Patterns and Determinants of Structural Change:

(1)

University of Groningen and University of Göttingen

Patterns and Determinants of Structural Change:

Economic transformation in Egypt and Morocco since 1960 and an assessment

of underlying drivers across 39 countries between 1970 and 2005

(2)

Abstract

(3)

List of Figures

1 GDP per capita development in Egypt, Morocco, Korea, Thailand and Ghana between 1960 and 2008 . . . 26 2 Relative productivity of manufacturing and agriculture in Morocco between 1960

and 2012 . . . 29 3 Relative productivity of manufacturing and agriculture in Egypt between 1960

and 2012 . . . 30 4 Egyptian duality in 2006 . . . 30 5 Decomposition of labor productivity growth in Egypt between 1960 and 2006 . . 35 6 Decomposition of labor productivity growth in Morocco between 1960 and 2004 35 7 Relative agricultural productivity in dierent regions between 1975 and 2005 . . 39 8 Pull factors: partial correlation of relative productivity of agriculture and the

labor reallocation contribution . . . 41 9 Pull factors: partial correlation of manufacturing productivity growth and the

labor reallocation contribution . . . 41 10 Excess labor: partial correlation of the initial agricultural employment share and

the labor reallocation contribution . . . 42 11 Push factors: partial correlation of agricultural productivity growth and the

labor reallocation contribution . . . 42 12 Undervaluation: partial correlation of score on undervaluation index and the

labor reallocation contribution . . . 43 13 Natural resources: partial correlation of rents of natural resources in total GDP

and the labor reallocation contribution . . . 43 14 Labor reallocation costs: partial correlation of score on LAMRIG index and the

labor reallocation contribution . . . 44 15 Pull factors: partial correlation of relative manufacturing productivity and the

labor reallocation conribution . . . 72 16 Pull factors: partial correlation of relative trade services productivity and the

labor reallocation conribution . . . 72 17 Pull factors: partial correlation of productivity growth in trade services and the

(5)

List of Tables

1 Comparison of decomposition methods . . . 18

2 Egypt and Morocco: Sectors, variables and time period covered . . . 21

3 Full sample: countries covered between 1970 and 2005 . . . 23

4 Descriptive patterns of structural transformation in Egypt between 1960 and 2006 32 5 Descriptive patterns of structural transformation in Morocco between 1960 and 2004 . . . 33

6 Determinants of the magnitude of structural change: OLS estimation . . . 45

7 Determinants of the magnitude of structural change: dierent pull measures in OLS estimation . . . 47

8 Determinants of the magnitude of structural change after 1990: OLS estimation 48 9 ISIC Rev. 3.1 classication . . . 65

10 Morocco and Egypt: Sources . . . 67

11 Correlation table . . . 74

(6)

Glossary

ASD - African Sector Database EMP - Employment

GDP - Gross domestic product

GGDC 10SD - Groningen Growth and Development Center 10-Sector Database ISI policy - Import-substitution industrial policy

ISIC Rev.3.1 - International Standard Industrial Classication of All Economic Activities, Re-vision 3.1

LA - Latin America

LAMRIG - Labor market rigidity and reform index LFS - Labor force survey

LIML - Limited information maximum likelihood

lnGDPpc - Logarithmic gross domestic product per capita MENA - Middle East and North Africa

PPP - Power purchasing parity PWT - Penn World Table SSA - Sub-Sahara Africa TFP - Total factor productivity

UNOCD - United Nations Ocial Country Data UNYB - United Nations Statistical Yearbook VA - Value added

(7)

1 Introduction

Modern economic growth has gained momentum since 1820 and led to large income dierences across countries (see Maddison, 2007). One way to think about economic growth is the entry and exit of rms: older, less productive rms exit the market, while new and more productive rms enter the market (see e.g. Aghion and Howitt, 1992). Those achieve increasing market shares and resources are reallocated to those activities. On the sectoral level, this process has been modeled in two-sector models in which, along the development process, less productive sectors decline, while new activities in the modern sectors are created and resources are reallo-cated towards these sectors (see e.g. Lewis, 1954).

Recent ndings do in fact suggest that this sectoral reallocation is a major determinant in ex-plaining dierences of labor productivity growth between regions indicating why some regions may be less developed than others (see e.g. McMillan and Rodrik, 2011). The approach is to decompose labor productivity growth into a within and a between term, or interchangeably called reallocation term or structural change. The within term depicts the contribution of la-bor productivity growth within sectors to aggregate lala-bor productivity growth and the between term the contribution of the reallocation of labor.

The amount of studies investigating such sectoral shifts has increased in recent years due to the increasing availability of consistent cross-country data, such as the Groningen Growth and De-velopment Center 10-Sector Database (hereafter GGDC 10SD) and the African Sector Database from the University of Groningen (hereafter ASD). Data availability and quality, however, re-main crucial issues when assessing such processes over long time horizons.

Furthermore, the determinants behind such shifts are less well understood. The main approach taken to explore the determinants of sectoral shifts is by testing the empirical t of dual economy models. Within that literature, two main hypotheses evolved that aim at explaining structural change: the labor pull and the labor push hypothesis (see e.g. Alvarez-Cuadrado and Poschke, 2011). The former suggests that the modern sectors are pulling labor to these more productive sectors, whereas the latter suggests that productivity growth in the unproductive sector is the main driver of structural change. Further topics highlighted in the literature are labor reallo-cation costs, dependence on natural resources and international competitiveness (see Ciccone and Papaioannou, 2008; McMillan, Rodrik and Verduzco-Gallo, 2013).

(8)

that extent the GGDC 10SD and ASD, and hence enable further research. Both countries provide interesting cases, because they experienced varying industrial policies and add a new pattern of growth to the analysis. The region developed faster than the least developed coun-tries, mainly located in sub-Saharan Africa, but slower than those who gained high- and middle income status, the Asian and Latin American countries. Second, the analysis takes a next step in exploring the determinants of structural change. Motivated by the above hypotheses and recent ndings (see e.g. McMillan and Rodrik, 2011), it is among the rst to explore the eects in a regression framework. It is attempted to identify links that help explaining why labor moves to more productive sectors in some countries, while it does not in others.

(9)

To put the ndings of Egypt and Morocco in perspective, the second part of the analysis uti-lizes a larger dataset and analyzes the drivers of structural change. This analysis is based on a panel dataset covering seven 5-year periods between 1970 and 2005 and 39 countries from ve regions: Asia, Europe and North America, Latin America, the Middle East and North Africa and sub-Sahara Africa. The main variables are based on the GGDC 10SD, the ASD, the constructed data series for Egypt and Morocco and World Bank (2013).

The statistical analysis reveals four main points. First, across dierent specications, the ini-tial employment share in agriculture is positively associated with the magnitude of the between term: the larger the initial structural gap, the more labor is shifting. This indicates that the extent of excess labor is associated with structural change, as suggested in two-sector economy models. Second, robust determinants of structural change could not be found. There does not seem to be a universal mechanism between the tested variables and structural change across dierent time periods. Third, related to that, higher productivity of the pulling sectors does not imply structural change. That means that even if a country has promising sectors that are competing internationally and have growing productivity levels, structural change is not an automatic process. The same holds for agriculture based productivity growth. It thus needs active policies that promote structural change. And fourth, currency undervaluation seems to be such a policy. For the post-1990 periods, it is found that currency undervaluation policies are positively associated with structural change. Thus, such policies appear to be helpful in times of increasing fragmentation of the production process in which international competitiveness becomes more important. Furthermore, within the two resource-rich regions Latin America and sub-Saharan Africa, also natural resource dependence is negatively associated with structural change after 1990. This may be linked to increasing commodity prices which raises the detri-mental eects of a specialization in natural resources. Moreover, that suggests that growth based on a boom in commodity prices may not be sustainable, because it is not necessarily accompanied by structural transformation.

