Long-term economic growth in Egypt and Turkey

Hele tekst

(1)University of Groningen, 16-12-2013. Long-term economic growth in Egypt and Turkey. Natasja M. Massink. Research Master Thesis University of Groningen, Faculty of Economics and Business. Abstract The aim of this paper is the evaluation of the determinants of (long-term) economic growth in Egypt and Turkey. In particular, an attempt is made to unravel the mystery of the widening gap between Egypt’s and Turkey’s GDP per capita levels. In order to do so, two approaches are followed. First, growth accounting is used to examine the proximate causes of growth. Second, the two-step accounting approach suggested by Subramanian and Roy (2003), using cross-country growth regression results to decompose growth, is used to examine the ultimate sources of growth. The results of the growth accounting exercise suggest that capital accumulation is mainly responsible for the divergence visible when comparing GDP per capita levels of Egypt and Turkey over the period 1950-2010. Moreover, the results of the second approach suggest that, initial GDP per capita and relative growth of the economically active population have been factors narrowing the gap in GDP per capital levels between 1960 and 2010, and during that same time institutional quality differences have had a widening effect. Finally, the liberalization efforts of both countries seem to have had a positive impact on economic growth through increasing openness to trade and the improvement of institutional quality..

(2) 1 University of Groningen, 24-11-2013 1. Introduction The sources of economic growth have long been part of the academic debate. The main motive behind it being that the identification of the factors underlying sustained economic growth is key in designing sound economic policy. Here, a distinction can be made between two types of growth factors, proximate and ultimate. “The proximate causes of economic growth are the effort to economize, the accumulation of knowledge, and the accumulation of capital” (Sir William Arthur Lewis). The proximate causes can be quantified relatively easy, whereas the ultimate causes, such as institutions and economic policy, cannot. Due to their interdependence, however, both are important for understanding economic growth performance and will be addressed in this paper. To some extent, the effect of the ultimate causes on economic growth runs through the proximate causes. The aim of this paper is the evaluation of the determinants of (long-term) economic growth in Egypt and Turkey. But why these two countries? That has to do with the puzzle that arises when GDP per capita performance of these two countries is combined with information about their initial conditions. Figure 1 is a visual representation of GDP per capita for Egypt and Turkey from 1820 till 2010. It shows that around halfway the 1940s, a divergent pattern of GDP per capita arises, with a widening gap between Egypt and Turkey that still continues to exist today. The next section discusses the similarity of the initial conditions in these two countries in terms of their shared historical, cultural, and geopolitical context as well as their (initially) similar reform strategies. Figure 1 – GDP per capita 9000. 1990 international Geary-Khamis $. 8000 7000 6000 5000 Egypt 4000. Turkey. 3000 2000 1000 1820 1923 1927 1931 1935 1939 1943 1947 1951 1955 1959 1963 1967 1971 1975 1979 1983 1987 1991 1995 1999 2003 2007. 0. Notes: GDP per capita is in 1990 international Geary-Khamis dollars. Data from 1820 till 2002 was collected directly from Maddison (2003). Additional data (2002-2010) was estimated by using growth rates of GDP per capita numbers from the Penn World Tables (Version 7.1).. The similarity of the initial conditions for Egypt and Turkey are both context and policy related. First, from a historical point of view, Egypt and Turkey share a common background as part of the Ottoman Empire and had a common role in the world economy at the start of the 20th century as.

(3) 2 University of Groningen, 24-11-2013 (net) agricultural exporters. Furthermore, related to their ‘cultural’ heritage, they are both secular Islamic countries with over 90 percent of their population being (mainly Sunni) Muslim. Also, in relation to their status in the world economy, Keyder and Öncü (1994) mention that these two countries were the two most populous, oil-poor, labor-surplus economies of the region. Moreover, Hansen (1991) mentions that in the 1920s (before they started their economic reform programs) they had similar population size and initial levels of income and development. Figure 1 indeed shows that, initially, GDP per capita was relatively similar in Egypt and Turkey. Second, until the (early) 1980s their reform strategies have been quite similar. Due to increasing European penetration of the region and (roughly at the same time) the falling into bankruptcy of the Ottoman Empire and Egypt at the end of the 19th century, both started their economic reform attempts relatively early in the 20th century (Moursi and Wahba, 1994). Moreover, they both followed an etatist development strategy based on Import Substitution Industrialization (ISI) (Hansen, 1991). By the end of the 1970s, Egypt and Turkey were faced with more or less the same economic problems – growing inflation, troublesome fiscal deficits, artificially low exchange rates, and depletion of their foreign reserves – that were mainly caused by their own protectionist (reform) policies. The ISI model thus became unsustainable. As a result, both countries changed to a form of export-based strategy. Turkey started its new reform strategy in 1980. Egypt, having been shielded from the worst effects of its policies due to its ability to export crude oil and its receipts of foreign capital through aid and loans, changed its course over a decade later (Moursi and Wahba, 1994). In addition, Adly (2012) mentions that, this new economic liberalization allowed them to reiterate the notion of export restructuring as well as upgrade their niche in the international division of labor. However, with regards to export restructuring, Egypt and Turkey have not followed the same path of (state) economic institution-building. Where Turkey prioritized in export expansion and diversification through institutional arrangements and policy changes, Egypt did not (Adly, 2012). The main question that arises from the above story is: how can it be that Egypt and Turkey, having similar initial conditions, end up at such different economic development levels? In order to answer this question, the following two research questions were formulated: First, what are the sources of growth that explain differences in GDP per capita (growth) in Egypt and Turkey from 1950 to 2010? Second, in terms of these sources of growth, is there a difference between the preliberalization and liberalization periods in these countries? Section 2.5 discusses the specific timing of these liberalization efforts in both countries as well as some additional sub-questions that were formulated in relation to the difference between proximate and ultimate sources of growth. In order to answer these questions, this paper makes use of growth accounting and a twofold accounting approach using cross-country growth regressions suggested by Subramanian and Roy (2003). The results of the growth accounting exercise suggest that capital accumulation is mainly responsible for the divergence visible when comparing GDP per capita levels of Egypt and Turkey over the period 1950-2010. Moreover, the results of the second approach suggest that, initial GDP per capita and relative growth of the economically active population have been factors narrowing the gap in GDP per capital levels between 1960 and 2010, and during that same time institutional quality differences have had a widening effect. Finally, the liberalization efforts of both countries seem to have had a positive impact on economic growth through increasing openness to trade and the improvement of institutional quality. The rest of the paper is structured as follows. First an overview of the literature is given in section 2, discussing proximate sources and growth accounting in 2.1-2.2 and the ultimate sources and cross-country growth regressions in 2.3. Next, section 3 provides an overview of the methods.

