Sustaining Long-Run Growth and Macroeconomic Stability in Low Income Countries - The Role of Structural Transformation and Diversification - Background Notes
Author: International Monetary Fund
Publication Date:
03 Jul 2014
eISBN:
9781498343664
Language:
English
Keywords:
PP; private sector; country authorities; low income; natural resources; output diversification
Download PDF (3 MB)

Sustaining Long-Run Growth and Macroeconomic Stability in Low Income Countries - The Role of Structural Transformation and Diversification - Background Notes

Abstract

Sustaining Long-Run Growth and Macroeconomic Stability in Low Income Countries - The Role of Structural Transformation and Diversification - Background Notes

Diversification: A Growth Determinant in Low-Income Countries1

A. Introduction

1. A fundamental paradigm of economic theory and empirics is that economic development involves structural transformation, the dynamic reallocation of resources from less productive to more productive sectors (see McMillan and Rodrik, 2011, and Lin, 2012). This section explores the link between structural transformation and economic growth. Specifically, it examines the effects of export and production diversification on economic performance in low-income countries (LICs). Historically, LICs have depended heavily on a narrow range of traditional primary products. Recent theories suggest that such limited diversification reflects market and government failures which limit technology spillovers and hamper productivity and economic growth. Imbs and Wacziarg (2003) provide empirical support suggesting that increases in per capita income are first associated with diversification and then with reconcentration in production. Cadot et al., (2011) go one step further and argue that this nonlinear diversification pattern “is an inherent feature of the economic development process.”

2. While economic transformation and diversification are correlated with development, it remains unclear whether a causal relationship exists. After all, growth may actually drive diversification to generate the observed positive correlation between the variables. For policy considerations, the issue of causality is of prime importance to answer the question whether policies should target growth or diversification. A second unresolved problem in the previous literature is whether the development-diversification relationship survives when alternative determinants of growth are included in formal econometric models that go beyond bivariate scatter plots. Previous analyses of diversification and development did not include many traditional growth determinants in regressions such as investment, education, and population growth.

3. The approach presented here represents the first integrated empirical treatment of the diversification-growth debate. To establish a causal effect of diversification on growth, it is imperative to control for endogeneity and include all previously relevant candidate growth determinants motivated by theory and to examine the growth-diversification relationship using novel statistical tools and data. The underlying dataset uses the export and output diversification measures developed by Papageorgiou and Spatafora (2012), and the empirical approach leverages Instrumental Variable Bayesian Model Averaging (Eicher, Lenkoski, and Raftery, 2009), IVBMA, a method specifically designed to allow for a potentially large set of growth determinants when causality is drawn into question.

4. This paper builds on Durlauf et al., (2008; DKT thereafter) seminal panel growth study. First, it extends the time dimension of the DKT data and then introduces trade diversification as a potential growth determinant. Second, it extends Durlauf et al.’s methodology to fully account for a large set of growth determinants in the presence of potential reverse causality.

5. The key finding is that the longer panel confirms Durlauf et al.’s earlier results that aggregate trade measures are not robust growth determinants. Once export diversification is introduced however, the results show that it is a crucial determinant of economic growth for LICs. The effect is not only statistically significant but also economically important: a one standard deviation increase in export diversification is shown to increase the average annual growth rate by 0.8 percentage points for LICs. Therefore, export diversification should be an important growth policy target for LICs. Aside from trade diversification, the growth determinants suggested by the approach are those central to all previous studies: initial GDP, population growth and investment reflecting neoclassical models; governance quality and government expenditures reflecting new growth theories.

6. Output diversification, measured by value added of real sectors, also matters for growth. LICs could greatly benefit from diversifying their real sectors. More specifically, the estimates imply that a one standard deviation increase in output diversification in LICs raises their average annual growth rate by about 1.4 percentage points.

B. Conceptual Framework

7. There exist a multitude of theories that link diversification to growth and with potentially distinct channels at different stages at development. One channel is that diversification helps to achieve stable growth by reducing growth volatility as implied by portfolio selection theory. Diversification could also enable a gradual allocation of resources to their most productive uses to increase growth (Acemoglu and Zilliboti, 1997). The impact of export diversification on growth is ultimately an empirical question, and the Theil index of export diversification developed in Papageorgiou and Spatafora (2012) can be used to examine this relationship. Conceptually their aggregate diversification measure is composed of two diversification dimensions: the extensive and intensive margins. Intuitively, the extensive margin measures the number of different export sectors, while the intensive margin represents the diversification of export volumes across active sectors. The intensive margin measures is therefore a less intuitive aspect of diversification, as it identifies countries as rather less diversified when GDP or export revenues are driven only by a few sectors (although the country might export/produce many different goods).

8. As noted previously, quantifying any link between export diversification and economic growth is complicated by the fact that there are likely to be numerous feedback effects between export diversification and growth. Growth may affect diversification as the country advances and expands its product space and exports. Thus, the dynamic development process renders it difficult to identify whether growth drives diversification or the other way around. An example for the simultaneity of growth and diversification is the dynamic reallocation of resources from less productive to more productive sectors and activities as outlined by McMillan and Rodrik (2011). Prime examples are the development experiences of the East Asian Tigers and Tiger Cubs in the 1970s and 1980s and many ex-Soviet bloc economies in the 1990s as they transformed from relatively agrarian to manufacturing economies. LICs still remain largely specialized in agriculture and other resource-based activities with limited potential for quality upgrading. Structural transformation will inevitably involve diversification, both in terms of domestic production and, given small domestic market size, external trade.

9. To counter potential simultaneity issues in this note, it is necessary to instrument for export diversification with a number of geographical features. In the spirit of Frankel and Romer (1999), the instruments in the empirical analysis are the log of a country’s land area, a dummy taking the value one for landlocked countries, and the log of a country’s population. While having the advantage of being predetermined with respect to growth rates, geographical features are also important drivers of export diversification. A country with a large population can develop and produce more types of products while a country with large land area is more likely to have specialization clusters across the country. At the same time, a landlocked country is less likely to engage in international trade and will thus have lower export diversification.

C. Econometric Methodology

10. The plethora of growth theories and their associated candidate regressors has given rise to a sizable literature seeking to identify robust growth determinants. Early approaches used Leamer’s (1978) Extreme Bound Analysis (Levine and Renelt, 1992, and Sala-i-Martin, 1997), which suffers from arbitrary robustness thresholds (“Extreme Bounds”). Extreme Bound Analysis examines specific combinations of all possible growth determinants in millions of regressions and examines if estimates change signs for individual growth determinants. If a sign change is observed, the variable is said to be not robustly related to growth.

11. The problem with Extreme Bound Analysis is not only the arbitrary search for regressors and regressions in which sign changes occur, but also the notion that all regressions should carry identical weight. Clearly some regressions that omit key growth determinants are grossly misspecified and suffer from rampant omitted variable bias. Subsequent approaches employ Bayesian Model Averaging, a methodology specifically designed to address model uncertainty empirically (Fernández at al., 2001, Brock and Durlauf, 2001, Sala-i-Martin et al., 2004, Ciccone and Jarocinski, 2010, Eicher et al., 2011). However, none of these approaches tackled endogeneity.

