1 Introduction

Recent history has shown that high growth on its own does not necessarily lead to good social outcomes. It matters if growth is inclusive or not. Thus, inclusiveness is an essential ingredient of any successful growth strategy. The concept has yet to be rigorously documented, as it is used to convey several aspects of growth. Numerous papers have proposed different definitions. For instance, the Commission on Growth and Development (2008) notes that inclusiveness of growth is associated with equity, equality of opportunity, and protection in market and employment transitions. In a similar vein, Ianchovichina and Lundstrom Gable (2012) define inclusive growth as rapid, broad-based across sectors and sustained growth that is inclusive of a large part of a country’s labor force. According to this definition, to be inclusive a growth path needs to be strong, pro-poor and redistributive, and further geared toward improving productive employment. Recently, Anand et al. (2013) refer to both the pace and distribution of economic growth to define inclusive growth, on the ground that for growth to be sustainable and effective in reducing poverty, it needs to be inclusive.

This paper argues that all these aspects of inclusive growth have a common denominator, dubbed the “quality of growth”. A necessary condition to achieve all these different approaches of inclusive growth is “good quality growth”. Good quality growth is seen as high, durable, and socially-friendly growth. There is a consensus that high growth over the long run is necessary to achieve lasting improvements in social outcomes, but it is increasingly evident that high growth alone may not be sufficient in many cases. For instance, over the past few decades, many developing countries experienced strong growth episodes in the context of relative macroeconomic stability, sound policies, and strengthening institutions. However, relatively few posted significant declines in poverty, inequality and unemployment (see Dollar and Kraay 2002; Dollar et al. 2013). It is therefore relevant for policy makers and academic professionals to assess whether the underlying “quality” of growth has been good, and has fundamentally been pro-poor.

This paper builds on Martinez and Mlachila (2013) who explored the quality of the recent high-growth episode in sub-Saharan Africa. They delved into the concept of good quality growth, defining it as one that is strong, stable, sustainable, increases productivity and leads to socially desirable outcomes, like improved standards of living, especially in the reduction of poverty. They draw on an extensive literature that discusses the various aspects of growth and how they matter (e.g., Berg et al. 2008; Johnson et al. 2007 on growth sustainability; Badinger 2010; Ramey and Ramey 1995 on growth volatility; Loayza and Raddatz 2010 on sectoral aspects of growth and poverty reduction; Yang 2008 on outward orientation etc.).

The main objective of this paper is to introduce an index of the quality of growth. The proposed quality of growth index (QGI) encompasses both the intrinsic nature of growth and its social dimensions. Following Martinez and Mlachila (2013), a good quality of growth—more than just its (high) level—is important to enhance living standards and welfare, and to create opportunities for better employment. Thus how growth is generated is critical for its sustainability and for accelerating employment creation and poverty reduction. Our index attempts to capture the multidimensional features of growth. That said, it also probably matters how the fruits of growth are managed by the state, notably through redistribution and public services.

Our QGI goes beyond the well-known Human Development Index (HDI) developed by the United Nations by concentrating not just on the levels of incomes, but the very nature of growth.Footnote 1 We argue that it does matter how one reaches from level L1 to L2 of income for various theoretical and empirical reasons that are elaborated on below. Arguably, since it is income level-based, the HDI is the result of millennia of growth. On the other hand, our index facilitates the assessment of the quality of various episodes of growth both within a country and across regions. There is ample evidence that not all growth is created equal: growth that is strong, stable, sustainable, increases total factor productivity, is broad-based sectorally, and export-oriented is likely to be more efficient in fostering socially desirable outcomes.

Our QGI is also different from the recently developed Social Progress Index (Scott et al. 2014), as the latter focuses more on aspects that are close to the social dimension of the QGI, without accounting for the growth fundamental aspect, which is the core of the QGI.Footnote 2 That said, it is worth mentioning that there is a dynamic and complementary relationship among social indicators such as education and health, and growth. Empirical evidence confirms this two-way relationship between investment in education and health, and growth (Bils and Peter 2000).

The paper’s key contribution is the rigorous development of the QGI, covering a wide panel of developing and emerging countries over 1990–2011. This allows us to explore how it has evolved over time and whether there are important regional variations in the quality of growth. It thus enables us to determine whether there has been some convergence in the quality of growth over time, or whether there exists a growth “quality trap”. The paper also explores whether the quality of growth is related to other development indicators identified in the literature.

The baseline QGI is robust to a number of weighting schemes and inclusion of additional measures of social and inequality outcomes. While our preferred index is based on the idea of simplicity, transparency, and use of readily available data for the maximum number of developing countries, we conducted a number of robustness tests of various alternatives. The alternative weighting schemes include zero weighting of some elements of the index to address issues of possible correlation among the elements themselves. We generally find that the various computed alternative QGIs are highly correlated, thereby validating our baseline index.