(4) 3 University of Groningen, 24-11-2013 and results of the (traditional) growth accounting exercise done to determine the proximate sources of growth for Egypt and Turkey. Section 4 then discusses the methods and results of the two-step growth accounting approach used to determine the relative effect of the ultimate sources of growth. Finally, section 5 provides an overview of the results as well as an interpretation of the results in relation to the research questions posed above.. 2. Literature review 2.1 Proximate and ultimate sources of growth The discussion of the growth determinants will be structured on the basis of the idea of ultimate and proximate sources of economic growth, as distinguished by Angus Maddison and as presented in the explanatory growth framework in Appendix A. Maddison is known for his construction and analysis of GDP time series, see e.g. his historical statistics of the World Economy presented in Maddison (2003). Maddison’s analyses, aimed at isolating the determinants of economic growth, distinguished between two components. The first, containing the ultimate sources of growth, considers the importance of immeasurable factors such as institutions, policies, and culture. The second, containing the proximate sources of growth, considers the importance of ‘measurable’ variables such as labor, capital, and productivity. This latter category, the proximate causes, can be more easily quantified and therefore it is more convenient to assess their relative importance. However, as Maddison himself said, “if we are to explain why the economic growth experience of nations has been so diverse, and why income spreads are now so wide, it is necessary to go beyond proximate and measurable elements of causality and consider institutional, social or policy influences which may retard or encourage economic development” (Maddison, 1997). Similarly, Rodrik (2003) states that accumulation (of physical and human capital) and productivity growth, the proximate causes of economic growth, are themselves endogenous. Thus, to an extent, these sources are themselves driven by other (ultimate) sources that are not directly captured by the standard growth-accounting framework. Hence, they can indeed best be thought of as proximate determinants of growth. Figure 2 – growth economics.

(5) 4 University of Groningen, 24-11-2013 Figure 2 gives the schematic representation of growth economics presented in Rodrik (2003). In terms of the ultimate sources of growth, the model is slightly different from the framework of Appendix A. First, although Rodrik mentions the existence of a wide array of ultimate determinants, such as those presented in the Maddison explanatory growth framework, he only makes a distinction between three (main) groups: geography (exogenous in his model), institutions, and trade (both partly endogenous in his model). Second, the representation of growth economics presented in figure 2 gives additional insight into how these different types of economic growth factors interact with each other. Finally, in order to determine these sources of growth, researchers had to developed statistical procedures to link them to economic development. Section 2.2 will provide an introduction to the accounting approaches that have been developed to examine the effect of the proximate sources of growth on economic development. Section 2.3 will then provide a more detailed description of the ultimate sources of growth and of the use of cross-country growth regressions in estimating the relation between these ultimate sources of growth and economic development. 2.2 Introduction to growth accounting based on the ‘Solow’ model1 Growth accounting was developed to derive the proximate sources of growth and to assess variation in economic growth rates2. The growth accounting method was introduced by Solow (1957). After that, the approach was further developed by Denison (1962) and Jorgenson and Griliches (1967). It is a framework for breaking down observed output into components of growth, more specifically into changes related to factor inputs and (indirectly measured) those related to technological progress. It can be viewed as an initial step in the analysis of the fundamental determinants of economic growth, namely in assessing the impact of the different proximate sources of growth on output. The next step would then be to create a link between the results found on these proximate sources and the earlier mentioned ultimate sources of growth. Section 2.3.2 will provide a more detailed description of the cross-country growth regressions that are generally used for this second step in the analysis. As e.g. Barro (1998) mentions, growth accounting can be a particularly useful (first-step) procedure if the fundamental determinants that influence factor growth are largely independent from those that influence technological change. Furthermore, knowing whether it is increases in inputs or increases in efficiency that have caused output growth is useful because “economic growth that is based on expansion of inputs, rather than on growth in output per unit of input, is inevitable subject to diminishing returns” (Krugman, 1994). Thus, only when efficiency increases economic growth can be sustainable. Moreover, improvements in the level of technological efficiency are seen as less costly when compared to the investments of valuable resources that are required to increase capital input. This is exactly what Krugman (1994) means with his distinction between, respectively, inspiration and perspiration. For these reasons, economic growth due to technological change is preferred. It would therefore be of interest to researchers to see which fundamental determinants of growth promote technological advancement. 1. The new growth theories can be recognized as a potential alternative to the neoclassical (Solow) model. In short, the basic idea behind these theories is that continuous growth can be generated endogenously instead of requiring exogenous technological progress (as in the neoclassical growth model). However, due to its limited relevance and applicability for developing countries and the limited scope of this paper, this method will not be discussed further. 2 An additional approach, used by e.g. Caselli (2005) and Hsieh and Klenow (2010), compares variances across a group of countries. This approach is not suitable for comparing two countries to each other as will be required here and hence will also not be discussed further..

(6) 5 University of Groningen, 24-11-2013 In order to fully understand the growth accounting approach one has to start with the basic (neoclassical) Solow model. In the 1950s, Robert Solow was the first to attempt analytical modeling of long-run growth. The theoretical framework he created then is still the one most commonly used to validate the (functional) form of growth regressions and their core variables. The mathematical representation of the model, the aggregate production function expressing the relationship between inputs and output, is used as a point of departure. 2.2.1. Basic model. The Solow model states that there are two inputs to the economy – capital and labor – which are rewarded their marginal product. In addition, savings rates, population growth and technological progress are taken to be exogenous. Assuming a Cobb-Douglas production function, production at time t, is given by: = , where 0 < < 1.. (1). Here, Y is aggregate output, K is capital, L is Labor, and A is multifactor or total factor productivity (TFP). TFP consists of all factors - other than capital and labor - that have an effect on economic growth. Examples are technological advancement and improvements in health, education or institutional quality. α is the extent of diminishing (since 0 < α < 1) marginal returns to capital (1-α the extent of those to labor). Dividing both sides of equation (1) by L leads to a function of output per worker or labor productivity (y): = ,. (2). where, capital stock per worker (k) is given as = / . Labor productivity thus depends positively on TFP and the capital stock per worker. Assuming population growth equals the growth rate of the labor force, the growth rate of output per person equals the growth rate of output per worker (y). Moreover, economic growth equals the growth rate of output per person and thus (under this assumption) the growth rate of output per worker. Therefore, while (also) making use of the property of logarithms that makes it possible to sum proportional increases in y caused by the individual factors, we find that the proportional growth rate of output is3: ∆ = ∆ + ∆ .. (3). The output per worker growth rate of country i between time t and t-1 (∆ )4 consists of two components: the rate of TFP growth (∆ ) and the rate of “capital-deepening” ( ∆ ). The purpose behind growth accounting is determining the relative size of these two components. 2.2.2. Model including human capital. Later on human capital has also been acknowledged to be of significance in explaining economic growth (e.g. Lucas, 1988; Mankiw et al., 1992). As Easterly and Levine (2001) put it, if improvements The following procedure was used. First, taking logs changes equation (2) to : = + . Subtracting ln yi,t-1 from ln yit and rearranging terms then leads to: − , = − , + ( − , ). Finally, writing − , as ∆ leads to equation (3). 4 By approximation the change in logs equals the growth rate; i.e. ≈ ( − )/ . 3.

