2.1 Methodology

The method of growth accounting is built on Solow (1956), and an aggregate production function is used to decompose the growth rate of output into its components: a weighted average of the growth rates of the factor inputs and an remainder term – the Solow residual, which is commonly referred to as total factor productivity (TFP) and captures technological progress and other components of GDP growth that cannot be attributed to labor, human and physical capital.

The starting point is a Cobb–Douglas production function with constant returns to scale

where Yt is real GDP at time t, At is total factor productivity, Kt is the real capital stock, Lt is total employment, and Ht is an index of human capital. LtHt is therefore taking into account the education of the employed labor force, by weighting employment with a measure of human capital.

In particular, the skill-adjusted employment variable, LtHt, is generated using

where ϕ are average returns to education and st are mean years of education of the working-age population.

The real capital stock Kt is computed by first approximating the initial capital stock, and second employing the perpetual inventory method. A more detailed description of the construction of the human as well as physical capital variable can be found in the Appendix.

Taking the natural logarithm of the production function given in eq. [1] results in

and differentiating with respect to time t yields

where yˆt, aˆt, kˆt, lˆt and hˆt denote growth rates of real output, productivity, capital stock, employment, and the human capital index, respectively. α and (1−α) are factor shares in total income. ^[2]

Given real GDP, the capital stock, the labor force, and the human capital of an economy, one can determine the contributions of the production factors to output growth. The remaining Solow residual is generally attributed to the technology/productivity of a country. Determining the factors that cause TFP to change over time is the main focus of the following analysis.

2.2 Data and Results

The main data source for the growth decomposition is the World Bank’s World Development Indicator (WDI) database, which provides data on GDP, employment, and investment. For data on educational attainment, we use the Barro and Lee (2013) dataset. Table 1 shows the average growth rates of GDP, physical capital and human capital-augmented employment as well as average growth rates of TFP for the 12 MENA countries under consideration. Tables 1–3 in the Appendix illustrate yearly contributions of factor inputs and TFP to GDP growth. The decomposition shows that as the contributions of factor inputs were relatively stable over time, the considerable fluctuations of GDP growth rates are mainly attributed to TFP.

This result holds for Gulf Cooperation Council (GCC) countries, as well as for Mashreq and Maghreb countries. Physical and human capital and employment appear to have limited power in explaining output growth, and the better part of variation in output growth remains unexplained and is summarized in the residual TFP. In order to shed light on the determinants of TFP, a panel regression framework is employed that links the countries’ ability to adopt technology from abroad, their potential for domestic innovation as well as their openness to changes in TFP.

3.1 Theoretical Background and Econometric Specifications

As specified in eq. [1], a country’s output growth depends on the growth rates of physical capital, human capital, and the labor force. After accounting for these inputs of production, the remaining residual (Solow residual, TFP)

can be (partially) attributed to the level of technology of the country. The growth decomposition shows that in the MENA region this residual accounts for a considerable part of GDP growth rates (Table 1 and Figures 1–3), and changes in the growth rate of output are mainly attributed to the growth rate of TFP. In the following, we introduce the setting for estimating the determinants of TFP in order to find sources for its volatility and further explain what output growth depends on the MENA region.

Figure 1:

GCC countries: growth decomposition.

Figure 2:

Mashreq countries: growth decomposition.

Figure 3:

Maghreb countries: growth decomposition.