The paper’s main findings are fourfold. First, the quality of growth has been improving in the majority of countries over the past two decades. Second, the rate of convergence is relatively slow. Third, there are considerable cross-country variations across income levels and regions. Finally, empirical investigations show that political stability, public pro-poor spending, macroeconomic stability, financial development, institutional quality and external factors such as FDI, are associated with higher QGI.

The paper is organized as follows. The next section describes in detail the steps involved in the creation of the index and introduces the dataset. Section 3 presents the computed index itself, shows some stylized facts and the country ranking, tests the presence of convergence in the QGI dynamics and explores the potential drivers of the QGI. Section 4 checks the robustness of the index by examining various computation approaches. Section 5 presents concluding remarks.

2 Methodology and Data

In order to evaluate the evolution of the quality of economic growth across countries and over time, we build a Quality of Growth Index (QGI). This section first sets forth the methodology used to derive the QGI along with its rationale, then discusses the sensitivity of the index to alternative assumptions, and introduces the dataset.

The QGI is a composite index, resulting from the aggregation of two building blocks: the intrinsic nature of the growth sub-index (“growth fundamentals”) and the social dimension sub-index representing the desired social outputs from growth, as illustrated by Fig. 1.

Fig. 1
figure 1

Conceptual representation of the quality of growth index. Note α and β represent the weights ascribed to growth fundamentals and social dimension in the QGI. γ 1, γ 2, γ 3 and γ 4 stand for the respective weights of the strength, stability, sectoral composition, and demand composition of growth in the growth fundamentals sub-component, while δ1 and δ2 are the weights assigned to health and education in the social dimension sub-component

Full size image

2.1 Growth Fundamentals

The sub-index for the intrinsic nature of growth encompasses four dimensions aiming at capturing the extent to which a given growth episode can be considered as of high quality, with regard to its (1) strength, (2) stability, (3) diversification of sources, and (4) its outward-orientation.

The strength of growth is measured by the annual change in real GDP per capita. We resort to GDP per capita instead of GDP, as the former is more in line with the concept of pro-poor growth which underlies the concept of quality of growth.Footnote 3 The strength of growth is an important aspect of the quality of growth since high growth is a necessary ingredient to put a dent to poverty (Dollar and Kraay 2002; Dollar et al. 2013). Accordingly, higher growth rate is expected to lead to substantial dent in poverty, and hence to a better QGI.

Stability can be approached from the perspective of the fluctuation of the level of growth. The literature offers several alternative proxies of growth stability. For instance, Guillaumont (2009) and Cariolle et al. (2016) build the Economic Vulnerability Index (EVI) using supply or demand shocks such as the instability of agricultural production or instability of trade volumes. We focus on the volatility of economic activity measured as GDP because we are mostly interested in the quality of the overall growth. This is in line with the mainstream literature on volatility and growth (e.g., Ramey and Ramey 1995). More specifically, the stability of growth is computed as the inverse of the coefficient of variation (CV) which is the ratio of the standard deviation over the mean. A 5-year rolling window is used to derive time-varying CVs.

Several papers have preferred the standard deviation or deviations from the trend (e.g., Collier and Dehn 2001; Cariolle et al. 2016). However, the standard deviation is scale-variant and does not facilitate cross-country comparisons. For our benchmarking purpose, a normalized and scale-invariant measure as captured by the CV is more appropriate. The CV offsets the apparent dispersion related to the level and allows a straight comparison of countries irrespective of the growth levels. It therefore allows smoothing out the influence of outliers such as small open economies that tend to be structurally more volatile or large countries with low growth that tend to be structurally less volatile. The higher the CV, the lower is the inverse of CV, and the less stable is the growth episode. However, growth instability is well-known to worsen poverty and equity, through a “hysteresis mechanism”. Indeed, swings in growth trajectories are particularly harmful to the poor, since the erosion of their human capital in “bad times” is not made up when the economy pulls out of the shock (Behrman et al. 1999; Ames et al. 2001; Guillaumont and Kpodar 2006). Accordingly, a stable growth episode is expected to feed positively into the QGI.

The diversification of sources of growth captures the extent to which growth is considered as generated by diversified sources. It is proxied by a diversification index computed as one minus a Herfindahl–Hirschman index (HHI) using exports data. The higher the index of diversification of export products, the more diversified are the sources of growth. The rationale of using the diversification of exports lies in the findings that that export diversification is associated with stronger growth and lower output volatility (Papageorgiou and Spatafora 2012), which are both conducive to higher QGI, as argued above.Footnote 4

The degree of outward orientation of a growth momentum is proxied by the share of net external demand—in percentage of GDP,Footnote 5 as opposed to the share of domestic demand. The rationale for this dimension builds on the fact that an outward orientation of growth is more likely to raise productivity growth through several mechanisms, including learning-by-doing processes, importation of more advanced technology, transfer of knowledge, the discipline of the world market, competition, and foreign direct investment (Diao et al. 2006). Note, however, that such an outward orientation of growth may increase the country’s vulnerability to external environment fluctuations and as such results in a more volatile and ultimately lower quality growth.Footnote 6

2.2 Social Dimensions of Growth

As mentioned above, a strong, stable, diversified, and outward oriented growth may prove insufficient in alleviating poverty substantially and improve living standards. In this paper, we focus on the level rather than the change because several authors show that it is the stock of better social conditions that matters most than the change when it comes to reducing poverty (see e.g., Anand and Sen 2000; Scott et al. 2014). They argue that human development, in the form of people being better educated, healthier, less debilitated, and so on, is not only constitutive of a better quality of life, but it also contributes to a person’s productivity and her ability to make a larger contribution to the progress of material prosperity. Hence, one could argue that what matters for the quality of growth is mostly the stock.

