THE AGRICULTURAL SECTOR AS THE NEW
GROWTH ENGINE
June 2, 2016
Jennienke Kamphuis
University of Groningen Faculty of Economics and Business
Supervisor: dr. R.K.J. Maseland
Author: Jennienke Kamphuis, Blinde Banisweg 23a, 7462 VH Rijssen, The Netherlands. Student number: s2350769
2 Abstract
The deindustrialization trend observed in developing regions is appointed as premature by Rodrik (2016). He states that these countries shift towards services-led growth, which limits their growth suspects significantly. Although this is the case for Sub-Saharan Africa, Rodrik overlooks a second growth possibility, which is growth through the agricultural sector. Due to the industrialization of Asia, the demand for agricultural products increased, which pushed up prices. These developments provide new growth possibilities for Latin American countries, which makes industrializing redundant for the time being. Therefore we see a trend of delayed or less deindustrialization in Latin America.
3 1. Introduction
Deindustrialization in the Western economies is a well-known phenomenon for the last decades. Broadly speaking there are two ways for measuring deindustrialization. The first is the manufacturing industries’ value-added share of total GDP. The second, which is most prominently used in the literature, is the manufacturing industries’ share of total employment. (Rodrik, 2016). Deindustrialization measured in value terms was moderate for the Western world, compared to the measurement in employment terms (Lawrence & Edwards, 2013, Rodrik, 2016).
Several explanations have been offered for the decreasing employment share of manufacturing. The research of Kehoe, Ruhl and Steinberg (2013) investigate if it was U.S. borrowing or productivity growth in the manufacturing sector which could explain the decline in employment for the manufacturing sector and conclude that rapid productivity growth has been the most important driver. Furthermore, Wolman, Wial and Hill (2015) point towards a broader range of explanations for the decline in manufacturing employment, such as globalization, technology, the shifting of consumer demand away from goods towards services and international differences in taxes or workforce skills. Overall, the conventional explanation for employment deindustrialization focuses on the differential rates of technological progress (Lawrence & Edwards, 2013).
However, Lawrence and Edwards (2013) offer a more intertwined explanation, which devotes deindustrialization to the combination of productivity growth, demand and trade. Manufacturing productivity growth has outpaced other sectors’ productivity growth, possibly because this sector is more dynamic, it faces more competition due to the tradability of goods, which are more easily automated as well. Therefore, manufacturing prices fell, while there was a shifting demand from goods towards services, which decreased manufacturing employment and value relative to other sectors. Trade is defined by Lawrence and Edwards as the final small explanation in the case of the U.S., since the U.S. deficit captures value and employment which could be used domestically as well. Furthermore, they indicate that their results for the U.S. are comparable to other developed economies, which completes the picture for deindustrialization in the developed world.
4
2016). Empirical evidence is illustrated by the case of Asia (McMillan & Rodrik, 2011). However surprisingly, they also show that since 1990 labor in Africa and Latin America flew from high-productivity activities to low-productivity activities, thus the other way around. Since the manufacturing sector faces a higher productivity growth than the traditional sectors like agriculture and (informal) services, these results imply deindustrialization for those developing countries as well. Rodrik (2016) calls his type of deindustrialization ‘premature deindustrialization’, since those “developing countries are turning into service economies without having gone through a proper experience of industrialization”. In the absence of this proper experience of industrialization, he proposes services-led growth as a new growth model. However, it should be kept in mind that the service sector suffers from several shortcomings; they are often technologically not very dynamic or they are non-tradable, which limits rapid expansion. Of course there are services like IT and finance, which are dynamic and tradable, but contrary to manufacturing those services often require high-skilled labor and are not able to absorb the type of labor which is abundant in low- and middle income economies.
Nevertheless, next to the explanation of developing countries turning into service economies too soon, there is another explanation of the deindustrialization of developing countries, which focuses on the increased input prices of the manufacturing sector, supplied by the agricultural sector. These prices increase due to the higher demand for agricultural raw materials, which is caused by the fast development of Asian industries. Since the higher prices for agricultural products provide developing countries with a more profitable primary sector, industrialization is just delayed or less needed.