12. DKT (2008) addresses endogeneity within the BMA context by producing fitted values for endogenous regressors via OLS in a first stage. The fitted values are then used in a second stage that is subjected to BMA. Subsequently, Eicher et al., (2009) develop a comprehensive two-stage extension of BMA to allow for model selection in both stages. Similar approaches have been suggested by Moral-Benito (2012) and Chen et al., (2009) who introduce BMA Generalized Method of Moments (GMM). Koop et al., (2012) develop a Bayesian IV methodology that does not rely on Eicher et al., (2009) approximations to integrated likelihoods and Karl and Lenkoski (2012) introduce conditional Bayes factors to resolve mixing difficulties associated with Koop’s et al., (2012) search algorithm. Details of the IVBMA methodology are provided in Appendix II.

13. The intuition behind IVBMA is that an efficient search algorithm explores the model space spanned by all candidate growth determinants. The methodology then averages coefficients over all empirical models, while weighting each model by its quality. Hence, highly misspecified models are weighted down. The approach has highly appealing statistical prosperities. In contrast to single regression approaches or Extreme Bound Analysis, IVBMA does not suffer from inflated t-statistics or artificially narrow confidence bands. Hence, it also delivers the best predictive performance and the lowest mean square error compared to these approaches. (See Raftery (1995) and Raftery and Zheng (2003).) For the policy maker, IVBMA produces a key statistic of interest: the posterior inclusion probability (PIP). Inclusion probabilities provide a probability statement regarding the importance of a particular growth determinant regressor that directly addresses what is often the policy maker’s prime concern: what is the probability that the regressor has an effect on the dependent variable? The general rule developed by Jeffreys (1961) and refined by Kass and Raftery (1995) stipulates effect thresholds for posterior inclusion probabilities. Posterior inclusion probabilities lower than 50 percent are seen as evidence against an effect, and the evidence for an effect is either weak, positive, strong, or decisive for posterior inclusion probabilities ranging from 50-75 percent, 75-95 percent, 95-99 percent, and higher than 99 percent respectively. In this analysis, a regressor is “effective” if its posterior inclusion probability exceeds 50 percent.

D. A Brief Look at the Data

14. Using non-overlapping five-year periods, the dataset includes 84 countries from the period 1965 to 2009 and comprises 583 country-period observations. Because the focus is on the relationship between diversification and growth, resource-rich economies that generate more than 20 percent of their GDP with resource rents (as reported by the World Development Indicators) are excluded from the sample. Resource-rich countries represent sizable outliers with unusually low export diversification relative to their income levels. Removing resource-rich countries therefore focuses of the empirical analysis on understanding whether the development of diversified export structures and broad-based comparative advantages are advantageous for growth. Small states were also removed small states from the sample. None of the above described changes to the dataset impact however, the qualitative results. The only country belonged Small States (with population less than 1.5 million) in the regression sample is Sierra Leone. The regression results remain the same significance level and the coefficients hardly change without this country.

15. The dependent variable is the average growth rate of GDP per capita during each five-year period. Per capita income data are obtained from Penn World Tables 7.1. All empirical specifications include period and regional dummies (Sub-Saharan Africa, East Asia, Latin America and the Caribbean) to control for spatial and time effects on growth. The primary measure of export diversification is the Total Theil index; results using the decomposition of the Total Theil into its extensive margin (between Theil) and intensive margin (Theil) components are also included. Finally, the analysis is extended to the more general concept of output diversification. As mentioned previously, all measures of diversification (external and real) are from Papageorgiou and Spatafora (2012).2

16. All additional covariates and instruments which are used in the empirical analysis are obtained from the growth determinants study of DKT (2008), which was recently updated by Henderson et al., (2011). DKT base their variable selection on Barro (2003), one of the most comprehensive approaches to growth determinants. Specifically, DKT introduce proxies for seven different growth theories:

  1. i)Regressors suggested by neoclassical growth theory include initial per capita income and the per-period averages of population growth, the investment to GDP ratio, and education (share of the working population with secondary schooling times the rate of successful completion of secondary school). The analysis follows DKT and instruments for these four variables with one-period lagged values.
  2. ii)Regressors that serve as proxies for demographic change include the reciprocal of life expectancy at age one and the logarithm of the total fertility rate, which are both assumed to be exogenous.
  3. iii)Theories that link macroeconomic policies to growth are proxied by the average ratio of government consumption to GDP, openness (exports + imports) over GDP filtered for land mass and population, and the average change in the CPI. All three variables are instrumented with their respective lagged values.
  4. iv)Theories that link geography to growth are proxied by the land area within 100km of an ice-free coast and the percentage of tropical land area, which are assumed to be exogenous.
  5. v)Theories linking institutions to growth are proxied with the risk of expropriation, constraints on the executive, and the World Bank governance index. Dummy variables for the English and French origin of a country’s legal system are included. Lagged values of the expropriation risk are used to instrument for the current value of the same variable. All other variables are treated as exogenous.3
  6. vi)The relation of religion to growth is proxied by the shares of the population adhering to Eastern, Hindu, Jewish, Muslim, Orthodox, Protestant, and other religions. As in DKT, the respective shares in 1900 as used instruments.
  7. vii)Regressors proxying for the impact of fractionalization within a country on growth are two linguistic fractionalization and ethnic tension indices. Both variables are assumed to be exogenous. The dataset used in the empirical analysis is described in more detail in the appendix, which also lists the variable sources and definitions.

17. The empirical strategy involves three steps:

  • Introduce export diversification as a potential growth determinant;

  • Address model uncertainty due to the large number of growth theories that predict different candidate regressors and/or opposing effects of trade on growth;

  • Examine the importance of controlling for endogeneity in growth regressions.

18. Table 1 presents the IVBMA results. A linear diversification term is included in column 1, and nonlinear diversification effects are introduced in column 2. In columns 3 and 4, the Total Theil export diversification measure is replaced with the intensive (within) and extensive (between) Theil indices. In both cases, the regression specifications allow for nonlinear diversification effects as described in the previous section. In addition to posterior inclusion probabilities (PIP) (see also Figure 1), the conditional means and standard deviations for the coefficients are reported. The coefficients can be interpreted as in standard OLS and 2SLS estimation (see Appendix Table 3). Complete results see Appendix Table A1.

Table 1.

IVBMA Regressions for Growth on Export Diversification, Developing Countries, 1965-2009

article image
Notes: ♠ Composite coefficient reported, based on the joint posterior distribution of Diversification and Diversification*CountryIncome interaction. Since the PIP is not defined for the composite, w e report the percentage of the joint posterior distribution of Diversification*CountryIncome interaction that is non-zero.

“Diversification” in this table is measured by different Theil indexes with lower values indicating higher levels of diversification.

Figure 1.
Figure 1.

Posterior Inclusion Probabilities from BMA Growth Regression, Developing Countries, 1965-2009

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

19. The first set of estimates in column 1 indicates that the traditional growth determinants exhibit the highest effect thresholds: Initial GDP, Government Quality, Investment, Population Growth, and Government Expenditure followed by Protestant Fraction, Sub Saharan Africa dummy, and Inflation. Export Diversification does not have an effect on growth volatility in the global sample, which may be due to the existence of nonlinearities that countries become less diversified after they achieve certain income level.

20. Export diversification has a decisive impact on growth for LICs once nonlinearities are introduced in the specification (see column 2). Using the results in column 2, a one standard deviation increase in LICs’ export diversification raises their growth rate by about 0.8 percentage points. As discussed above, the IVBMA-Sargan test outlined in Eicher et al., (2009) indicates instrument validity in all IVBMA specifications in Table 1. This suggests export diversification is crucial for growth in LICs.

21. The set of growth determinants identified by IVBMA is parsimonious but expected. With Initial GDP, Government Quality, Investment, Population Growth, and Government Expenditure, the results provide support for both the neoclassical growth model as well as new growth theories that rely on productive government expenditures and the quality of institutions.