The pro-poor aspect of the quality of growth is therefore factored in its social dimension sub-index, through indicators capturing two of the most basic dimensions of human capital building. These include (1) a long and healthy life, and (2) an access to a decent education/knowledge,Footnote 7 which are both commonly acknowledged in the literature as key drivers of the changes in poverty levels (Schultz 1999).Footnote 8

The health component captures the extent to which a country’s population can enjoy a long and healthy life through the aggregation of two sub-components, namely: (1) the reverse of infant mortality rate; and (2) life expectancy at birth. Both these health measures are considered as key poverty symptoms, consistently with Amartya Sen’s biological approach of measuring poverty (Sen 2003).

Education is captured by the primary school completion rate. The main motivation for using only this indicator is data availability. Several variables could also be good proxies for a country’s educational level, including inter alia, the average years of schooling and the net primary school enrollment rate. However, the lack of observations on these variables over a long period for many countries, either in the well-known Barro and Lee (2013) database or the WDI dataset, prevents us from employing such variables. It is worth noting that missing observations also do exist in the primary school completion rate variable, but we retain this variable, as a “lesser evil”.Footnote 9 Nevertheless, to avoid reducing considerably the sample size owing to missing values from the primary school completion rate, we made some assumptions allowing us to generate and fill up these missing values, consistently with the approach set forth in “Appendix 2”.

2.3 Construction of the Index

The construction of the QGI follows a two-step approach: first the variables are standardized into indices of same scale, and then are aggregated into a single index using different weights.

2.3.1 Standardization of the Components

The different variables presented above and representing the different components of the QGI are not expressed in the same unit, which makes their aggregation into a single index comparable to a “mixing apples and oranges” problem. Two main approaches allow us to deal with this issue, namely, the centered-reduced normalization or Z-score approach, and the Min–Max approach. The former consists of transforming a given variable X characterized by its mean µ and standard deviation σ, into an index or Z score expressed as follows:\(Z = \frac{{\left( {X - \mu } \right)}}{\sigma }\). If X is normally distributed, then Z follows a centered-reduced normal distribution, with a zero mean and a unity standard deviation. With this standardization, all variables are expressed in the same unit, namely the standard deviation, and can therefore be meaningfully aggregated into a single index. But one matter of concern related to this approach is the sensitivity of the transformed Z variable to the presence of outliers. For example, small open economies (predominant in our sample) have much more volatile growth, implying higher values for σ compared to the rest of countries. This leads to a highly dispersed distribution of Z-score (unbounded by definition) and renders the standardization strategy less appropriate to rank countries.

The Min–Max approach also consists of transforming the variable X into an index Z′, according to the following formula: \(Z^{{\prime }} = \frac{{\left( {X - X_{\hbox{min} } } \right)}}{{\left( {X_{\hbox{max} } - X_{\hbox{min} } } \right)}}\), where X min and X max stand for the minimum (min) and the maximum (max) of X, respectively. Unlike the aforementioned Z variable, Z′ is bounded, ranging from 0 to 1, and is consequently less likely to have a highly dispersed distribution, rendering it more fit for the country ranking perspective of this paper. But a key issue in building Z′ relates to the choice of the minimum (X min ) and maximum (X max ) of X. What should be taken as the maximum value of life expectancy at life for example? Should it be what ideally desired (positive argument) is or rather the highest value actually observed in the considered panel of countries (objective argument)? Given the potential controversy surrounding the subjective or positive-based choice of “ideal” max and min for X, we base our standardization on the max and min actually observed in the sample. However, note that this choice has consequences on the construction of the QGI, since the QGI will be heavily influenced by how far countries stand relatively to the sample’s maximum or minimum. The results would therefore depend strongly on the sample used. This is somewhat mitigated by the use of a broad sample of developing and emerging countries with available data.

2.3.2 Weighting Approach

We assign equal weights (50 % each) to the intrinsic nature of growth (α) and to the social dimension of growth sub-indices (β), respectively. Equal weight (γ1 = γ2 = γ3 = γ4 = 25 %) is also given to the four sub-components of the intrinsic nature sub-index, so is for the two sub-components of the social dimension of growth, namely health and education (δ1 = δ2 = 50 %). Equal weight (50 %) is also assigned to the two sub-components of the health sub-index.Footnote 10 The main rationale for this weighting option, which is used in other well-known indices such as the Human Development Index (HDI) or the Economic Vulnerability Index (EVI), lies in its simplicity and transparency. But as well stressed by Guillaumont (2009), this equal weighting does carry a dose of arbitrariness, since the weights are determined by the number of components, and hence depends heavily on the components retained themselves.