(10)

2 Literature review and theory

This section provides an overview on theoretical concepts related to two-sector models and main ndings in the eld. These concepts motivate the assessment of the patterns and the determinants of structural change.

2.1 Theoretical background

The thesis' theoretical antecedents lie in dual economy models with the basic idea of two or more non-summable sectors with diering production functions (e.g. Lewis, 1954), as opposed to the neoclassic approach in which all economic sectors can be aggregated (e.g. Solow, 1956). The neoclassical approach rests on the assumption of perfect markets and thus assumes contin-uous equilibrium growth which is fueled by factor accumulation and total factor productivity growth. The dual economy approach, however, acknowledges disequilibiria on factor markets and therefore adds the reallocation of resources to more productive sectors as sources of growth (see Chenery, Robsinson and Syrquin, 1986).

This approach can also be applied on the rm-level (for an overview, see Bartelsman and Doms, 2000). Similarly as in the analysis based on sectors, shifts of resources between unproductive and productive rms within sectors inuence the sector's aggregate productivity. Moreover, new and innovative rms enter the market, gain larger market shares and resources are reallo-cated towards those rms, while older rms exit the market. Therefore, the question that arises is whether it is actually useful to use sectoral boarders or whether there are simply productive and unproductive rms in the economy. Productivity dierences between sectors, however, suggest that there are at least on average more productive rms in certain sectors.

One of such models based on sectoral boarders is the two-sector model by Lewis (1954). It models how an economy transforms from a traditional one into an advanced one. It is assumed that there are two sectors: one is unproductive with excess labor and one is highly productive and absorbs excess labor.

Due to higher productivity levels and capital accumulation in the modern sector, wages in that sector raise and that attracts excess labor: the pull factor. In the same vein argue Harris and Todaro (1970) who present a model in which labor movements depend on the expected dierence of urban and agricultural wages. This process is also modeled in Nelson and Pack (1999) and more recently in Vollrath (2009).

(11)

output in the unproductive sector and thus, the extent of excess labor is associated with greater potential benets from reallocation. Female labor force participation and population growth may contribute positively to the extent of excess labor.

A prominent extension of this model is Ranis and Fei (1961). In this model, the speed and timing of reallocation depends on productivity dierences between the sectors and the extent of excess labor, as in the Lewis model (1954), but also on productivity growth in the unproductive sector: the push factor.

One way to think about the labor push hypothesis is a solution of the food problem, as identi-ed by Schultz (1953). This is also recently modeled in Gollin, Parente and Rogerson (2002), for instance. Their model assumes that households have a constant demand for agricultural products in order to satisfy their basic needs. Only once this demand is satised, goods from the modern sectors are consumed and produced. Thus, if this demand is not satised no labor will leave the agricultural sector, but as soon as that is the case, labor moves to the modern sector. This process is stimulated by productivity growth in the agricultural sector, because then less labor is needed in order to satisfy basic needs.

A second way to think about the role of the agricultural sector is presented in Badiane, Ulimwengu and Badibanga (2012) and is based on Johnston and Mellor (1961). The basic idea is that rising gross domestic product (hereafter GDP) per capita and population growth actually increases demand for agricultural products. If agricultural output growth is slower than demand growth, food prices will raise. This also impacts the modern sectors, as wages will accordingly raise as well. Thus, a more productive agricultural sector keeps food prices low and through that wages in the modern sectors.

A second theme that is linked to agricultural productivity growth is the convergence of sectoral productivity levels. In the long-run economies evolve eventually from duality to one-sector economies in the neo-classical sense (see Ranis, 2012).

Thus, the theoretical literature on the determinants of labor reallocation highlights two eects based on two-sector models. One is the idea that the modern sectors determine the movement of labor, the labor pull hypothesis and the other one is its rivalry hypothesis that it is the agricultural sector which induces shifts out of that sector, a hypothesis sometimes referred to as the labor push hypothesis (see e.g. Alvarez-Cuadrado and Poschke, 2011).

(12)

the costs of acquiring the skills to work in the modern sector or the time and bureaucratic costs of leaving one sector and entering a new one. More generally, if productive rms can quickly hire workers and they are quickly released from unproductive rms, the process of structural change goes naturally faster. As Ciccone and Papaioannou (2008) show for the manufacturing sector, if ring costs are high, rms rather tend to upgrade their capital goods instead of hiring more workers. However, excess labor in the agricultural sector may be less constrained by such formal costs of labor reallocation than labor in already more productive sectors. Therefore, such factors may play more important roles in later stages of the development, respectively are more important between more formalized sectors. Reallocation costs are also implicitly modeled in Nelson and Pack (1999) who argue that an institutional framework can provide incentives to catch opportunities in the modern sector.

(13)

of productivity and prices (see e.g. Baldwin, 2006). If the exchange rate is lower, the tradable sector increases its output share as international demand increases and that in turn, attracts labor in order to produce that output. Hence, undervaluation policies in particular might be benecial for structural change.

2.2 Findings and stylized facts on patterns of structural change

This section describes the main ndings in the literature on patterns of structural change. First, general trends have been documented concerning the economic structure and sectoral relative productivity. And second, the contribution to aggregate labor productivity growth of those changes in the economic structure have been quantied. At the end of the section I will issue a notice on methodological issues.

A rst stylized fact is that across countries, the economic structure has changed over time. Agricultural employment and output shares decrease while other sectors gain in importance. For large parts of the world, it has mainly been the manufacturing sector that absorbed large shares of the labor force (e.g. Kuznets, 1971), whereas the service sector also seems to take that role in some countries (on India, see Bosworth and Collins, 2008).

This changing structure has already been documented by Kuznets (1971) and Chenery, Robin-son and Syrquin (1986). Kuznets (1971) uses cross-sectional data for the 1960s and longitudinal data starting in 1900 for mainly currently developed European countries, their oshoots and Japan. Chenery, Robinson and Syrquin (1986) show similar descriptive results while consid-ering the period 1960 to 1980. Both analyses include Egypt and and their ndings on the aggregate sectoral shares actually conform to the ones in the analysis at hand.

(14)

semi-Those changes in employment shares and relative productivity levels aect aggregate labor productivity, because labor is dierently distributed across sectors with dierent productivity levels. This is assessed by decomposing labor productivity growth and has been mainly done to explain the post-war period since 1950. Such analyses typically nd that the most developed countries have beneted from reallocation of labor in early phases of the development, but are now mainly dependent on productivity growth within sectors (see e.g. Ocampo, Rada and Taylor, 2009; McMillan and Rodrik, 2011). Furthermore, Latin American countries have ex-perienced positive contributions of reallocation in earlier time periods and especially in phases of import-substitution policies. However, this shift is currently only marginally contributing to labor productivity growth, if labor is not even moving to less productive sectors on aver-age. Furthermore, Asian countries have been able to sustain large contributions of structural change over time (on Latin America and Asia, see e.g. Escaith, 2007; Timmer and De Vries, 2009; McMillan and Rodrik, 2011). African countries, on the other hand, were able to shift large shares of the labor force to more productive sectors in early time periods and experienced decreasing contributions thereafter. However, this positive development regained momentum at the latest since the 2000s (see De Vries, Timmer and De Vries, 2013; McMillan, Rodrik and Verduzco-Gallo, 2013).

(15)

As the analysis of Roncolato and Kucera (2013) shows, one obstacle to the eld of research is its data intensity. More detailed data enables a more detailed analyses. Moreover, the decom-position methods itself range from relatively simple ones to ones that intend to capture also rm entry and rm exit (see Iskasson, 2009; De Vries, Timmer and De Vries, 2013 and further discussion in section 3.1).

