We examine the effect of both CAF/CAD development strategy, and AfT on structural change in production by drawing from the relatively scant literature on the macroeconomic determinants of structural change (e.g., Dabla-Norris et al. 2013; Duarte and Restuccia 2010; McMillan et al. 2014; Jha and Afrin 2017; Herrendorf et al. 2014; Martins 2018). In addition to the two variables of interest in the analysis, namely Aid for Trade and the indicator of the development strategy based on comparative advantage, the analysis also considers a number of control variables that are expected to influence the effect of these two variables of interest on the dependent variable. These controls include the trade policy stance (“TP”); the development level (“GDPC”) measured by the real per capita income, which also captures the domestic demand pattern; the education level (“EDU”); the financial development depth (denoted “FINDEV”); countries’ physical fundamentals, such as the share of arable land in total land in a given country (“SHARABLE”), the population size (“POP”), and the institutional and governance quality (“INST”).
Dabla-Norris et al. (2013) have examined the determinants of structural change in production by using countries’ fundamentals as well as policy and institutional factors to explain the differences in output structures. In particular, the authors have documented stylized facts on the process of structural transformation around the world by using output shares (sectoral real value added by economic activity, including agriculture, manufacturing and services) as a proxy for structural change. Along the same lines, Jha and Afrin (2017) have examined the patterns of structural transformation in African countries, including by considering the evolution and determinants of the shares of agricultural, manufacturing and services in total output for 53 African countries. McMillan et al. (2014) have subsequently considered the determinants of structural change, but also provided empirical evidence that structural change contributed positively to Africa’s overall productivity growth. The authors have proxied structural change by the labour reallocation effect. Specifically, using data on economy-wide, and sectoral labour productivity, the authors have decomposed the change in the economy-wide productivity into two components. The first component is the ‘‘within” component of productivity growth, i.e., the weighted sum of productivity growth within individual sectors, where the weights are the employment share of each sector at the beginning of the time-period. The second component referred to as ‘‘structural change” term reflects the productivity effect of labour re-allocations across different sectors. In the spirit of the work by McMillan et al. (2014), Martins (2018) has examined the determinants and patterns of structural change by using the Shapley decomposition developed by Shorrocks (2013) to extract the indicator of structural change. Herrendorf et al. (2014) have studied structural transformation by developing a multi-sector extension of the one-sector growth model (that encompasses the main existing theories of structural transformation), which serves as a natural benchmark to study structural transformation and that it is able to account for many salient features of structural transformation. Duarte and Restuccia (2010) have developed a model to measure structural transformation so as to examine the role of structural transformation (the secular reallocation of labour across sectors) on aggregate productivity.
The current analysis has used two measures of the extent of structural change in production across several production sectors in the economy. The first and main measure (henceforth referred to as SCINAV) is a metric-based structural change index, which relies on the Norm and Absolute Value Index. The second indicator used for robustness check analysis is the Modified Lilien Index of structural change, henceforth referred to as SCIMLI. Both indicators have been computed using the disaggregation of United Nations Data on Sectoral Value Added. In particular, we use the following sectoral disaggregation (four sectors are considered). Sector 1: agriculture, hunting, forestry and fishing; Sector 2: manufacturing; Sector 3: mining and utilities and Sector 4: construction, wholesale, retail trade, restaurants and hotels, transport, storage and communication and other activities, which we sum up to obtain the “Service” sector.
The SCINAV indicator, which has also been used in other studies (e.g., Productivity Commission 1998; Bacchetta and Jansen 2003; Cortuk and Singh 2011; Dietrich 2012; and Fiorini et al. 2013) reflects the extent of structural change across several production sectors in the economy. In the literature, it is also referred to as the Michaely Index (Michaely 1962) or Stoikov Index (Stoikov 1966). It is the most prominent and also simplest measure of structural change, and is derived from a metric-based approach of measuring structural change. It summarizes here the changes in sectoral composition of an economy between two points in time.
The SCINAV indicator is defined as follows: \({\text{SCINAV}} = 0.5\sum\nolimits_{i = 1}^{n} {|x_{it} - x_{is} |}\), where \(x_{it}\) is the share of sector i at time t and \(x_{is}\) is the share of sector i at time s. Hence, the differences of the sector shares \(x_{i}\) are first calculated between two points in time (s and t). The absolute amounts of these differences are summed up and divided by two (since each change is counted twice). The SCINAV indicator has been computed using non-overlapping 3-year averages of data on sectoral shares (of total output) in order to capture medium term effects over the period 1996–2016 (see the sub-periods below). As such, the SCINAV values range between zero and unity, which facilitates its interpretation. For example, the amount of structural change exactly equals the share of the movements of the sectors as a percentage of the whole economy (see Dietrich 2012). If the structure remains unchanged, the index is equal to zero, and if the whole economy undergoes a total change (the change in all sectors is at its highest), then the index is equal to unity.