(7) 6 University of Groningen, 24-11-2013 in the quality of education and health and thus the quality of labor inputs are not taken into account, these improvements would incorrectly be assigned to improvements to TFP growth. Human capital can be incorporated in two ways: explicitly, as a (direct) input in the production function ( =. ". ! " ) – also mentioned as the Augmented-Solow approach – or as an augmentation of labor (L). The latter results in equation (1) becoming: = ( # ) , where 0 < < 1.. (4). = # .. (5). Here, human capital is embodied in labor, # is a skill or quality-adjusted measure of labor input (workforce), and # is the average level of human capital per worker. Equation (2) will become:. Equation (3) will then become: ∆ = ∆ + ∆ + ( − ) #.. (6). In order to see how much of the variation in y can be explained by the ‘observed’ variables, k and h, and how much is ‘residual’ and should thus be attributed to A, data is needed on y, k, h, and α. Considering there are no direct measures of the extent of diminishing marginal returns (α) and TFP growth, the following two-step procedure is necessary to deal with this issue. First, estimate α using data on factor income shares5. Second, estimate TFP growth as a ‘residual’. 2.2.3. General results found in the literature. The growth accounting research done in the 1950s and 1960s indicated a change in research regarding the factors that determine the wealth and economic growth of nations. Initial research by Solow (1957), using the type of growth accounting presented earlier (in section 2.2.1) on Net National Product (NNP) data from 1909 to 1949, showed that it was not capital accumulation that drove long-run improvements in standards of living. Instead it was technological advancement. More specifically, his results showed that changes in the residual factor (TFP) were responsible for more than half of United States (US) economic growth. This is in line with the idea that societies eventually reach their limit of capital and labor deployment and that TFP thus will be the decisive factor in longterm growth. These findings were later on reinforced in a larger (world) setting by Kuznets’ (1971) international comparisons. The approach used by both authors is commonly referred to as the traditional approach. Subsequent authors, such as Denison (1962) and Jorgenson and Griliches (1967), suggested refinements to this approach. Respectively by incorporating elements of (labor) input quality (related to e.g. variations in working hours and education) to account for part of the residual and by correcting for aggregation errors and changes in the labor and capital utilization rate. Furthermore, Jorgenson and Grilliches also started using Gross National Product (GNP) data instead of Net National Product (NNP) data, no longer excluding depreciation. The results of Jorgenson and Griliches (1967), showing that 96.7 percent of the US growth rate from 1945 to 1965 was accounted for by input growth, were strikingly different from the (old) results of the traditional approach. For a similar time period (1950-1973), Maddison (1987) also points out, for a group of OECD countries, that 5. Assuming that the inputs are paid the value of their marginal products, the elasticity of substitution (in the text mentioned as the extent of diminishing marginal returns, α) equals factor income share..

(8) 7 University of Groningen, 24-11-2013. productivity growth was mainly caused by improvements in efficiency and production factor quality (incorporated in the new approach) rather than disembodied technical change (as suggested by Solow, 1957). Still these early growth accounting studies were mainly focused on the US and (later) the rest of the industrialized (OECD) countries. Experiences in terms of the sources of growth are not typically the same across countries. A striking example of the new findings is provided by East-Asia. In countries such as Korea, Singapore and Taiwan the contribution of capital deepening to the rapid catch-up experienced by these countries in the period 1960 to 1990 has been impressive and has by far exceeded the contribution of TFP growth (Krugman, 1994). More recent is the interest in information technology (IT) as a source of growth. Jorgenson et al. (2008) e.g. show that the driving forces behind the accelerated labor productivity growth in the US during the mid-1990s were the productivity growth in IT sectors and the capital-deepening effect of IT investment. Taken together, they accounted for almost 80 percent of productivity growth increases between 1995 and 2000. TFP growth outside of the production of IT gained importance since 2000, but is still mainly confined to IT-intensive industries. TFP growth remains to be an important factor in US productivity growth, although its sources vary over time. Moreover, Jorgenson and Vu (2010), in their study of potential growth of the World economy, show that US growth trends have counterparts in the rest of the world. Making use of the new (more sophisticated) data they compiled on investment in IT equipment and software as well as the growth in labor quality, their findings show that investment in assets, including IT but dominated by non-IT investment, was most important for growth. With working hours outweighing quality, labor input followed next in magnitude of importance. Finally, TFP was only a dominant source of growth in Eastern Europe and the former Soviet Union (1989-1995). These ‘new’ findings thus show supremacy of inputs over TFP as sources of GDP growth as well as an increase in the role of IT investment. 2.2.4. Limitations. There are, however, some limitations that come to mind when discussing the growth accounting. First, growth accounting does not measure causality. Moreover, factor accumulation could increase overall growth as well as TFP growth. The accounting procedures do not take into account the possible interactions between causes. When factor accumulation and efficiency influence each other, which is likely, it is difficult to assign separate significance to each factor. Moreover, as Fagerberg (1988) puts it, with interdependent variables and no theory of how they interact, decompositions are only illustrations of the growth process and cannot themselves be regarded as causal factors. Second, these approaches to growth analysis use a mix of hard and soft evidence. Components of the accounting equations are based on assumptions about unknown parameters and for some of them there are several theoretically legitimate options. Two of those unknown parameters are the depreciation rate and the factor weight(s). Regarding the first, little information is available about depreciation rates in the developed world and hardly any information is available about the developing world, which forces researchers to make imperfect choices. An example of researcher’s choices with regards to the depreciation rate in the developing world can be found section 3.1.3. With regards to the second, the choice of e.g. excluding the factor ‘land’ (as done in the model used here) could lead to an overestimation of the share of capital in the production function. This, since the share of capital and the share of labor in the production function add to one. In turn this could lead to an underestimation of the rate of TFP growth. Furthermore, the share of capital is calculated by subtracting a measure of the labor share from their combined value (is equal to one). Several alternatives are available for this labor share measure, as can be seen in Appendix D. Even the ‘harder’ evidence, such as measures of GDP and capital stock, are subject to e.g. the.

(9) 8 University of Groningen, 24-11-2013 refinement of the available data. Furthermore, there are no standard measurement methods for estimating growth rates of capital and labor. Also, when not taken into account directly in the labor and capital measures, all qualitative changes in these factors of production are captured indirectly in TFP. Moreover, when comparing data between countries, there are additional issues arising due to differences in assumptions made by these countries in terms of asset lives, depreciation formulas, scrapping patterns, benchmark years, and price deflation techniques (Maddison, 1987). Third, since TFP growth is measured as a residual, i.e. that part of growth that cannot be accounted for by growth in the input factors, it is not only technological progress that is captured. For example, besides the technological progress (in the Solow model), the residual also captures allocation efficiency, economies of scale, omitted variables, and measurement error. The TFP residual reflects all dynamic effects and hence does not provide insight into the nature of TFP growth. Fourth, growth accounting only uncovers the proximate causes of growth. Thus not the ultimate ones. It does not say anything about the causes of low factor accumulation or low levels of TFP growth. As Caselli (2005) says, the most likely scenario is that the same ultimate causes explain both factor accumulation and TFP growth levels. Finally, as Maddison (1997) mentions, proximate analysis has resulted in a clear view of what the driving forces of successful capitalist development have been, but has not provided an answer to many other cases, such as the failure of the communist experiment or why Latin America performs worse than North America. Regardless of the limitations, these approaches are still very helpful as a diagnostic tool as the approach does identify the facts that need more, as Maddison (1987) says it, ‘ultimate’ explanation. 2.3 Introduction to the ultimate sources of growth literature Up to this point the proximate elements of growth have been discussed as well as their measurability and the growth accounting framework used to estimate them. This section will discuss the literature regarding the ultimate sources of growth. It will not go into detail about all the factors presented in Appendix A, but will provide an insight into the main strands of thought in the literature – regarding the role of geography, institutions and openness to trade as determinants of the steady state – as well as discuss the idea of convergence towards this steady-state. This will be done in respectively, section 2.3.1 and 2.3.2. The general results found in the literature will be discussed in relation to these first two sections. Section 2.3.3 will then present the main issues and limitations of the type of analysis that is used. 2.3.1. Main strands of thought. Figure 2 (below the dotted line) presents the three main strands of thought in the literature regarding the fundamental causes of growth (i.e. those ultimate causes of growth, as presented in appendix A, that are said to be the most important for economic growth); one emphasizing the role of geography, one the role of international trade, and one the role of institutions. The first theory focuses on the physical and geographical environment or simply said a country’s geography. Geography determines (in large part) the climate, natural resource endowments, and disease environment of a country, its transport options and costs, and, moreover, the potential for technological diffusion from more advanced countries. Several sub-categories of the geography factor, such as landlockedness, tropical climate, and natural resource abundance, have been linked to economic growth in the past. Sachs and Warner (1997) e.g. find a significant negative relation between these sub-categories and economic growth. The ideas behind it are that a landlocked country faces higher transport costs, tropical climate negatively affects agricultural productivity and labor productivity (through its impact on e.g. soil and the disease environment), and natural resource.