De Vries, Timmer and De Vries (2013) show the crucial role of accurate data and the inuence of dierent decomposition methods. The authors repeat the same decomposition methodology, as is applied in McMillan and Rodrik (2011), but with their newly introduced ASD and have dierent results due to dierences in the data. One mentioned problem is female labor force participation which also needs attention when considering data for Egypt and Morocco and is discussed in the appendix. Furthermore, the authors also show that using dierent decompo-sition methods also alters the results.

Another point of discussion is the measurement of productivity, which can be measured as total factor productivity (hereafter TFP) or labor productivity: value added per worker. This is further discussed in section 6. The measure applied here is labor productivity.

2.3 Findings on determinants of structural change

The mentioned studies show that some economies gain from labor reallocation while others do not, put dierently, some had positive structural change and others not. Gaining from labor reallocation means that a larger share of the labor force shifts from unproductive to productive sectors or that productivity dierences between sectors are larger. Therefore, economies with-out a dual economy structure will not gain from reallocation and rather follow the neo-classical model. If a dual structure exists, then the reallocation of labor from less productive sectors to more productive ones is an easy way to increase welfare, because the long way of factor accumulation, such as human capital, does not need to be taken (see also Rodrik, 2013a). The crucial question is thus how and why some countries have witnessed positive structural change while others have not. There are two streams of empirical work that aim at identifying the determinants of structural change: growth modeling and econometric analysis. Both streams do provide relatively little evidence to identify determinants of structural change.

(16)

con-to labor reallocation by assuming a constant absolute level of employment in the agricultural sector.

The ndings are not uniform. Concerning push factors, Gollin, Parente and Rogerson (2002) highlight its importance in driving structural change. However, Gollin, Parente and Rogerson (2007) also nd that push factors are important in the short-run, whereas the pulling sectors gain in importance in the long-run for development. Somewhat contrarily, Alvarez-Cuadrado and Poschke (2012) nd for 12 industrialized countries that pull factors matter more in early stages of the development, and push factors in later ones. Matsuyama (1992), on the other hand, nds that agricultural productivity growth induces structural change in closed economies, but that it has the opposite eect in small, open economies. Concerning labor reallocation costs, Hayashi and Prescott (2002) nd furthermore that labor reallocation costs, conceptualized as informal barriers to labor allocation, did slow down the progress of structural change in pre-war Japan.

The second approach is to identify determinants in econometric analyses, which has only started to evolve. In particular, the approach is to identify determinants of the magnitude of the con-tribution of structural change to labor productivity growth.

The rst guidance through the data is oered by McMillan and Rodrik (2011). In a cross-sectional study, the authors nd that natural resource dependence and high labor reallocation costs are both associated with negative structural change, whereas countries that apply cur-rency undervaluation policies are associated with positive structural change. Moreover, the authors nd that there is conditional convergence: controlling for natural resource dependence turns their variable that depicts the initial structural gap signicant. The analysis at hand follows this approach and analyzes structural change in a regression framework.

2.4 Hypotheses

Based on the discussion above, several hypotheses can be derived.

First, in dual economy models, excess labor is labor with a very low marginal productivity and thus it can be reallocated without signicantly reducing output in the traditional sector. Therefore, the more excess labor is available, or the larger the initial structural gap, the larger is the potential gain from reallocation (e.g. Lewis, 1954). Hence, the extent of excess labor is expected to be positively linked to structural change.

(17)

and therefore, higher productivity in the modern sectors is expected to be positively linked to structural change, the pull factor.

Third, extended dual economy models highlight the inter-connectivity between the modern and the agricultural sector (e.g. Ranis and Fei, 1961). Demand for modern goods only occurs once demand for basic goods, produced in the agricultural sector, is satised (e.g. Gollin, Parente and Rogerson, 2002). Higher demand for modern goods is linked to its production and thus, the hiring of additional workers in these sectors. Hence, agricultural productivity is expected to be positively linked to structural change, the push factor.

Fourth, barriers to the free allocation of labor will slow down the process of structural change (e.g. Hayashani and Prescott, 2002). To give an example, if labor reallocation costs are higher than the expected wage dierence, labor is not pulled towards the modern sector. Hence, higher labor reallocation costs are expected to be negatively linked to structural change.

Fifth, specialization on natural resources my have detrimental eects on other tradable sectors and this sector cannot absorb a lot of labor itself (e.g. Krugman, 1987). Therefore, a special-ization in natural resources is expected to be negatively linked to structural change.

Sixth, on the other hand, currency undervaluation reduces prices of tradable goods on the global market and reduces competition from abroad on the domestic market. Demand and production may thus rise for domestic tradable goods. Therefore, currency undervaluation is expected to be positively linked to structural change.

3 Methodology and Data

This section describes the methods and the data on which the analysis in section 4 and 5 is based. The rst two section describe the decomposition method and the regression model and the latter two sections describe the main data sources. 3.3 focuses on the GGDC 10SD and ASD and highlights the principles which were closely followed when constructing the data series for Egypt and Morocco. 3.4 describes the additional data sources and the construction of the variables used in the regression analysis.

3.1 Decomposition method

(18)

change or reallocation term). Intuitively, the process is as follows. Aggregate productivity in-creases when employment shares stay constant, but sectors become more productive. However, it also increases when sectors' productivity remains constant, but a larger employment share is engaged in sectors with higher productivity levels. The calculated between term increases when larger shares of the labor force shift from unproductive to productive sectors, and when productivity dierences between sectors are large, because it is weighted by productivity levels. Hence, it also takes into account whether these shift happen in a dual economy structure and thus whether they matter for aggregate productivity.

The canonical decomposition method which originates in Fabricant (1942) is the basis for most recent publications in the eld. One decomposition method is the following:

4 P =X i (P_it− P_it−1)Si+ X i (S_it− S_it−1)Pi (1)

where P is labor productivity, S employment shares, i indicates sectors and t and t-1 are two points in time and the bar indicates the arithmetic average over the time period. Labor productivity is dened as P = V A

EM P where V A is value added and EMP absolute employment. The rst term in equation (1) represents the within term and the second one the between term. In this decomposition both the within and between term are weighted with the average between start and end points of employment shares, respectively sectoral productivity levels. This decomposition method is used for instance in Escaith (2007). McMillan and Rodrik (2011), however, prefer the following one:

4 P = X i (P_it− Pt−1 i )S t−1 i + X i (S_it− St−1 i )P t i. (2)

Thus, McMillan and Rodrik (2011) and also McMillan, Rodrik and Verdusco-Gallo (2013) weight the within contribution with start year employment shares and the between term with end year productivity levels. As noted in Haltiwanger (2000), this method makes the within contribution look larger, at the expense of the structural change contribution. Alternatively, also the opposite can be applied:

4 P = X

i

(P_it− P_it−1)S_it+X i

(19)

Furthermore, there are more complex decomposition methods. As suggested in Haltiwanger (1997) and applied in De Vries, Timmer and De Vries (2013) in a more recent contribution, it is possible to take into account whether workers actually move to sectors with positive produc-tivity growth or to sectors with negative producproduc-tivity growth. This is done by an interaction eect of changes in productivity and employment shares and this is positive in the former and negative in the latter case. Yet, another renement is to take the actual net entry of rms into account by adding two additional terms that capture entry and exit of rms (see Foster, Haltiwanger and Krizhan, 2001).

The focus of the analysis at hand is the general movement of labor from low productive sectors to more productive sectors and less on the eects on the rm and the plant level. Therefore, equation (1) is applied here, although it needs to be acknowledged that more sophisticated decomposition methods rene the measurement and also impact the results. The question is rather whether these eects are big enough to inuence the conclusions about the within and between term (see Isaksson, 2009).