Although Rodrik (2016) is quite convinced of the existence of premature deindustrialization, he never did proper research to check whether it really is premature deindustrialization or whether it could be delayed or less industrialization. This will be the aim of this paper, phrased in the following research question:
5 2. Literature review
Since the deindustrialization in the developed countries is particularly noticeable through the change in employment (Rodrik, 2016), we present a simple deindustrialization model explaining the declining employment share of manufacturing. This model is very common in the development literature (Rowthorn & Wells, 1987, Rowthorn & Ramaswamy, 1997) and will be used as a benchmark model. Its main purpose in this paper is to illustrate the usual path that the shares of employment followed during the development of the developed countries. In this model the economy is divided into the agriculture (a), manufacturing (m) and service (s) sector. It uses the commonly used concept of productivity growth differentials between those sectors as the explanation for deindustrialization and shows the rising importance of manufacturing employment during the early phase of development, as well as the transition to a service economy in which manufacturing employment declines. Appendix A shows the model in detail.
6 Figure 1: the changing structure of employment
↓ Employment share 1
0
Level of development →
This model only explains the employment deindustrialization and neglects output deindustrialization. However, employment deindustrialization is what we mainly observe in the developed countries, therefore a technology-based story fits the patterns in the developed countries very well (Rodrik, 2016).
For developing countries, though, the technology-based argument is not applicable in the same way. Crucially, this argument is based on the adjustments in domestic relative prices. Provided that the productivity growth outweighs the demand growth, higher productivity growth in manufacturing depresses the domestic relative price of manufacturing such that the demand for labor in the manufacturing sector decreases. Contrary to the developed world, developing countries are often small in world markets for manufactures, which makes them essentially price takers. In the limit, where domestic relative prices are fully determined globally, higher domestic productivity growth in manufacturing would cause industrialization instead of deindustrialization, in both employment and output terms. A plausible alternative for the technology-based story relevant for the developing countries would be globalization and trade (Rodrik, 2016, Wolman, Wial & Hill, 2015, Lawrence & Edwards, 2013). Following this argument, manufacturing sectors in most developing countries were even double hit when they opened up to trade. Countries without a strong comparative advantage in manufacturing became net importers of manufactures, which implies deindustrialization both in terms of employment and output. Furthermore, developing countries suddenly faced lower relative prices of manufacturing due to the price trends originating from the developed countries, which implies employment and output deindustrialization as well. This is consistent with the empirics that
Agriculture
Manufacturing
7
show a reduction in both employment and output shares in developing countries, especially for countries without a strong comparative advantage in manufacturing. Those with a comparative advantage in manufacturing, for instance Asian countries, have mostly been spared those trends (Rodrik, 2016). So while technological progress largely explains employment deindustrialization in the developed countries, globalization and trade are more important in explaining deindustrialization trends in developing countries.
Now why would the deindustrialization process in the developing countries be called premature? He distinguishes two senses in which the shrinking manufacturing sector in developing countries can be seen as premature. First, since 1990 developing countries start to deindustrialize at levels of income that are around 40 percent of their earlier counterparts who deindustrialized before 1990. Second, early deindustrialization may have significant negative effects on economic growth. Manufacturing activities have some features that makes them very useful in the process of growth. First, the manufacturing sector tends to be a technologically dynamic sector. Rodrik (2013) even shows that manufacturing sectors exhibit unconditional labor productivity convergence, unlike other sectors in the economy. Second, the manufacturing sector is able to absorb high quantities of unskilled labor. The combination of high labor productivity and the ability to absorb lots of unskilled labor makes the manufacturing sector rather unique. Furthermore, this sector is tradable, which extends the home market demand to worldwide demand. Taken together, Rodrik (2014) fears that due to deindustrialization, the historical engine behind rapid growth is taken away.
Instead of the manufacturing engine, Rodrik (2016) assumes future growth will be driven by the service sector. In that case, growth will probably diminish since the service sector does not contain growth stimulating features comparable to the manufacturing sector. If we follow this line of thinking, the decline in the manufacturing share is accompanied by an incline in the service sector. This idea is captured in hypothesis 1:
If there is a trend of premature deindustrialization, the positive relation between the share of services and income has grown stronger over the last decades.
8
increased (Rodrik, 2016). Especially the manufacturing sector of the Asian countries developed significantly. This happened at the expense of the agricultural share of output and employment in these countries (McMillan, M.S. & Rodrik, D, 2011, Alderson, A.S. 1997). However, due to the increasing industrialization of especially the Asian economies, the demand for raw material and commodities of the agricultural sector increased. Due to this increase in demand, the price for agricultural products probably increases and the agricultural sector becomes more profitable for the developing countries. This idea is captured in hypothesis 2a:
If the industrialization of Asia pushed up demand for agricultural raw materials, there is a positive relation between the price of raw materials and explanatory variables for Asian
manufacturing employment share.