Human capital is assumed to be one of the most important drivers for technological progress. Mankiw, Romer, and Weil (1992) augment the standard Solow model and include human capital as an additional input factor in the production function. They show that a larger share of the variation in output can be explained by accounting for human capital as an input factor of production. We adapt this methodology and include human capital in the production function using a specification suggested by Hall and Jones (1999). Equation [5] shows that human capital is included in the growth decomposition framework as an input factor. Accounting for human capital as a production factor, however, does not imply that human capital is not a driver of productivity at the same time. The well-known contribution from Benhabib and Spiegel (1994) provides a theoretical background for this reasoning. They argue that the level of human capital increases the ability of a country to develop technologies domestically and, based on Nelson and Phelps (1966), to adopt new technologies from abroad. Furthermore, they state that the technology growth rate of a country depends on the gap between the technological leader and the country’s level of development. The following equation shows this relationship.

where Ait is TFP in country i in year t, and g(⋅) and c(⋅) are two functions with ∂g∂H≥0 and ∂c∂H≥0. The first term on the right-hand side in eq. [6] measures the extent to which human capital enhances TFP growth via domestic innovations, and the second term captures the catch-up effect with the leading country.

In order to test in a regression framework whether these effects apply to the MENA countries, we estimate the following relationship

where μi and νt are country and year specific fixed effects, respectively, Yi,t−1Ni,t−1 is lagged GDP per capita, and Xi,t−1 an additional set of lagged control variables that vary by specification. In order to reduce possible problems stemming from endogeneity, all explanatory variables enter in lags. ^[3] The coefficient of the human capital term, β3, is expected to be positive whenever the level of human capital in a country enables it to produce domestic innovations that boost productivity. The interaction term of human capital and the level of initial development, approximated by GDP per capita, allows for a nonlinear effect of human capital on productivity. The sign of β4 is expected to be negative, assuming that countries with initially low levels of development gain more from increases in the human capital stock.

The choice of additional control variables X is based on the findings of previous studies. Miller and Upadhyay (2000) study the effect of openness and trade orientation on TFP and find that openness increases TFP, if however, trade affects deviations from purchasing power parity the effect on TFP is negative. We use an index of globalization to proxy the outward orientation of a country and interact this variable with a country’s human capital. This specification allows to determine whether globalization increases TFP, conditional on a certain (threshold) endowment of skills/knowledge in an economy.

Danquah, Moral-Benito, and Ouattara (2014) apply a purely data-related approach to estimate the determinants of TFP. Similarly to growth regressions, a number of models can be applied when trying to assess the question of what drives TFP, and the “true” model is a-priori unknown. Different theoretical models suggest different variables to be included in the regression, and it is hardly possible to test for the validity of all the models in one estimation. For this reason, the authors apply the framework of Bayesian Averaging of Classical Estimates (BACE) to gain information about the main determinants of TFP. In the BACE approach, all possible models are estimated to obtain a distribution of the parameters. The effect on TFP attributed to a variable is then a weighted average of the effects that variable has in each single model. The results from Danquah, Moral-Benito, and Ouattara (2014) provide valuable insights, as they highlight the importance of controlling for country-fixed effects that capture country characteristics which do not change over time. Examples for such constant characteristics of a country are geographical or climactic factors, or cultural and institutional environments to the extent that they are persistent. Their findings further accentuate the importance of controlling for the openness of a country and the lagged level of GDP per capita.

3.2 Data

The dependent variable, the logarithm of TFP, is based on the growth decomposition discussed in Section 2.2 and in the Appendix. Data on GDP per capita is taken from the World Bank’s World Development Indicators. The share of individuals with secondary and tertiary education in the total working-age population (age group 15–64) is obtained from the Barro and Lee (2013) dataset on educational attainment. The IIASA-VID dataset (Lutz et al. 2007) would be preferable, but the coverage of the MENA countries is slightly better in the Barro and Lee (2013) dataset. To measure the dimension of globalization we use the KOF Index of Globalization (Dreher 2006) obtained from the Quality of Governance dataset (Teorell et al. 2011). This index accounts for the three main dimensions of globalization: (i) economic globalization, measured by flows of capital, goods, and services, (ii) political globalization, characterized by the diffusion of government policies and (iii) social globalization, that captures the spread of ideas, information, images, and people (Dreher 2006). Table 4 in the Appendix provides the means and standard deviations of the explained and the explanatory variables by country.