22. Replacing the Total Theil export diversification measure by the intensive and extensive Theil indices (see columns 3 and 4) results in very similar conclusions for LICs. The extensive Theil index indicates export diversification by expanding to new products; the intensive Theil index indicates diversification by equalizing the shares of different products. Thus, LICs can stimulate growth by diversifying their exports both at the extensive and intensive margins. There is also some evidence in column 3 that lower- and upper-middle income countries can benefit from increasing the diversification of their exports at the extensive margin.

E. Output Diversification and Growth

23. Export diversification and output diversification are in principle interlinked, the former reflecting diversification in the external sector, and the latter capturing diversification in the domestic production process across sectors. This section examines if the previous results hold on a broader level. Instead of focusing on export diversification, a measure of output diversification is now included in the empirical specification. The total Theil index of output diversification is calculated using value-added shares in seven real subsectors reported in the UN sectoral database. Table 2 presents IVBMA results with a linear output diversification term in column 1, while nonlinear output diversification effects are introduced in column 2. Note that the total Theil index in the present context measures the inequality between sectoral production shares in each country. Complete results see Appendix Table A2.

Table 2.

IVBMA Regressions for Growth on Output Diversification, Developing Countries, 1965-2009

article image
Note: ♠ Composite coefficient reported, based on the joint posterior distribution of Diversification and Diversification*CountryIncome interaction. Since the PIP is not defined for the composite, w e report the percentage of the joint posterior distribution of Diversification*CountryIncome interaction that is non-zero.

“Diversification” in this table are measured by different Theil indexes, with the lower value means more diversified.

24. Two results emerge when comparing the output diversification results (Table 2, column 1) tothe one linear export diversification result (Table 1, column 1). First, in contrast to export diversification, output diversification is significant in the linear specification. And second, in the output diversification specification, there is strong support for neoclassical and institutional growth determinants. When allowing for a nonlinear output diversification effect in column 2, the results show that LICs are the likely driver of the aggregate effect (column 1). Overall, the results shown in column 2 indicate that LICs can greatly benefit from diversifying their production structure. More specifically, the estimates imply that a one standard deviation increase in output diversification in LICs raises their average annual growth rate by about 1.4 percentage points.4 This effect is even greater than the one found for export diversification.

F. Conclusions

25. This section has examined the impact of diversification on economic growth. In the five-year period panel ranging from 1965 to 2009, there is decisive evidence that export diversification is a substantial driver of growth in LICs. The findings are robust to the two biggest caveats encountered in growth regressions, endogeneity and model uncertainty, which are addressed through the use of the Instrumental Variable Bayesian Model Averaging (IVBMA) estimator. The results also show that both export diversification at the intensive and extensive margins are drivers of economic growth in LICs. These conclusions carry over to the more general concept of output diversification.

26. Overall, the results suggest that countries at early stages of development could benefit considerably by diversifying their exports. At later stages of development, export diversification seems to be rather a by-product of prosperity rather than its cause. Export diversification could be the driver of a country’s early development through several channels. For one, a more diversified economy offers an insurance against idiosyncratic sectoral shocks, especially at low stages of development when countries produce only few goods for export, such as agricultural products and natural resources. And second, countries with greater export diversification at early development stages are more likely to be able to move into new products and spur development further. Hausmann and Hidalgo (2011) and Kali et al., (2013) offer a detailed discussion of this point from an economic network’s perspective.

Diversificiation and Volatility9

A. Introduction

27. Macroeconomic stability has been instrumental to sustained growth and development in developing countries. This section examines the relationship between limited diversification in the export sector and the volatility of economic growth.

28. The existing literature provides some evidence that economic diversification can increase a country’s resilience to external shocks. Openness to trade is often a source of output growth volatility. The more open an economy is, the more susceptible it is to external shocks; on the other hand, openness to trade helps insulate against domestic growth slowdowns by providing access to additional markets. As Hadad et al., (2013) point out, countries that have a more diversified basket of export goods are less likely to be impacted negatively by external shocks. Export diversification could reduce volatility by reducing dependence on particular products, especially primary products and commodities which tend to be associated with higher risk. Stanley and Bunnag (2001) show that for four countries in Central America, greater export diversification leads to lower income instability within the 1974 to 1995 period. In addition, they argue that different product combinations can have different impacts on income stability. The effects of reducing instability are stronger if the new products are less volatile or negatively co-vary with the current exports. Agosin (2007) shows negative correlation between export diversification and export growth variance. He suggests export diversification could influence growth by reducing the variance of export growth. Koren and Tenreyro (2007) show evidence that the productive structure moves to less volatile sectors and the degree of sectoral concentration declines as countries develop. Mobarak (2005) finds that higher levels of diversification lower growth volatility. Bertinelli et al. (2009) use a panel data set of developing countries to estimate the trade-off between export earnings and its variability based on modern portfolio theory. They find there are welfare gains from export diversification structure.

29. With this in mind, the focus is on developing countries and three key questions.10 First, are episodes of significant, sustained diversification associated with increased macroeconomic stability? Second, does export diversification of products help reduce growth volatility, and as a follow-up to this question, does this effect happen through extensive or intensive diversification? Lastly, does output diversification have an effect on reducing growth volatility?

Stylized Facts

30. Developing countries have experienced a higher level of growth volatility (Figure 2). While growth volatility declined after 1995 in both EMs and LICs, it still remains higher than in advanced countries.

Figure 2.
Figure 2.

Growth Volatility, by IMF Income Groups, 1962-2010

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

Sources: World Bank WDI, IMF staff calculations.

31. A higher level of export diversification is generally associated with lower growth volatility. Figures 3 and 4 show the relationship between the Theil index and output volatility. The Theil index is a measure of inequality; here it is used as a measure of concentration for exported products, and a lower Theil index indicates higher diversification in exports. As shown in Figure 3, overall, LICs and EMs have a higher level of output volatility and a higher Theil index than AMs; Figure 4 shows the positive relationship between higher output volatility and lower diversification, particularly for LICs. This relationship continues to hold even after controlling for other determinants of growth volatility (Lederman and Maloney, 2012). Indeed, recent evidence suggests that industry diversification helped attenuate the impact of the global financial crisis (da Costa Neto and Romeu, 2011).

Figure 3.
Figure 3.

Growth Volatility and Export Diversification by IMF Income Group, 1962-2010

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

Sources: UN Comtrade; World Bank WDI; IMF staff calculations.
Figure 4.
Figure 4.

Export Diversification and Growth Volatility, 1962-2010

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

Sources: UN Comtrade; World Bank WDI; IMF staff calculations.

32. The link between diversification and volatility is easiest to observe in the context of large diversification spurts. A total of 61 diversification spurts11 in 51 developing countries were identified in the post–1962 period. Diversification spurts occurred more frequently in the 1960s and 1990s and were evenly distributed across regions (after controlling for the relative number of countries). For the sample as a whole, the spurts lasted 13 years on average; in the East Asia and Pacific region, spurts lasted 20 years on average. Diversification spurts are associated with a reduction in the volatility of output growth in developing countries. The decrease is especially pronounced in LICs, where growth volatility decreases 1.5 percentage points in the wake of diversification spurts (Figure 5).

Figure 5.
Figure 5.

Growth Volatility and Diversification Spurts in LICs, 1962-2010

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

Sources: UN Comtrade; World Bank WDI; IMF Staff calculations.