In this way, developing countries just react on the increasing demand for agricultural raw materials by moving labor and thus output from the manufacturing sector to the agricultural sector. Further industrialization might just be delayed or not needed as long as the agricultural sector provides growth. This positive view is formulated in hypothesis 2b:
If there is a trend of delayed or less industrialization, the negative relation between the share of agriculture and income has grown weaker over the last decades.
3. Data and Methods
To check whether the price of agricultural raw materials is positively related to the Asian development, we first have estimate the Asian manufacturing employment as a function of income and demographic trends. Then, the estimated income and demographic coefficients are combined and related to the price of agricultural raw materials and commodities.
9
like large economies in Latin-America, Asia and sub-Saharan Africa. Appendix B contains more details on the data set. Constant-price series are at 2005 prices, with the exception of West Germany2. The GGDC dataset includes data on GDP and is complemented by data on population per country published by the Department of Economic and Social Affairs from the United Nations (2016).
To relate the price of agricultural raw materials with the dynamics of the booming Asian manufacturing employment share, the latter first needs to be obtained by repeating Rodrik’s analysis. This analysis is based on a quadratic estimation which relates the sector share with income and population, estimated by OLS regressions. Furthermore, time is controlled for. The coefficients on income and population are saved, after which they are regressed with the price of agricultural raw materials.
The baseline results are based on a comparable quadratic estimation which relates the sector share with income, estimated by OLS regressions. In the same way as the
manufacturing share, the increase or decrease in the service and agricultural sector can be measured. For completeness, both ways are used, thus services real value added (SVA) and agricultural real value added (AVA) as a share of GDP, as well as services employment and agricultural employment as a share of total employment1. Again, population size and time are controlled for, as well as the manufacturing share. A variable for time is included in two different ways, first as a single continuous variable, which captures of time effects in one coefficient. Second, as period dummies, since these capture the dynamics of the relation between income and the increase or decrease of the sector share. The baseline OLS regression is specified in the next section.
1 Rodrik (2016) also used a third measure; manufacturing value added in current prices (nommva). This measure contains inseparable information about prices and quantities, which makes it less useful in disentangling the structural change patterns.
10 4. Empirical Results
The agricultural raw materials price issue plays an important role in explaining the reasoning behind the hypothesis of delayed or less industrialization. Therefore we start with explaining hypothesis 2a in section 4.1, after which we continue with the baseline results related to hypothesis 1 and 2b in section 4.2.
4.1 Relation between manufacturing share and price of raw materials
Hypothesis 2a states that due to the Asian development of the manufacturing sector, the demand for agricultural raw materials, which serve as inputs for the industrial sector, increases significantly. Due to this increase in demand, prices of agricultural raw materials increase as well. To check this, we ran the recession explained in the previous section. As table 1 shows, the estimated coefficients for the income and population effect are both positive, although only the population effect is statistically significant. This implies that the increase of the Asian manufacturing share, as explained by the income and demographic trends, is positively related to the price level of agricultural raw materials, which underlines hypothesis 2a.
Table 1: Relation between worldwide price of raw materials and explanatory variables for Asian
manufacturing employment share
Coefficients Income effect on Asian manufacturing employment
Population effect on Asian manufacturing employment
67.577 (150.114) 67.430* (20.452) Standard errors are reported in parentheses
Levels of statistical significance: *99%, ** 95%, *** 90%
4.2 Baseline results
11 (1) 𝑠𝑒𝑐𝑡𝑜𝑟𝑠ℎ𝑎𝑟𝑒𝑖𝑡 = 𝛽0+ 𝛽1ln 𝑝𝑜𝑝𝑖𝑡+ 𝛽2(ln 𝑝𝑜𝑝𝑖𝑡)2+ 𝛽 3ln 𝑦𝑖𝑡+ 𝛽4(ln 𝑦𝑖𝑡)2+ 𝛽5𝑚𝑎𝑛𝑠ℎ𝑎𝑟𝑒𝑖𝑡+ 𝛽6𝑦𝑒𝑎𝑟𝑡+ 𝜖𝑖𝑡, (2) 𝑠𝑒𝑐𝑡𝑜𝑟𝑠ℎ𝑎𝑟𝑒𝑖𝑡 = 𝛽0+ 𝛽1ln 𝑝𝑜𝑝𝑖𝑡+ 𝛽2(ln 𝑝𝑜𝑝𝑖𝑡)2+ 𝛽3ln 𝑦𝑖𝑡+ 𝛽4(ln 𝑦𝑖𝑡)2+ 𝛽5𝑚𝑎𝑛𝑠ℎ𝑎𝑟𝑒𝑖𝑡+ ∑ 𝛽𝑇 6𝑇𝑃𝐸𝑅𝑇+ 𝜖𝑖𝑡.