The various variables describing the business environment of the countries used in Section 4 are taken from the World Development Indicator database, so are the research and development-related variables and the measures that are used as proxies for infrastructure. The Corruption Perception Index (Transparency International) as well as the indices for regulatory quality, political stability and government effectiveness (Kaufmann, Kraay, and Mastruzzi 2009) are taken from the Quality of Governance dataset (Teorell et al. 2011).

The core econometric analysis relies on data for 11 MENA countries, ^[4] and covers the period between 1980 and 2009. Several gaps in the data cause the panel to be unbalanced. For Kuwait, data is missing during 1990 and 1994 as well as from 2008 onwards, and due to missing data on gross fixed capital formation, the United Arab Emirates and Yemen are not included before the early 1990s. The specifications that involve the KOF Index of Globalization are restricted to the period 1980–2006. In spite of the gaps, an appropriate number of countries representing the Maghreb, Mashreq, and GCC regions in combination with a sufficient length of the time series allow a robust empirical analysis of the drivers of TFP in the region.

3.3 Results

The results of the estimation of different model specifications are displayed in Table 2. All estimations control for country-specific fixed effects and include a full set of time dummy variables. Column (1) shows that the log of TFP is strongly persistent and thus the inclusion of a lag of the dependent variable as a regressor is required. The specification in Column (2) tests for the effect of income convergence in TFP dynamics by including the level of income per capita as an explanatory variable.

The stationarity of TFP indicates that TFP converges within countries, ^[5] but the level of income per capita does not appear to be a significant driver of productivity growth in the sample.

In Column (3) human capital is introduced as a potential driver for TFP. Human capital does not have a significant impact on TFP, a result that suggests that domestic innovations are not the driving force of productivity when the whole MENA region is considered. Benhabib and Spiegel (1994) find similar results for low- and middle-income countries. Only in high-income countries are domestic innovations found to provide significant contributions to productivity. They argue that for technologically advanced economies domestic innovation is more effective than technology adoption from abroad. Likewise, less developed countries do not have the prerequisites for effective domestic innovation and it is more efficient for them to adopt existing technologies from abroad. This is further confirmed in Column (4). Allowing for a nonlinear relationship between the share of the working-age population with secondary or tertiary education and TFP shows that only after reaching a certain threshold level of education, its contribution via domestic innovations turns positive. To compute this threshold value, the estimated equation

is differentiated with respect to H. Setting the first derivative to zero allows computing the minimum of the quadratic function, H∗, i.e. the value of the education level that is theoretically necessary to effectively develop technologies leading to productivity gains,

This threshold lies at 67% of secondary and tertiary education, on average. Figure 4 in the Appendix shows the relationship graphically. How this threshold relates to the countries under consideration is summarized in Table 3. The GCC countries (in our sample: UAE, Bahrain, Kuwait, and Saudi Arabia) show the highest average level of educational attainment in the working-age population since the 1980s. However, while Bahrain reached the critical level of education already in the mid-1990s, the UAE did so only during the last few years. Educational attainment in Kuwait is below the level that is necessary to positively contribute to productivity through the channel of domestic innovation in the country. The average educational attainment in the Mashreq region (in our sample: Egypt, Jordan, Syria and Yemen) since the 1980s is clearly below the GCC average; however, the sharp increase in educational attainment in Jordan enables that country to effectively adopt and implement new technologies. Syria and Yemen both exhibit education levels that do not suffice for a positive contribution of domestic innovation on TFP. Concerning the Maghreb region, which is represented by Algeria, Morocco and Tunisia in our sample, our model predicts that the level of human capital in those countries is not high enough to effectively innovate domestically.

The specification in Column (5) includes an interaction term of the level of a country’s development (approximated by GDP per capita) and the human capital variable. This allows to test whether human capital speeds up technology adoption (that is, whether countries with relatively low-income levels tend to benefit more from human capital accumulation in terms of technology adoption). The results show that, holding the level of development constant, countries with a higher level of education can adopt technology faster and thus are better able to catch up with leading countries. To put these results differently, countries with low levels of development can benefit from improvements in human capital to a larger extent.