(20)

Table 1: Comparison of decomposition methods

Morocco Egypt

Period Within Between Interaction Total Period Within Between Interaction Total 1960-1971 1.5% 0.8% -0.1% 2.2% 1960-1966 3.0% 0.3% -0.4% 3.0% 1.4% 0.8% 2.2% 2.8% 0.1% 3.0% 1971-1982 -1.2% 1.8% -0.5% 0.1% 1966-1976 2.8% 0.3% 0.4% 3.5% -1.4% 1.5% 0.1% 3.0% 0.5% 3.5% 1982-1994 1.2% 0.5% 0.1% 1.7% 1976-1986 5.3% 0.3% -0.1% 5.6% 1.2% 0.5% 1.7% 5.3% 0.3% 5.6% 1994-2004 0.2% 0.3% -0.1% 0.5% 1986-1996 1.1% 0.6% -0.1% 1.6% 0.2% 0.3% 0.5% 1.0% 0.6% 1.6% 1996-2006 3.3% -0.4% -1.0% 1.9% 2.8% -0.9% 1.9%

Source: own calculations based on equation (1) and De Vries, Timmer and De Vries (2013) based on data series described in main body and appendix.

Note: Sums may not exactly add up die to rounding error.

3.2 Regression model

This section describes the empirical specication that is applied in order test the hypotheses outlined in section 2.4. The dataset, which is further described in the next sections, is a panel dataset with 39 countries and seven time periods.

The basic specication is a pooled ordinary least squared (hereafter OLS) model with country and time dummies to account for unobserved heterogeneity (least squares dummy variable regression): yit= N X j=1 αjCji+ T X s=2 γsPst+ βXit+ eit (4)

(21)

Moreover, heterosceasticity was detected and therefore heteroscedasticity robust standard errors were applied. Applying the Wooldridge (2002) test for autocorrelation in panel data, autocor-relation was not detected. Econometric issues that exceed the baseline model are discussed in 5.3.

3.3 GGDC 10-Sector Database and African Sector Database

The GGDC 10SD and ASD are the main data sources used in the analysis. This section high-lights the main features: what is included in the databases, how are they constructed, why are they needed and what are they used for. Detailed descriptions of the characteristics of the databases, however, can be found in De Vries, Timmer and De Vries (2013) concerning the ASD and in Timmer and De Vries (2007, 2009) concerning the GGDC 10SD. These databases are also the basis for the construction of the Egyptian and Moroccan data series.

The databases include series for value added (hereafter VA) and employment and cover broadly the period 1960 to 2010 and together 39 countries from Asia, Europe and North America, Latin America and sub-Saharan African. VA series are available in current and in constant prices in local currencies and employment levels are preferably conceptualized as persons engaged in economic activity above 15 years. In both databases economic activity is disaggregated into nine to ten sectors based on the International Standard Industrial Classication of All Eco-nomic Activities, Revision 3.1 (hereafter ISIC Rev. 3.1). Table 9 in the appendix describes the ten sector in more detail and summarizes the used names and abbreviations for the sectors in the main body.

Depending on the database and the countries, the level of detail and renement varies. For countries included in the ASD also values of female and male employment is available and 2005 VA values in international $ in Purchasing Power Parities (hereafter PPP). For the most developed countries covered in the GGDC 10SD also a series of hours worked is available which renes the measurement of employment.

(22)

To construct the VA series, ocial sources are obtained from releases of National Statistical Institutes or the United Nations database that collects ocial country data published online (UN, 2014b) and in Statistical Yearbooks (United Nations Statistical Yearbooks). These data series have several gaps, breaks and revisions which leads to inter-temporal inconsistency as older classications cover dierent activities than newer ones. Therefore, the series are linked via extrapolation. The latest ocial source and thus the latest revision serves as the benchmark classication. From that year backwards, the series is extrapolated by making use of the growth rates in the other ocial sources. Only in cases where no ocial sources have overlapping years, it must be relied on estimates. This assures inter-temporal consistency within the VA series. Concerning employment data, it is proceeded similarly. The ocial sources on sectoral employ-ment levels are population censuses which are mostly held decade-wise. These censuses serve as benchmark levels and the periods in-between are interpolated by making use of surveys or estimates, such as from Labor Force Surveys (hereafter LFS) mainly obtained from the Inter-national Labor Organization (ILO, 2014a; ILO, 2014b), establishment surveys, data from the Food and Agricultural Organization of the Untied Nations (FAOSTAT database; FAO, 2014) or by assuming constant labor productivity growth. However, since only the growth rates of survey data is used, inter-temporal consistency is assured.

Internal consistency is assured by basing the sectoral VA and employment classications on ISIC Rev. 3.1 and assuring that the same activities are covered, and international consistency is assured by sticking to the same concepts of VA and employment across countries covered in the databases.

There are other databases that provide sectoral VA and employment series (e.g. World Bank, 2013), but there are several reasons to prefer this dataset over other ones. First of all, those series often only cover periods from 1980 or 1990 onward and are therefore not suited for his-torical long-term analyses.

(23)

result stems from an urban sample in the LFS. Business surveys, on the other hand, often cover not all activities, because threshold levels in terms of employees or annual turnover are applied. The population census, however, covers all kinds of economic activity. Therefore, there are no threshold levels, no or less sample bias and it also covers informal activities ,which is especially important in developing countries.

Second, intertemporal consistency is also not assured in other data series. The concepts and samples of employment series obtained from LFS and business surveys do not only collide with sectoral VA data, but are also not consistent over time. Therefore, there are sudden jumps in the data when concepts change (see Timmer and De Vries, 2009). The same holds for VA series. As indicated, the UN (2014b) data has several gaps and revisions and thus it is needed to smooth these series for intertemporal comparison. Naturally, the concepts applied in LFS and business surveys do also vary across countries in dierent time periods. Therefore, also international consistency is not assured.

Another great advantage of the provided series is the exclusion of the VA shares of owner occu-pied dwellings from the business services sector. Owner occuoccu-pied dwellings appear in VA data, but do not have an employment equivalent. That renes the calculation of labor productivity of the business services sector. For instance, labor productivity in the business services sec-tor in the Egyptian data series is on average about 22% higher, if that is not taken into account. Conclusively, the databases provide consistent data that cover all kinds of employment, includ-ing informal employment, and are needed for analyses of labor productivity, here dened as value added per worker. Based on the above guidelines, the data series for Egypt and Morocco were constructed. This is described in greater detail in the appendix. Moreover, the appendix includes a discussion relating to potential problems of the obtained ocial data due to female labor force coverage and a territorial change in Morocco due to the annexation of Western Sahara in the 1970s. Table 2 shows the variables, sectors and time period covered in the series of Egypt and Morocco that were constructed as part of this thesis.

Table 2: Egypt and Morocco: Sectors, variables and time period covered

Egypt Morocco

Sectors AtB, C, D, E, F, G+H, I, J+K, LtP, dwellings AtB, C, D, E, F, G+H, I, J+K, LtO, P, dwellings Variables EMP as persons engaged > 15; VA in current and constant prices (local currency)

Time Period 1960-2012

(24)

3.4 Measures for potential determinants in regression analysis

This section describes the sources of the variables used in the regression analysis on the deter-minants of structural change in section 5. Potential deterdeter-minants of structural change identied in the literature and highlighted in section 2.4 are the extent of excess labor, pull factors, push factors, labor reallocation costs, natural resource dependence and currency undervaluation. Furthermore, other measures of potential determinants are highlighted that were tested in the econometric analysis.

A general diculty in operationalizing the stated hypotheses is the availabilty of measures that cover the broad set of countries and the long time horizon of the analysis. Therefore, only rather broad measures could be used (see below), as otherwise the least developed countries and early time periods could not have been covered. That would have biased the sample towards the remaining most developed countries.