These regressions measure whether the sector share increases or decreases over time, beyond the share that is explained by population, income and the manufacturing share. This is captured by the time variables 𝑦𝑒𝑎𝑟𝑡and 𝑃𝐸𝑅𝑇, which contain all still unexplained effects.
Furthermore, the interaction period regressions are included because they specifically trace the change in the explanatory effect of income on the services share over time. Contrary to the simple period analysis, the interaction variables only contain unexplained effects caused by a change in the explanatory effect of income. The remaining unexplained effects are captured by the single year variable. So the interaction period analysis accentuates the change in the relationship between income and the sector share, while the simple period analysis pools this change with the rest of the unexplained effects. The interaction period regressions are expressed as follows:
(3) 𝑠𝑒𝑐𝑡𝑜𝑟𝑠ℎ𝑎𝑟𝑒𝑖𝑡 = 𝛽0+ 𝛽1ln 𝑝𝑜𝑝𝑖𝑡+ 𝛽2(ln 𝑝𝑜𝑝𝑖𝑡)2+ 𝛽3ln 𝑦𝑖𝑡+ 𝛽4(ln 𝑦𝑖𝑡)2+ 𝛽5𝑚𝑎𝑛𝑠ℎ𝑎𝑟𝑒𝑖𝑡+ 𝛽6𝑦𝑒𝑎𝑟𝑡+ 𝛽7(year ∗ ln 𝑦𝑖𝑡) + 𝛽8(year ∗ (ln 𝑦𝑖𝑡)2) + 𝜖 𝑖𝑡, (4) 𝑠𝑒𝑐𝑡𝑜𝑟𝑠ℎ𝑎𝑟𝑒𝑖𝑡 = 𝛽0+ 𝛽1ln 𝑝𝑜𝑝𝑖𝑡+ 𝛽2(ln 𝑝𝑜𝑝𝑖𝑡)2+ 𝛽 3ln 𝑦𝑖𝑡+ 𝛽4(ln 𝑦𝑖𝑡)2+ 𝛽5𝑚𝑎𝑛𝑠ℎ𝑎𝑟𝑒𝑖𝑡+ ∑ 𝛽𝑇 6𝑇𝑃𝐸𝑅𝑇+ ∑ 𝛽𝑇 7𝑇(𝑃𝐸𝑅𝑇∗ ln𝑦𝑖𝑡) + (∑ 𝛽𝑇 8𝑇𝑃𝐸𝑅𝑇∗ (ln 𝑦𝑖𝑡)2) + 𝜖𝑖𝑡.
Sectorshare denotes the employment or value added measure for the service sector.
12
The agricultural, manufacturing and services share add up to one. Since 𝛽5 controls for the manufacturing share in all regressions, the results for the services share are the exact opposite of the results for the agricultural share. When the manufacturing share is controlled for, the agricultural share has to decrease in order for the services share to increase. Therefore, despite only regressing the services share, we can still draw conclusions on the agricultural share. The period dummies (𝑃𝐸𝑅𝑇) are expressed per decade, namely the 1960s, 1970s,
1980s, 1990s, and 2000s years. Data before the 1960s are used as base years, while data after 2010 are captured by the 2000s dummy variable.
The sample will be divided into 5 regions; Developed Countries, Asia, Latin-America, Sub-Saharan Africa and Sub-Sub-Saharan Africa excluding Mauritius. Appendix B shows detailed information about the included countries. The reason Mauritius is excluded in the last region is because it has important differences with the other SSA countries, such that it is a strong manufactures exporting island country which is densely populated with a majority of Asian inhabitants. Including this country in the Sub-Saharan Africa region influenced the results of Rodrik (2016) significantly by turning the estimated coefficients for the manufacturing share from negative to positive. Since this issue might also be relevant when estimating the services share, this 5th region is included in this paper as well. The regions of our interest are the
developing regions, in this sample represented by Latin-America and Sub-Saharan Africa (both including and excluding Mauritius).
