.212*** 0.00928 -0.0726** -0.0784** 0.0843 (0.0393) (0.0324) (0.0322) (0.0315) (0.0681) lngdppc_ppp2 -0.551*** -0.552*** -0.694*** -0.820*** -0.768*** (0.0737) (0.0599) (0.0614) (0.0725) (0.129) teco 3.390*** 3.130* 3.294*** 2.796*** 0.175 (0.901) (1.639) (0.369) (1.010) (0.777) ieco -2.341*** 1.144 -1.166* -0.671 -0.442 (0.492) (1.744) (0.636) (0.436) (1.091) va_mu -0.0787*** -0.0906*** -0.139*** -0.0198 0.238*** (0.00722) (0.0105) (0.00834) (0.0342) (0.0202) lnlandarea -0.545*** -0.473*** -0.779*** -1.120*** -1.007*** (0.112) (0.0871) (0.0786) (0.115) (0.159) lnpopulation 0.962*** 1.350*** 1.350*** 1.326*** 0.718*** (0.205) (0.131) (0.115) (0.182) (0.142) arable -0.0530*** -0.0727*** -0.0461*** -0.0577*** -0.0312* (0.0158) (0.00642) (0.00908) (0.0118) (0.0178) agedep_young -0.117*** -0.137*** -0.0993*** -0.0576*** -0.0337** (0.0139) (0.0106) (0.00950) (0.0112) (0.0155) agedep_old -0.0318 -0.0364* -0.0962*** 0.0487 0.171*** (0.0389) (0.0220) (0.0198) (0.0338) (0.0411) Constant 8.976 -24.01*** -49.38*** -57.15*** -27.25*** (5.821) (4.463) (3.939) (5.439) (6.832) Observations 1,566 1,566 1,566 1,566 1,566 Number of countries 87 87 87 87 87
Time dummies YES YES YES YES YES
Pseudo R2 0.288 0.307 0.272 0.212 0.253
Notes: Robust standard errors in parentheses. The quantiles represent the relative share of manufacturing *** p<0.01, ** p<0.05, * p<0.1
The effect has changed over time, showing some significant effects of China's import on the share of manufacturing for the 1992-2009 period. Even though this effect is still negative for the lowest quantile, some of the other quantiles show a significantly positive effect. Since the observed differences are rather counter intuitive, we use to Table A9 to gain more insight on the effect divided per level of development. It appears that the effect on low income countries has changed differently from the effect on advanced and emerging economies.
19
type of import from China. In sum, even though the results for the full sample are still relevant, it also matters what type of country is being looked at. For example, certain quantiles can be dominated by low income countries due to their share of manufacturing. Especially for emerging economies the theory that increased imports lead to deindustrialization holds.
3.3.3. Robustness Checks
The previous sections provided the baseline results contrasted with a number of robustness checks expressed in terms of nature of the sample and time periods. We now move to a different type of diagnostic check expressed in terms of different measure of imports from China and a different sample of economies (see Section 3.2 Source Data). The results of a pooled OLS regression with this data are displayed in Table 7.