B. Export Diversification and Volatility

33. The following specification for the growth volatility estimations is used:
Voli,t=αVoli,t−1+βDivi,t+ρxi,t+γt+ϵi+εi,t

The data cover the time period from 1962-2010. Voli,t denotes the growth volatility in country i at time t. It is calculated as the standard deviation of GDP growth using a five-year window. Divi,t denotes the diversification index, and four different diversification indices were tried in separate regression specifications. The first two indices, Total Theil and the Herfindahl index, capture the effect a country’s overall level of diversification has on volatility. The second two indices, the extensive and intensive margins, can be obtained from a decomposition of the overall Theil index. Extensive diversification occurs when a country exports new product lines, while intensive diversification occurs when a country exports a more balanced mix of existing products. Lower values for all four indices indicate a higher level of diversification.

34. Openi,t denotes the trade openness level for each country/year, defined as total exports and imports as a share of GDP. Several regressions include interaction terms between the diversification index and a measure of trade openness (Openi,t *Divi,t denotes the interaction term). xi,t denotes other control variables such as terms of trade volatility, inflation volatility, and exchange rate volatility. γt is time effect. ϵi is unobserved time-invariant country-specific effects. εi,t is residual error. The data are five-year averages for each variable in order to exclude extreme values and business cycles; thus, t denotes each five-year period.

35. Regressions are estimated using the two-step GMM model because of the dynamic nature of the regression equation. Since there is a lagged dependent variable in the estimation, fixed effects model estimates are biased. Following Arellano and Bond (1991), the GMM estimator thus is necessary to obtain consistent estimates.

C. Empirical Results

36. Export diversification helps to reduce growth volatility. The GMM regression results are reported in Tables 3 to 4. Table 3 shows the regression results based on the sample of developing countries. Table 4 shows the results for all countries. Note that a lower Theil index (total, intensive, or extensive) or Herfindahl index (HFI) means a higher level of diversification in export products. The regression results in first two columns of Table 3 show positive coefficients on diversification measures Theil and HFI, which suggests product diversification helps reduce growth volatility. From column (3), the coefficient on the intensive Theil index is also significant and positive, suggesting the effect of diversification on growth volatility is through equalizing the export shares of the current export basket (see Figure 6). Similar results hold when all countries are included in the regressions.

Table 3.

System GMM Regressions for Growth Volatility on Export Diversification of Products, Developing Countries, Panel of 5 Year Average, 1962-2010

article image
Robust z-statistics in brackets. *** p<0.01, ** p<0.05, * p<0.1Note: Dependent variable is Volatility of GDP Per Capita Growth.Period dummies and a constant were included, but not reported.
Table 4.

System GMM Regressions for Growth Volatility on Export Diversification of Products, All Countries, Panel of 5 Year Average, 1962-2010

article image
Robust z-statistics in brackets *** p<0.01, ** p<0.05, * p<0.1Note: Dependent variable is Volatility of GDP Per Capita Growth.Period dummies and a constant were included, but not reported.
Figure 6.
Figure 6.

Intensive Diversification and Growth Volatility

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

Sources: UN Comtrade; World Bank WDI; IMF Staff calculations.

37. The effects of diversification on growth volatility hold after including trade openness and other control variables. Columns (4) to (6) in Table 3 show the regression results after adding the trade openness variable. The HFI and intensive Theil diversification measures are still significant with positive signs; the sign on total Theil is still positive but the coefficient is no longer significant. These results suggest that even after accounting for trade openness, increased export diversification reduces growth volatility. Additional control variables (terms-of-trade volatility, exchange rate volatility, and inflation volatility) are added to the baseline specification. Now, only the HFI diversification measure remains positive and significant. The control variables follow the paper by Haddad et al. (2013). Results for all countries are similar.

38. Export diversification reduces growth volatility when a country is more open to trade. Columns (10) to (12) in both tables show the regression results when the interaction terms of diversification and trade openness are included. The overall pattern of results mirrors what Haddad et al. (2013) find: negative coefficients on both trade openness and the measure of concentration, and a positive and significant coefficient on the interaction term of trade openness and diversification. For developing countries, while the signs are the same as Haddad et al., (2013), the coefficients for trade openness, total Theil index, and HFI coefficients are not significant. When all countries are included, again, the coefficients on trade openness and HFI are not significant. Now though, the total Theil index is significant and the intensive Theil is not. Figure 7 presents a graphic representation of the positive relationship between trade openness and diversification interaction term and output growth volatility.

Figure 7.
Figure 7.

Interaction of Diversification and Openness vs. Growth Volatility

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

Sources: UN Comtrade; World Bank WDI; IMF Staff calculations.

39. There is some evidence that output diversification also helps lower growth volatility in developing countries. The output diversification measures are calculated based on value added data from UN databases. There are seven sectors available, and the data start in 1970. (See Table 5 and Figure 8.) The results in columns (7) and (8) of Table 5 show that the interaction term of real diversification and trade openness is significant; thus, a country that is more diversified and has a higher level of trade openness has lower growth volatility.

Table 5.

System GMM Regressions for Growth Volatility on Output Diversification, Panel of 5 Year Average, 1970-2010

article image
Robust z-statistics in brackets. *** p<0.01, ** p<0.05, * p<0.1Note: Dependent variable is Volatility of GDP Per Capita Growth.Period dummies and a constant were included, but not reported.
Figure 8.
Figure 8.

Output Diversification vs. Growth Volatility

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

Sources: UN Value added database, World Bank WDI, IMF Staff calculations.

D. Conclusions

40. Export diversification matters for macro-stability in developing countries. This is particularly true for vulnerable LICs where increasing export diversification will help reduce growth volatility. Export diversification could happen through either the extensive or intensive margin or both; however, the results show that intensive diversification is very important for reducing volatility. Increased intensive diversification could lower output growth volatility. When an economy becomes less concentrated in specific products, especially those products with volatile prices or high demand volatility such as primary commodities, the country could experience a decrease in growth volatility. There is also evidence on the impact of output diversification on growth volatility.

Structural Transformation and Sectoral Productivity in Low-Income Countries12

A. Introduction

41. The past two decades have seen unprecedented growth and rapid catch-up convergence in low-income countries (LICs). Growth in per capita output for many LICs rebounded in the early 1990s, and since then has surpassed that in many advanced and even emerging market economies, particularly in the period after the global financial crisis. The frequency of growth takeoffs—sustained high growth episodes—in LICs has also risen markedly during this time, and takeoffs have lasted longer (IMF, 2013a). Part of this solid performance can be attributed to favorable commodity prices in the 2000s, but even non-commodity exporting LICs have done well (IMF, 2013b). Improved macroeconomic stability through better policy making, healthier economic and political institutions, and the undertaking of wide ranging economic and structural reforms are all potential contributors to the recent growth acceleration.

42. Underpinning this solid growth performance is robust productivity dynamics and labor reallocation at the sector level. Historical growth takeoffs in LICs, on average, have been accompanied by productivity surges in broad economic sectors—agriculture, industry, and services (Figure 9). Productivity gains can also come from better reallocation of resources across and within sectors of the economy, i.e.,, structural transformation, and labor reallocation out of the low-productivity agricultural sector has contributed to this process. As economies diversified their production, trade diversification as measured by changes in the type and quality of export products also increased (Papageorgiou and Spatafora, 2012). This section seeks to analyze the dynamics of sectoral reallocation and sectoral productivity13 in the recent past, understand their contribution to economy-wide productivity growth and structural transformation, and examine their policy and institutional drivers with a view to draw policy implications for the future.