For regressions (1) and (2), the variable of interest is the single time variable,
estimated by 𝛽6 and 𝛽6𝑇. The first hypothesis of premature industrialization expects a stronger positive relationship over time between the services share and income, implying 𝛽6 > 0 and 𝛽6𝑇 > 0. Since regression (2) also captures the dynamics over time relative to the pre-1960
years through the dummy variables, the first hypothesis also expects 𝛽6𝑇 to increase as T increases. Hypothesis 2b of less or delayed industrialization expects the negative relation between the agricultural share and income to grow weaker. Since the results of the
agricultural share are the exact opposite of the results of the services share, 𝛽6 < 0 and 𝛽6𝑇 < 0 count as evidence for hypothesis 2b. Furthermore, hypothesis 2b expects 𝛽6𝑇 to decrease as
T increases.
13
interaction variables, estimated by 𝛽7 and 𝛽7𝑇, which capture the change over time of the linear effect of income on the sector share, and estimated by 𝛽8 and 𝛽8𝑇, which capture the
change over time of the u-shaped or hump shaped effect of income on the sector share. When the combined effect of income is positive, such that for regression (3) (𝛽7+𝛽8) > 0 and for regression (4) (𝛽7𝑇+𝛽8𝑇) > 0, this is evidence for the first hypothesis. Furthermore, the combined effect of 𝛽7𝑇 and 𝛽8𝑇 should be increasing in T to count as evidence for hypothesis
Table 2a Simple period analysis, Services – Employment shares
Standard errors are reported in parentheses
Levels of statistical significance: *99%, ** 95%, *** 90% All countries All developing
countries
Developed countries
Asia Latin-America Sub-Saharan Africa Sub-Saharan Africa ex. Mauritius ln population ln population squared ln income per capita
15
Table 2b Simple period analysis, Services – Employment shares
Standard errors are reported in parentheses
Levels of statistical significance: *99%, ** 95%, *** 90% All countries All developing
countries
Developed countries
Asia Latin-America Sub-Saharan Africa Sub-Saharan Africa ex. Mauritius ln population ln population squared ln income per capita
16
Table 3a Simple period analysis, Services – Real Value Added Shares
Standard errors are reported in parentheses
Levels of statistical significance: *99%, ** 95%, *** 90% All countries All developing
countries
Developed countries
Asia Latin-America Sub-Saharan Africa
Sub-Saharan Africa ex. Mauritius ln population
ln population squared ln income per capita
17
Table 3b Simple period analysis, Services – Real Value Added Shares
Standard errors are reported in parentheses
Levels of statistical significance: *99%, ** 95%, *** 90% All countries All developing
countries
Developed countries
Asia Latin-America Sub-Saharan Africa Sub-Saharan Africa ex. Mauritius ln population ln population squared ln income per capita
18
Table 4a Interaction period analysis, Services – Employment shares
Standard errors are reported in parentheses
Levels of statistical significance: *99%, ** 95%, *** 90% All countries All developing
countries
Developed countries
Asia Latin-America Sub-Saharan Africa Sub-Saharan Africa ex. Mauritius ln population ln population squared ln income per capita
ln income per capita squared Manufacturing share
Year
19
Table 4b Interaction period analysis, Services – Employment shares
All countries All developing countries
Developed countries
Asia Latin-America Sub-Saharan Africa
Sub-Saharan Africa ex. Mauritius ln population
ln population squared ln income per capita
ln income per capita squared Manufacturing share 1960s 1970s 1980s 1990s 2000s+
1960s * ln income per capita 1960 * ln income per capita squared COMBINED EFFECT 1960s 1970s * ln income per capita
1970 * ln income per capita squared COMBINED EFFECT 1970s 1980s * ln income per capita
1980 * ln income per capita squared COMBINED EFFECT 1980s 1990s * ln income per capita
1990 * ln income per capita squared
20 Standard errors are reported in parentheses
Levels of statistical significance: *99%, ** 95%, *** 90 COMBINED EFFECT 1990s
2000s+ * ln income per capita
21
Table 5a Interaction period analysis, Services – Real Value Added Shares
Standard errors are reported in parentheses
Levels of statistical significance: *99%, ** 95%, *** 90%
All countries All developing countries
Developed countries
Asia Latin-America Sub-Saharan Africa Sub-Saharan Africa ex. Mauritius ln population ln population squared ln income per capita
ln income per capita squared Manufacturing share
Year
Year * ln income per capita
22
Table 5b Interaction period analysis, Services – Real Value Added Shares
All countries All developing countries
Developed countries
Asia Latin-America Sub-Saharan Africa Sub-Saharan Africa ex. Mauritius ln population ln population squared ln income per capita
ln income per capita squared Manufacturing share 1960s 1970s 1980s 1990s 2000s+
1960s * ln income per capita
1960s * ln income per capita squared COMBINED EFFECT 1960s 1970s * ln income per capita
1970s * ln income per capita squared COMBINED EFFECT 1970s 1980s * ln income per capita
1980s * ln income per capita squared
23 Standard errors are reported in parentheses
Levels of statistical significance: *99%, ** 95%, *** 90% COMBINED EFFECT 1980s
1990s * ln income per capita
1990s * ln income per capita squared COMBINED EFFECT 1990s 2000s+ * ln income per capita
The results point towards regional differences. Especially when analyzed using interaction effects, the results between developed (Western) and developing countries (Latin America & Sub-Saharan Africa) differ significantly. Furthermore, within the developing regions the differences are striking as well. The detailed discussion of results starts with the developed countries.