Column (6) relates the level of a country’s globalization to its TFP. The insignificant coefficient of the measure of globalization suggests no relationship between technology and a country’s TFP growth. When interacting the globalization variable with the human capital variable, it turns out that the effect of the globalization intensity appears to depend positively on the level of human capital (Column 7). Comparing two countries with the same level of human capital, the one with the higher index of globalization is able to adopt more technology from abroad and thus, holding other factors constant, has a higher level of TFP.

Applying the same logic as above when analyzing Column (4), we estimate that whenever the share of the secondary or tertiary educated working-age population exceeds approximately 27%, globalization speeds up the process of technology adoption and thus, increases TFP and finally output growth. Although the threshold number must be interpreted with care as it changes when the model specification is changed, the educational attainment necessary is comparably low and all countries in our sample, with the exception of Yemen, have the required level of human capital to benefit from globalization (see Table 3 for an overview of the educational attainment of the MENA countries).

4.1 Business Environment and Infrastructure

The business environment is of particular interest when explaining productivity and technological progress of a country. In the last years, great efforts have been made to quantify business environment-related factors in a way that allows comparison across countries. However, the number of observations is not large enough to estimate the impact on TFP in an econometric model and thus we cannot establish causal relationships. Simple scatter plots that show the correlation between the (post-2000 country means of) TFP growth and the cost of, and the time needed for, starting a new business, and furthermore the World Bank’s logistics performance index and the country’s road density are displayed in Figure 5 in the Appendix. Countries in which starting a business is relatively cheap and not very time-intensive tend to exhibit higher growth rates of TFP, whereas countries such as Yemen and Djibouti, where business start-ups are both expensive and time-intensive, are associated with lower growth rates of TFP. ^[11] The logistics performance index used in the analysis evaluates the trade and transport-related infrastructure (e.g. ports, railroads, roads, information technology). Low values of this index indicate a poor infrastructure, high values a good one. The correlation with TFP growth is slightly positive, although not as strong as for the variables measuring the ease of starting a business. The lower right graph displays a plot of TFP growth against the mean of the road densities and shows a highly positive relationship.

4.2 Political Stability and Institutions

Figure 6 displays scatter plots and linear trends that relate TFP growth to political and institutional quality indices, all obtained from the Quality of Governance Dataset (Teorell et al. 2011). The upper-left figure shows the positive relationship of TFP growth and the corruption perception index from Transparency International that assigns higher values to countries in which the perceived extent of corruption is low. The three remaining figures relate institutional characteristics measured by a selection of the World Bank Governance Indicators (Kaufmann, Kraay, and Mastruzzi 2009) to changes in TFP. The scatter plots show that less corruption, better regulatory quality, a higher degree of government effectiveness and a stable political environment are associated with higher growth rates of TFP. As argued by Edwards (1998), better institutions and a sound political environment encourage innovations and promote productive activities.

Further investigation of the scatter plots reveals that the indices themselves are highly positively correlated and countries with high levels of perceived corruption are those exhibiting low regulatory quality, low government effectiveness and only limited political stability. Examples are Yemen or Algeria, countries that are placed in the lower left part in each of the graphs, or Bahrain, on the other hand, that is placed in the upper right corner of each graph where data is available.

4.3 Research and Development

The last set of graphs, Figure 7, links TFP growth to the share of researchers and technicians working in research and development as well as to the research and development expenditures (as % of GDP). The relationship between the former two is slightly positive, but mainly driven by the country of Jordan, that has a particularly high share of researchers and technicians. When it comes to research and development expenditures, Morocco and Tunisia invest the highest share of their GDP in R&D, but they do not reach the TFP growth rates obtained by Kuwait or Jordan.