The Modified Lilien Index of structural change (SCIMLI) is derived from an axiomatic analysis of SCI (see details on Dietrich 2012). The Lilien (1982) Index originally measured the standard deviation of the sectoral growth rates of employment from period s to period t. Stamer (1998) modified this index in order to fulfil the characteristics of a metric. The SCIMLI indicator is defined as follows: \({\text{SCIMLI}} = \sqrt {\sum\nolimits_{i = 1}^{n} {x_{it} .x_{it'} .\left( {\ln \frac{{x_{it} }}{{x_{it'} }}} \right)^{2} } }\), with \(x_{it}\) > 0 and \(x_{it'}\) > 0. \(x_{it}\) is the share of sector i at time t and \(x_{it'}\) is the share of sector i at time t′. According to this index, the influence of sector i grows in proportion to its size as well as to the value of its relative growth. Like for the SCINAV indicator, the SCIMLI indicator has also been computed using non-overlapping 3-year averages of data on sectoral shares so as to capture medium term effects. A rise in the values of SCINAV and SCIMLI indicators reflects a greater extent of structural change across production sectors, while a decrease in the values of these indicators indicate a lower extent of structural change across production sectors.
The variable “TCI” stands for the proxy for the development strategy based on comparative advantage (i.e., CAF or CAD) adopted by a given country (see for example Lin and Liu, 2004 and Bruno et al. 2015). Following for example, Lin and Liu (2004) and Bruno et al. (2015), we calculate TCI using the formula: \({\text{TCI}}_{it} = \frac{{{\raise0.7ex\hbox{${{\text{AVM}}_{it} }$} \!\mathord{\left/ {\vphantom {{{\text{AVM}}_{it} } {{\text{LM}}_{it} }}}\right.\kern-0pt} \!\lower0.7ex\hbox{${{\text{LM}}_{it} }$}}}}{{{\raise0.7ex\hbox{${{\text{GDP}}_{it} }$} \!\mathord{\left/ {\vphantom {{{\text{GDP}}_{it} } {{\text{L}}_{it} }}}\right.\kern-0pt} \!\lower0.7ex\hbox{${{\text{L}}_{it} }$}}}}\), where \({\text{AVM}}_{it}\) is the added value of manufacturing industries of a given country i, at time t; \({\text{GDP}}_{it}\) is the total added value of the country i; \({\text{LM}}_{it}\) stands for the labour in the manufacturing industry, and \({\text{L}}_{it}\) is the total labour force. For a given country, a rise in the values of the TCI indicator reflects the fact that this country follows a CAD development strategy by investing in heavy (capital-intensive) manufacturing industries, whereas declining values indicate that the country follows a CAF development strategy. The numerator of the TCI (i.e., \({\raise0.7ex\hbox{${{\text{AVM}}_{it} }$} \!\mathord{\left/ {\vphantom {{{\text{AVM}}_{it} } {{\text{LM}}_{it} }}}\right.\kern-0pt} \!\lower0.7ex\hbox{${{\text{LM}}_{it} }$}}\)) is relatively larger when manufacturing firms experience large market shares due to the government’s intervention (with access to subsidized credit and inputs, and very high profits lead to higher investment into capital) and as a result, where the value added generated by the sector is above what would be generated otherwise. Concurrently, as such a strategy leads to a distorted sector where capital-intensive technologies are the government’s priority, the sector would employ less labour. As noted by Bruno et al. (2015, p. 134), the TCI indicator reflects a situation where a government tries to kick-start economic growth through policies supporting a capital-intensive manufacturing sector.
