I. Introduction

There is a strong interest among policy makers in understanding and achieving diversification in low and middle-income countries. This is especially true for commodity exporters, who are highly susceptible to commodity price fluctuations, and who face significant macroeconomic challenges and struggle to grow in low commodity price environment. In addition, with natural resources wealth likely to run out in the future many commodity exporters are often hard pressed to find an alternative engine of growth.

Conceptually, the objective of diversification appears to go against the benefits of specialization driven by comparative advantage, which is the cornerstone of trade theory. However, greater openness to trade and specialization increase exposure to sector-specific external shocks which may be difficult to insure against. Diversification, therefore, can help to reduce macroeconomic volatility and empirical evidence shows that this is indeed the case (IMF (2014), Haddad et al. (2013), and Koren and Tenreyro (2007)). Moreover, there is evidence that diversification can support growth in low income countries. IMF (2014) and IMF (2017), which focused on making a case for diversification, find that diversification in both exports and output is a key determinant of growth for low-income countries (LICs).

In this paper we build on Cadot et al. (2011b), IMF (2014) and IMF (2017). First, like Cadot et al (2011b), we focus on the drivers of diversification, but identify the key factors robustly associated with export diversification from a much larger set of potential drivers.² Second, we put a lens on export diversification in resource rich developing economies, something not explored in the three aforementioned studies; as emphasized above the need for and benefit from diversification is potentially the greatest in such economies. Third, given the potentially large set of factors that can affect diversification and the lack of consensus on the channels through which these effects get transmitted, we use Bayesian Model Averaging (BMA) to address model uncertainty. Importantly, the BMA approach also ranks variables in order of importance vis-a-vis their explanatory power. IMF (2014) and IMF (2017) also employ BMA in a growth regression framework.

The BMA analysis highlights some key factors that, expectedly, predispose countries toward lower export diversification – size of the economy and natural resource abundance. Smaller and resource abundant countries tend to be less diversified. However, resource abundance impedes diversification not only at the intensive margin – distribution of export earnings across a fixed set of products -, but more importantly also at the extensive margin – change in distribution of export earnings due to (net) addition of products. In other words, even among the commodity exporters, countries with higher natural resource rents are likely to export fewer products after controlling for other factors, including quality of institutions. The negative effect on the extensive margin is akin to the Dutch disease effect of resources on growth discussed in the context of the literature on the resource curse.

Among the set of actionable policy areas, BMA reveals the following to be most robustly associated (in order of importance vis-a-vis their explanatory power) with higher export diversification: (i) higher levels of human capital, with primary education being the key driver via the extensive margin, but secondary education being more important in commodity exporters mainly through the extensive margin; (ii) greater openness to trade especially through the extensive margin; and (iii) higher quality institutions, proxied by quality of governance.

In addition, we find that more developed financial sector can help diversification, especially among commodity exporters, at the intensive margin; better infrastructure, measured by phone connectivity, is associated with higher diversification through the extensive margin, but this association is weak among commodity exporters; and greater openness to capital flows also helps diversification by positively affecting the intensive margin of diversification, but not among commodity exporters.

We also augment our baseline set of macro variables, with World Bank’s Doing Business indicators. However, this comes at the cost of reducing the sample size due to a shorter time series availability of these indicators. With the truncated sample, we find that lower costs of trading across borders are associated with greater diversification.

The rest of the paper is organized as follows. Section II reviews the related literature, section III discusses the data and methodology, section IV presents some stylized facts, Sections V and VI discuss the results on the drivers of diversification for all countries and commodity exporters, respectively, section VII presents the results with Doing Business indicators, and section VIII draws policy lessons from the analysis. The last section concludes.