Some alternative aggregation options (such as regression-based approach) exist in the literature but present several inconveniences for our purpose. The validity of the regression-based approach depends heavily on the quality of the regressions, including notably issues related to the endogeneity of the regressors. That said, in the robustness section, we make use of alternative weights for the intrinsic nature of growth and the social dimension of growth sub-indices.

2.3.3 Aggregation Approach

The QGI is calculated as the arithmetic mean—with equal weighting, but with alternative weighting subsequently, for robustness purpose, as mentioned before—of the intrinsic nature of growth and the social dimension of growth sub-indices. The same averaging approach is applied for these sub-indices themselves.Footnote 11 Simplicity and transparency are once again the main rationale for the choice of this aggregation strategy. But this strategy implicitly assumes the absence of interactions between the different components of the QGI, i.e., there is a substitutability relationship between the various components of the QGI. However, some complementarities may exist between the different components of the QGI, which would make the geometric averaging strategy more appropriate.Footnote 12 Indeed, a country’s level of human capital— education and health—may influence the productivity of its economy, and hence its growth pace, and vice versa. Such complementarities may also be at work between the sub-components of each of the two major sub-indices of the QGI. For the sake of ensuring that the construction of the QGI and its associated country ranking is not skewed by the chosen averaging approach, we resort, in addition to the arithmetic averaging, to a geometric approach.

To sum up, the calculation of the QGI can be formally written as:

$$QGI = \alpha \left( {Fundamentals} \right) + \beta \left( {Social} \right)$$
(1)

with “growth fundamentals” dimension defined as Fundamentals = γ 1 Level + γ 2 Stability + γ 3 Diversification + γ 4 Orientation and the “social dimension” defined as Social = δ 1 School + δ 2 Health.

Under the geometric averaging strategy carried out for robustness purpose, the QGI is defined as \(QGI = (Fundamentals)^{{\alpha_{1} }} (Social)^{{\alpha_{2} }}\) where \(Fundamentals = (Level)^{{\gamma_{1} }} (Stability)^{{\gamma_{2} }} (Diversification)^{{\gamma_{3} }} (Orientation)^{{\gamma_{4} }}\) and \(Social = (School)^{{\delta_{1} }} (Health)^{{\delta_{2} }}\).

2.4 Data

For this paper, we use a panel data covering 93 developing countries between 1990 and 2011. The sample includes 57 middle-income countries and 36 low-income countries. To smoothen out the effects of short-term fluctuations on macroeconomic variables, each variable has been averaged over 5-year (1990–1994, 1995–1999, 2000–2004 and 2005–2011). The variables used in this study are drawn upon from various databanks, including the IMF World Economic Outlook database, the World Bank’s World Development Indicators (WDI) database, COMTRADE, the International Country Risk Guide database, Barro and Lee (2013) and Sala-i-Martin (2006). Detailed sources and definitions of variables are provided in “Appendix 1”. “Appendix 2” elaborates on the specific case of dealing with missing observations in the primary school completion rate.

3 Results

In this section, we highlight our key findings. First, we present out some stylized facts of the QGI. Second, we rank the QGI and categorize countries based on their performances. We also assess the convergence hypothesis in the QGI. Third, we put the QGI into perspective with the existing development and living standard indicators, and explore the potential drivers of the QGI.

3.1 Some Stylized Facts

This section builds on the benchmark computed QGI to rank the full sample countries over 1990–2011. The ranking results are reported in Table 1 below. Over the most recent sub-period, namely 2005–2011, Bulgaria emerges as the top performer, with a QGI of 0.843, followed by China (0.842) and Argentina (0.830), while Chad (0.334), Central African Republic (0.402) and Niger (0.415) are the poorest performers, respectively.

Table 1 Full sample QGI-based ranking
Full size table

The QGI varies markedly across periods, countries, and income levels. The QGI has also improved over time (Fig. 2a). The average value of the QGI stands at 0.604 (Fig. 2). The minimum QGI is 0.258 for Niger over 1990–1994 and the maximum of 0.849 for China over 2000–2004. The QGI increases from 0.556 in 1990–1994 to 0.656 in 2005–2011. Moreover, a density plot shows that the distribution of the QGI is shifting to the right over time (Fig. 3a). At the same time, distributions have become narrower, denoting a certain level of convergence among countries over time.