Figure 9.
Figure 9.

Sectoral Dynamics Around Growth Takeoffs

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

Sources: UN National Accounts database, ILO, GGDC, WDI, and IMF staff calculations.Notes: The event study includes 29 growth takeoff episodes in LICs between 1990 and 2011, taken from IMF (2013a). Productivity is real value added per worker.

43. Notwithstanding significant heterogeneity across LICs, some key patterns emerge. There is a tight relationship between sectoral productivity growth and structural transformation/diversification. Some economies benefit from a virtuous and mutually enforcing cycle characterized by rapid sectoral productivity growth, productivity-enhancing labor reallocation and increasing export diversification. Others are stuck in a less optimal equilibrium characterized by slow growth, stagnant productivity, little diversification, and productivity-reducing sectoral shifts. The process of structural transformation is not automatic and tends to be strongly influenced by the policy, business, and institutional environment. The results also show that a number of structural reform measures, such as removing trade barriers and reforming the banking and networks industries, have proven effective in kick-starting sectoral productivity growth and structural transformation in low-income countries.

44. Many LICs are experiencing rapid changes in the structure of the economy, but the pattern is uneven across countries. Over the past four decades in LICs, the agricultural share in total value added has continued to decline, accompanied by a commensurate increase in service share, whereas the share of manufacturing has not changed dramatically and remains at a low level (about 12 percent) for the average LIC (Figure 10, top left panel). Both the pace and nature of sectoral reallocation of resources in LICs have been different compared to the historical experience of advanced and emerging market economies.

Figure 10.
Figure 10.

Dynamics of Sectoral Shares in LICs

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

45. Dabla-Norris et al., (2013b) showed that many LICs, particularly those in Sub-Saharan Africa (SSA), have higher agricultural shares and lower manufacturing shares than as predicted by the level of economic development and country fundamentals. In addition, the high and growing share of services at low levels of development is a striking feature of many LICs, and marks a departure from the development path of many dynamic economies in industrial Asia, who transformed largely through expansion of low-wage manufacturing. However, the service sector in LICs tends to be dominated by low-skilled and less productive activities such as retail, social, and personal services, although construction, transportation, and communication have recently been gaining value added shares from a low level. In manufacturing, there has been a gradual movement towards higher skill-intensive activities, but the majority of manufacturing value added continues to be generated by low-skilled industries such as food, clothing, and footwear (Figure 10, top right panel).

46. The agricultural employment share tends to decline much more rapidly in economies with a more diversified export base, possibly thanks to the availability of alternative productive opportunities that greater diversification provides, and similarly in the group of non-commodities exporters (Figure 10, bottom left panel). There is also significant heterogeneity in the regional distribution of sectoral employment (Figure 10, bottom right panel).14 While the employment share of agriculture has declined across regions between 1990 and 2007, the change has been more pronounced in Asia but less so in SSA. In the average SSA country, over 60 percent of the workforce continues to be employed in agriculture despite the sector’s low value-added share, pointing to very low agricultural productivity.

47. Sectoral productivity growth in LICs experienced a strong rebound in the 2000s (Figure 11, top left panel). The productivity surge was most pronounced in agriculture and services, reflecting marked improvements in the terms of trade, robust investment rates, and significant reforms in both real and financial sectors. In addition, countries that have experienced greater reduced agricultural shares on average exhibited strong and broad-based productivity growth at the sector level, in marked contrast to those who have not (Figure 11, top right panel).

Figure 11.
Figure 11.

Sectoral Productivity Growth in LICs

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

48. As with sectoral shifts, the average pattern of sectoral productivity masks considerable disparities at the regional level (Figure 11, bottom left panel). Recent productivity performance was strongest for countries in Asia and Europe and Central Asia (ECA)—the former led by frontier markets such as Vietnam, and the latter by Albania and Georgia, whereas performance was mixed in Latin America and the Caribbean (LAC), Middle-East and North Africa (MENA) and SSA. LAC economies registered high productivity growth in agriculture, but productivity declined in industry and services. For MENA and SSA, agriculture and services have performed well, while industry productivity stagnated.

49. Despite recent improvements, productivity gaps between LICs and advanced economies remain large and show little sign of narrowing. Figure 11 (bottom right panel) illustrates the evolution of labor productivity as a percentage of the U.S. level in agriculture, manufacturing, construction, and market services (broadly defined as non-governmental services) for several SSA economies where PPP-adjusted sectoral labor productivity data are available. Not surprisingly, agriculture in SSA economies exhibits the widest productivity gap (averaging less than five percent of U.S. level), with the gap remaining roughly unchanged over the fifteen-year period. Stagnation in industry is driven by low productivity in manufacturing—the sector in which gap with the U.S. has actually been widening, whereas relative labor productivity in construction is higher at about 25 percent of U.S. level. Mining productivity (not shown here) often exceeds that in the U.S. given the important role that the resource sector plays in many SSA economies and the capital intensive nature of the sector.

50. Manufacturing value added and employment in LICs are concentrated in low-technology, labor-intensive activities, e.g., agro processing industries (Figure 12). For example, food and beverages on average contributed 32 percent to total manufacturing value added in 2006, and 22 percent to employment. Other large employment creating industries within manufacturing include textiles and apparel, but value added creation from these industries is rather limited, contributing to low average productivity. On the other hand, average productivity is relatively higher in mineral products, chemical products, and basic metals, whose value added contribution exceeded employment contribution. There is very little reliable information on informal manufacturing activities, whose size can be substantially larger than the formal manufacturing sector analyzed here.

Figure 12.
Figure 12.

Distribution of Value Added and Employment in Manufacturing

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

B. Contribution to Aggregate Productivity Growth

51. LICs that experienced larger reduction in agricultural share and more diversification have reaped economy-wide productivity gains (Figure 13). Larger reduction in the agricultural share, on average, is associated with faster aggregate labor productivity increases during 1990-2010 period. This is intuitive since agriculture tends to be less productive than the rest of the economy, so that reallocation of labor from a less to more productive sector raises economy-wide labor productivity (and hence temporarily boosts productivity growth). Similarly on the trade side, economies with a more diversified export base have registered stronger productivity gains.

Figure 13.
Figure 13.

Aggregate Productivity Growth, 1990-2010

(Annual Average, Percent)

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

Sources: UN National Accounts database, Conference Board, and IMF staff calculations.Notes: Diversification is measured by the Theil index, averaged over 1990-2010.Structural change is measured by the reduction in agricultural share between 1990 and 2010.

52. The contribution of sectoral shifts to aggregate productivity growth varies greatly across regions (Figure 14). A shift-share analysis decomposes aggregate labor productivity growth during 1990-2007 into relative contributions of within-sector productivity growth and a sectoral shifts component (see e.g., McMillan and Rodrick, 2011).15 It shows that inter-sectoral labor reallocation has been most productivity-enhancing in Asia, whereas it contributed negatively to aggregate productivity growth in LAC and SSA. In addition, while within-sector productivity growth has been the main driver of economy-wide productivity, there was considerable heterogeneity in the relative importance of different sectors. For Asia’s LICs, tradable sectors such as agriculture and manufacturing were the primary sources of productivity growth. Mining played a relatively dominant role in LAC, ECA, and to an extent SSA. Services, in particular wholesale and retail trade, were a relatively important part of SSA economies and made a sizable contribution to aggregate productivity growth during 1990-2007.

Figure 14.
Figure 14.