II. Related Literature

The process of economic development involves sustained growth in GDP per capita along with structural change, where the latter refers to the reallocation of factors of production and output across different activities or sectors. Most of the literature has investigated structural change at an aggregate level, which typically entails reallocation from agriculture to industry and then to services (see Herrendorf et al. (2014) for an exhaustive review).³

While this aggregate analysis of structural transformation is encompassed in the concept of diversification, it was the seminal work of Imbs and Wacziarg (2003) (IW henceforth) that analyzed the process of reallocation of output and labor along the development path at a more granular level – cross-country data across 9 broad sectors as well as 28 manufacturing industries. They found that there exist “stages of diversification” along the development path, which refers to the U-shaped relationship between specialization and income – starting with low levels of income countries diversify as their income rises, and then start re-concentrating/specializing once they reach the income level of upper-middle income and advanced-economies. This was later corroborated by Klinger and Lederman (2006) and Cadot et al. (2011a) for diversification across export products, with the latter using highly disaggregated export data.⁴ Focusing on introduction of new export products – extensive margin of exports -, both studies find that diversification in poor countries is mostly due to the extensive margin. IMF (2014) also finds evidence for this among the LICs, with a broader move from agriculture to manufacturing exports.⁵

IW interpret the stages of diversification through the lens of a Ricardian trade model, where an interaction of rising productivity (increases the range of goods produced) and declining trade costs (increases concentration) delivers the U-shaped relationship between diversification and income. One such combination is where closing the technological gap becomes increasingly difficult, but trade costs decline linearly. Klinger and Lederman (2006), on the other hand, emphasize the importance of market failures in new product discoveries, as put forth by Hausman and Rodrik (2003), which suggests that successful new product discoveries create valuable social knowledge. However, this type of innovation cannot be protected from imitation, and hence entrepreneurs will underinvest in the experimentation necessary to discover new export opportunities, causing the process of productive diversification to stagnate. More recently, Samaniego and Sun (2016) develop a multi-sector closed economy model in which different productivity growth rates across industries and sectors lead to structural transformation as well as the U-shaped diversification pattern in a closed economy setting.⁶

IMF (2014) (as well as IMF (2017)) does not take a stand on the mechanism(s) behind the process of diversification. Instead it focused on making an empirical case for diversification by providing evidence for a causal effect of diversification on growth in LICs. Given the multiplicity of potential mechanism affecting diversification and the resulting uncertainty in model specification, we adopt the BMA methodology used in IMF (2014).⁷ It allows us to parse through the potentially numerous proximate determinants of diversification, and ranks them based on a posterior inclusion probability.

We go beyond the link between diversification and income and explore a much richer set of determinants of diversification than existing papers (including Cadot et al. (2011b), IMF 2017), which includes the following country features – size, resource wealth, trade, market access, trade costs, FDI, human capital, public investment and expenditure, exchange rate misalignment, terms of trade, financial market development, infrastructure, and quality of institutions. Furthermore, we also bring in information from Doing Business Indicators into our analysis, which has not been examined so far in the literature.

The paper also contributes by focusing specifically on resource dependent economies, which are exposed to commodity price fluctuations, and have struggled to diversify away from natural resource dependent industries. Though IMF (2014) excludes commodity exporters from its analysis of diversification among LICs, it documents that commodity exporters tend to be less diversified, have lower quality exports and larger agricultural employment as compared to diversified exporters.

The literature on resource dependence has focused on the link between resource abundance and growth. It found that resource rich economies experienced slower growth, which came to be known as the natural resource curse – Auty (1993), Sachs and Warner (1995), Sala-i-Martin and Subhramanian (2003) and Kaldor, Karl and Said (2007). However, recent evidence questions the presence of a curse. Resource wealth can create incentives to block institutional development and thereby prevent checks and balances on excessive rent seeking – Collier and Hoefler (2009), Bulte and Damania (2008), Caselli and Cunningham (2009). Brunnschweiler and Bulte (2008) argue that measuring natural resource dependence by the share of primary products in either total exports or in GDP makes it endogenous to bad policies and institutional breakdowns. The intuition is that resource sectors are well protected enclaves that survive institutional breakdowns whereas manufacturing doesn’t, and hence bad policies and poor institutions can mechanically raise the importance of the resource sector while also causing poor subsequent growth. We explore the importance of various factors that are important for diversification among commodity exporters, including size of resource rents, rate of resource depletion, quality of institution, change in terms of trade and exchange rate overvaluation.