Fig. 2
figure 2

QGI: Average values. a QGI by period, b QGI by region, c QGI by level of income, d QGI by fragility and resource-rich criteria. Note: AP, Asia and Pacific; CEE, Central and Eastern Europe; LA, Latin America; MENA, Middle East and North Africa; SSA, Sub-Saharan Africa; LIC, low-income countries; LMIC, lower-middle income countries; UMIC, upper-middle income countries

Full size image
Fig. 3
figure 3

QGI: Kernel density. a By period, b by region, c by income level, d fragility and resource-rich criteria

Full size image

Moreover, there are significant differences across regions. LA countries exhibit the highest QGI scores whereas sub-Saharan Africa lags behind with the lowest QGI (Fig. 2b). The density analysis shows that sub-Saharan African countries exhibit the flattest and leftmost density in their QGI distribution, with thick distribution tails (Fig. 3b). This suggests that sub-Saharan Africa is the poorest performing country group in terms of quality of growth. In addition, the flatter density along with the thick tails tend to signal the presence of greater inequalities in the QGI scores, with a few countries performing quite well, namely above the full sample average score, while the bulk of sub-Saharan African countries—roughly more than 60 % of observations—are left behind the full sample average score (0.604).

Furthermore, the QGI also varies by income levels (Fig. 2c).Footnote 13 The QGI is positively correlated with countries’ income level. The upper-middle income countries record the highest QGI score, followed by the lower-middle income countries and the low-income countries, respectively. A density analysis suggests that the richer is a country group, the more in the right-hand side of the figure its density curve stands, confirming a positive association between countries’ level of development and their ability to draw upon a better quality of growth from their growth momentum (Fig. 3c).

We also focus on the fragility status and resource endowment.Footnote 14 It emerges that fragile countries significantly underperformed the sample average by almost 16 % point (Fig. 2d). This suggests that fragile countries face tougher structural impediments when it comes to achieving better quality of growth. Likewise, resource-rich countries have QGI scores standing slightly lower than their non-resource rich peers, which may fuel the natural resource curse debate (Sachs and Warner 2001). From the density analysis (Fig. 3d), it clearly transpires that most fragile countries are concentrated in the left side of the distribution, with around 25 % of observations corresponding to a performance gap of as high as 0.2 point with regard to the full sample average, and around 40 % of observations corresponding to a QGI score higher than that of the former group, but lower than the full sample average (0.604). This confirms the previously-underscored message from Fig. 2d that fragility weighs severely on countries’ underperformance in terms of QGI. The distribution of QGI also shows that below the sample average, the density curve for resource-rich countries’ stands above the curve for non-resource rich countries, while the reverse is observed from the full sample average onwards. This therefore suggests that endowment in natural resources worked more as a curse rather than as a blessing for countries when it comes to achieving a better quality of growth.

3.2 Convergence Hypothesis

We first investigate the presence of convergence in the QGI process. To this end, we report in Table 2 below simple pooled OLS estimates linking the change in countries’ QGI to their past QGI. The results show that some convergence is at play in the QGI process. Past QGI performance, expressed either as the lagged (one period) value of the QGI or the initial (1990–1994) value of the QGI, is found to be negatively associated with the growth rate of the QGI. This is reflected by the negative and significant estimated coefficient of the lagged QGI and the initial QGI. This result therefore suggests that the least performing countries tend to catch up the best performers over time.

Table 2 Test of convergence hypothesis
Full size table

Second, building on this above-evidenced presence of convergence in the QGI process, we propose a categorization of countries, based on their QGI dynamics between the initial period (1990–1994) and the final period (2005–2011), as reflected in Fig. 4.

Fig. 4
figure 4

Quality of growth over time

Full size image

The sample average QGI score is underlined by the red horizontal and vertical lines representing the QGI values of the initial period and the final period, respectively. The dashed line represents the 45 degree line; the higher above this line, the greater the improvement in the level of the QGI. The intersection of these three lines yields 6 non-overlapping regions. It is worth mentioning that virtually all the countries have improved their quality of growth over the past two decades. Region 2 includes countries whose QGI scores in the starting period as well as in the final period stand below the sample average, but did improve between the two periods. This group is made up of low-income and/or fragile countries, mostly from sub-Saharan Africa and incidentally from MENA, which are labelled as the “hopefuls”, in that they have reasonably good prospects of converging progressively toward the sample average country. Region 3 includes countries that were able to improve their QGI from below to above the sample average between the starting and the final period, and as such are dubbed as the “contenders”, with a reference to the idea that they are contending to be among the high performers. This group includes mainly countries from Asia Pacific (for example Bangladesh, Laos and Nepal), from sub-Saharan Africa (for example Ghana,Tanzania and Zambia), from Middle East and North Africa (for example Algeria, Iran and Morocco), from Central and Eastern Europe (for example Azerbaijan, Tajikistan and Uzbekistan) and a few from Latin America (Guatemala and Nicaragua).

Region 4 encompasses countries that not only recorded a QGI score superior to the sample average, in the initial as well as in the final period, but also experienced an improvement in their QGI between the two periods. This country group is labelled as the “club of best performers”, and includes chiefly upper-middle and lower-middle income countries. A noticeable finding is that a handful of sub-Saharan African countries belongs to this club of best performers, including notably Kenya, Namibia and South Africa. Finally, region 5 is characterized by countries that performed well above the sample average in terms of QGI in both the initial and the final periods, but have the particularity of having experienced a mild drop in their QGI between the two periods. This group comprises only two countries, namely Botswana and Malaysia, and is categorized as the “club of superior performers”.