(10) 9 University of Groningen, 24-11-2013 abundance, although open to question according to Sachs and Warner (1997), creates some sort of ‘Dutch disease’ effect (where manufacturing becomes internationally less competitive due to an ‘inflated’ exchange rate) and greater incentives for rent-seeking. As we can see in figure 2, geography is the only variable that is truly exogenous in the growth model. As can be seen, it can potentially work through (three) different channels. Taking the subcategories presented above as an example, it can be seen that each of them has an effect on economic growth via a different channel. More specifically, climate directly affects the productivity (agricultural and labor) of a country, being landlocked – reflecting a relative distance from (international) markets – affects growth through integration and trade, and finally, natural resources could potentially affect a country’s exchange rate (Dutch disease) and hence trade factor as well as its institutions (by providing incentives for rent-seeking)6. The second strand of thought regards openness to trade theories and is also referred to as the integration view. This, due to the important role these theories give to market integration in fostering economic convergence. Integration into the world economy has been pointed out as a key driver of growth by many international policymakers. In the academic literature regarding the sources of growth, it is a widespread idea that, once certain pre-requisites (e.g. institutional) have been met, trade can indeed be an underlying source of growth. Several authors have found a positive effect of trade openness on growth7. Sachs and Warner (1997) used information on trade policies to determine the number of years a country has been ‘open’ to trade and then regressed this on changes in GDP over the same period. The idea behind the positive relation they found is that openness to trade can lead to productivity improvement by allowing for competition and technology transfers across borders. Some research papers, however, point to a more extreme perspective in that international trade is a key determinant of economic development. In their paper, Dollar and Kraay (2004) e.g. suggest that one of the best ways for developing countries to increase economic growth is to pursue trade liberalization. Moreover, their results, using decadal changes in trade volume, support the view that globalization leads to faster growth and to reduction of poverty in developing countries. There are, however, some issues in relation to the measurement of trade openness and its relation to economic growth. First, the issue of interrelatedness between the trade openness variable and other (potentially omitted) variables. For example, using trade policies as a proxy for openness makes it difficult to identify the impact of trade on economic growth. This, since these policies are potentially related to other policies, such as fiscal or monetary policies, that also affect growth, but are omitted from the estimation. Second, the issue of reverse causality (as represented by the arrows going both ways between trade and growth in figure 2) points to the fact that a simple relation between trade and income cannot provide answers regarding causality. Authors have therefore tried to look for ‘instruments’ that allow them to surpass these issues. Frankel and Romer (1999) e.g. use an instrumental variable (IV) measure based on geography. They mention that geography (in terms of relative distance to each other) is a powerful determinant of bilateral trade as well as overall trade. Moreover, geography is not affected by country income or policies and hence exogenous (as mentioned earlier). Although Frankel and Romer (1999) use the idea that it is difficult to even think of other effects of geography on income (than through trade) as a reason to use their 6. Sala-i-Martin and Subramanian (2003) e.g. show a link between natural resource endowment and low levels of institutional development. For the case of Nigeria, they show that natural resources exert a negative (nonlinear) impact on economic growth, which is caused by their negative effect on the institutional quality (and thus not a Dutch disease effect). 7 See e.g. Sachs and Warner (1997), Edwards (1998), and Subramanian and Roy (2003)..

(11) 10 University of Groningen, 24-11-2013 IV measure, this obviously is a matter of opinion. As discussed in the previous part on geography, a direct effect of geography through the climate would also be a relevant option. They additionally control for country size since it’s negatively related to relative proximity to other countries and thus markets (due to a larger internal market)8. Their results, although moderately significant, suggest that trade has a large and robust positive effect on income. Furthermore, Dollar and Kraay (2004) tried to cope with these issues by using a technique suggest by Caselli et al. (1996), namely be estimating in differences and using lagged values of the explanatory variables – among which the trade openness variable – as instruments. They also mention that since their estimates reflect the effect of changes in trade on changes in growth, the estimates do not reflect geography-based differences (such as in Frankel and Romer, 1999). The third theory regards the primacy of (formal) institutions. Formal institutions, qualified as the codified structures and written rules that govern actions through incentives (North, 1990), have been at the center of the academic debate on economic growth and development. The idea is that the rules of the game (ultimately determined by how humans decided to organize society) and their contribution to desirable economic behavior is what matters most for economic growth. The research has mainly focused on economic institutions, in particular the roles of the rule of law and property rights. Acemoglu et al. (2005) mention that “there must be enforcement of property rights for a broad cross-section of society so that all individuals have an incentive to invest, innovate and take part in economic activity. There must also be some degree of equality of opportunity in society, including such things as equality before the law, so that those with good investment opportunities can take advantage of them”. The importance of ‘good’ economic institutions has been widely acknowledged9. However, when looking at specific institutions or their context the results are less clear. Most studies use a general institutional quality index, such as the one published by the Political Risk Services10, which includes indices of (the willingness of citizens to accept) the rule of law, bureaucratic quality (autonomy from political pressure), corruption in government, risk of expropriation, and government repudiation (modification) of contracts. Acemoglu et al. (2001) take a different approach and instead use settler mortality (believed to influence current institutions through the potential for colonial settlement and the establishment of early institutions) as an instrumental variable to proxy for institutions11. The idea is that the historical disease environment only affects growth today through its effect on current institutions and thus can be used as an exogenous source of variation in current institutions. Using this instrumental variable they find that the largest part of the gap between the developed and developing countries today is due to differences in economic institutions. They also show that once institutions are controlled for, geographical factors, such as latitude, being landlocked, or the current disease environment, no longer hold any explanatory power for current economic development. Market imperfections or the complete absence of markets have also been mentioned as a potential strain on economic growth and development. In practice both developed and developing countries do not really experience the states of perfect competition in their factor and product 8. Alesina et al. (2005) also combine country size and international trade. Their results suggest a strong effect of both size and openness on growth in a sample of countries since 1960. Country size matters for economic performance and is endogenous to economic factors such as free trade, public goods provision and preference heterogeneity. 9 See e.g. Acemoglu et al. (2001); Glaeser et al. (2004); Rodrik et al. (2004). 10 The index is constructed from data of the International Country Risk Guide by the Center for Institutional Reform and the Informal Sector (IRIS). 11 Like Frankel and Romer (1999), they use a geography based instrumental variable to control for issues of reverse causality..