Our paper attempts to shed light on the question: how should a country try to diversify? Callen et al. (2014) finds that improving incentive structure for workers and firms is key to diversification in the Gulf countries. Cherif et al. (2016) notes the importance of industrial policy to develop and diversify the economy. Our paper does not delve into labor market reforms or industrial policy due to data limitations.

III. Methodology and Data

The diversification index of a country is a Theil index following Cadot et al. (2011a) and calculated as follows for each year:

where x_ik is the value of product k exported by country i,n is the total number of export products, and μ is average exports defined as

A desirable property of the Theil index is that it can be decomposed into a between group (D_i,between) and a within group (D_i,within) component.

where j = 1, … , J represents J partitions or groups of export products, n_j is the number of products and μ_j is the average export value in group j Note that the term in the big parenthesis on the right-hand side of the equation above is the Theil index for group j The between group component captures diversification resulting from the extensive margin of trade, i.e. net addition of exported products. The within group component captures diversification due to a more even distribution of export sales across the existing set of exported products. By construction, lower value of the indices corresponds to higher degree of diversification.⁸ We follow IMF (2014) to construct the index using the 6-digit HS product level data on exports of each country and define groups at the 4-digit level.

Following, Cadot et al. (2011b), we estimate the following model for data spanning 1990 to 2015:

where D_i,t is the average of the diversification index for country i over a five-year period t⁹. In (1), Y_i,t is the average log of real GDP per capita and P_i,t is average log of population for country i, in period t. λ_t are time specific effects captured by including time fixed effects through period specific dummies. v_i is set of country specific variables that do not vary over time. v_i includes latitude of a country and its 1970 GDP per capita to control for geography and convergence. ¹⁰

Compared to Cadot et al. (2011b), there are two main enhancements – (i) the number of determinants in X_i,t is much bigger, and (ii) instead of OLS we use Bayesian Model Averaging (BMA).

The vector of potential determinants of diversification, X_i,t, includes – resource rents and resource depletion, openness (goods and capital markets), market access, foreign direct investment (FDI), human capital, public expenditure and investment, exchange rate misalignment, terms of trade, financial market development, infrastructure, and quality of institutions (the complete list, including sources of key variables are shown in Table 1 in Appendix 1). Physical and human capital variables are lagged one period in (1).

With the large number of potential determinants of diversification there is significant uncertainty about the appropriate model. Hence we apply Bayesian Model Averaging (BMA) technique to determine which of the variables in vector X_i,t are robust to model uncertainty and are important in explaining the variation in diversification.¹¹ In the BMA exercise, we keep $$Y_{i,t},Y_{i,t}^2,P_{i,t}$$ and v_i fixed (always included) in the model, while allowing the algorithm to iterate over the potential determinants in X_i,t. Thus, if there are N variables in X_i,t , there are 2^N possible models to estimate with us being agnostic about the “true” model from this model space. The estimates of γ are the weighted average of the parameter across different estimated models, with the weights being the posterior inclusion probability (PIP) of the variable in the “true” model. In interpreting the BMA results, following the literature we will consider variables with PIP above 0.5 to be robustly correlated with export diversification.¹² The PIP also provides a way to rank the variables with regards to their importance in explaining variation in diversification across countries and over time. Higher PIP signals that a variable is more robustly associated with diversification. We also estimate (1) using OLS.

In addition to estimating (1) for total Theil index, we also estimate it for between group Theil index (export diversification at the extensive margin) and within group Theil index (export diversification at the intensive margin). Lastly, we report the results for 4 different samples -All countries (92), Emerging markets and developing economies (EMDE) (73), EMDE commodity exporters (EMDE-CE) (28) and EMDE diversified exporters (EMD-DE) (45).¹³

The Doing Business indicators from the World Bank which measure various aspects of the business climate can also signal potentially binding constraints on diversification. However, data for these indicators are only available from 2004, which significantly truncates our sample. Hence, we run two separate sets of regressions, one without the Doing Business indicators, and one with the Doing Business indicators.