First, both the simple period analysis (Table 2a and 2b) and the interaction period analysis (Table 4a and 4b) show a positive estimate of the single year variable or increasing estimates of the period dummies when measured for developed countries in employment terms. However, the results for developed countries in terms of real value added are not that clear-cut. The simple analysis in terms of real value added (Table 3a and 3b) shows a negative year variable and decreasing estimates of the period dummies, indicating a decreasing
services share, while the interaction analysis (Table 5a and 5b) shows the opposite estimates, indicating increasing services shares in recent decades. Nevertheless, these results
complement Rodrik’s story very well. He only found evidence for deindustrialization of developed countries when measured in employment terms. So for developing countries, the deindustrialization in employment terms (beyond what would have been expected by income and demographic trends) is compensated for by the increasing employment share of the service sector.
Turning to Asia, the positive year estimate and the increasing estimates of the period dummies point towards an increasing services share in employment terms and in real value added terms, both for the simple period analysis and the interaction period analysis. In our analysis, we control for the manufacturing sector, therefore the results of the services sector are opposite to the results of the agricultural sector. Since the services share increases, it follows directly that the Asian agricultural sector decreases. According to Rodrik (2016), Asia is industrializing. When combining this result with the increasing services share, it is clear that both the Asian secondary and tertiary sector flourish at the expense of the primary sector.
25
employment measure are not as pronounced, since the simple period analysis shows positive period dummy estimates, implying an increase of the employment services share at the expense of the agricultural sector, while the interaction period analysis gives negative period dummy estimates, indicating a decrease of this same share. Due to this difference in results two forces can be distinguished. Since the interaction period analysis specifically analyzes the relation between income and the sector share, these negative estimates point towards a weaker relation over time between income and the service sector, and thus a stronger relation between income and the agricultural sector. The simple period analysis contains all unexplained effects including the relation between income and the sector share. The positive estimates of the simple period analysis show there are other forces at work which overrule the income related effect and therefore turn the decreasing services share into an increasing services share in terms of employment. Since the aim of this research is to investigate what happens with the relation between income and the services and agricultural shares of developing countries, the interaction period results are decisive, and the hypothesis of less or delayed industrialization is supported by this.
26 5. Conclusions
The decreasing manufacturing share in the developing regions Latin America and Sub-Saharan Africa is described by Rodrik (2016) as premature deindustrialization, since he supposed that the missing manufacturing in terms of employment and real value added automatically flows to the services sector. However, this paper shows that only for the
developing region of Sub-Saharan Africa, when measured in employment terms, there is clear evidence for the existence of premature deindustrialization. This might interfere the
development of the Sub-Saharan African countries, because the services sector suffers from several shortcomings like the inability to absorb high quantities of unskilled labor and the non-tradability of most of its products. However, deindustrialization measured in real value added should be seen as less or delayed industrialization rather than premature
deindustrialization, because the East-Asian development provides them with the possibility to grow rich by supplying the industrializing world with agricultural input products. This implies that the productivity per worker in the agricultural and services sector respectively increased and decreased, since labor shifted from the agricultural sector to the services sector while real value added shifted inversely.