3.3 Putting the QGI into Perspective with Existing Development Indicators

To have an idea about where our proposed QGI stands compared to existing development indicators, we display in Fig. 5 correlations between the QGI and a selection of living standards variables. It appears that the QGI is positively correlated with the well-established United Nations (UN) HDI and real GDP per capita (albeit non-linearly), and negatively with the poverty rate and income inequality. These findings imply that the QGI could be another legitimate part of the toolkit available for gauging countries’ progress toward inclusvie growth.

Fig. 5
figure 5

Correlation between the QGI and existing development indicators

Full size image

3.4 Drivers of the QGI: An Appraisal

3.4.1 Pairwise Correlation

We investigate the key factors driving the QGI scores. We first adopt a pairwise correlation approach. We focus on living standards, politico-institutional indicators and external financing conditions. Figure 6a points to a strong correlation between the QGI and politico-institutional factors. On the one hand, institutional quality, as measured by the quality of bureaucracy, the rule of law or the control of corruption, is positively associated with the QGI, but the correlation is less marked with the latter. On the other hand, more political stability, as proxied by government stability goes hand-in-hand with a higher QGI.

Fig. 6
figure 6figure 6

Quality of growth index: key correlations. a QGI and politico-institutional factors, b QGI and domestic macroeconomic environment, c QGI and social spending, d QGI and external environment

Full size image

A sound and stable macroeconomic environment as well as a better access to credits for financing good projects may be conducive to higher QGI scores. Indeed, Fig. 6b shows that the volatility of inflation and credit to the private sector are negatively and positively correlated with the QGI, respectively. Figure 6c points out a relevance of social spending in achieving a good quality of growth. The QGI is positively linked with public spending in education and health sectors. The correlations with external conditions, portrayed in Fig. 6d reveal somewhat a weak association between the QGI and both FDI and remittances. The correlation between the QGI and foreign aidFootnote 15 is markedly negative, likely owing to the fact that foreign aid is mostly allocated to low-income countries.Footnote 16

3.4.2 Econometric Analysis

Altogether, the results above are simple unconditional correlations and should not be viewed as causal links between the QGI and the variables considered. In what follows, we attempt a preliminary appraisal of the factors driving potentially the QGI by carrying out some conditional estimates of the influence of these factors on the QGI. We perform OLS-based estimates using lagged (one period) values of the explanatory variables to account for likely delays in the influence of these variables on the QGI, and mitigate endogeneity issues. Table 3 reports the estimation results.Footnote 17

Table 3 Determinants of Quality of Growth over 1990–2011
Full size table

The baseline results focus on domestic macroeconomic factors, and are depicted in column 1 of Table 3. First, it emerges that devoting more public resources to social sectors allows improving the quality of growth. Second, stable government is found to be conducive to a better quality of growth. This suggests that less frequent changes of governments, which implies a lower uncertainty faced by government members as to whether they will be left out, may increase their focus toward implementing effectively the country’s development agenda and hence achieve a good quality of growth rather than getting involved in rent-seeking activities. Third, the coefficient of the volatility of inflation is negative and significant, implying that macroeconomic stability is a necessary ingredient for attaining pro-poor growth outcomes. Another striking result is the positive impact of financial development on QGI. This suggests that greater financial development, especially if accompanied by higher access to credit, may help unleash the private sector’s potential for creating wealth and decent jobs for the population, and hence achieving a good quality growth ultimately. Last, but not the least, column 1 sheds light on the importance of structural factors such as institutional quality in steering up countries’ QGI scores. The quality of institutions is measured by the average of quality of bureaucracy, the rule of law and the control of corruption. However, as already evidenced in Fig. 5b above, the quality of bureaucracy seems to matter the most for improving the QGI, since the effect of rule of law on the QGI is positive but not significant whereas the coefficient of control of corruption does not have the expected positive sign but also is not significant (columns 5 to 7). Columns 2 to 4 of Table 3 allow assessing the role of external conditions for the QGI. FDI is found to be positively related to the QGI. This suggests that FDI, by contributing to close the domestic saving gaps, play a pivotal role in achieving a good quality of growth in developing countries. The coefficient of remittances is also positive but not significant. For foreign aid the result is somewhat puzzling (column 4), as its effect is significantly negative. As indicated earlier, this is not necessarily a statement about aid ineffectiveness, but may rather point to a mere empirical regularity that aid is mainly allocated to poorer countries.Footnote 18

Finally, it is worth noting that our empirical investigation is based on a partial correlation analysis after controlling for the simultaneity bias. Further research is warranted to uncover the causal links—when longer time coverage of the QGI will be available, before pretending to draw robust causal links between the QGI and these potential factors.

4 Robustness of the QGI to Alternative Specifications

We assess the sensitivity of the QGI in several aspects.