Decomposition of Aggregate Productivity Growth, 1990-2007

(Annual Average, Percent)

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

Source: IMF staff calculations.Notes: Within X refers to the contribution of productivity growth in sector X to aggregate productivity growth. Sectoral shifts refer to the contribution of intersectoral labor reallocation to aggregate productivity growth.

53. Productivity-enhancing labor reallocation is closely linked to reduced employment in agriculture. Changes in the sectoral composition work to enhance aggregate productivity if labor moves to activities with relatively higher productivity. In economies where sectoral shifts have contributed positively to aggregate productivity growth (e.g.,, Vietnam, Ethiopia, and Albania), employment has migrated out of agriculture – often the sector with lowest average productivity and largest employment share – and into more productive activities such as construction, manufacturing, and services (Figure 15). Thus in these economies, there was a positive correlation between sectoral productivity and employment changes. At the other spectrum are countries that have experienced productivity-reducing sectoral shifts (e.g., Nigeria, Zambia, and Bolivia). In Nigeria and Zambia, for example, the employment share of agriculture has increased between 1990 and 2007. In Bolivia, less productive services have gained employment share at the expense of more productive industries such as transportation and communication.

Figure 15.
Figure 15.

Sectoral Productivity and Employment Changes in Selected Economies

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

Source: IMF staff calculations.Notes: Marker size denotes employment share of the sector circa 1990.Employment shifts are annual changes in employment shares (percent) between 1990 and 2007. Relative productivity (log) is productivity of the sector relative to average, at the end of the period.

54. High productivity growth in agriculture and a diversified export base are associated with productive sectoral shifts (Figure 16). Part of the process of moving labor out of agriculture would occur unconditionally, i.e., countries with an initially large agricultural sector have more scope to reduce the agricultural share and therefore are more likely to benefit from structural transformation. But improving productivity growth in agriculture, according to one theory of structural transformation16, is critical to facilitate labor movements. Countries that have gained from structural transformation (e.g., Vietnam, Albania) tend to have high productivity growth in agriculture during the period.17 The contribution of sectoral shifts to aggregate productivity also tends to be higher in more diversified versus less diversified economies, and in the group of non-commodity exporters compared to the group of commodity exporters.

Figure 16.
Figure 16.

Correlation with Sectoral Shifts Component

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

Sources: IMF staff calculations.Notes: Sectoral shifts refer to the contribution of intersectoral labor shifts to aggregate productivity growth. Diversification is measured by the Theil index (low values imply more diversified).High corresponds to values below 25th percentile, low is above 75th percentile.

55. Productivity gaps between agriculture and non-agricultural sectors remain large in LICs, implying considerable scope for further productivity gains from either within-sector productivity growth or labor reallocation. The gap is measured as agricultural productivity as a percentage of labor productivity in non-agricultural sectors (i.e., industry and services).18 It averaged about 30 percent in 2005 for LICs—little changed from a decade ago, compared to 40 percent in emerging market economies, but with considerable dispersion across countries (Figure 17, top left panel).

Figure 17.
Figure 17.

The Agricultural Productivity Gap in LICs

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

56. Less diversified economies (in terms of exports) tend to exhibit much wider productivity gaps compared to more diversified economies; similarly, the productivity gap is larger on average for commodity exporters (Figure 17, top right panel). There is also a clear relationship with the level of development. From a cross-country perspective at a point in time, LICs tend to have lower relative productivity in agriculture (Figure 17, bottom left panel). However, as McMillan and Rodrick (2011) found (and replicated here), there is a U-shape relationship across countries and over time, so that relative agricultural productivity would initially worsen (as the non-agricultural sectors expand) before improving once the economy achieves a certain level of development (Figure 17, bottom right panel).

C. Drivers of Sectoral Productivity and Structural Transformation

57. This section presents cross-country evidence on the policy drivers of within-sector productivity growth and sectoral shifts. The stylized facts presented in the previous section show considerable heterogeneity in countries’ experience; some LICs have managed to reallocate labor towards productive sectors and kick-start sectoral productivity growth, whereas others have been less successful. While there is a general consensus that the country’s structural and institutional settings matter, little empirical evidence is available on the type of policy measures that can help remove the impediments to a successful transformation, particularly in the LICs context. Notwithstanding challenges in assessing the impact of polices on performance as well as questionable data quality for LICs, this paper try to fill this gap, focusing on de jure type of reform measures that are at the disposal of policy makers.19

Model and Data

58. The following specification for the behavior of sectoral productivity growth is postulated:
Δyi,t=α+βyi,US,t−1+γXi,t−1+μt+vi+εi,t

Here, yi,t denotes the logarithm of average labor productivity in either agriculture, manufacturing, or services in country i at time t. Thus the dependent variable is the annual productivity growth rate at the sector level. There is also interest in explaining resource shifts across sectors of the economy, in which case yi,t denotes the value added or employment share of agriculture, manufacturing, or services.

59. The goal is to identify the policy and institutional variables that matter for sectoral productivity growth and structural transformation in LICs. A large literature (e.g., Prati et al., 2013; Buera and Shin, 2011) has discussed the role of structural reforms in removing distortions and boosting productivity growth. The model includes a range of reform indices capturing reforms in various areas such as international trade, domestic financial sector, and product markets.20 This set of key structural reforms is complemented with other key variables such as labor cost, labor market and business regulation, infrastructure stock, and education. These variables (lagged one period) enter the model one-by-one as Xi,t−1. In addition, the initial sectoral productivity/share gap with the US, yi,US,t−1 = yi,t−1 − yUS,t−1, is included to capture possible convergence effects. μt and νi denote year and country fixed effects, the latter controlling for time-invariant country characteristics that may affect sectoral productivity/shares as well as the adoption of reforms.

60. The two-step GMM estimator proposed by Arellano and Bond (1991) is used to estimate the econometric model. The model above can be rewritten as:21
yi,t=α+(1+β)yi,t−1+γXi,t−1+μt+vi+εi,t

It is well-known that in dynamic models with the lagged dependent variable included as a regressor, fixed effect OLS estimates are inconsistent since the lagged dependent variable is correlated with the lagged error term. The problem diminishes as the number of time periods increases, but there are only 15 years in the sample. Thus the GMM estimator, which takes the first differences of the above equation to remove country-specific unobserved heterogeneity and uses two or more lags of the dependent variable as instrument, is necessary to obtain consistent estimates. The model is estimated using a de-trended measure of yi,t, with country-specific linear trends. The panel consists of 28 LICs for the period 1995-2010. While the time window is primarily constrained by availability of sectoral data, it corresponds to the period of unprecedented growth in LICs.

D. Results

Sectoral productivity

61. A number of structural reforms are associated with boosting productivity growth at the sector level, but different sectors require different policy focuses (Table 6). For example, removing tariff barriers to international trade and financial sector reform in the area of interest rate controls have a positive effect on agricultural productivity growth in LICs. In theory, tariff liberalization can improve the efficiency of farming through better market and technology access, cheaper imported inputs, and greater competition with imports. However, the effect of other reform measures, including agricultural reform, on agricultural productivity is not statistically significant.

Table 6.

Policy Determinants of Sectoral Productivity

article image
Note: Estimation method is two-step GMM, with 3 lags of dependent variable as instrument. Reforms indices are from Prati and others (2013) and are normalized between 0 and 1. Labor cost is the average tax wedge. Fraser regulation index ranges from 0 to 10, with higher score indicating less regulation. Road infrastructure is the log of per capita road network. Tertiary education is the percentage of high school. Policy determinants enter each regression one by one. All specifications include year fixed effects. Significant at * 10%, ** 5%, and *** 1%.