Due to data availability, not all countries and time periods in the GGDC 10SD and ASD could be covered. Data for Taiwan, for instance, is lacking on a range of indicators and this coun-try is therefore not included in the sample. Also, West Germany is excluded since the series naturally ends in 1991. Similarly, earlier time periods for some countries had to be excluded, because some of the key variables were not collected. One of those key variables, the measure for natural resource dependence is not available before 1970 and therefore, time periods for all countries start in 1970. Hence, the whole sample covers 39 countries and seven 5-year periods between 1970 and 200. Table 3 provides an overview of the countries covered in the sample and from which year onward complete data was available. Furthermore, two observations were dropped from the sample, which is discussed in 5.3. A correlation table and summary statistics are provided in the appendix.

(25)

Table 3: Full sample: countries covered between 1970 and 2005

Developed LA SSA ASIA MENA

Denmark Argentina Botswana Hong Kong (1975) Egypt France Bolivia Ethiopia (1980) India Morocco

Italy Brazil Ghana Indonesia (1975)

Netherlands Chili Kenya Japan

Spain Colombia Malawi Malaysia (1975) Sweden Costa Rica Mauritius Philipines (1975) United Kingdom Mexico Nigeria Singapore

United States Peru Senegal South Korea Venezuela South Africa Thailand

Tanzania (1985) Zambia

Note: Deviating start years are indicated in parentheses. LA is Latin America, SSA is sub-Saharan Africa, ASIA is Asia and MENA is Middle East and North Africa.

agricultural sector, as modeled in two-sector models (e.g. Lewis, 1954), then every movement out of agriculture increases average productivity within that sector. Thus, agricultural pro-ductivity growth may not only push labor out of agriculture and thus lead to a larger between term, but structural change itself (the movement of labor out of agriculture) may link back to higher average productivity in agriculture and thus to agricultural labor productivity growth. Endogeneity problems are discussed in 5.3. However, this variable did in fact not pose the main problems. The impact of dierences in marginal and average productivity on the decomposition method is discussed in section 6.

Labor pull factors are operationalized in several ways. The preferred measure is annual manu-facturing labor productivity growth, MProdG. The manumanu-facturing sector is chosen, because it was largely the manufacturing sector that attracted most of the labor that was released from the agricultural sector in the past. Moreover, the productivity growth rate is chosen, because that depicts a dynamic relationship: do changes in productivity pull labor? The same vari-able was constructed for the trade services sector, GHProdG, because this sector also absorbed large shares of the labor force. However, for the same reasons as above, these two variables may also be endogenous. If the between term is large, then productivity growth in the highly productive sectors may tend to be smaller, if the marginal productivity of the additional labor in these sectors is smaller than average productivity. Three further measures for pull factors were constructed that relate more closely to the Lewis (1954) model. Lewis (1954) argues that wage dierences and therefore dierences of productivity levels pull labor to the modern sec-tors. This is measured as the relative productivity of the pulling sectors and of the agricultural sector. This is calculated as follows. AgRelP rod = LPAg

(26)

productivity of the other two sectors is calculated accordingly, ManuRelProd and GHRelProd. It is expected that the lower the agricultural relative productivity, the greater the gain from reallocation. Accordingly, the higher the manufacturing and trade services relative productiv-ity, the stronger the pull and thus the greater the gain from reallocation.

Furthermore, the initial employment share in agriculture, EMPshare, was obtained. The inclu-sion of this variable aims at capturing the eect of the initial structural gap. The more labor is employed in the unproductive sector, the larger the extent of excess labor and the larger the potential gain from reallocation. This relationship has been modeled in classic dual economy models. As the agricultural sector is the least productive sector across almost all countries and time periods, this sector relates to the unproductive sectors in such models. In chapter 5.3, it is further discussed why GDP per capita was not included as an alternative measure.

All data needed to construct the above measures is found in the GGDC 10SD, ASD and in the constructed data series for Egypt and Morocco.

The main measure for labor reallocation costs is labor market regulations. A newly developed index is the LAMRIG (Labor Market Rigidity and Reform Index) index which is constructed in the framework of the ILO, is available in ve-year periods between 1960 and 2005 and covers up to 140 countries (see Campos and Nugent, 2012). However, this index is a de jure measure of labor regulations and thus, it might be expected that these regulations play more important roles in more developed countries and in later phases of the development. First, more developed countries may be expected to have more formalized laws. And second, the movement out of agriculture might be less restricted by labor regulation laws, as employment in that sector is less formalized. Nevertheless, it is the only measure to my knowledge that consistently cap-tures labor market characteristics for a large cross-country panel covering the post-war period, as needed for the analysis at hand.

(27)

in the Penn World Table 8.0 (Feenstra, Inklaar and Timmer, 2013), GDP per capita in current US $ (World Bank, 2013), population growth rates (World Bank, 2013) and access to credit as private credit in total total GDP (World Bank, 2013) were obtained and tested.

4 Structural change in Egypt and Morocco

This section discusses the patterns of economic growth in Egypt and Morocco and detects sources of labor productivity growth. The rst section provides an overview on Egypt and Morocco and the second section aims at explaining that pattern and discusses implications.

4.1 Economic and political development since 1960

In 1960, Morocco and Egypt had similar income levels as Korea, Thailand and Ghana. Fig-ure 1, however, shows that growth paths diered substantially. Korea's growth spurt began already during the 1960s and Thailand's towards the 1980s. Ghana was on a much less promis-ing growth path with long periods of recession, but is now recoverpromis-ing. This is an experience comparable to the less successful sub-Saharan African countries. Egypt and Morocco, on the other hand, developed relatively moderately across the whole time period with slowly but con-stantly increasing GDP per capita. Especially Morocco developed very gradually. However, several growth phases can be identied. GDP per capita evolves relatively constant until the early 1990s with a phase of slightly higher growth rates in the 1970s (see Bolt and Van Zanden, 2013). Between the early 1990s and early 2000s, however, the Moroccan economy grew very little. Egypt, on the other hand, had a less constant development and more concrete phases of growth and stagnation. It saw constant, but relatively small growth rates until the mid-1970s, followed by a growth spurt until the mid-1980s in which Egypt caught up with Morocco. There-after came again a phase of stagnation which re-ignited in the mid-1990s. The graph further highlights that growth was again stronger since the 2000s in Egypt and Morocco, as also in other African economies. This growth ignition coincided with increasing commodity prices and one might argue that therefore, this is not a sustainable growth path (see Arbache and Page, 2009).

(28)

Figure 1: GDP per capita development in Egypt, Morocco, Korea, Thailand and Ghana between 1960 and 2008 0 500 1,000 1,500 2,000 2,500 3,000 3,500 4,000 1960 1963 1966 1969 1972 1975 1978 1981 1984 1987 1990 1993 1996 1999 2002 2005 2008 G DP p er ca p it a (19 90 In t. G K $) Year EGY MOR KOR THA GHA

Source: retrieved from Bolt and Van Zanden (2013)

Note: Int.GK$ is the Geary-Khamis dollar. EGY is Egypt, MOR is Morocco, KOR is Korea, THA is Thailand, GHA is Ghana.

Both countries took part in several military conicts (see Uppsala Conict Data Program, 2014). Morocco had a military confrontation at the boarder to Algeria in 1963, the so called Sand war. In 1967, Egypt tool part in the Six-Days War followed by three years of military conict at the boarder to Israel. Three years later, Egypt and Morocco, both were involved in a war against Israel, the Arab-Israeli war. Throughout the 1970s, the most important conict in Moroccan modern history came up: the annexation of Western Saharan. Since the mid-1970s, Western Sahara is de facto annexed by Morocco. In 1977, Egypt had a short military boarder conict with Libya, but took part in the more inuential First Gulf war in 1990.