IV. Stylized Facts

This section presents facts on the diversification index and some of the important potential drivers. Due to better availability of data for these key drivers as well as the data on exports, we present the facts for the period 1975–2015 in contrast to the shorter period of 1990–2015 employed in the regressions.¹⁴ Table 1 shows the distributional statistics for the Theil index (across countries) by country groups for 1975 and 2015.

Table 1.

Distribution Statistics of Diversification Index

Table 1.

Distribution Statistics of Diversification Index

Statistic	All countries		AE		EMDE		EMDE – CE		EMDE – DE
	1975	2015	1975	2015	1975	2015	1975	2015	1975	2015
Median	3.74	3.49	1.96	2.20	4.05	3.86	4.35	4.53	3.73	3.16
Mean	3.75	3.56	2.12	2.38	4.12	3.85	4.54	4.53	3.84	3.43
Inter-Quartile Range (IQR)	1.68	2.04	1.15	0.78	1.51	1.92	1.61	1.66	1.37	1.77

View Table

Table 1.

Distribution Statistics of Diversification Index

Statistic	All countries		AE		EMDE		EMDE – CE		EMDE – DE
	1975	2015	1975	2015	1975	2015	1975	2015	1975	2015
Median	3.74	3.49	1.96	2.20	4.05	3.86	4.35	4.53	3.73	3.16
Mean	3.75	3.56	2.12	2.38	4.12	3.85	4.54	4.53	3.84	3.43
Inter-Quartile Range (IQR)	1.68	2.04	1.15	0.78	1.51	1.92	1.61	1.66	1.37	1.77

The first thing to note is the difference in the level of diversification across country groups -AEs are more diversified than EMDEs, and EMDE-DEs are more diversified than EMDE-CEs. Over time, the median (as well as mean) diversification level for All countries has increased (Theil index decreased). However, trends differ between advanced economies (AEs) and EMDEs – AEs experienced a decrease whereas EMDEs experienced an increase in diversification. Among the EMDEs, the commodity and diversified exporters also behave differently – EMDE-CE have become less diversified whereas EMDE-DE have become more diversified. The dispersion in diversification, as measured by the IQR, increased between 1975 and 2015, except for AEs.

Figure 1 shows the relationship between diversification and development, proxied by GDP per capita. For the All country sample diversification increases as income increases, but the eventual decrease in diversification with higher income is not observable as was found by IW.¹⁵ This becomes apparent when we split the sample into AEs and EMDEs. There is clear non-linearity observable among both AEs and EMDEs.

View Full Size

Diversification and Development

Citation: IMF Working Papers 2019, 105; 10.5089/9781498313087.001.A001

• Download Figure
• Download figure as PowerPoint slide

Figures 2 and 3 show the relationship (unconditional correlation) between the Theil index and select potential drivers of diversification, averaged over five-year periods. Human capital, size of a country (population), quality of institutions (International Country Risk Guide (ICRG) indicator of Quality of Government (QoG)), FDI, and level of financial sector development (Credit-to-GDP) are negatively correlated with the Theil index (and hence positively correlated with, level of diversification). However, FDI exhibits weak association with diversification, particularly for EMDEs. On the other hand, trade (% of GDP) and public investment (% of GDP) exhibits a weak positive relationship. Exchange rate undervaluation exhibits weak positive relationship for All countries, but a weak negative relationship for EMDEs. Exchange rate undervaluation is constructed following Rodrik (2008), with positive values indicating the currency is undervalued and negative values indicating the currency is overvalued. Thus, for All countries undervaluation is inversely correlated with diversification, for EMDEs it positively correlated with diversification.

View Full Size

Diversification and Potential Determinants

Citation: IMF Working Papers 2019, 105; 10.5089/9781498313087.001.A001

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Diversification and Potential Determinants

Citation: IMF Working Papers 2019, 105; 10.5089/9781498313087.001.A001

• Download Figure
• Download figure as PowerPoint slide

We have a clearer intuition for the mechanism as well as the expected effect for some of these potential drivers of diversification, while for others the expected relationship is harder to ascertain. For instance, larger economies, proxied by population, will tend to be more diversified due to larger domestic markets and greater abundance of factors of productions; human capital is likely to encourage diversification through extensive margin (development of new products); better quality institutions are likely to help diversification through numerous channels, including better enforcement of contracts, limiting rent seeking and keeping barriers to entry low by imposing checks on corruption and excessive concentration of market power. A well-developed financial sector can help to increase access to credit which in turn could encourage new product creation through an increase in entrepreneurial activity and risk taking. It can also help to reduce distortions in credit allocation responsible for insufficient credit allocation to some sectors and overallocation to some others.