Table 7 – Pooled OLS based on World Bank Data: All countries
VARIABLES Agriculture Manufacturing Services
lnCHNman -1.391** -0.645*** 1.597*** (0.635) (0.224) (0.589) lngdppc_ppp -45.44*** 12.87*** 31.74*** (2.540) (0.832) (2.226) c.lnCHNman#c.lngdppc_ppp 0.189*** 0.0755** -0.179*** (0.0657) (0.0295) (0.0652) lngdppc_ppp2 2.001*** -0.843*** -1.198*** (0.163) (0.0630) (0.149) teco -3.677*** 1.484** 3.513*** (0.551) (0.754) (0.894) ieco 0.829 -4.235*** 3.109*** (0.538) (0.436) (0.597) va_mu -0.0782*** -0.0769*** -0.799*** (0.0153) (0.0102) (0.0179) lnlandarea 1.052*** -0.515*** -0.324*** (0.0843) (0.0837) (0.113) lnpopulation -2.046*** 1.113*** 0.487*** (0.176) (0.117) (0.183) arable 0.0843*** -0.0279*** -0.0358** (0.0154) (0.00772) (0.0139) agedep_young -0.0342** -0.126*** 0.195*** (0.0151) (0.0106) (0.0134) agedep_old -0.0469* -0.0471** 0.128*** (0.0244) (0.0202) (0.0305) Constant 269.8*** -36.08*** -131.3*** (11.69) (3.772) (10.01) Observations 1,800 1,800 1,800 R-squared 0.866 0.447 0.836 Number of countries 100 100 100
Time dummies YES YES YES
20
The results based on this type of import competition seem to be much more distinct. For example, the effect on all sectors is now significant and the effect on deindustrialization is now larger than before (column 5 of Table 5, -0.49, compared with the -0.65 here). Intuitively, this makes sense as the relationship is now not affected by imports of goods other than manufactures. A rise of 1 percent of this type of import thus results in a relative decline in the agricultural and manufacturing sector by 1.39 and 0.65 percentage points, respectively. . It also confirms that it will shift output to the services sector, resulting in an increase of 1.6 percent. Looking at Tables A10, the results show a negative effect of import of manufactures from China on the manufacturing sectors of all countries. Remarkably, the manufacturing sector in advanced economies is affected the most.
Table 8 – Quantile regression for the manufacturing sector using World Bank data: All countries VARIABLES Q10 Q25 Q50 Q75 Q90 lnCHNman -2.023*** -0.720*** -0.686*** 0.240 0.408* (0.243) (0.248) (0.187) (0.269) (0.217) lngdppc_ppp 6.940*** 11.00*** 13.78*** 15.64*** 8.864*** (0.897) (0.825) (0.534) (0.818) (0.930) c.lnCHNman#c.lngdppc_ppp 0.207*** 0.0660** 0.0589** 0.00622 0.0299 (0.0319) (0.0323) (0.0244) (0.0329) (0.0307) lngdppc_ppp2 -0.527*** -0.714*** -0.878*** -1.006*** -0.569*** (0.0696) (0.0615) (0.0451) (0.0688) (0.0665) teco 3.683* 2.058 2.128*** 0.781 -1.363*** (1.974) (1.859) (0.406) (1.060) (0.437) ieco -4.929*** -4.908*** -6.525*** -2.576*** -1.317*** (0.427) (0.378) (0.374) (0.403) (0.259) va_mu -0.0955*** -0.0980*** -0.0919*** 0.0463 0.213*** (0.0104) (0.00959) (0.00928) (0.0328) (0.00356) lnlandarea -0.291*** -0.311*** -0.485*** -1.053*** -0.555*** (0.106) (0.0888) (0.0761) (0.0889) (0.0914) lnpopulation 0.849*** 1.186*** 1.181*** 1.259*** 0.389*** (0.157) (0.122) (0.104) (0.156) (0.0945) arable -0.0358*** -0.0456*** -0.0228*** -0.0348** 0.0225** (0.0103) (0.00607) (0.00824) (0.0148) (0.0104) agedep_young -0.0723*** -0.116*** -0.129*** -0.0821*** -0.0299*** (0.0133) (0.0108) (0.00802) (0.0148) (0.00551) agedep_old -0.0298 -0.0485** -0.0618*** 0.115*** 0.0912*** (0.0300) (0.0219) (0.0226) (0.0432) (0.0319) Constant -12.41*** -33.95*** -40.09*** -51.04*** -26.69*** (3.830) (3.893) (2.211) (3.873) (3.827) Observations 1,800 1,800 1,800 1,800 1,800 Number of countries 100 100 100 100 100
Time dummies YES YES YES YES YES
Pseudo R2 0.300 0.331 0.306 0.230 0.287
21
Comparing the quantile regressions might provide an in-depth way of comparing both datasets. Table 8 shows the results for the full sample and Table A11 the results as split by stage of development. The results are mixed. However, apart from the highest quantile, the results are similar to the benchmark model, with the lowest quantile being affected the most. The 'odd' result of an increasing manufacturing sector in the highest quantile completely disappears when looking at the different levels of development. Advanced economies with a relatively low share of manufacturing (Cyprus and Greece, but also Norway) are still affected the most, but those with a higher level of manufacturing are not significantly affected anymore. Low income countries and emerging economies are affected at all levels of manufacturing with the latter being affected the most, just as in previous estimations. In sum, the type of Chinese import does matter, but in a broader sense, the effect is still the same: deindustrialization increases as a result of import competition and the effect is the largest for emerging economies.8
4.