First, as flagged in the methodology section, the benchmark QGI (QGI1) used so far stems from an equal weighting aggregation approach. In the following, we consider alternative weighting scenarios. Four alternative QGI (QGI2, QGI3, QGI4 and QGI5) are considered. QGI2 refers to the index derived from the following weighting combination, namely, \(\upalpha = {2 \mathord{\left/ {\vphantom {2 3}} \right. \kern-0pt} 3},\;\upbeta{1 \mathord{\left/ {\vphantom {1 3}} \right. \kern-0pt} 3}\), and by symmetry for QGI3 \(\upalpha = {1 \mathord{\left/ {\vphantom {1 3}} \right. \kern-0pt} 3},\;\upbeta{2 \mathord{\left/ {\vphantom {2 3}} \right. \kern-0pt} 3}\). QGI4 corresponds to the specification that assigns respectively \(\upalpha = {3 \mathord{\left/ {\vphantom {3 4}} \right. \kern-0pt} 4}\) and \({\beta = }{1 \mathord{\left/ {\vphantom {1 4}} \right. \kern-0pt} 4}\) as weights for the intrinsic nature and the social dimension of growth, and conversely for QGI5. Remark that the great emphasis put on the intrinsic nature sub-component (\(\upalpha = {3 \mathord{\left/ {\vphantom {3 4}} \right. \kern-0pt} 4}\)) compared the social dimension sub-component in QGI5 allows us to highlight to some extent the distinct nature of the QGI with respect to the UN HDI, which focuses primarily on human development while the QGI rather focuses overridingly on growth fundamentals. Key findings are as follows. First, key trends with all indices are not quantitatively and qualitatively different from the benchmark ones (QGI1). This is confirmed by the correlation matrix of the five QGIs (“Appendix 5”), which underscores that these indices are highly and significantly correlated, with a correlation coefficient ranging from 0.91 to 0.99.

Second, to capture possible interactions between the different components of the QGI, we consider a geometric averaging approach, instead of the arithmetic one previously used. This strategy does not alter substantially the ranking, as provided in “Appendix 3”. This is also pinned down in “Appendix 6”, which reports a Spearman’s rank order correlation test. The P value for a lack of correlation between the arithmetic mean-based and geometric mean-based country ranking is zero, and the test statistic for the correlation between the two sets of ranking is quite large, standing at 0.995, suggesting that both these ways of deriving the QGI lead to consistent results, from a ranking standpoint.

Third, we rebased the Min–Max normalization of the different components of the QGI on region-specific minimum and maximum values. As underscored in the methodology section, given that the min and max used to standardize the different variables are endogenous to the retained sample, the performance gap may vary tremendously. “Appendix 7” displays the results of the Spearman’s rank correlation test between this ranking based on a region-specific standardization and the benchmark one. It appears that both sets of ranking lead significantly to comparable results. Furthermore, the correlation matrix of these two sets of QGI leads to the same conclusion, as it clearly points out that these two series are highly and significantly correlated (“Appendix 8”).

Fourth, so as to gauge the additional information brought by the stability, diversification and outward-orientation sub-components, we compute three alternative QGIs, wherein we assign, respectively a: (1) zero weight to growth stability; (2) zero weight to diversification; (3) zero weight to outward-orientation. Put simply, we use the following weighting scheme for the intrinsic nature building bloc of the QGI, respectively: (\({1 \mathord{\left/ {\vphantom {1 3}} \right. \kern-0pt} 3}\), 0, \({1 \mathord{\left/ {\vphantom {1 3}} \right. \kern-0pt} 3}\), \({1 \mathord{\left/ {\vphantom {1 3}} \right. \kern-0pt} 3}\)), (\({1 \mathord{\left/ {\vphantom {1 3}} \right. \kern-0pt} 3}\); \({1 \mathord{\left/ {\vphantom {1 3}} \right. \kern-0pt} 3}\); 0; \({1 \mathord{\left/ {\vphantom {1 3}} \right. \kern-0pt} 3}\)), and (\({1 \mathord{\left/ {\vphantom {1 3}} \right. \kern-0pt} 3}\), \({1 \mathord{\left/ {\vphantom {1 3}} \right. \kern-0pt} 3}\), \({1 \mathord{\left/ {\vphantom {1 3}} \right. \kern-0pt} 3}\), 0). These alternative specifications did not change qualitatively the baseline result (see “Appendices 10 and 11”). For the sake of further robustness checks, we computed additional QGI, using the following alternative weighting scheme for γ1, γ2, γ3 and γ4:(\({1 \mathord{\left/ {\vphantom {1 2}} \right. \kern-0pt} 2}\), \({1 \mathord{\left/ {\vphantom {1 2}} \right. \kern-0pt} 2}\), 0, 0), (\({1 \mathord{\left/ {\vphantom {1 2}} \right. \kern-0pt} 2}\), 0, \({1 \mathord{\left/ {\vphantom {1 2}} \right. \kern-0pt} 2}\), 0), (\({1 \mathord{\left/ {\vphantom {1 3}} \right. \kern-0pt} 3}\), \({1 \mathord{\left/ {\vphantom {1 3}} \right. \kern-0pt} 3}\), \({1 \mathord{\left/ {\vphantom {1 6}} \right. \kern-0pt} 6}\), \({1 \mathord{\left/ {\vphantom {1 6}} \right. \kern-0pt} 6}\)), (\({1 \mathord{\left/ {\vphantom {1 3}} \right. \kern-0pt} 3}\), \({1 \mathord{\left/ {\vphantom {1 6}} \right. \kern-0pt} 6}\), \({1 \mathord{\left/ {\vphantom {1 3}} \right. \kern-0pt} 3}\), \({1 \mathord{\left/ {\vphantom {1 6}} \right. \kern-0pt} 6}\)). These alternative specifications did not change qualitatively the baseline result as well (see “Appendices 10 and 11”). In addition, we use the principal component methodology and derive the following weighting schemes: (γ1 = 0.38, γ2 = 0.27, γ3 = 0.23, γ4 = 0.12); (δ1 = 0.90, δ2 = 0.10); and (α = 0.75, β = 0.25). The resulting QGI is not qualitatively different from the baseline QGI (see “Appendices 10 and 11”).