(12) 11 University of Groningen, 24-11-2013 markets that are often assumed in international trade and economic growth models. Imperfect markets are bad for competition, which, to a certain extent, is good for growth. On the one hand due to its positive effect on consumer choices, prices and product quality. On the other, due to providing firms with the incentives to become more efficient and innovative. In terms of competition and innovation Aghion et al. (2005) suggest an inverted u-shape, where competition may increase the incremental profit from innovating (in particular in industries where companies have similar technological levels and there is neck and neck competition), but competition may also reduce innovation incentives by decreasing post-innovation rents (especially in industries with laggard companies that already have low initial profits). The effect of competition in a particular country thus depends on where a country stands on this u-shaped curve, which in turn depends on the fraction of industries in which there is neck and neck competition. Market distortions may thus be harmful for growth to the extent that competition fosters growth. Distortions in factors market are often linked to government policy, such as a government’s control over energy prices or their stringent employment regulations, and can result in inefficient resource allocation and levels of output which are below a country’s potential. However, according to e.g. Acemoglu et al. (2005), markets are endogenously determined as an outcome of systems of property rights and political institutions and thus not unalterable forces responsible for cross-country growth differences. Related to this third strand is the ‘cultural’ explanation – formal versus informal institutions – where culture is a key determinant of the values, preferences and beliefs of individuals and societies that play a role in shaping economic performance. This is related to Weber’s (1930) theory that different religions have different implications for prosperity. A famous example is the idea that the origins of the industrialization in Western Europe can be found in Protestant reformation (Calvinism). Culture, an informal institution12, has been strongly associated with economic growth in the literature13. The reasons for this link is the ability of culture to affect the success of formal rules and constraints. However, as Tabellini (2008) correctly points out, culture and formal institutions influence each other, and hence culture (at least partly) evolves endogenously. A related thought is that culture and (formal) institutions are imperfect substitutes and that culture could thus especially be relevant in cases where institutional quality is low or where institutions are lacking altogether. Finally, the problem lies in comparing these views. What are the partial effects of these three ‘factors’ and does one of these theories indeed explain economic growth best. In terms of openness to trade, it could be that the positive results found in e.g. Dollar and Kraay (2004) at least partly reflect the omission of institutional quality. Several research papers have been written to discuss this issue. However, the conclusions made are very different from each other due to the use of different estimators and due to the fact that most of the papers only incorporate one or two views in their empirical model. An additional problem, as Dollar and Kraay (2003) state, is that some of the existing attempts to isolate the partial effects institutions and trade suffer from identification problems. More specifically, it is a problem that the historical and geographical instruments used in the literature have strong predictive power for both institutions and trade. Hence, although they work well on their own, they are not strong enough to indentify partial effects. The most extensive example of this type of study, including all three views and a wide selection of robustness checks. 12. North (1990) defines institutions as the ‘rules of the game’ that govern actions through incentives. Moreover, informal institutions are said to include the cultures, norms, and conventions enforced by society. 13 See e.g. Guiso et al. (2006); Tabellini (2010); Williamson and Mathers (2010)..

(13) 12 University of Groningen, 24-11-2013 related to other comparison studies14, is Rodrik et al. (2004). Their results show a (robust) primacy of institutions in explaining economic development over geography as well as integration. 2.3.2. Distance to the ‘steady-state’. The previous section is focused on the average effect across countries and discusses the main determinants of the steady-state. However, it is possible that effects differ between countries because of the existence of different development paths towards this ‘steady-state’. The assumptions is made in the neoclassical model that countries move towards their steady-state, but that they might take different paths to get there. That is where the notion of convergence comes in. Convergence issues have been playing a key role in the body of growth literature for many years. Researchers have tried to find out whether developing countries are able to catch up with the technological leaders of today and what factors may cause or aid this catching-up process. The simple catching-up hypothesis, expressed in Abramovitz (1986), is that being backward in the level of productivity carries a potential for rapid advance. This occurs through the ability of developing countries to import technology from more technologically advanced leader countries. In addition, the potential for rapid growth weakens as technology levels converge; the larger the gap in terms of productivity, the larger the possible leap that can be made. However, why is it that (many) formal models of economic growth portray that poor countries should grow faster than rich ones (and hence the income gap should be closing over time) and yet when international comparisons of income growth are made over recent decades, there is no overall tendency visible of the poorer regions in the world catching up with the richer regions? This situation does not appear to be very promising for the least developing countries. It seems that these developing countries do not catch up with the technological leaders of the Western World, but fall behind even more in terms of per capita income. What needs to be mentioned here first is that convergence can be understood in two ways: (1) in terms of growth, where all countries grow at the same rate in their steady state (which differs across countries), and (2) in terms of income level, where aggregate production functions and thus steady state income levels are assumed equal across countries. It is the first type that is of interest here since the neoclassical growth model predicts that countries will converge to their balanced growth path or steady state. Per capita growth rate is thus expected to be inversely related to initial GDP per capita. Basically, what is expected is ‘conditional’ convergence, where convergence occurs after controlling for certain determinants of a country’s steady state. In an attempt to explain the lack of cross-country evidence on absolute convergence, Sachs and Warner (1995) for example suggest that a country’s trade regime might be the answer. A country’s trade regime is thus suggested as one of the determinants of its steady state. According to this paper15, the lack of convergence is due to the fact that developing regions have been (more) closed and that it is openness to trade that leads to convergence. The basic idea of conditional convergence is that certain factors, such as the three main strands of thought discussed above, determine a country’s steady state level of income. The results in the literature, taking into account those different determinants, have shown a consistent appearance of conditional convergence in OECD countries of (in most cases) approximately 2 percent per annum16 and hence the use of initial GDP measures in regression analysis is widespread. However, when these (developed) OECD countries are combined 14. Among others: Alcala and Ciccone (2004) – trade view; Sachs (2003) – geography view. Later papers of these two authors are more focused on geography as an explanation of growth patterns. 16 See e.g. Baumol (1986), Barro (1992), Mankiw et al. (1992), and Sala-i-Martin (1996). 15.

(14) 13 University of Groningen, 24-11-2013 with other less developed countries the results are no longer consistent with the convergence hypothesis following from the neoclassical model(s). One of the stylized facts of (world) economic growth is that it is not conditional convergence but divergence that occurs. Differences between the developed and developing world are increasing (Easterly and Levine, 2001). Cross-country growth regression analysis is used to find the (ultimate) determinants of differences in growth rates. The method is usually credited to Barro (1991) and typically based on a derivation of the Solow model, like the one presented in section 2.2.2 (including human capital). The sub-sequent cross-country growth regression for per capita GDP growth (y) would then look like: ,,$% = &' + & ()* + &+ (∗) + -)* ,. (7). where ,,$% is the growth rate of per capita GDP for country i between time t and t+T, ()* is per capita GDP for country i at time t (a proximate for initial GDP), (∗) is the steady-state value of per capita GDP for country i, and -)* is the error term. In terms of the convergence theory, & can be viewed of as the measure of convergence predicted by the neoclassical growth model. Hence the often used term β-convergence. As mentioned earlier, the idea is that a higher income reduces the scope for catching up with the technological leader and thus a negative relation between initial GDP and growth is expected. Moreover, the steady state term ((∗) ) can be viewed of as the vector of determinants (approximated by the explanatory variables in cross-country growth regressions)17 that need to be controlled for in order to determine conditional convergence. Taken together, in this model, conditional convergence is thus said to occur when & < 0. 2.3.3. Limitations. There are three main issues or limitations that are commonly mentioned in discussions of crosscountry regression literature: the issue of endogeneity, the issue of model uncertainty, and the issue of parameter heterogeneity. First, as mentioned earlier, the variables of the growth models are themselves (partly) endogenous. This means that, without moderations to the model, claims regarding the causality between these variables and economic growth may be regarded as invalid. As discussed in section 2.3.1, researchers have tried to overcome this issue by using either lagged values or instrumental variables18. A note, however, is that there are no theoretical grounds to assume that these instruments are statistically unrelated to the model errors (Brock and Durlauf, 2001). The second issue is related to the compatibility of the different strands of thoughts (theories) and the numerous variables that have been linked to growth in the past. There is no theoretical justification for variables outside those presented in the neoclassical model and if one factor matters for growth it does not mean another, per definition, does not. The issue of model uncertainty related to this can easily be explained by using a quote from Durlauf et al. (2005): “If one has a set of K potential growth theories, all of which are logically compatible with one another (and all subsets of theories), there exist 2K – 1 potential theoretical specifications. Two options that have been used in the literature to find a way in this wide array of theoretical specifications are: (1) the identification of variables whose empirical significance is robust across different specifications (e.g. Levine and Renelt, 1992; Sala-i-Martin, 1997) or (2) the use of a general-to-specific-modeling strategy (e.g. Bleaney and Nishiyama, 2002). The first approach, however, can be very time consuming and (also) has not been 17 18. These explanatory variables are related to the main strands of thought as discussed in section 2.3.1. See e.g. Frankel and Romer (1999) or Acemoglu et al. (2001)..