Concluding remarks
Previous literature described how Chinese import competition, among other factors, affects industrialization and that growth patterns have changed during the past few decades. It showed that countries need quality upgrading in order to avoid the drawbacks from competition from China and benefit from its demand. Not all countries have been able to do so, due to weak policy, a lack of education or other shocks. The reallocation that is required will take a long-term approach, as workers are not perfectly mobile across sectors.
Testing empirically, we find that Chinese import competition indeed negatively affects the manufacturing sector. This effect is smaller for countries with higher levels of GDP. By using a quantile regression, it is possible to see what countries are affected the most. It shows that countries with a comparative advantage in manufacturing are affected more than those with lower shares of manufacturing. However, this is where the North differs from the South, as advanced economies with a relatively low share of manufacturing are affected more than those with a high initial share of manufacturing. Splitting the sample by stage of development, we find that emerging economies (the South) deindustrialize the most as a consequence of Chinese import competition. Thus, the "made in China" effect is more of a threat to the South than it is to the North and has led to rapid deindustrialization.
Some limitations should be noted. Firstly, the description of the variable lnCHNimUS in the benchmark model is somewhat inconsistent. Fouquin and Hugot (2016) indicated that this variable presents data on merchandise trade, excluding trade in services, bullion and species, whenever possible. However, the exact imports that are included in this variable might slightly
8 Other robustness checks to confirm this result were performed as well. 1. A comparison between the benchmark
22
differ per country or per time period. Secondly, the type of manufactures can be disaggregated even further. This would allow examining differences within the manufacturing sector and showing the relevance of quality upgrading. This topic of discussion might be relevant for further research. Thirdly, the R-squared in our models does not reach above 0.5 for the manufacturing sector, implying that there are many other factor that affect the size of the manufacturing sector that were not included in our models, possibly affecting the effect Chinese import competition has on deindustrialization.
23
References
Acemoglu, D., Autor, D., Dorn, D., Hanson, G.H. & Price, B. (2016). Import Competition and the Great US Employment Sag of the 2000s. Journal of Labor Economics, 34(1), 141-198. Autor, D. H., Dorn, D., & Hanson, G. H. (2016). The china shock: Learning from labor-market adjustment to large changes in trade. Annual Review of Economics, 8, 205-240.
Bah, E. H. M. (2011). Structural transformation paths across countries. Emerging Markets Finance and Trade, 47(2), 5-19.
Cheremukhin, A., Golosov, M., Guriev, S., & Tsyvinski, A. (2015). The economy of People’s Republic of China from 1953. (NBER Working Paper No. w21397).Retrieved from
http://www.nber.org/papers/w21397
Colantone, I., & Stanig, P. (2017). The Trade Origins of Economic Nationalism: Import
Competition and Voting Behavior in Western Europe. (BAFFI CAREFIN Centre Research Paper No. 2017-49). Retrieved from
https://www.bancaditalia.it/pubblicazioni/altri-atti-seminari/2016/14_novembre_Colantone_Stanig
Dabla-Norris, M. E., Thomas, M. A. H., Garcia-Verdu, M. R., & Chen, M. Y.