Furthermore, Latin American countries show this trend for both measures. For these countries, deindustrialization measured in real value added and employment terms is only delayed or less needed due to East-Asian development. This implies that deindustrialization need not be growth destroying for developing countries, but just replaces the traditional growth channel of industrialization with a new untraditional one. The shortcomings of the services sector as an engine for growth are less relevant for the agricultural sector, since this sector requires relatively unskilled labor, which is usually abundant in developing countries, and the agricultural products are more tradable.
27 Appendices
A. Benchmark model of deindustrialization
The simple deindustrialization model which is presented here is very common in the development literature (Rowthorn & Wells, 1987, Rowthorn & Ramaswamy, 1997). In this model the economy is divided into the agriculture (a), manufacturing (m) and services (s) sector. It uses the commonly used concept of productivity growth differentials between those sectors as the explanation for deindustrialization and is based on the following propositions:
(1) The demand for food is income-inelastic, which is Engel’s law;
(2) Real demand for services increases roughly in line with real national income; (3) Labor productivity growth is higher in the manufacturing than in the service sector.
Using these propositions, the rising importance of manufacturing employment during the early phase of development is shown, as well as the transition to a service economy in which manufacturing employment declines.
For simplicity, we assume a closed economy and real output is given by
1) Y = Ya + Ys + Ym
where Ya, Ys and Ym stand for output, measured at constant prices, in agriculture, services and
manufacturing respectively. The consumption of the agricultural product food, is constant, just as the labor force L, which captures the whole population. Since the economy is closed, this implies that
2) Ya = bL
where b is a constant. The output of services is a constant fraction of real output:
3) Ys = cY
As mentioned earlier, it is realistic to assume that productivity growth in
manufacturing is higher than in services. For sake of simplicity it is assumed that agriculture and manufacturing face the same productivity. Although most probably untrue, it won’t affect the main conclusions. Productivity growth rates remain constant over time and output per worker is the same in each sector of the economy at time zero. With these assumptions, output per worker per sector is
28
4) ys = y0eαt
ym = y0eλαt
where ya, ys and ym reflect output per worker in agriculture, services and manufacturing
respectively, and λ > 1, y0 > 0 and α > 0 are constants. The parameter λ is an index of uneven
productivity growth between manufacturing and services. Total employment is given by
5) L = La + Ls + Lm
Output per worker in each sector is
ya = Ya / La
6) ys = Ys / Ls
ym = Ym / Lm
Using equation (2) – (6), we get
7) 𝐿 = 𝑌
𝑦0[𝑐𝑒
−𝛼𝑡+ (1 − 𝑐)𝑒−𝜆𝛼𝑡]
The share of the labor force employed in each sector is denoted as
Pa = La / L
8) Ps = Ls / L
Pm = 1 - Pa - Ps
which can be rewritten as
𝑃𝑎 = 𝑏 𝑦0𝑒 −𝜆𝛼𝑡 9) 𝑃𝑠 = 𝑐 𝑐+(1−𝑐)𝑒−(𝜆−1)𝛼𝑡 𝑃𝑚 = 1 − 𝑏 𝑦0𝑒 −𝜆𝛼𝑡− 𝑐 𝑐+(1−𝑐)𝑒−(𝜆−1)𝛼𝑡
It is clear from (9), as t tends to infinity,
Pa → 0
Ps → 1
29
For the agricultural and service sector, convergence to the final limit is uniform:
the agricultural share in total employment declines steadily to zero, while the service share increases steadily to 1, such that everyone is employed in the service sector. However, the case of the manufacturing sector requires further analysis.