(15) 14 University of Groningen, 24-11-2013 very influential in the growth literature. The third issue is related to a quote by Harberger (1987): “What do Thailand, the Dominican Republic, Zimbabwe, Greece, and Bolivia have in common that merits them being put in the same regression analysis?”. Basically, different countries cannot be represented as draws from a common (linear) growth model. Here, again researchers have tried to find ways to circumvent this issue. The use of e.g. fixed effects – allowing for a time invariant intercept – in panel analysis, is widely used. Generally, it makes sense to at least test for neglected parameter heterogeneity by using interaction terms and/or diagnostic tests (Durlauf et al, 2005). Longitudinal case studies, which are an alternative to cross-country regressions, offer more in-depth research into the economic growth determinants of specific countries. A major criticism of this approach is, however, that due to the specificity of this type of research, the results often cannot be applied elsewhere. An approach in between, which is also followed here, would be looking at a set of countries that are similar in terms of geography, historical development, and/or policy (reform) experience, to identify common causes behind their growth performance. 2.4 Combined approach This paper combines the growth accounting and cross-country growth regression analysis (discussed above) in a two-step approach. The findings of the two approaches are used as complements. In step one, growth accounting is used to see what the contribution of (human and physical) capital accumulation and (as a residual) the contribution of TFP – contributions of the proximate sources – are to the per capita economic growth rates of Egypt and Turkey. It can e.g. be seen whether either of these countries is suffering from low factor or low efficiency growth. Furthermore, in this growth accounting analysis each country can be evaluated separately; no international comparisons are necessary. However, as mentioned in section 2.2.4, this approach does not provide insight into the nature of TFP growth nor an understanding of why TFP growth (or capital accumulation rates) may differ across countries or over time. Moreover, the individual country growth accounting approach taken is also not able to evaluate one of the important implications of the Solow model, namely the occurrence of convergence; evaluating the convergence hypothesis requires data comparisons across a range of countries. The combined approach with cross-country growth regressions can partly account for these shortcomings. In step two, a cross-country growth regression analysis is done to see which factors can explain growth differences between Egypt, Turkey and two sets of reference countries as well as differences over time within Egypt and Turkey. Unlike growth accounting, the regression analysis done allows for differences in productivity growth (TFP) to be explained. Regression analysis can add factors, such as the quality of political institutions or the integration into the ‘World’ market(s), that are left out in the growth accounting approach and hence are captured in the residual factor (TFP). Basically, it allows for a look beyond the proximate sources of growth to the ultimate causes that (potentially) underlie them. At the same time this analysis is also not restricted by the assumptions of the growth accounting framework with regards to e.g. perfect competition and factor shares that were mentioned earlier. It does not take the production function as a point of departure, but instead tries to identify reasonable determinants (of growth). However, it does not make a distinction between Krugman’s ideas of inspiration and perspiration (mentioned in section 2.2) as can be done within the growth accounting framework. The idea is that economic growth can only by sustainable when efficiency increases (inspiration). By combining the two approaches some of the inherent weaknesses of each method are thus overcome..

(16) 15 University of Groningen, 24-11-2013 2.5 Research questions In relation to the puzzle arising from figure 1 in the introduction, two main research questions were formulated. First, what are the sources of growth that explain differences in GDP per capita (growth) in Egypt and Turkey from 1950 to 2010? Second, in terms of these sources of growth, is there a difference between the pre-liberalization and liberalization periods in these countries? Section 2.5.1 will present a set of additional sub-questions in relation to the difference between proximate and ultimate causes of growth as well as the earlier discussed methods of estimating these sources. After that section 2.5.2 will provide an overview of the specific timing of the liberalization efforts in Egypt and Turkey. 2.5.1. Sub-questions. In order to answer the main research question discussed above the following sub-questions were formulated: (1) Which proximate sources of growth explain economic growth in Egypt and Turkey in the period from 1950 until 2010? (2) Moreover, is there a difference between the pre-liberalization and liberalization periods in terms of the proximate sources that explain economic growth in these two countries? In relation to sub-question 2, the specific sub-periods with regards to liberalization efforts are:. Egypt. Turkey. Pre-Liberalization Periods. Liberalization Periods. 1950-1956 1956-1991. 1991-2010 2004-2010. 1950-1991. 1991-2010. 1950-1961 1961-1980. 1980-2001 2001-2010. 1950-1980. 1980-2010. The choice for these specific sub-periods are discussed in section 2.5.2. Furthermore, sub-questions 1 and 2 will be evaluated by using the results of the growth accounting procedure that is outlined in section 3. (3) Which ultimate sources of growth explain economic growth differences between Egypt and Turkey, as well as differences between these countries and other (reference) country groups, in the period from 1960 until 2010? The three main strands of thought with regards to the ultimate causes, as discussed in section 2.3.1, will be used as a way to structure the results. Furthermore, the specific country reference groups that will be used are: the (other) MENA countries and the (other) developing countries. (4) Moreover, is there a difference between the pre-liberalization and liberalization periods in terms of the ultimate sources that explain economic growth in Egypt and Turkey?.

(17) 16 University of Groningen, 24-11-2013 In relation to sub-question 4, the specific sub-periods with regards to liberalization efforts are: Pre-Liberalization Periods. Liberalization Periods. Egypt. 1960-1991. 1991-2010 2004-2010. Turkey. 1960-1980. 1980-2001 2001-2010. The choice for these specific sub-periods differs slightly from the periods presented above in relation to sub-question 2 since data is available from 1960 onwards. Moreover, sub-questions 3 and 4 will be evaluated by using the results of the procedure to breakdown growth by using cross-country regression data that is outlined in section 4. (5) Finally, is there growth left unexplained by the cross-country growth model, and if so, what are the potential (idiosyncratic) causes of growth that are not included but might be important for Egypt or Turkey? This question basically asks to what extent the ultimate sources of growth in the standard crosscountry growth model used explain economic growth differences between Egypt, Turkey, and other (reference) groups. This questions is evaluated by looking at the cross-country growth regression results as well as reviewing the literature with regards of the factors of growth that have been found to be relevant for Egypt and Turkey. The evaluation of all these sub-questions will be presented in section 5.1-5.3. Section 5.4 will combine the results of these questions to evaluate the main research questions. 2.5.2. Timing of liberalization efforts. The start of the liberalization period is defined as a switch from import substitution industrialization (ISI) to market orientation. The following section discusses the specific dates at which this switch occurs for each country and whether there can be sub-periods distinguished within the preliberalization and liberalization period. Hansen (1991) mentions that, after an initial preference for (but not actively pursued) market-based industrialization, both countries turned to etatism19 and ISI due to externally generated crises. For Turkey, the Great Depression of the 1930s caused its external markets for agricultural exports to collapse, resulting in a significant drop in national income. In order to promote economic recovery, the Turkish government stepped in and reoriented towards etatism and import-substitution. Furthermore, Özcan et al. (2007) mention that the 1960-1961 constitution (formulated by the revolutionary government) increased the importance of the state by combining import-substitution with economy-wide planning. The period of ISI in Turkey could therefore itself potentially be split into two parts: pre- and post-1961. For Egypt, the 1955-56 Suez crisis can be seen as the starting point of a change in state and economic policy. The Egyptian government, that came to power after the 1952 revolution, moved from encouraging the private sector, to gradually increasing restrictions and controls, to nationalizing French and British assets after the Suez War (1956). Afterwards, they continued by also nationalizing 19. It is the belief that a government (to some degree) should control economic policy, social policy, or both..