The variable “AfT” is the measure of the real gross disbursements of AfT flows. It could either be real total gross disbursements AfT flows (constant US dollar 2016 prices), and denoted “AfTTOTCST” or its components, namely the real gross disbursements Aid for Trade flows for economic infrastructure (constant US dollar 2016 prices) denoted “AfTINFRACST”, the real gross disbursements of Aid for Trade flows for building productive capacity (constant US dollar 2016 prices) denoted “AfTPRODCST”, and the real gross disbursements of Aid for Trade flows allocated for trade policies and regulations (constant US dollar 2016 prices) denoted “AfTPOLCST”. Data on gross disbursements of AfT is available in OECD/CRS database from 2002 onwards. In particular, when the current study was being written, the data available covered the period 2002–2016. As this period is relatively short to capture the extent of structural change in production (it takes time for countries to change the sectoral structure of output) in AfT recipient countries, we rely on AfT commitment data that covers the period 1995 onwards, and adopt the approach used in Selaya and Sunesen (2012, p. 2158) to expand our AfT data so that it covers now the period 1996–2016 (this approach has also been used in Clemens et al. 2012; Thiele et al. 2006). The approach assumes that the proportion of AfT actually disbursed to sector “x” (\({\text{AfT}}_{x}\)) (for example, AfT disbursed for economic infrastructure; productive capacity building; and trade policies and regulations) during a given period is equal to the proportion of aid committed to sector x during this period, and is hence given by \({\text{AfT}}_{x} = \frac{{{\text{Commit}}_{x} }}{{\mathop \sum \nolimits_{x} {\text{Commit}}_{x} }}\mathop \sum \nolimits_{x} {\text{AfT}}_{x}\), where \({\text{Commit}}_{x}\) stands for the amount of real AfT commitments (constant US dollar 2016 prices) to sector x; \(\mathop \sum \nolimits_{x} {\text{AfT}}_{x}\) is the total amounts of AfT commitments and disbursements (constant US dollar 2016 prices) received during each period, respectively. While there may be some concerns about the approximation of sectoral disbursements with sectoral commitments because of differences in definitions and statistical record (see Clemens et al. 2012 for more details), Odedokun (2003) and Clemens et al. (2012) have noted that this problem is likely to be small since aid disbursements and commitments (both on the aggregate and sectoral levels) are highly correlated. Using this formula and based on AfT commitments and disbursements (constant US dollar 2016 prices) extracting from the OECD/CRS database (see Appendix 1 for more details), we have calculated for each country in the sample, and for each year, from 1996 to 2001, data on gross disbursements of AfT for economic infrastructure, gross disbursements of AfT for productive capacity building, and gross disbursements of AfT for trade policies and regulations. This data has been merged with the available dataset on OECD/CRS database on these three types of AfT flows over the period 2002–2016, so as to obtain the dataset of 81 countries over the period 1996–2016, used in the current analysis.
Against this background, we posit the following model:
$$\begin{aligned} { \log }\left( {SCI} \right)_{it} & = \varphi_{0} + \varphi_{1} { \log }\left( {\text{SCI}} \right)_{it - 1} + \varphi_{2} {\text{TCI}}_{it} + \varphi_{3} { \log }\left( {\text{AfT}} \right)_{it} + \varphi_{4} {\text{TP}}_{it} + \varphi_{5} {\text{EDU}}_{it} \\ & \quad + \varphi_{6} { \log }\left( {\text{GDPC}} \right)_{it} + \varphi_{7} {\text{FINDEV}}_{it} + \varphi_{8} { \log }\left( {\text{POP}} \right)_{it} + \varphi_{9} {\text{SHARABLE}}_{it} + \varphi_{10} {\text{INST}}_{it} + \vartheta_{i} \\ & \quad + \omega_{t} + \varepsilon_{it} , \\ \end{aligned}$$
(1)
where i represents a country’s index; t denotes the time-period. The panel dataset used to estimate model (1) contains 81 AfT recipients over the period 1996–2016. The variable “SCI” is the dependent variable, which is primarily measured by the indicator “SCINAV”, and for robustness check by the indicator “SCIMLI”. “TCI” is the indicator of the development strategy. “AfT” is the measure of real gross disbursements of AfT flows. The description and source of all control variables are provided in Appendix 1. Data on variables has been averaged over non-overlapping sub-periods of 3-year average so as to reduce the influence of business cycles on variables, as well as to obtain medium-term effects of regressors on the dependent variable. The sub-periods include 1996–1998; 1999–2001; 2002–2004; 2005–2007; 2008–2010; 2011–2013; and 2014–2016. The dependent variable “SCI” represents the extent of structural change in production. We have applied the natural logarithm to this variable so as to limit its high skewness. Likewise, the natural logarithm has been applied to the variables “AfT”, “GDPC”, and “POP” in order to limit their high skewness. The one-period lag of the dependent variable has been included in model (1) as a regressor so as to capture the initial level of structural change in production as well as the persistence of this variable over time. Model (1) features countries’ fixed effects (\(\vartheta_{i}\)) and time effects (\(\omega_{t}\)). The latter represent global shocks that could affect the process of structural change in production in all countries together. \(\varphi_{0}\) to \(\varphi_{10}\) are parameters to be estimated. \(\varepsilon_{it}\) is an error term. Appendix 2 shows descriptive statistics on the variables used in the analysis, while Appendix 3 presents the list of countries used in the analysis.