(2013). Benchmarking structural transformation across the world. (International Monetary Fund Working Paper No. 13-176). Retrieved from
https://www.imf.org/en/Publications/WP/Issues/2016/12/31/Benchmarking-Structural-Transformation-Across-the-World-40847
David, H., Dorn, D., & Hanson, G. H. (2013). The China syndrome: Local labor market effects of import competition in the United States. American Economic Review, 103(6), 2121-68. Feenstra, R. C., & Sasahara, A. (2017). The ‘China Shock’, Exports and US Employment: A Global Input-Output Analysis. (NBER Working Paper No. w24022). doi:10.3386/w24022 Fouquin, M., Hugot, J. (2016). Two Centuries of Bilateral Trade and Gravity Data: 1827-2014. (CEPII Working Paper No 2016-14). Retrieved from
http://www.cepii.fr/PDF_PUB/wp/2016/wp2016-14.pdf
Giovannetti, G., Sanfilippo, M., & Velucchi, M. (2012). The impact of China on manufacturing exports of Italy and Germany. (EUI Working Paper No. 2012/26). Retrieved from
http://hdl.handle.net/1814/22387
Hanson, G. H., & Robertson, R. (2008). China and the manufacturing exports of other developing countries. (NBER Working Paper No. w14497). Retrieved from
24
Hanson, G. H. (2016, May 5). US Labor Market Adjustment to China Trade Shocks. Presentation at PIIE, California, San Diego.
Haraguchi, N., Cheng, C.F.C., & Smeets, E. (2017). The Importance of Manufacturing in Economic Development: Has This Changed?. World Development, 93, 293-315.
Harchaoui, T.M., Beuckens, M., & Ben Chaâben, M. (2017). Tunisia's Idiosyncratic Path to Development: The Good, the Bad and the Ugly, Paper presented at the 67th Congress of the French Association of Economics, Paris, May 2018.
https://afse2018.sciencesconf.org/191951/document
Iacovone, L., Rauch, F., & Winters, L. A. (2013). Trade as an engine of creative destruction: Mexican experience with Chinese competition. Journal of International Economics, 89(2), 379-392.
Koenker, R., & Hallock, K.F. (2001). Quantile Regression. Journal of Economic Perspectives, 15(4), 143-156.
Marin, D. (2017, September 19). The China shock: Why Germany thrived while the US struggled. World Economic Forum. Retrieved from
https://www.weforum.org/agenda/2017/09/the-china-shock-why-germany-thrived-while-the-us-struggled
McMillan, M., Rodrik, D., & Verduzco-Gallo, Í. (2014). Globalization, structural change, and productivity growth, with an update on Africa. World Development, 63, 11-32.
Nickell, S., Redding, S., & Swaffield, J. (2008). The uneven pace of deindustrialisation in the OECD. The World Economy, 31(9), 1154-1184.
Rodrik, D. (2016). Premature deindustrialization. Journal of Economic Growth, 21(1), 1-33. Wood, A., & Mayer, J. (2011). Has China de-industrialised other developing countries?. Review of World Economics, 147(2), 325-350.
Wood, A. (2017). Variation in structural change around the world, 1985–2015: Patterns, causes and implications. (WIDER Working Paper No 2017/34). Retrieved from
Appendix
Table A1 – List of countries included in benchmark dataset
Advanced Emerging Low income
Name Freq. Name Freq. Name Freq.
Australia 30 Algeria 30 Bangladesh 30
Austria 30 Argentina 30 Benin 30
Canada 30 Bahrain 30 Burkina Faso 30
Cyprus 30 Brazil 30 Burundi 30
Denmark 30 Chile 30 Cameroon 30
Finland 30 Colombia 30 Central African Republic 30
France 30 Costa Rica 30 Congo, Rep. 30
Greece 30 Dominican Republic 30 Gambia 30
Iceland 30 Ecuador 30 Ghana 30
Ireland 30 Egypt 30 Guinea-Bissau 30
Italy 30 Fiji 30 Honduras 30
Japan 30 Gabon 30 Kenya 30
Malta 30 Guatemala 30 Liberia 30
Netherlands 30 Hungary 30 Madagascar 30
New Zealand 30 India 30 Mali 30
Norway 30 Indonesia 30 Mozambique 30
Portugal 30 Iran 30 Nicaragua 30
Spain 30 Jordan 30 Niger 30
Sweden 30 Malaysia 30 Nigeria 30
Switzerland 30 Mauritius 30 Papua New Guinea 30
United Kingdom 30 Mexico 30 Rwanda 30
United States of America 30 Morocco 30 Senegal 30
Oman 30 Sudan 30 Pakistan 30 Togo 30 Panama 30 Vanuatu 30 Paraguay 30 Zambia 30 Peru 30 Philippines 30 Saudi Arabia 30 Sri Lanka 30 Suriname 30 Syria 30 Thailand 30 Trinidad 30 Tunisia 30 Turkey 30
United Arab Emirates 30
Uruguay 30
Table A2 – List of countries included in World Bank Dataset
Advanced Emerging Low income
Name Freq. Name Freq. Name Freq.