Differentiating equation of (9) for the manufacturing sector, results in
10) 𝑑𝑃𝑚
𝑑𝑡 = 𝜆𝛼𝑃𝑎− (𝜆 − 1)𝛼𝑃𝑠(1 − 𝑃𝑠)
Hence, 𝑑𝑃𝑚
𝑑𝑡 > 0 if and only if
11) 𝜆𝛼𝑃𝑎 > (𝜆 − 1)𝛼𝑃𝑠(1 − 𝑃𝑠)
30
B. Country and variable coverage in the GGDC 10-Sector Database
Acronym Country Value added in
constant prices Employment by sector Sub-Saharan Africa BWA ETH GHA KEN MWI MUS NGA NGA(alt) SEN ZAF TZA ZMB Botswana Ethiopia Ghana Kenya Malawi Mauritius Nigeria Nigeria (2014 GDP revision) Senegal South Africa Tanzania Zambia 1964-2010 1961–2010 1960-2010 1964-2010 1966-2010 1970-2010 1960–2010 2010–2013 (in 2010 prices) 1970–2010 1960–2010 1960–2010 1965–2010 1964-2010 1961–2010 1960-2010 1969-2010 1966-2010 1970-2010 1960–2011 1970–2010 1960–2010 1960–2010 1965–2010 Asia CHN HKG IND IDN JPN KOR MYS PHL SGP TWN THA China Honk Kong India Indonesia Japan South Korea Malaysia Philippines Singapore Taiwan Thailand 1952-2010 1974–2011 1950-2012 1960-2012 1953-2011 1953-2011 1970-2011 1971-2012 1960-2012 1961-2012 1951-2011 1952-2011 1974–2011 1960-2010 1961-2012 1953-2012 1963-2011 1975-2011 1971-2012 1970-2011 1963-2012 1960-2011 Latin America ARG BOL BRA CHL COL CRI MEX PER VEN Argentina Bolivia Brazil Chile Colombia Costa Rica Mexico Peru Venezuela 1950-2011 1950-2011 1950-2011 1950-2011 1950-2011 1950-2011 1950-2011 1950-2011 1950-2012 1950-2011 1950-2010 1950-2011 1950-2012 1950-2010 1950-2011 1950-2012 1960-2011 1950-2011 Developed Countries USA DEW DNK ESP FRA GBR ITA NLD SWE
United States of America West Germany Denmark Spain France United Kingdom Italy The Netherlands Sweden 1947-2010 1950-1991 (1991 prices) 1947-2009 1947-2009 1950-2009 1949-2009 1951-2009 1949-2009 1950-2009 1950-2010 1950-1991 1948-2011 1950-2011 1950-2011 1948-2011 1951-2011 1950-2011 1950-2011
31 References
Alderson, A. S. (1997). Globalization and Dcindustrialization: Direct
lnvcstmcnt and the Decline of Manufacturing Employment in 17 OECD Nations.
Journal of World-Systems Research 3, 1-34.
Atkinson, R. (2013, March 14). Why the 2000s Were a Lost Decade for American
Manufacturing. Retrieved from http://www.industryweek.com/global-economy/why-2000s-were-lost-decade-american-manufacturing.
IndexMundi. (2016) Commodity Agricultural Raw Materials Index Monthly Price – Index
Number. Retrieved from http://www.indexmundi.com/commodities/.
Kehoe, T.J., Ruhl, K.J. & Steinberg, J.B. (2013). Global Imbalances and Structural Change in the United States (Working Paper No. 19339). Retrieved from National Bureau of Economic Research website: http://www.nber.org/papers/w19339.
Kollmeyer, C. & Pichler, F. (2013). Is Deindustrialization Causing High Unemployment in Affluent Countries? Evidence from 16 OECD Countries, 1970-2003. Social Forces, 91(3), 785-812.
Lawrence, R.Z. & Edwards, L. (2013). US Employment Deindustrialization: Insights from History and the International Experience. Peterson Institute for International Economics, Policy Brief No. PB13-27.
McMillan, M.S. & Rodrik, D. (2011). Globalization, Structural Change, and Productivity Growth (Working Paper No. 17143). Retrieved from National Bureau of Economic Research website: http://www.nber.org/papers/w17143.pdf.
Rodrik, D. (2016). Premature Deindustrialization. Journal of Economic Growth, 21(1), 1-33.
Rodrik, D. (2014). The past, present, and future of economic growth. Challenge, 57(3), 5-39. Rodrik, D. (2013). Unconditional convergence in manufacturing. Quarterly Journal of
Economics, 128(1), 165–204.
Rowthorn, R.E. & Ramaswamy, R. (1997). Deindustrialization: Causes and Implications (Working Paper No. 97/42). Retrieved from International Monetary Fund website: https://www.imf.org/external/pubs/ft/wp/wp9742.pdf.
Rowthorn, R.E. & Wells, J.R. (1987). De-industrialization and foreign trade. Cambridge: Cambridge University Press.
Timmer, M.P., de Vries, G.J., & de Vries, K. (2014). Patterns of Structural Change in
Developing Countries. Groningen Growth and Development Center, Research Memorandum 149.
United Nations – Department of Economic and Social Affairs. (2016). Total Population –
32
Wolman, H.H., Wial, H. & Hill, E.D. (2015). Introduction to Focus Issue on
Deindustrialization, Manufacturing Job Loss, and Economic Development Policy. Economic