(18) 17 University of Groningen, 24-11-2013 Egyptian banks and companies and by implementing state intervention throughout the economy (Ikram, 2006). These policies were later justified as Arab Socialism in the 1962 National Charter20. Both countries eventually switched to a more market-oriented approach to economic development. For Turkey, this switch was made after the inability of the Turkish government to respond sufficiently to the 1973-1974 increases in world oil prices and the subsequent crisis it found itself in. The (extended) foreign exchange crisis that hit Turkey in the late 1970s made clear that major changes were need in their ISI approach to sustain economic growth. As a result, in 1980, the government made a switch from their ISI approach to market orientation with the adoption of a new and liberal policy package (Altuğ et al., 2008). Moreover, the 2001 crisis spurred a new IMF and World Bank program based on market orientation as well as openness to the World Economy. According to Adly (2012), the dynamics of state economic institution-building have changed considerably after that, which would justify an additional split within this market-orientation period, namely between the pre- and post 2001 financial crisis periods. For Egypt, a switch towards more market-oriented policy was caused by external pressure from international (financial) players. “Due to the cumulative effects of the heavy external borrowing (since 1975)… Egypt found itself in a debt trap in which capital inflows were increasingly consumed by debt servicing” (Ikram, 2006). In order to meet the terms of the international lenders and donors, in 1991, a stabilization and privatization program was introduced in accordance with the IMF and World Bank (Ikram, 2006). This reform policy package marks the switch from ISI to marketorientation. Adly (2012) makes an additional split at 2004 for Egypt. The main dividing line he mentions for Egypt is the formation of Ahmed Nazif’s cabinet in 2004, which lasted till the fall of the Mubarak regime in February 2011. In terms of the latter choice, he refers to Rutherford (2008) in saying that this cabinet change marked the rise of Mubarak’s junior economic team, showing much more ideological cohesion and homogeneity, and having a firm and pronounced commitment to economic liberalism.. 3. Accounting exercises on the proximate sources of growth in Egypt and Turkey This section presents the growth accounting approach used to answer sub-questions 1 and 2. The first regarding which proximate sources are responsible for growth in Egypt and Turkey and, in particular, which ones might be responsible for the differences in their growth performance (i.e. the divergent pattern visible in graph 1). The second regarding the differences between the preliberalization and liberalization periods. More specifically, section 3.1 describes which data is used and how the different variables are calculated from that data. Section 3.2 then discusses the specific approaches followed and section 3.3 presents the main results. 3.1 Data The following section will describe how the different variables are calculated, which choices and assumptions were made in calculating them, and where the ‘raw’ data was found. The raw data is available from 1950 to 2010, with the exception of labor share data. That data is only available from. 20. However, it has been suggested, by e.g. Hansen (1991) and Ikram (2006), that it was the neutralization of political opponents that was at the base of these reforms (instead of Arab Socialism)..

(19) 18 University of Groningen, 24-11-2013 1996 till 2008 for Egypt and from 1987 till 2006 for Turkey. Appendix B provides an overview of the data used, its sources and availability. 3.1.1. Income (per worker). Following Caselli (2005), GDP per worker in international dollars (PPP adjusted) is used as a measure of income per worker (y). Data on GDP per worker at constant 2005 prices, calculated using the chain series method21 (RGDPWOK), was collected from PWT 7.1. In order to calculate the capital labor ratio ( = / ) in equation (6), data on the number of workers is also needed. Following Caselli (2005), first the number of workers (L) is calculated as follows: real GDP per capita (RGDPCH) times population (POP) times a thousand22 divided by real GDP per worker (RGDPWOK). RGDPCH is PPP converted GDP per capita (chain series method) at 2005 constant prices. 3.1.2. Human capital. Ideally, a measure of human capital should include a diverse set of aspects, including, but not limited to formal and informal education, job-training, social skills, and even health. In practice, however, such a comprehensive measure is unavailable and most researchers use education statistics to proxy for human capital. The following formula, that is used to measure human capital, is derived by Caselli (2005) from Hall and Jones (1999): # = ./(0) ,. (8). where s is average years of schooling. Based on the summary of Mincerian wage regressions by Psacharopoulos (1994), the function Φ(s) is assumed to be piecewise linear, with: '. 45 0, 6 ≤ 5 '. 45 ∗ 5 + '. ' (0 − 5), 4 < 0 ≤ 9< 1(2) = 3 '. 45 ∗ 5 + '. ' ∗ 5 + '. ':9 (0 − 9), 6 > 9. (9). Caselli (2005) uses data on the average years of schooling in the population over 25 years old from Barro and Lee (2001). According to Inklaar and Timmer (2013) this is defendable only when a (major) part of the population between 15 and 25 is still in school and thus not contributing to GDP. Inklaar and Timmer (2013) themselves use data on the population of 15 years and older. In their opinion, when assuming the years of schooling increases over time excluding the 15 to 25 year-old category likely understates the amount of human capital. Using UNESCO statistics they indicated that, although in a growing share of countries part of the working-age population will still be in school, this remains a minority. Furthermore, they find in the PWT 8.0 version that the average years of schooling for the 25+ category is lower than that of the 15+ category. In addition, they found this to be most profound in countries with relatively low average years of schooling, so where only few of the 15 to 25 year olds will be in school. Here data was collected from Barro and Lee (2012) on both the average years of schooling for the 15+ (S15) and 25+ (S25) age categories. The data is available 5yearly and the additional years are estimated using linear interpolation. The average years of. 21. The chain series method is a technique used to change raw data (reflecting both production volume and price changes) to (real) data that only reflect production volume. This is done by computing the production volume for each year in the prices of the preceding years and (chain) linking them together to create one time series free of price changes. 22 Since population data is in thousands..