(29)

In terms of industrial policy, the two countries experienced several dierent approaches in order to generate growth. Both countries had early phases that can be characterized as ISI policies and later ones that were more closely linked to WC policy advises. Moroccan industrial policy can be split in two main phases with two sub-periods. Harabi (2008) characterizes the period between 1960 and 1982 as a phase of ISI policies which were even more active in the 1970s. The second period, 1983 to 2005, is judged as a more liberalized phase with a second liberalization wave in the early 1990s.

The literature does not divide the Egyptian industrial policy in such distinct phases, but rather indicates gradual movements from ISI policies to WC policies (see e.g. Said, Chang and Sakr, 1995; Galal, 2008; Benner, 2013). However, changes may be identied in 1974 and in the early 1990s.

The periods that are analyzed in the next section are chosen on the basis of distinct growth phases, periods of diering industrial policies and the properties of the data series. Census years provide the most reliable employment data (see appendix for details).

Concerning Egypt, four growth phases were identied: 1960 to 1970, 1970 to mid-1980, mid-1980 to mid-1990 and since the mid-1990s. Policy changes were not that distinctly dierentiated, but one policy change is observed in the mid-1970s and one in the early 1990s. These growth phases and policy changes t the benchmark years of population censuses rela-tively well. They were carried out in 1960, 1966, 1976, 1986, 1996 and 2006.

Concerning Morocco, growth phases did not vary that substantially. However, one stagnant period is identied in the 1990s. Moreover, the analysis of labor productivity growth does in fact reveal that the chosen periods do dier in that concern. On the other hand, policy changes were relatively clearly distinguished: 1960 to early 1970s, early 1970s to 1982, 1982 to early 1990s and since then. The policy changes do t the population censuses quite well: 1960, 1971, 1982, 1994 and 2004.

4.2 Labor productivity and labor reallocation

(30)

4.2.1 Labor reallocation, productivity convergence and duality

Table 4 and 5 contain information on value added shares, on employment shares and on rela-tive productivity: sectoral productivity level to weighted average economy-wide productivity. Table 4 provides the data for Egypt and table 5 for Morocco.

Table 4 and 5 show that the agricultural sector is decreasing in importance in terms of value added and in terms of employment and thus a transformation of the economies took place. Especially in Morocco, the speed of transformation is not uniform across periods. During the rst 22 years, the agricultural employment share decreased by almost 23 % points and its out-put share by about 22 % points, whereas in the following 22 years the agricultural employment share only decreased by about 8 % points and its output share by about 2 % points. In contrast, Egypt's development is more constant: the agricultural employment share decreased by about 5 to 7% points every ten years throughout all time periods.

Both countries experienced stalled industrialization. In Morocco, the employment share of the manufacturing sector increased by 6.3% points until 1982, peaked at 15.4% and dropped again by 1.1% until 2004. Value added shares followed the same pattern. In Egypt, the employment share of the manufacturing sector increased by 4.7% points until 1976, peaked at 14.3%, but decreased by 2% points until 2006. Value added shares, however, continued to increase. To put that in international perspective, nowadays more developed countries have reached that peak with larger shares of employment and at higher income levels. For instance, West Germany peaked at a manufacturing employment share of about 35% and Korea of about 28%, and both were already much further developed in terms of income levels than Egypt and Morocco around 1980 (see Rodrik, 2014).

Trade services played an important role in terms of labor absorption in both countries, espe-cially since manufacturing declined. The employment share of trade services increased by 3.5% points since 1982 in Morocco and by 3.9% points since 1976 in Egypt. Other sectors that cap-ture large employment shares are the construction sector and non-market services. The latter is especially large in Egypt, capturing 25.4% in 2006.

(31)

productivity. However, considering the manufacturing sector, two contrary developments can be observed. In Morocco, relative productivity of the manufacturing sector converges towards the economy-wide average, and this trend can also be observed for most other sectors. This development only seems to be dierent in mining, utilities and transport services. However, as they only capture 5% of total employment in 2004, a trend of overall productivity convergence seems to be the case. In Egypt, on the other hand, relative productivity does not seem to con-verge. Productivity levels in the manufacturing sector actually depart from the economy-wide average since the mid-1970s. This non-convergence of productivity levels can be observed in Egypt for all sectors, except the business and trade services sectors.

Figure 2: Relative productivity of manufacturing and agriculture in Morocco between 1960 and 2012 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2 19 60 19 63 19 66 19 69 19 72 19 75 19 78 19 81 19 84 19 87 19 90 19 93 19 96 19 99 20 02 20 05 20 08 20 11 R e lat ive p ro d u ct ivi ty Year D AtB

Source: own calculations based on data series described in main body and appendix.

Note: D refers to manufacturing and AtB refers to Agriculture. For sector names, abbreviations and classications, see table 9 in the appendix.

(32)

Figure 3: Relative productivity of manufacturing and agriculture in Egypt between 1960 and 2012 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 19 60 19 63 19 66 19 69 19 72 19 75 19 78 19 81 19 84 19 87 19 90 19 93 19 96 19 99 20 02 20 05 20 08 20 11 R el ati ve Pr o d u cti vi ty Year D AtB

Source: own calculations based on data series described in main body and appendix.

Note: D refers to manufacturing and AtB refers to Agriculture. For sector names, abbreviations and classications, see table 9 in the appendix.

Figure 4: Egyptian duality in 2006

0 0.5 1 1.5 2 2.5 3 Re lati ve Pr o d u cti vi ty i n 2006

Employment share in percent

F LMNOP AtB GH E I D JK C - 76.54 100 61.8 35.1

Source: own calculations based on data series described in the main body and appendix.

Note: Relative productivity of sector C exceeds the scale, but reaches a relative productivity of 76.54. For sector names, abbreviations and classications, see table 9 in the appendix.

(33)

The two most productive sectors in Egypt, business services and mining, only make up for 3.7% of the labor force but on the other hand capture large shares of value added: 21.6% (see ta-ble 4). The strong reliance on natural resources might cause the Dutch Disease and thus reduce the manufacturing capacity, whereas the mining sector itself does not absorb labor due to its capital intensity. In Morocco, the most productive sector is the business services sector. This sector has quite low shares of employment, but makes up for quite a large share of value added, only 1% point less than trade services. However, this sector is relatively human capital-intense and may therefore not be suited for large labor absorption (see Rodrik, 2014). In accordance to this nding, the OECD (2011) reports that skill mismatch is one of the main obstacles to employment creation in Morocco. Thus, the non-convergence indicates that the dual economy structure is more apparent in Egypt than in Morocco and the large share of the labor force in unproductive sectors suggests that Egypt may have a greater potential to gain from reallocation than Morocco. However, also in Morocco still 35% of the labor force is employed in agriculture and that sector is still distant from average productivity levels.