Trade and FDI can help diversification through investment in new products, but also deepen existing comparative advantage and thereby reduce diversification.¹⁶ Exchange rate overvaluation could undermine the tradable sectors’ competitiveness, especially of manufacturing goods, which in turn could lead to lower diversification. This is the Dutch disease phenomenon highlighted among resource abundant countries.

Higher natural resource dependence is expected to reduce diversification by deepening the endowment driven comparative advantage. This what is captured in Figure 2, where resource dependence is measured by total natural resources rents (% of GDP). Total natural resource rents are the sum of oil rents, natural gas rents, coal rents (hard and soft), mineral rents, and forest rents.

V. Drivers of Export Diversification: All Countries and Emdes

Table 2 (Appendix 1) shows the results of estimating (1) for the three Theil indices using BMA and OLS for the samples of All countries and EMDEs. Note that some of the variables – GDP per capita, GDP per capita squared, population size, latitude and GDP per capita in 1970 – are always included in the model (as discussed above) and hence have PIPs have one by construction.¹⁷ The object of interest are other variables and their inclusion probabilities after controlling for the fact that GDP per capita, GDP per capita squared, population size, and fixed features such as latitude and GDP per capita in 1970 are already in controlled for in the model.

VI. Drivers of Export Diversification: EMDE Commodity Exporters Versus Diversified Exporters

Table 3 (Appendix 1) presents the results for the baseline specification (of Table 2) for commodity and diversified exporters within the group of EMDEs.

VII. Doing Business Indicators and Diversification

We augment our baseline set of macro variables with World Bank’s enterprise survey based Doing Business (DB) indicators, which measure various aspects of the business climate and can signal potentially binding constraints on diversification. However, data for these indicators are only available from 2004, which significantly truncates our sample. Table 4 (Appendix 1) shows the results of estimating equation 1 after including the Doing Business indicators.²⁶

Among the Doing Business indicators (DBI) costs of trading across border appears to be robustly correlated with diversification – lower costs of trading across borders (higher DTF values) are associated with greater diversification. Furthermore, the cost of trading across borders affects the intensive margin of diversification. None of the other DB indicators exhibit robust association with diversification. Among the core set of macro variables, compared to the baseline specification, we observe that enrollment rates lose statistical significance. This is mainly due to loss in observations. Trade-to-GDP also loses its statistical power, except for the extensive margin for the All countries sample. This may be due to the inclusion of the cost of trading across border, which also captures trade openness.

VIII. Policy Implications

The analysis in this paper has important implications for policy makers who are seeking to diversify their economies. It highlights some of the key factors that predisposes them toward lower export diversification – size of the economy, geography and natural resource abundance. Moreover, even among the commodity exporters, countries with higher natural resource rents are likely to export fewer products after controlling for other factors, including quality of institutions. It also highlights areas where policy actions to improve export diversification are most likely to bear fruit, based on the statistical robustness of their association with diversification.

• Human capital accumulation. Education is a crucial factor for export diversification. Primary education is robustly associated with export diversification, with the dominant effect being on the set of products a country exports. Improvement in human capital can increase the number of products a country can potentially produce, and export competitively, and hence directly impacts its comparative advantage vis-à-vis trade partner, which in turn affects export diversification. For commodity exporters secondary education is the more important dimension to improve export diversification. Again, it is robustly associated with the number of products a country exports. It is only out-ranked by openness among the policy variables tested in our model. While we do not explore the importance of quality of education, the literature has established the importance of quality of education for growth and labor market outcomes in addition to the importance of quantity of education. Thus, policies aimed at improving education outcomes must focus on quantity as well as quality of education.