(20) 19 University of Groningen, 24-11-2013 schooling for the 15+ age category will be used in the main results and, as a robustness check, these results will be compared to the results using the 25+ age category. 3.1.3. Physical capital. For the physical capital measure (K) the perpetual inventory method (PIM)23 is used as a means to cumulate and depreciate past investments. Capital stock at time t can be calculated using the following perpetual inventory equation: = = + ( – ?) . (10). where It is investment and δ is the depreciation rate. Real aggregate investment in PPP (It) is computed, as in Caselli (2005), as real income per capita (RGDPL)24 times population (POP) times a thousand25 times investment share in total income (KI). RGDPL is PPP converted GDP per capita (obtained with the Laspeyres method26). Data are collected from PWT 7.1. Moreover, Caselli (2005) sets δ = 0.06 for the aggregate investment (constant across countries and time). The choice of δ is important for the rest of the exercise, since the rate of depreciation affects the initial and final capital stock levels. If e.g. the estimate would be too high, the initial capital stock would be less and so would the capital stock in the years that follow. Data on the rate of depreciation is unfortunately not widely available. More specifically, data for the developing world is lacking altogether. Hence, when studying developing countries, researchers make assumptions regarding the rate of depreciation. Nehru and Dhareshwar (1993) mention a level between 0.03 and 0.04 for developed countries (based on Romer, 1989) and a level between 0.04 and 0.05 for developing countries (rough estimate), because they expect higher total depreciation to output ratios in developing countries. A reason for expecting higher depreciation rates in developing countries (compared to developed countries) is a lack of proper maintenance27 and the subsequent premature disposal of capital. Since variation in the depreciation rate may cause variations in capital stock estimates, the estimates of Nehru and Dhareshwar (1993) will be used as a robustness check. Furthermore, since capital stock in year t (Kt) depends on capital stock in year t-1 (Kt-1), the initial capital stock (K0) needs to be calculated separately. Following the approach of Harberger (1978), the most popular approach among researchers, Caselli (2005) computes the initial capital stock K0 as I0/(g + δ), where I0 is the value of the investment series in the first year it is available, and g is the average geometric growth rate28 for the investment series between the first year with available data and 1970 (average growth). Here, Caselli deviates from the original Harberger method and applies the modification suggested by Nehru and Dhareshwar (1993)29. The rationale behind the formula is that I/(g + δ) is the expression for capital stock in the steady state of the Solow model. This means that the assumption is made that all economies were in a steady state in the first year of data availability and that investment shows a sensible steady-state 23. Goldsmith (1951) pioneered with this method, where stock estimates are derived from long-term investment series at constant prices. 24 Real income per capita (RGDPL) is obtained using the Laspeyres method. 25 Since population data is in thousands. 26 The total value of GDP is calculated holding prices constant at their first year levels. 27 A common problem in developing countries, especially for public capital (Worldbank, 1994). 28 Geometric growth is growth for which the increases per time period are constant. 29 Nehru and Dhareshwar (1993) themselves used data till 1973 to derive the level of investment in the initial period. They made this choice because Chow tests showed that 1973 represented an important structural break for 82 of the 92 countries in their sample. Here Caselli (2005) will be followed..

(21) 20 University of Groningen, 24-11-2013 growth rate. Inklaar and Timmer (2013) deviate from this steady state relationship of the Solow growth model and suggest the usage of the initial capital/output ratio instead. This, because according to them this method leads to more plausible results in earlier years (for all countries) and for transition economies (for all years). Initial capital stock K is then measured as K0 = Y0 k, where Y0 is GDP in the initial year and k is the assumed capital/output ratio K/Y. Inklaar and Timmer (2013) also deviate from Caselli (2005) by using investment by asset instead of aggregate investment and by using a depreciation rate that varies across assets. Because asset-specific data are not widely available for Egypt and Turkey, this paper will not try to attempt a replication of their approach. As a comparison, over the entire period of available data (1960-2010) K/L growth rates of Egypt calculated with the PWT8.0 capital (rkna) and employment data (emp)30 are approximately equal to the growth rates calculated using the method presented here. The respective average annual rates being 4.83 and 4.80 percent31. For Turkey, over the period 1950-2010, the rates using the PWT8.0 methodology are lower than those calculated with the method used here, respectively annual growth rates of 4.31 and 6.43 percent32. Holding other factors constant, higher capital growth rates, would mean a larger calculated contribution of capital accumulation to growth and hence a lower calculated value of the residual (TFP). The choice of the approach of this paper over the one presented in Timmer and Inklaar (2013) thus potentially underestimates the level of TFP growth. 3.1.4. Data on α. Traditionally, the capital share (α) is measured as aggregate income minus worker compensation (labor share = 1 - α), making the assumption that output elasticity’s equal factor income shares. Caselli (2005) uses US time series data on the capital-share (calculated using worker compensation measures), whose long-run (and roughly constant) average value is 1/3. However, when it comes to cross-country variation, factor shares are not as stable as in the US. Moreover, the labor shares that were (traditionally) used to measure capital shares exclude the labor component of the selfemployed. Therefore, the approach of this paper will deviate from Caselli (2005) and (where possible) follow the suggestion of Gollin (2002) to use labor share measures including the labor component of self-employment income. Inklaar and Timmer (2013) mention that labor force data suggest that the exclusion of this component may be particularly important in poor countries due to the large fractions of small enterprises and self-employed in their workforce. They also provide adjusted calculations themselves of what is often called the ‘naïve’ (traditionally calculated) share and find that these adjustments are consistent with earlier claims made that factor shares are more or less constant across time and space33. Appendix D provides an overview of the different adjustments to the ‘naïve’ share that are suggested in the literature and used in this paper. 3.2 Accounting model Following the production function including embodied human capital, as presented in section 2.2.2, growth accounting will be used to decompose the growth of GDP per worker. GDP per worker 30. Rkna is the capital stock at constant 2005 national prices (in mil. 2005US$) and emp (employees) is the number of persons engaged (in millions). 31 For the sub-periods 1960-1991 and 1991-2010, the PWT8.0 calculations were respectively lower (4.72 versus 5.81) and higher (4.93 versus 3.24) than the calculations using the method of this paper. 32 For the sub-periods 1950-1980 and 1980-2010, the PWT8.0 calculations were respectively approximately equal to (4.57 versus 4.72) and lower (4.06 versus 8.02) than the calculations using the method of this paper. 33 For more information on the specific adjustments made and their procedure in selecting the best alternative for each country see Inklaar and Timmer (2013), page 16-22..

(22) 21 University of Groningen, 24-11-2013 growth of Egypt and Turkey will be decomposed into the contribution of physical capital, human capital, and TFP. Following equation (6): @ABCDA AE F#0GH GHFH = ,. (11). @ABCDA AE #DIH GHFH = ( − ) # ,. (12). @ABCDA AE %JK = = − − ( − ) # ,. (13). where t represents the different time periods as discussed in section 2.4.2. 3.3 Main results growth accounting This section will present the main results of the accounting exercises as well as a brief comparison to the results in the existing literature. A more extensive comparison of the results and methodology to literature will be done in Appendix H. The main results of the growth accounting exercise are presented in table 1 (Egypt) and 2 (Turkey) and are based on the (preferred) use of LS2 for Egypt and LS4 for Turkey (see Appendix D), human capital levels calculated using average schooling of the 15+ age category, and a depreciation rate (δ) of 0.0634. As a start, figure 3 and 4 present the indexed growth figures (based on this preferred setting) of y, k, and hc for, respectively, Egypt and Turkey.. Figure 3 – Growth of y, k, hc, and TFP in Egypt (indexed values) 1600.00 1400.00. Index, 1950=100. 1200.00 1000.00 y 800.00. k hc. 600.00. tfp 400.00 200.00. 34. 2010. 2007. 2004. 2001. 1998. 1995. 1992. 1989. 1986. 1983. 1980. 1977. 1974. 1971. 1968. 1965. 1962. 1959. 1956. 1953. 1950. 0.00. Changes in terms of the depreciation rate (to either 0.04 or 0.05) alter the results only slightly and do not lead to any changes in terms of the interpretation. Hence these robustness results will not be presented in the paper..

No results found