(34)

Table 4: Descriptive patterns of structural transformation in Egypt between 1960 and 2006

Value Added Employment Relative Productivity

1960 1976 1986 1996 2006 1960 1976 1986 1996 2006 1960 1976 1986 1996 2006 Agriculture 44,5% 30,6% 19,2% 16,8% 14,9% 55,0% 45,0% 38,9% 32,0% 26,7% 0,8 0,7 0,5 0,5 0,6 Industry 22,9% 30,9% 39,9% 38,5% 38,2% 12,6% 19,8% 22,1% 23,1% 23,5% 1,8 1,6 1,8 1,7 1,6 Mining 5,6% 13,6% 20,3% 16,1% 13,8% 0,3% 0,4% 0,4% 0,4% 0,2% 18,5 38,2 53,4 41,8 76,5 Manufacturing 16,1% 14,0% 14,6% 17,1% 18,3% 9,6% 14,3% 13,3% 13,2% 12,3% 1,7 1,0 1,1 1,3 1,5 Utilities 0,2% 0,7% 1,1% 1,2% 1,7% 0,5% 0,7% 0,9% 1,1% 1,4% 0,3 1,0 1,2 1,2 1,2 Construction 1,0% 2,6% 3,9% 4,1% 4,4% 2,2% 4,5% 7,5% 8,4% 9,7% 0,5 0,6 0,5 0,5 0,5 Services 32,6% 38,5% 40,8% 44,7% 46,9% 32,4% 35,2% 39,0% 45,0% 49,8% 1,0 1,1 1,0 1,0 0,9 Market services 21,8% 24,5% 28,2% 30,5% 33,5% 13,7% 15,2% 15,5% 19,9% 24,3% 1,6 1,6 1,8 1,5 1,4 Business services 3,6% 4,5% 5,3% 7,7% 7,8% 0,9% 1,0% 2,1% 3,2% 3,5% 4,0 4,7 2,5 2,4 2,2 Trade services 14,5% 13,3% 14,1% 13,8% 14,9% 9,0% 9,0% 7,6% 10,6% 12,9% 1,6 1,5 1,9 1,3 1,2 Transport services 3,6% 6,7% 8,8% 9,0% 10,8% 3,8% 5,2% 5,8% 6,1% 7,9% 1,0 1,3 1,5 1,5 1,4 Non-market services 10,8% 14,0% 12,6% 14,2% 13,4% 18,8% 20,0% 23,4% 25,1% 25,4% 0,6 0,7 0,5 0,6 0,5

Note: Relative productivity is the sectors' productivity to the weighted economy-wide average. For sector names, abbreviations and classications, see table 9 in the appendix.

(35)

Table 5: Descriptive patterns of structural transformation in Morocco between 1960 and 2004

Value Added Employment Relative Productivity

1960 1971 1982 1994 2004 1960 1971 1982 1994 2004 1960 1971 1982 1994 2004 Agriculture 31,0% 30,2% 19,1% 20,7% 17,0% 63,6% 55,7% 40,9% 37,6% 34,2% 0,5 0,5 0,5 0,6 0,5 Industry 26,6% 29,3% 31,7% 26,0% 27,4% 13,4% 18,1% 25,6% 24,7% 25,0% 2,0 1,6 1,2 1,0 1,1 Mining 4,1% 3,2% 2,6% 1,7% 1,8% 1,5% 1,7% 1,2% 0,8% 0,5% 2,7 1,8 2,1 2,1 3,8 Manufacturing 15,8% 15,8% 17,9% 16,9% 16,3% 9,1% 10,8% 15,4% 15,2% 14,3% 1,7 1,5 1,2 1,1 1,1 Utilities 1,1% 1,6% 1,8% 2,0% 2,9% 0,3% 0,3% 0,4% 0,5% 0,5% 3,6 4,7 4,1 4,1 6,4 Construction 5,6% 8,7% 9,4% 5,3% 6,4% 2,4% 5,3% 8,5% 8,2% 9,8% 2,3 1,6 1,1 0,6 0,7 Services 42,4% 40,5% 49,1% 53,3% 55,6% 23,0% 26,2% 33,5% 37,6% 40,7% 1,8 1,5 1,5 1,4 1,4 Market services 25,7% 24,3% 27,9% 32,0% 34,4% 11,9% 12,5% 16,5% 19,6% 22,3% 2,2 2,0 1,7 1,6 1,5 Business services 6,6% 6,5% 7,8% 12,9% 13,2% 0,5% 0,5% 0,8% 1,3% 1,5% 14,0 12,4 9,5 10,1 8,5 Trade services 16,1% 15,1% 16,4% 15,0% 14,3% 8,7% 8,8% 12,9% 14,9% 16,4% 1,8 1,7 1,3 1,0 0,9 Transport services 3,1% 2,8% 3,8% 4,0% 6,9% 2,7% 3,1% 2,8% 3,4% 4,4% 1,1 0,9 1,4 1,2 1,6 Non-market services 16,7% 16,1% 21,2% 21,3% 21,2% 11,1% 13,7% 17,1% 18,1% 18,5% 1,5 1,2 1,2 1,2 1,1 Gov. services 15,1% 14,7% 19,3% 19,4% 19,5% 9,5% 10,8% 15,1% 16,0% 16,6% 1,6 1,4 1,3 1,2 1,2 Pers. services 1,5% 1,5% 1,9% 1,9% 1,7% 1,6% 2,9% 2,0% 2,1% 1,9% 1,0 0,5 0,9 0,9 0,9

Note: Relative productivity is the sectors' productivity to the weighted economy-wide average. For sector names, abbreviations and classications, see table 9 in the appendix.

(36)

4.2.2 Decomposition, industrial policies and premature industrialization

The following paragraphs regard the formal decomposition of labor productivity growth in or-der to identify the role of labor reallocation. This is shown in gure 5 and 6.

Egypt's labor productivity growth does rely more on productivity growth within sectors than in shifting labor to more productive activities. Moreover, labor even moved in the wrong direc-tion, to less productive sectors, in the latest period. As discussed, the large movements to the construction sector and non-market services were certainly not benecial. This is also the main reason for low or even negative structural change. As argued in OECD (2011), the government services attract large parts of the labor force, because it oers save jobs. This relatively skilled labor force, however, is missing in relatively productive sectors. Relating these ndings to the growth pattern depicted in gure 1 indicates that the source of the growth spurt between the mid-1970s and mid-1980s were high rates of productivity growth within sectors. The relatively low productivity growth rates within sectors in the following periods also account for the low GDP per capita growth rates. Structural change, on the other hand, did not substantially aect labor productivity growth.

Morocco, on the other hand, beneted much more from labor reallocation. The contribution of that reallocation is relatively large in all periods, but declining. This may be linked to Morocco's converging productivity levels. As modeled in dual economy models, Morocco has gained from reallocation in early periods in which productivity dierences were large, but dur-ing this process of development, productivity levels converged and reallocation gains decreased. At the same time, the speed of transformation also slowed down considerably. As mentioned earlier, with still 35% of the labor force in agriculture, there is still potential to reallocate labor. However, productivity growth within sectors was much smaller than in Egypt and largely neg-ative in the 1970s. Thus, in Morocco, labor seems to move to more productive sectors, but the main obstacle to productivity growth is rather that the sectors are not becoming much more productive. Labor productivity growth is generally relatively small in Morocco and was almost stagnant in the 1970s. Regarding gure 1, Morocco experienced relatively constant GDP per capita growth rates in the 1970s. That might be related to a large increase in labor force participation in that period (see Feenstra, Inklaar and Timmer, 2013).

(37)

Figure 5: Decomposition of labor productivity growth in Egypt between 1960 and 2006 -2% 0% 2% 4% 6% 1996-2006 1986-1996 1976-1986 1966-1976 1960-1966

Annual labor productivity growth

Ti m e p e ri o d Within Between

Source: own calculations based on equation (1) and data series described in the main body and appendix.

Figure 6: Decomposition of labor productivity growth in Morocco between 1960 and 2004

-2% -1% 0% 1% 2% 3%

1994-2004 1982-1994 1971-1982 1960-1971

Annual labor productivity growth

Ti me p e ri o d Within Between

Source: own calculations based on equation (1) and data series described in the main body and appendix.

(38)

shift also contributed positively to labor productivity growth in that period.