62. Domestic financial sector reforms, capital account liberalization (FDI), and improvement in road infrastructure and tertiary education matter for manufacturing productivity. Reforms in the domestic financial sector encompass several aspects, i.e., removal of credit and interest rate controls, privatization, and entry liberalization in the banking sector. By removing distortions and forcing banks to be efficient, these reforms can improve the allocation of capital in the economy, which has been found to boost productivity (Banerjee and Duflo, 2005; Hsieh and Klenow, 2009). In a similar vein, several empirical studies have found positive “spillover” effects of service sector reforms on manufacturing productivity, given that services form an increasing proportion of the inputs used in manufacturing as the economy develops22. Improving the quantity and quality of the road network is essential, particularly for LICs, to foster connectivity to markets and facilitate the production and distribution of goods and services. More education at the tertiary level also benefits skill-intensive manufacturing.

63. Liberalization of network industries, specifically telecommunication, is found to generate productivity gains in the service sector. These reforms aim to enhance competition among the providers of telecommunication services, enabling greater innovation, more FDI, and better access to these services at lower prices for both consumers and businesses. Productivity gains from networks liberalization have been found for OECD countries (Boylaud and Nicoletti, 2001; Bena et al., 2011), and it is encouraging that similar findings are extended to LICs context. Improving road infrastructure also has a positive effect on service productivity growth, especially given that distribution services (e.g., transportation, wholesale and retail trade) are becoming increasingly important for many LICs.

Sectoral shifts

64. Policies and institutions can also have direct impacts on sectoral shifts (Table 7). This section consider shifts in both the (real) value added23 and employment shares of the sectors. While employment shifts may be the preferred measure of structural transformation, sectoral employment data for LICs are particularly patchy, and need to be complemented with value added shifts. As discussed above, policies may have indirect impacts on sectoral shifts through their impact on sectoral productivity, which are not considered in this simple framework.

Table 7.

Policy Determinants of Sectoral Shifts

article image
Note: Estimation method is two-step GMM, with 3 lags of dependent variable as instrument. Reforms indices are from Prati and others (2013) and are normalized between 0 and 1. Labor cost is the average tax wedge. Fraser regulation index ranges from 0 to 10, with higher score indicating less regulation. Road infrastructure is the log of per capita road network. Tertiary education is the percentage of high school. Policy determinants enter each regression one by one. All specifications include year fixed effects. Significant at * 10%, ** 5%, and *** 1%.

65. There is some evidence that removing tariff barriers to international trade is associated with resources moving from agriculture to manufacturing and services. Structural shifts occur both in terms of employment and value added shares (although the latter mostly in manufacturing). The role of trade openness in generating structural transformation has recently gained attention in the theoretical literature, given the weak ability of traditional theories in explaining the observed patterns of structural shifts (see e.g., Matsuyama, 2009; Yi and Zhang, 201124). The intuitive explanation is that trade openness changes the relative prices across sectors, inducing resources to move into the sectors with relative comparative advantage. Cross-country empirical evidence is more mixed and varies with the set of trade liberalization episodes used, the level of sectoral disaggregation, among others.25 However, individual case studies have found an important role of trade policies in increasing labor demand within manufacturing, thus contributing to moving labor from agriculture into manufacturing activities (see e.g., McCaig and Pavcnik, 2013 for Vietnam).

66. Structural transformation also responds to the infrastructure and regulatory environment, which affects labor mobility, and the supply of skilled labor. In particular, liberalization of the electricity market, easing credit and labor market and business regulations (as captured by the Fraser regulation index) are linked to higher industry employment share. Improving tertiary education is found to increase the value added shares of manufacturing and services.

E. Conclusions

67. Sectoral shifts and sectoral productivity growth are two key features of the growth and development process. This section documents stylized facts on the patterns of sectoral shifts and productivity in LICs, and empirically examines their policy and institutional drivers. Despite many years of progress, the state of structural transformation in LICs remains low and uneven. Some economies, especially those in Asia, have been able to engender robust productivity growth across economic activities, continue to move labor out of traditional agriculture, and produce and export higher value-added and new products. Meanwhile, others have not been as successful; their economic structure continues to be concentrated in a small number of low value-added activities, with little technology and skill spillovers to the rest of the economy.

68. Structural transformation does not occur automatically and countries need to have the “right” conditions in place. This means having a structural and institutional setting that is conducive for productivity growth and labor mobility. The empirical exercise here, based on a sample of LICs reform experience over the past fifteen years, seeks to identify policy measures that can potentially remove distortions and provide the short-term impulse to sectoral productivity and sectoral shifts. A number of policy/reform measures have proven effective, including removing tariff barriers, reforming the financial and networks sectors, and improving education, infrastructure, and the regulatory framework, but challenges in different sectors require different focus.

Opportunities for Quality Upgrading in Low-Income Countries26

A. Introduction

69. Economic development is associated with the transformation of a country’s economic structure in two dimensions—both are important for LICs. These dimensions are horizontal (across sectors) and vertical (within a sector). Diversification into new higher value added sectors is the horizontal dimension. Quality upgrading is the vertical dimension and focuses on producing higher quality (and generally higher priced) products within existing sectors. Producing higher-quality varieties of existing products helps build on existing comparative advantages to boost export revenues and productivity. This section shows that both of these dimensions are important for LICs’ development. It focuses particularly on the quality upgrading dimension—which has been less explored to date—and its link to economic development.

70. The two dimensions—sectoral diversification and quality upgrading—are complementary. The potential for quality upgrading is considerably higher in some products than others (Khandelwal, 2010). Notably, it has been found to be higher in manufactures than in agriculture and natural resources. Among LICs, some currently remain specialized in products with limited quality upgrading potential. Consequently, diversification is a precondition for these countries to reap large gains from quality improvement. Meanwhile, many LICs are already engaged in sectors with large quality upgrading potential and could harness it as a driver of development (Hausmann et al., 2007; Sutton and Trefler, 2011).

71. Quality cannot be directly observed and needs to be estimated. Unit values, that is, average export prices for each product category, are the closest observable proxy and interestingly they increase with GDP per capita (Schott, 2004; Hummels and Klenow, 2005). This sparked an interest in estimating export quality, for which unit values are at best a noisy proxy, because they are also driven by a series of other factors, including production cost differences, firms’ pricing strategies, and the fact that shipments to more distant destinations typically consist of higher priced goods.

72. The multi-level export quality database used in this section is now available to Fund economists as an online toolkit.27 The database is developed in Henn, Papageorgiou, and Spatafora (2013) and provides quality measures that correct unit values for these above factors.28 Built with the motivation of achieving the best possible LIC coverage, including going back in time, this database is far more extensive than previous efforts. It covers 178 countries and 851 products over 1962–2010.29 At the most disaggregate SITC 4-digit level it consists of more than 20 million product-exporter-importer-year observations. Quality estimates are also supplied at the 1) SITC 3-, 2-, and 1-digit levels; 2), country-level for the BEC classification (which allows links to national accounts data); and 3) three broad sector level for the BEC classification (agriculture, manufactures, and non-agricultural commodities). To enable cross-product comparisons, all quality estimates are normalized to the world frontier quality, which is assumed to be the 90th percentile in each product-year combination. The resulting quality values typically range between 0 and 1.2. At each aggregation step, the normalization to the 90th percentile is repeated. This normalization implies that if a country’s quality measure is rising, it is upgrading quality faster than the world on average.