Australia 18 Algeria 18 Bangladesh 18
Austria 18 Argentina 18 Benin 18
Canada 18 Bahamas 18 Bolivia 18
Cyprus 18 Bahrain 18 Burkina Faso 18
Denmark 18 Belize 18 Burundi 18
Finland 18 Brazil 18 Cameroon 18
France 18 Chile 18 Central African Republic 18
Greece 18 Colombia 18 Chad 18
Iceland 18 Costa Rica 18 Comoros 18
Ireland 18 Dominican Republic 18 Congo, Rep. 18
Israel 18 Ecuador 18 Gambia 18
Italy 18 Egypt 18 Ghana 18
Japan 18 El Salvador 18 Grenada 18
Malta 18 Fiji 18 Guinea-Bissau 18
Netherlands 18 Gabon 18 Guyana 18
New Zealand 18 Guatemala 18 Honduras 18
Norway 18 Hungary 18 Kenya 18
Portugal 18 India 18 Liberia 18
Spain 18 Indonesia 18 Madagascar 18
Sweden 18 Iran 18 Malawi 18
Switzerland 18 Jordan 18 Mali 18
United Kingdom 18 Malaysia 18 Mozambique 18
United States of America 18 Mauritius 18 Nepal 18
Mexico 18 Nicaragua 18
Morocco 18 Niger 18
Oman 18 Nigeria 18
Pakistan 18 Papua New Guinea 18
Panama 18 Rwanda 18
Paraguay 18 Saint Lucia 18
Peru 18 Saint Vincent and the Grenadines 18
Philippines 18 Senegal 18
Saudi Arabia 18 Sierra Leone 18
Sri Lanka 18 Togo 18
Suriname 18 Vanuatu 18 Syria 18 Zambia 18 Thailand 18 Trinidad 18 Tunisia 18 Turkey 18
United Arab Emirates 18
Uruguay 18
Table A2.1 – White's test for Ho: homoskedasticity, real value added share agriculture Source chi2 df p Heteroskedasticity 1473.69 381 0.0000 Skewness 106.28 39 0.0000 Kurtosis 19.96 1 0.0000 Total 1599.93 421 0.0000
Table A3.2 – White's test for Ho: homoskedasticity, real value added share manufacturing
Source chi2 df p
Heteroskedasticity 1002.49 381 0.0000
Skewness 100.23 39 0.0000
Kurtosis 0.15 1 0.6954
Total 1102.87 421 0.0000
Table A3.3 – White's test for Ho: homoskedasticity, real value added share services
Source chi2 df p
Heteroskedasticity 1445.13 381 0.0000
Skewness 89.04 39 0.0000
Kurtosis 33.56 1 0.0000
Table A4 – Correlations between dependent and independent variables va_agr va_man va_ser CHN
import
GDP/c teco ieco va_mu landarea pop arable age dep.yo age dep.old va_agr 1 va_man -0.3754 1 va_ser -0.531 0.219 1 CHNimportUS -0.1309 0.0511 0.1782 1 gdppc_ppp -0.5584 0.0903 0.3878 0.3294 1 teco -0.0774 0.106 0.0706 -0.0088 0.0184 1 ieco -0.0298 -0.0273 0.2043 -0.0367 0.0423 -0.0237 1 va_mu -0.3957 -0.1856 -0.4958 -0.046 0.1698 -0.0141 -0.149 1 landarea -0.1447 0.0274 0.1067 0.3204 0.1493 -0.0466 -0.1018 0.049 1 population 0.015 0.1005 -0.0349 0.2038 -0.0425 -0.0274 -0.0757 -0.0257 0.3282 1 arable 0.0889 0.2017 0.113 0.0344 -0.055 0.2868 -0.0655 -0.2782 -0.1111 0.3383 1 agedep_young 0.6848 -0.4381 -0.532 -0.1991 -0.7165 -0.1439 -0.0701 -0.03 -0.1458 -0.0588 -0.134 1 agedep_old -0.5165 0.3123 0.6014 0.2166 0.6082 0.169 0.0197 -0.1699 0.0804 -0.0215 0.1904 -0.7924 1