We shed light on how the key variables of interest, namely structural change indices (SCINAV and SCIMLI), TCI and the total AfT flows, in constant values (AfTTOTCST), have evolved over the period 1996–2016. To do so, we use the 3-year average data to provide in Fig. 1 the evolution of the two indices of structural change in production and TCI, while Fig. 2 displays the evolution of the two indices of structural change in production as well as the total real AfT flows. Furthermore, Fig. 3 has used SCINAV as the measure of structural change in production (which is our primary measure of structural change in production), and shows the correlation pattern (in the form of scatter plot) between this variable and TCI on the one hand, and between SCINAV (in Logs to limit its skewness) and the total real AfT flows (also in Logs to limit its skewness), on the other hand. We observe in Fig. 1 that the developments of SCINAV and SCIMLI are quite similar: the two indices (the extent of structural change in production) have remained relatively stable from 1996–1998 to 2002–2004, but have declined from 2002–2004 to 2008–2010. They have subsequently decreased from 2008–2010 to 2014–2016, thereby reflecting a decline in the extent of structural change in production. Likewise, the TCI indicator has increased from 5.4 in 1996–1998 to 8.5 in 2002–2004 (which reflects a tendency for the adoption of a CAD development strategy), and subsequently declined to reach 5.14 in 2005–2007 (thereby suggesting the adoption of a CAF strategy). From 2008–2010 to 2011–2013, values of TCI have risen from 6.9 to 8.8, and then fallen to 6.3 in 2014–2016. Overall, TCI and the extent of structural change indicators have not always evolved in the same direction over the period 1996–2016. It is worth noting that as per statistics reported in Appendix 2, we note that values of TCI range between 0.475 (for Tunisia in the sub-period 2014–2016) to 231.7 for (for Niger in the sub-period 2002–2004) (Fig. 4).
Figure 2 suggests that the total real AfT flows and structural change in production indicators have evolved in opposite directions. In particular, AfT flows have declined from US$ 392 million in 1996–1998 to US$ 118.1 million in 2002–2004. Real AfT flows have subsequently embraced a rising trend from 2005 onwards—from 2007–2009 to 2014–2016—(i.e., after the launch of the AfT Initiative in 2005). Finally, with regard to Fig. 3, we observe in the left-hand graph a negative correlation pattern between the total real AfT flows and SCINAV, and in the right-hand graph a positive correlation pattern between TCI and SCINAV. These tend to indicate that AfT flows are associated with a lower extent of structural change in production, while the CAD strategy is correlated with a higher extent of structural change in production. However, these patterns reflect simple correlations, but not causality.
Expected effects of control variables
We expect greater trade policy liberalization to be conducive to structural change in production through, inter alia, the diffusion of knowledge and technology from the direct import of high-tech goods and higher productivity (e.g., Grossman and Helpman 1991; Barro and Sala-i-Martin 1997; Baldwin et al. 2005) that would promote output and employment in high-productivity sectors. However, if greater trade policy liberalization leads countries to lock their production into sectors in which they have a comparative advantage (which is particularly the case for poor countries exporters of primary products), this would likely result in lower extent of structural change in production. Real per capita income (which acts as a measure of development level) is likely positively associated with the extent of structural change, as richer countries have a greater capacity, including human, physical and productive capacities to promote structural change in production compared to relatively less advanced countries. Human capital characteristics, including the education level reflects the average skill level of the workforce, and could induce a greater extent of structural change in production if this workforce is used in a wide range of high value-added productive sectors. However, if the bulk of existing workforce is used in existing activities (e.g., Agosin et al. 2012), this might result in lower extent of structural change in production. Limited access to affordable finance would constrain the ability of firms to produce and generate higher employment growth (Martins 2018), which would reduce the extent of structural change in production. A rise in the depth of financial development could allow greater diversification, risk sharing, and investment in higher productivity activities, and hence facilitate resource allocation across the economy (Levine 2005). This could, in turn, facilitate the development of activities with high value addition, and promote structural change in production across the various sectors of the economy. In the meantime, one could expect financial development to be associated with a lower extent of structural change in production (i.e., for example a concentration of production activities in few sectors of the economy) if banks are willing to finance only the development of activities in which the country has a comparative advantage. The size of the population aims to capture possible scale effects that can affect structural change in production. Countries’ fundamentals are captured through the inclusion of the share of arable land in total land in model (1).
Weak institutions divert resources from productive sectors to unproductive sectors, and hence promote rent-seeking activities (Iqbal and Daly 2014). Similarly, widespread corruption, inefficient bureaucracy, and a high risk of expropriation of private property by the government can create uncertainty among producers and discourage them from investing and innovating over the long term (Faruq 2011). In this context, better governance and institutional quality would be positively associated with the extent of structural change in production. However, better institutional and governance quality might be associated with a lower extent of structural change in production (i.e., a concentration of production in few production sectors of the economy) if countries’ activities are developed in sectors of comparative advantage.