73. Based on this new database, this section highlights important stylized facts on quality upgrading in low- and lower-middle-income countries. Notably, on a regional basis, quality upgrading among low- and lower-middle-income countries was most strongly driven by East Asian success in the manufacturing sector. The apparel sector has constituted an important first beachhead in the manufacturing sector, and these countries have been quite successful in upgrading quality in this sector. Among LICs, only non-fragile countries managed to upgrade their quality considerably and have further potential to do so. In contrast, non-fragile LICs have thus far been left behind in many sectors, and some indeed will first need to achieve horizontal diversification into new sectors to enable considerable quality upgrading. At the quality levels currently produced by LICs, higher quality typically translates into higher export prices, so that these countries could experience an improvement in their terms of trade from such upgrading.

74. Higher export quality is associated with higher incomes, and quality upgrading is associated with growth. Quality upgrading is particularly rapid during the early stages of development, with quality convergence largely completed as a country reaches upper-middle-income status. This suggests that LICs may gain considerably from quality upgrading. There is wide variation in quality upgrading experiences across countries; thus, it is possible to identify its strong association with GDP per capita growth. LICs that converged fast in quality during 1995-2010 typically grew one percentage point per year more than countries that converged slowly. The link between quality upgrading and growth is strongest for manufacturing, and quality upgrading in manufacturing is accompanied by increasing share of the sector in LIC economies. Ample quality upgrading opportunities also exist in agriculture, but these are typically associated with a rebalancing of the sector toward higher value products, and increasing productivity typically leads the sector to set free resources.

75. Important policy implications are derived by investigating the determinants of quality upgrading. Given that quality upgrading is associated with growth, it is important for policymakers to be aware of its determinants to be able to harness it for development. An important result is that, once a country has entered an export product, quality tends to converge unconditionally over time. This suggests that policymakers should focus on facilitating entry into new markets, both for domestic and foreign firms, particularly if those are characterized by large quality upgrading potential. Moreover, improvements in institutional quality and human capital are associated with faster quality upgrading. Meanwhile, there is no evidence that lack of demand for quality in LICs’ and lower-middle-income countries’ existing destination markets constrains their quality upgrading prospects. Consequently, opening up new export markets could take a lower priority initially.

B. Export Quality: Stylized Facts

Quality Developments across Country Groups and Broad Sectors

76. There is a marked contrast between non-fragile and fragile LICs: only non-fragile LICs have upgraded their export quality considerably (Figure 18). In the late 1980s, average export quality of non-fragile LICs was among the lowest five percent worldwide. Since then, non-fragile LICs have upgraded substantially and their average quality level has risen to three-fourths of the world frontier quality, with the manufacturing sector of particular importance in underpinning this trend. Meanwhile, fragile LICs have not been able to converge in quality during the same period and remain among the lowest quality exporters in the world. This fragile/non-fragile pattern may partly underlie the marked difference in these two groups’ annual GDP per capita growth since 1990 (3.4 vs. 1.1 percent). Another important early takeaway is that quality tends to evolve gradually.

Figure 18.
Figure 18.

Export Quality in LICs Quality Over Time for All Sectors Relative to Quality Ladder

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

77. Quality upgrading among low- and lower-middle-income countries was most strongly driven by East Asian success in the manufacturing sector. Figure 19 illustrates this for regional aggregates of low- and lower-middle-income countries. In the manufacturing sector, there has also been some quality upgrading since 1990 in Sub-Saharan Africa, though very gradual. Meanwhile Latin America’s quality has been stagnant in the manufacturing sector and falling until the early 2000s in the agricultural sector. In agriculture, there are some indications that quality upgrading is now also underway since 2000 in East Asia and the Middle East and North Africa. Further analysis in shows that heterogeneity in quality upgrading experiences is not limited to regions, but is also strong among countries within those regions.

Figure 19.
Figure 19.

Export Quality by Region for LICs and Lower Middle Income Countries

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

78. Quality increases lead to higher export prices at the quality levels at which most LICs currently produce (Figure 20). The data suggest that quality increases translate into export price increases until a country’s quality level reaches about 80-85 percent of the world frontier. With LICs’ quality levels for most products being below that level, this implies that quality increases would likely result in terms of trade improvements for them. Quality increases beyond that 80-85 percent level tend to not to drive prices higher, possibly because higher efficiency in production may keep prices stable.30 Also, quality increases are particularly strongly correlated with price increases in agricultural goods, a key sector for both exports and employment in most LICs.

Figure 20.
Figure 20.

Quality and Unit Values

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

Notes: Each dot depicts an exporter-year combination. The 90th percentile is set to unity for both unit values and quality observations.

C. Quality in Important LIC Export Sectors

79. Fragile LICs have been lagging behind non-fragile LICs in quality upgrading in many important sectors (Figure 21). The figure illustrates three SITC 2-digit sectors which make up a high percentage of LIC exports: fruit and vegetables; coffee, tea, cocoa, and spices; and apparel. In all three cases, the better performance of non-fragile LICs is again apparent.31 The coffee, tea, cocoa, and spices sector is a traditional agricultural export sector for LICs. Non-fragile LICs have been successful in turning quality increases into market share gains in this sector. Since the early 2000s, fragile LICs have been reversing previous quality declines but have not yet achieved market share gains in response.

Figure 21.
Figure 21.

Quality, Export Prices, and Market Shares in Two Important LIC Export Sectors

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001

80. Fruits and vegetables is the agricultural subsector in which many African LICs have experienced new export success during the last decade. This again was driven by non-fragile LICs. They have been strongly upgrading quality in this sector since 1990, yielding steep world market share increases starting about five years on, which were later further boosted by price reductions. There are anecdotal success stories underpinning this success, often driven by integration into regional or global value chains through capacity building. For instance, Zambia become a net exporter in the sector shortly after the entry of foreign-owned supermarket chains boosted local producers’ capability to meet international food standards.32 Meanwhile, fragile LICs’ quality has been declining during the same time and they were unable to increase their market share.

81. The apparel sector is of particular importance for LICs because it is typically one of the first manufacturing sectors a country enters. Non-fragile LICs have been particularly successful in this sector, recording quality increases since the late 1980s and drastic increases in their market share. This trend has been mostly driven by the Asian LICs such as Bangladesh and Vietnam, which are today among the largest exporters in the sector. East Africa has also been increasing its world market share, though from a still low base. In contrast, LICs in other regions of Africa have generally not been able to enter this export industry on a scale worth mentioning. In this sector, fragile LICs have also been upgrading their apparel export quality since 2000, but their market share has not yet responded. Henn et al., (2013) expore further country-specific experiences and illustrate for instance that Korea and Thailand have entered and subsequently withdrawn from the apparel sector as they developed further.33 Moreover, they show that China’s (and to more limited extent India’s) success in the sector has come against a backdrop of substantial quality increases against stable prices.

D. Quality Ladders: Potential for Quality Upgrading

82. Countries’ positions on sectoral quality ladders indicate potential for quality upgrading in the existing product basket. Figure 22 illustrates such sectoral quality ladders at the relatively aggregated SITC one-digit level alongside the composition of export baskets in 2010 for a series of LIC groupings. The length of quality ladders varies considerably by sectors, and likewise a country’s relative position may vary considerably across sectors. This remains the case also when looking at the most disaggregated SITC four-digit level as demonstrated by Henn et al., (2013).

Figure 22.
Figure 22.

Quality Ladders by SITC1 Sector, 2010

Citation: Policy Papers 2014, 039; 10.5089/9781498343664.007.A001