Table A5 – Collinearity Test
Figure A1.1 – GDP per capita against real value added share agriculture
Notes: fitted values = linear fit; lowess va_agr lngdppc_ppp = line based on locally weighted regression
Figure A1.2 – GDP per capita against real value added share manufacturing
Notes: fitted values = linear fit; lowess va_agr lngdppc_ppp = line based on locally weighted regression
0 5 0 1 0 0 4 6 8 10 12 lngdppc_ppp
Value added - Agriculture share Fitted values lowess va_agr lngdppc_ppp 0 5 1 0 1 5 2 0 2 5 4 6 8 10 12 lngdppc_ppp
Figure A1.3 – GDP per capita against real value added share service
Notes: fitted values = linear fit; lowess va_agr lngdppc_ppp = line based on locally weighted regression
0 2 0 4 0 6 0 8 0 4 6 8 10 12 lngdppc_ppp
Table A6.1 – Pooled OLS model, robust standard errors: Advanced economies
VARIABLES Agriculture Manufacturing Services
lnCHNimUS -0.644 -1.007 1.680 (0.734) (1.110) (2.250) lngdppc_ppp -10.51 -17.45 65.95*** (6.588) (13.96) (23.17) c.lnCHNimUS#c.lngdppc_ppp 0.0200 0.131 -0.135 (0.0682) (0.115) (0.219) lngdppc_ppp2 0.439 0.745 -3.093** (0.382) (0.811) (1.305) ieco 1.425*** 1.502*** -1.463** (0.240) (0.430) (0.738) va_mu -0.133*** 0.126*** -0.829*** (0.0113) (0.0355) (0.0325) lnlandarea 0.439*** -0.666*** 0.0890 (0.0337) (0.0708) (0.103) lnpopulation -0.236* 0.366* 0.170 (0.123) (0.206) (0.337) arable -0.0152*** -0.0163*** 0.0633*** (0.00284) (0.00487) (0.00955) agedep_young -0.000995 -0.0878*** 0.265*** (0.0124) (0.0248) (0.0427) agedep_old -0.0161 -0.0587 0.265*** (0.0199) (0.0428) (0.0632) Constant 72.40** 112.6* -300.2*** (28.07) (59.61) (104.3) Observations 660 660 660 R-squared 0.658 0.260 0.718 Number of countries 87 87 87
Time dummies YES YES YES
Table A6.2 – Pooled OLS model, robust standard errors: Emerging economies
VARIABLES Agriculture Manufacturing Services
lnCHNimUS 0.0716 -4.589*** 4.995*** (0.400) (0.667) (0.705) lngdppc_ppp -34.71*** 6.051*** 30.63*** (1.291) (1.881) (1.927) c.lnCHNimUS#c.lngdppc_ppp 0.0156 0.567*** -0.623*** (0.0450) (0.0793) (0.0849) lngdppc_ppp2 1.685*** -1.090*** -0.703*** (0.0752) (0.115) (0.131) teco -6.607*** -2.913*** 9.620*** (0.562) (0.713) (0.792) ieco 4.247*** -1.870*** -0.645 (0.497) (0.441) (0.610) va_mu -0.0243* -0.0625*** -0.884*** (0.0124) (0.0156) (0.0188) lnlandarea 0.598*** -0.644*** 0.175 (0.0941) (0.143) (0.155) lnpopulation -0.685*** 1.006*** -0.805*** (0.146) (0.189) (0.224) arable 0.132*** -0.0226** -0.0794*** (0.0102) (0.0101) (0.0101) agedep_young 0.0410*** -0.157*** 0.136*** (0.0109) (0.0127) (0.0137) agedep_old 0.121*** 0.144*** -0.208*** (0.0294) (0.0435) (0.0457) Constant 173.3*** 39.42*** -127.5*** (7.693) (12.11) (12.16) Observations 1,170 1,170 1,170 R-squared 0.791 0.507 0.829 Number of countries 39 39 39