• Trade openness. Countries that trade more tend to have greater export diversification, both at the intensive and extensive margin. This effect is borne out even among the set of Doing Business indicators, where lower trade costs is the only variable that is important in explaining the variation in diversification. For the bigger – all country and EMDE – samples, the positive effect of openness operates mainly through the set of products exported. For commodity exporters among the EMDEs, openness can help to diversify exports along the extensive margin, and it is the most robustly associated determinant of diversification. However, this effect is countered by the greater specialization induced by trade at the intensive margin.

• Quality of institution. Higher quality of institution as measured by the quality of governance is associated with less concentrated export base, and higher overall level of diversification.

• Infrastructure. We find evidence that higher level of infrastructure provision, proxied by higher fixed telephone penetration, is important for the extensive margin of diversification. Higher mobile phone penetration is associated with diversification at the intensive margin.

• Capital account openness. Greater openness to capital flows also helps diversification by positively affecting the intensive margin of diversification.

• Financial sector development. Lastly, more developed financial sector can help to diversify EMDEs at the intensive margin, and this effect survives even among commodity exporters and diversified exporters. Thus, policies to improve both financial access and the allocation of credit across sectors (and across firms within sectors) would be beneficial, especially among commodity exporters given the significant Dutch disease effect.

Thus, resource abundant economies need to focus on overcoming the export product space limiting effect of resources wealth by having more open trade regimes, improving education outcomes for secondary and higher education, and developing their financial sector to increase access to and improving allocation of credit.

IX. Conclusion

This paper attempts to address the question of how to diversify. We identify key factors, from a pool of large number of potential determinants, that explain the variation in export diversification across countries and over time using Bayesian Model Averaging, which addresses the issue of model uncertainty. Our methodology allows us to rank variables in terms of their importance in explaining the variation in export diversification in the data and hence helps to prioritize areas for policy actions.

Expectedly, we find that natural resource abundance and smaller size (in terms of population) predispose countries towards being less diversified. However, more importantly, higher natural resource rent is associated with lower export diversification not only at the intensive margin but also at the extensive margin. The negative effect on the extensive margin is akin to the Dutch disease effect of resources on growth discussed in the context of the literature on the resource curse. Our analysis suggests that to diversify, policy makers should prioritize human capital accumulation and reduce barriers to trade. Other policy areas include (in order of importance) improving quality of institutions, quality of infrastructure and deepening financial markets. For commodity exporters: (i) secondary education is the most important driver of diversification; (ii) greater trade is associated with greater number of products exported; and (iii) improving access as well as allocation of credit

One must be careful in interpreting the findings as evidence of causality. Though we significantly expand the set of regressors, controlling for important time-invariant features of economies (initial GDP and geography), we follow the literature in adopting the specification. This implicitly views the regressors as the causal factors, but as is the case with most of the literature, our methodology does not address the issue of identification completely. While physical capital and human accumulation variables are included as lags, and trade openness is also proxied using tariff rates and Doing Business indicator on costs of trading across border, addressing causality adequately will require instruments for the full set of regressors. This is quite challenging given the large set of regressors considered.

There remain interesting avenues for further research. Our focus is on the level of diversification and not on the nature of diversification. For instance, we do not differentiate between diversification within agriculture from diversification from agriculture into manufacturing as long as the theil index changes by the same amount in both cases. But, this qualitative difference between various paths of diversification may be important from a growth stand-point, and therefore should be examined.

The role of industrial policies in enhancing diversification has garnered a lot of attention recently (see Aghion et al. (2011)), in part due to the rapid economic transformation of China. Cherif at al. (2016) and Cherif and Hasanov (2019) delve into this issue and make a case for the role of the state in the process of development. We do not address this issue mainly due to a lack of comparable data on different policy instruments across countries and over time. Leveraging firm level data to understand differences in firm level characteristics across certain groups of countries within the same industries, such as commodity versus diversified exporters is also an interesting area for future research. This could shed light on the key constraints that must be addressed to ensure desirable firm level dynamics and efficient allocation of resources (see Hsieh and Klenow (2014, 2009)). Finally, our paper did not explore labor and product market reforms, as data availability is poor especially among LICs, and these reforms could also be important for diversification.