Another potential avenue for further tweaking the robustness checks of our results could be to assess their sensitivity to using flow (as opposed to stock) variables for building the social sub-component of the QGI. One might indeed question the consistency of our composite QGI, in that most of the sub-indices under the growth fundamentals sub-component are based on flow variables (growth, volatility, etc.) while those under the social dimension building block are stock-based (education and health levels). However, we refrained from employing flow variables for constructing the social dimension sub-component, as this could have biased the results (scores and ranking), given the very nature of the variables underlying the social dimension building block. There are indeed upper bound limits to what can be achieved in terms of school enrollment or life expectancy (at least from a finite time horizon perspective for the latter), which signals that the stock of better social conditions matters most than the change when it comes to reducing poverty (see e.g., Anand and Sen 2000; Scott et al. 2014), thus warranting using stock-based variables instead of flow-based ones. For example, a country X that experiences a stagnation of its life expectancy at birth at around 80 years still over-performs a country Y that experiences an increase in its life expectancy at birth but from 40 to 45 over a given timeframe. Analogously, a country that maintains a 100 %primary school enrollment rate over a 5-year horizon remains in a better position than a country that improves the primary school enrollment rate to 70 %, up from 50 % over the same time horizon. In a nutshell, since the dynamics of these social dimension variables exhibit a marked convergence over time (as illustrated in Fig. 7), using flow-based measures of sub-component of the social dimension building block would lead to misleading conclusions about the quality of growth across countries and time.

Fig. 7
figure 7

Social dimension and QGI over time: convergence

Full size image

5 Conclusion

Improving living standards and reducing poverty are the ultimate goal of any growth strategy in developing countries. A nascent literature shows that countries with strong, stable, and broad-based growth are more likely to improve living standards and reduce poverty rates. This paper contributes to the literature and proposes a Quality of Growth Index (QGI) that captures both the intrinsic nature and the social dimension of growth, with the view that a “good quality of growth” is more than just the level of growth and should improve individuals’ welfare. Hence, “good growth” should feature sound fundamentals and notable achievements in social outcomes.

Based on data availability, the QGI is designed as a composite index of sub-indexes capturing the “growth nature” aspect and the “desirable social outcomes” aspect. For the “growth fundamentals” sub-index, four dimensions are considered of good quality: the strength of growth, the stability of growth, the diversification of the sources of growth, and the outward orientation of growth. For the “social outcomes” sub-index, the most basic indicators of human capital are used: health proxied by life expectancy at birth and infant survival rate at birth and education proxied by primary school completion rate. These variables are supposed to reflect countries’ progress toward better social indicators.

The paper has brought out the following stylized facts of the QGI. First, we find that the QGI scores well when compared with other welfare measures such as the UN HDI, poverty rates, income inequality and income per capita measures. Second, while the QGI has broadly improved over time, it has varied markedly across regions, income levels with LIC and sub-Saharan African countries lagging behind. In the same vein, structural factors such as fragility and resource endowment tend to be associated with lower QGI. Third, we explore the possibility of the existence of a “poor growth quality trap”. We do find that there is a convergence in the quality of growth over time, though at a slow pace.

Building on the QGI, we also investigate the main drivers of the quality of growth. Empirical estimates indicate that institutions and policies tend to be associated with good quality growth. We find that the QGI is higher in a stable political environment. The quality of institutions, especially the quality of bureaucracy that ensures the capacity of government to deliver good public services, is also positively associated with higher QGI. Sound macroeconomic policies that help produce stable and broad-based growth contribute to better QGI. These include the share of spending allocated to social sectors, price stability, financial development and inclusion, and FDI.

The QGI is a positive step in understanding how high growth could lead to better social outcomes. It provides a comprehensive measure of the very nature of growth despite the paucity of data. It is also a dynamic concept and allows cross-country comparisons. Potential future research, as data become available, could enhance and broaden the scope of the QGI by including labor market and inequality measures. These would undoubtedly improve the inclusiveness dimension of the QGI.

For now, the QGI developed in this paper could serve as a benchmarking tool to guide policies for an inclusive growth.