Time dummies YES YES YES
Table A6.3 – Pooled OLS model, robust standard errors: Low income countries
VARIABLES Agriculture Manufacturing Services
lnCHNimUS -1.005 0.926 -1.477 (2.596) (0.656) (2.158) lngdppc_ppp 11.15 -4.287 -8.206 (11.33) (2.641) (9.525) c.lnCHNimUS#c.lngdppc_ppp 0.258 -0.273*** 0.253 (0.374) (0.0952) (0.305) lngdppc_ppp2 -2.538*** 0.899*** 1.300** (0.783) (0.185) (0.624) ieco -2.660 -7.452*** 16.06*** (1.950) (0.678) (1.598) va_mu -0.0455 -0.155*** -0.759*** (0.0286) (0.0118) (0.0274) lnlandarea 1.010** -1.919*** 1.131*** (0.440) (0.186) (0.427) lnpopulation -5.453*** 3.773*** 1.761*** (0.555) (0.229) (0.544) arable 0.0735** -0.128*** 0.0536 (0.0350) (0.0144) (0.0336) agedep_young -0.243*** -0.0286* 0.279*** (0.0476) (0.0150) (0.0386) agedep_old -0.600* 0.502*** -0.0125 (0.307) (0.118) (0.317) Constant 148.2*** -17.77 -17.19 (48.85) (11.70) (42.59) Observations 780 780 780 R-squared 0.689 0.486 0.661 Number of countries 26 26 26
Time dummies YES YES YES
Table A7.1 – Quantile regression: Advanced economies VARIABLES Q10 Q25 Q50 Q75 Q90 lnCHNimUS -8.260*** -5.553*** 0.232 1.074 1.105* (0.913) (1.674) (1.113) (0.843) (0.565) lngdppc_ppp 29.44*** -0.921 -29.18*** -25.22** -8.904 (7.751) (19.38) (9.698) (10.42) (5.975) c.lnCHNimUS#c.lngdppc_ppp 0.916*** 0.622*** 0.0268 -0.128 -0.118** (0.0878) (0.175) (0.110) (0.0826) (0.0498) lngdppc_ppp2 -2.482*** -0.728 1.440** 1.429** 0.585* (0.463) (1.140) (0.594) (0.607) (0.340) ieco 4.012*** 3.881*** 1.021*** 0.133 -1.516*** (0.307) (0.614) (0.256) (0.266) (0.256) va_mu 0.0724*** 0.125** 0.339*** 0.646*** 0.828*** (0.0169) (0.0572) (0.0244) (0.0483) (0.0198) lnlandarea -0.249*** -0.170 -0.946*** -0.759*** -0.628*** (0.0454) (0.121) (0.0674) (0.0649) (0.0560) lnpopulation -0.482*** -0.197 0.203 0.813*** 0.350 (0.131) (0.344) (0.143) (0.174) (0.242) arable 0.0487*** 0.0373*** -0.0461*** -0.0326*** -0.0182** (0.00380) (0.00987) (0.00494) (0.00440) (0.00859) agedep_young -0.0274 -0.0851** -0.0625*** -0.0491*** -0.0140 (0.0177) (0.0431) (0.0158) (0.0144) (0.00990) agedep_old -0.0421** -0.00329 0.0987*** 0.0392 -0.0266* (0.0212) (0.0606) (0.0295) (0.0282) (0.0148) Constant -41.20 84.64 155.7*** 114.5*** 36.52 (32.13) (81.42) (38.67) (44.19) (26.72) Observations 660 660 660 660 660 Number of countries 22 22 22 22 22
Time dummies YES YES YES YES YES
Pseudo R2 0.314 0.190 0.256 0.345 0.448