Harabi (2008) describes the period after 1982 as a period that aimed at implementing WC poli-cies: e.g. macroeconomic stabilization programs, liberalization of trade and of selected domestic markets, and privatization. The privatization started more thoroughly in 1993, however, some entities remained in public ownership. Harabi (2008), however, highlights that certain selected industries were still picked, but the focus drifted away from manufacturing. One example is the tourism industry. Economic activity of the tourism sector is captured to some extent in the hotel and restaurant business which is aggregated in trade services (see table 9). As shown above, this sector's employment share also kept on rising throughout that period. Thus, active promotion of industries may be indeed linked to the reallocation of labor and as gure 6 shows, that also contributed to labor productivity growth.

Regarding Egyptian industrial policy, a similar link can be established, although the phases are not that clearly to distinguish (summarized in Said, Chang and Sakr, 1995; Galal, 2008; Galal and El-Megharbel, 2008; Benner, 2013). Most remarkable is that also Egypt started with sup-porting manufacturing industries, but then switched to trade services, and toursim in particular since the mid-1970s. This support also coincides with the numbers provided in table 4. In later time periods, Egypt followed more closely WC policies and the manufacturing sector became relatively more productive. However, the direction of labor reallocation was even negative since the latest phase of liberalization in 1993 (see gure 5). Thus, it seems that ISI policies tend to do a better job in allocating labor to more productive sectors than more liberalized policies, a link that is also established in Rodrik (2013a). On the other hand, at least in Egypt, phases of liberalization also coincide with a relatively more productive manufacturing sector.

(39)

regression framework and also nds a negative eect in Egypt. Related to the weak perfor-mance of the manufacturing sector, Galal (2008) investigated characteristics of that sector and nds that the sector did not diversify signicantly and is therefore performing weakly. A weak manufacturing sector can be linked to weak pull factors and the consequence is that the sector does not absorb labor. As suggested in Benner (2013), it is rather the tourism industry, or on the aggregate level, the trade services sector that needs to take that role in Egypt.

(40)

agricul-tural sector, but suggests that the obstacles to modern agriculture are likely to be the same as for modern manufacturing. The other scenario, growth based on tradable services may be problematic for a second reason apart from high human capital intensity: context-specicity of technologies (see Rodrik, 2014). That in turn may imply that African countries have to steadily increase their innovation capabilities and human-capital in order to build a competitive services sector base. That, however, does not suggest a fast path to high-income levels.

Conclusively, this section made four main points. First, transformation took place in both countries, but early movements to manufacturing were replaced by movements to trade services. This links to the phenomenon of pre-mature industrialization. Second, Egypt's economy is more characterized by a dual economy structure than Morocco, and with 61.8% in unproductive sectors, there are large potential benets from reallocation. Third, linked to that, Morocco has gained in the past from reallocation, whereas Egypt rather relied on productivity growth within sectors. Fourth, the movements of labor can be linked to industrial policies.

5 Determinants of structural change: regression analysis

As shown in the previous section, labor reallocation to more productive sectors played a more prominent role in Morocco than in Egypt, which was more reliant on productivity growth within sectors. Among others, one factor that coincided with labor reallocation was the pattern of industrial policy. This section further explores whether determinants can be identied that are linked to the magnitude of labor productivity growth due to labor reallocation. Section 5.1 shows trends and partial correlations in the full dataset to identify general relationships and section 5.2 performs a formal statistical analysis of the outlined hypotheses.

5.1 Trends and correlations

Referring to the theoretical approaches and empirical ndings in section 2, some trends and correlations are shown.

(41)

level of development in Latin America and the most developed countries, this is an expected pat-tern, but Latin American countries should also gain more from reallocation. That has not been documented (see e.g. McMillan, Rodrik and Verduzco-Gallo, 2013). For sub-Saharan African countries, the pattern diverges a little bit. Until the end of the 1980s relative agricultural pro-ductivity was not increasing and even decreasing to some extent. However, convergence seems slowly taking place since 1990. Asian countries seem to follow a very dierent pattern. There seems to be no convergence of agricultural productivity over time. Thus, Asian countries seem to uphold the dual economy structure in contrast to the other regions. This might thus help to explain why labor reallocation kept on contributing largely to labor productivity growth in Asia and not in other regions (see e.g. Escaith, 2007). Constructing an index of relative non-agricultural to relative non-agricultural productivity, Temple and Woessman (2006) provide a very similar nding. Convergence seems to take place in all regions, except in Asia. Recalling the Egyptian and Moroccon graph, they compare to the Latin American countries in that concern: both countries have a relative productivity of agriculture of about 0.5 to 0.6 in 2005.

Figure 7: Relative agricultural productivity in dierent regions between 1975 and 2005

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 19 75 19 77 19 79 19 81 19 83 19 85 19 87 19 89 19 91 19 93 19 95 19 97 19 99 20 01 20 03 20 05 R elati ve p ro d u ctiv ity o f ag ri cu ltu re Year Developed LA ASIA SSA

Source: own calculations based on GGDC 10SD and ASD. Note: regions classied as in table 3.

(42)

is not statistically signicant, but relatively close to statistical signicance. As mentioned in 3.1, part of that relation is by construction of the between term.

Hypothesis two can also be operationalized in other ways. Figure 9 plots manufacturing pro-ductivity growth against the between term. This also shows a positive, but weak relationship between productivity growth in the manufacturing sector and gains from labor reallocation. Moreover, the relationship may be driven by some of the very large values in manufacturing productivity growth (e.g. Botswana between 1970 and 1975). The slope suggests that a 1% point increase in annual productivity growth in manufacturing is associated with 0.04% larger contribution in the between term. Figure 15 to 17 in the appendix, show three further correla-tion plots that depict pull factors: relative productivity of trade services, relative productivity of manufacturing and productivity growth of trade services. In the rst two plots, the expected relationship is graphed, but the relationships appears to be relatively weak for manufacturing relative productivity. Somewhat surprisingly, the relative productivity of trade services seems to be statistically stronger correlated to the between term than the other measures. However, productivity growth in trade services, on the other hand, does not seem to be related to the between term.

Other key variables are plotted in the following gures. Hypothesis one postulated that the extent of excess labor is related to structural change. The extent of excess labor is measured as the initial employment share in agriculture and gure 10 plots this relationship. Across coun-tries and time periods, councoun-tries with a larger share of the labor force in agriculture do indeed have larger gains from reallocating labor. This relationship is also statistically signicant in this simple linear regression and also conrmed when controlling for country and time period xed eects in the next section.

(43)

Figure 8: Pull factors: partial correlation of relative productivity of agriculture and the labor reallocation contribution y = -0.01209x + 0.0084 p = 0.119 -15% -10% -5% 0% 5% 10% 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Co n tri b u ti o n o f rell o cat io n t o an n u al lab o r p ro d u ct iv it y gro wth

Relative productivity of agriculture

Source: variables based on own calculations described in main body (section 3.1 and 3.4).

Note: contribution of reallocation to annual labor productivity growth is the obtained "between" term in equation (1). y is the dependent variable, x the independent; slope, intercept and p-value are obtained in a simple linear regression.

Figure 9: Pull factors: partial correlation of manufacturing productivity growth and the labor reallocation contribution y = 0.042x + 0.0049 p = 0.181 -15% -10% -5% 0% 5% 10% -20% -10% 0% 10% 20% 30% C o n tri b u ti o n o f re al lo cat io n t o an n u al lab o r p ro d u ct iv it y gro wt h

Productivity growth in manufacturing

BWA (1970-1975)

Source: variables based on own calculations described in main body (section 3.1 and 3.4).

Note: contribution of reallocation to annual labor productivity growth is the obtained "between" term in equation (1). y is the dependent variable, x the independent; slope, intercept and p-value are obtained in a simple linear regression. BWA stands for

Patterns and Determinants of Structural Change:

University of Groningen and University of Göttingen