Table A7.2 – Quantile regression: Emerging economies VARIABLES Q10 Q25 Q50 Q75 Q90 lnCHNimUS -1.555 -2.201*** -3.847*** -5.432*** -6.891*** (1.051) (0.804) (0.537) (0.732) (0.784) lngdppc_ppp 10.05*** 9.868*** 6.851*** 12.30*** 9.551*** (2.575) (1.515) (2.141) (2.577) (2.586) c.lnCHNimUS#c.lngdppc_ppp 0.270** 0.334*** 0.460*** 0.662*** 0.849*** (0.125) (0.0931) (0.0613) (0.0893) (0.0919) lngdppc_ppp2 -0.939*** -1.063*** -0.993*** -1.575*** -1.709*** (0.192) (0.105) (0.121) (0.161) (0.169) teco -5.533*** -1.146 -2.739*** -3.266*** -5.605*** (1.491) (0.748) (0.758) (1.150) (0.891) ieco 0.776 -1.563*** -2.277*** -3.435*** -2.819*** (1.118) (0.416) (0.272) (0.697) (0.585) va_mu -0.157*** -0.0848*** -0.0921*** -0.0452* 0.230*** (0.0249) (0.0160) (0.0106) (0.0241) (0.0221) lnlandarea 0.469* 0.110 -0.332*** -1.012*** -1.432*** (0.255) (0.130) (0.119) (0.189) (0.220) lnpopulation 0.00774 -0.0494 0.733*** 1.524*** 1.396*** (0.394) (0.177) (0.114) (0.254) (0.211) arable 0.0192 -0.00320 0.0130 -0.0391*** -0.0474** (0.0279) (0.0101) (0.0170) (0.0151) (0.0194) agedep_young -0.233*** -0.186*** -0.141*** -0.178*** -0.0967*** (0.0191) (0.0104) (0.00857) (0.0162) (0.0106) agedep_old 0.0355 -0.0217 0.0849*** 0.0875 0.425*** (0.0777) (0.0409) (0.0204) (0.0789) (0.0687) Constant -6.447 7.568 28.66** 22.79 48.20*** (14.75) (11.59) (11.51) (14.85) (14.37) Observations 1,170 1,170 1,170 1,170 1,170 Number of countries 39 39 39 39 39
Time dummies YES YES YES YES YES
Pseudo R2 0.414 0.415 0.316 0.239 0.262
Table A7.3 – Quantile regression: Low income countries VARIABLES Q10 Q25 Q50 Q75 Q90 lnCHNimUS -0.122 0.552 0.928 -1.414* -2.069*** (0.730) (0.956) (0.711) (0.812) (0.675) lngdppc_ppp -0.834 -7.121** -8.050*** -9.179*** -2.787 (3.821) (3.330) (2.923) (3.516) (3.760) c.lnCHNimUS#c.lngdppc_ppp -0.131 -0.248* -0.301*** 0.119 0.229** (0.110) (0.143) (0.104) (0.118) (0.0916) lngdppc_ppp2 0.433** 1.072*** 1.229*** 0.739*** 0.0970 (0.204) (0.288) (0.180) (0.218) (0.279) ieco -4.774*** -6.691*** -7.836*** -5.577*** -4.954*** (0.499) (0.841) (0.777) (0.772) (0.658) va_mu -0.137*** -0.180*** -0.154*** -0.0428 0.0460*** (0.00894) (0.0129) (0.0177) (0.0361) (0.00954) lnlandarea -1.572*** -1.184*** -1.266*** -2.511*** -3.100*** (0.124) (0.162) (0.171) (0.221) (0.242) lnpopulation 3.243*** 3.114*** 3.398*** 4.102*** 4.520*** (0.180) (0.200) (0.228) (0.265) (0.238) arable -0.152*** -0.0747*** -0.0741*** -0.155*** -0.192*** (0.0100) (0.0207) (0.0144) (0.0121) (0.0201) agedep_young 0.00109 -0.0486*** -0.0371** 0.0500** -0.00434 (0.0137) (0.0154) (0.0185) (0.0225) (0.0279) agedep_old 0.0143 -0.0348 0.476*** 0.850*** 0.387** (0.0853) (0.0851) (0.140) (0.157) (0.165) Constant -17.70 -0.243 -6.232 10.88 4.518 (18.07) (14.39) (14.27) (15.63) (13.55) Observations 780 780 780 780 780 Number of countries 26 26 26 26 26
Time dummies YES YES YES YES YES
Pseudo R2 0.312 0.278 0.315 0.376 0.406
