## wpiea2019105

## Source details

**Canonical URL:** [wpiea2019105](https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019105.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2019/wpiea2019105.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2019/wpiea2019105.pdf.json)

---

### Introduction
- Purpose and extensions
  - Focus: drivers of export diversification, with explicit lens on resource rich developing economies (commodity exporters).
  - Methodological extension: use of Bayesian Model Averaging (BMA) to address model uncertainty and rank determinants by explanatory power.
  - Three main extensions relative to prior work: (i) identify key drivers from a much larger set; (ii) explicit attention to commodity exporters; (iii) use BMA.
- Key high-level findings previewed
  - Factors predisposing countries to lower export diversification: size of the economy and natural resource abundance.
  - Natural resource rents reduce diversification at both intensive and extensive margins.
  - Most robust policy-associated correlates of higher diversification (ranked by explanatory power): human capital, trade openness (and lower trade costs), quality of governance.

### Methodology and data
- Diversification measure
  - Total Theil Index constructed following Cadot et al. (2011a); decomposed into Between Theil Index (extensive margin: net addition of exported products) and Within Theil Index (intensive margin: distribution across existing exported products).
  - Lower Theil index values correspond to higher diversification.
  - Export data: 6-digit HS product level, groups at 4-digit level.
- Empirical specification and estimation
  - Baseline equation (averaged over five-year periods):
    - DDi,b = α + β1 Yi,b + β2 Yi,b2 + β3 Pi,b + γ Xi,b + λb + ηi + ei,b
    - DDi,b: average diversification index for country i over a five-year period.
    - Yi,b: average log of real GDP per capita; Pi,b: average log of population.
    - Time fixed effects λb; country-specific time-invariant controls ηi (including latitude and 1970 GDP per capita).
  - Model uncertainty: BMA using Magnus, Powell, and Prüfer (2010) and the bma command in Stata.
    - Yi,b, Yi,b2, Pi,b and ηi kept fixed (always included).
    - Posterior inclusion probability (PIP) used to rank variables; variables with PIP >= 0.5 considered robustly correlated.
  - Lags and samples
    - Physical and human capital variables are lagged one period in (1).
    - Periods used in baseline regressions: 1990-1994, 1995-1999, 2000-2004, 2005-2009, 2010-2014.
    - Stylized facts use 1975-2015.
    - Sample breakdowns: All countries (92), EMDE (73), EMDE-CE (28), EMDE-DE (45).
    - Doing Business indicators available only from 2004; two sets of regressions run (with and without Doing Business).

### Stylized facts (1975–2015)
- Cross-group patterns
  - Advanced economies (AEs) are more diversified than EMDEs; EMDE-DEs more diversified than EMDE-CEs.
  - Over 1975–2015 median and mean diversification for All countries increased (Theil index decreased); AEs experienced a decrease in diversification, EMDEs an increase; EMDE-CE became less diversified; EMDE-DE became more diversified.
  - Dispersion (IQR) increased between 1975 and 2015 except for AEs.
- Correlations (unconditional)
  - Positive association with diversification (i.e., negative with Theil index): human capital, population size, quality of institutions (ICRG Quality of Government), FDI (weak), credit-to-GDP.
  - Trade (% of GDP) and public investment (% of GDP): weak positive relationship with diversification.
  - Exchange rate undervaluation: sign varies — for All countries undervaluation inversely correlated with diversification; for EMDEs undervaluation positively correlated with diversification.
  - Resource dependence (total natural resource rents % of GDP) associated with lower diversification.

### Drivers of export diversification — key empirical findings
- Variables always included (PIP = 1 by construction): log GDP per capita, log GDP per capita squared, log population, latitude, and 1970 GDP per capita.
- Resource endowment
  - Natural Resource Rent is an important determinant:
    - OLS: coefficient 0.059***.
    - BMA: coefficient 0.057; variable has PIP = 1 across specifications.
    - Higher natural resource rent associated with lower diversification at both intensive and extensive margins; quantitative impact larger on the intensive margin.
  - Natural Resource Depletion (% of GNI) not robustly associated with overall diversification in BMA (mixed OLS results).
- Human capital
  - Primary education: robustly positively associated with overall diversification and affects both margins; dominant effect on extensive margin.
  - Secondary and tertiary education: not robust across all samples in BMA but matter in subsamples (see EMDE-CE vs EMDE-DE).
- Cross-border flows and external openness
  - Trade-to-GDP: OLS -0.003*; BMA shows strong association with greater diversification mainly through the extensive margin (PIP near 1 for extensive-margin specification).
  - Capital Account Openness: OLS -0.490***; BMA -0.572. Greater openness associated with higher diversification operating through the intensive margin.
  - Foreign Direct Investment (lagged): OLS 0.029*; BMA 0.010. Higher FDI associated with greater specialization (lower diversification) mainly at the extensive margin; correlation weaker in EMDE sample (PIP < 0.5).
- Institutions and infrastructure
  - Quality of Institution (ICRG): OLS -0.845**; BMA -0.697. Higher quality governance associated with higher diversification largely at the extensive margin.
  - Fixed phone subscriptions associated with extensive margin; mobile subscriptions associated with intensive margin. These infrastructure associations are weaker among EMDEs.
- Financial sector development
  - Credit to Private Sector: mixed results. For All countries not strongly associated; among EMDEs associated with increased diversification via the intensive margin.
- International price variables
  - Change in terms of trade and exchange rate undervaluation not significantly associated with diversification once trade flows are accounted for.

### Heterogeneity: EMDE commodity exporters (EMDE-CE) vs EMDE diversified exporters (EMDE-DE)
- Resource endowment
  - EMDE-CE: Natural Resource Rent strongly and robustly associated with lower diversification at both margins (PIP for intensive margin marginally < 0.5 in some specifications).
  - EMDE-DE: Natural Resource Rent depresses total and intensive margin diversification but is associated with higher extensive-margin diversification in BMA (extensive-margin channel with sufficiently high PIP).
  - Depletion rates: not robustly associated with higher diversification for commodity exporters.
- Human capital
  - EMDE-DE: primary and tertiary education tend to support intensive-margin diversification.
  - EMDE-CE: secondary enrollment aids extensive-margin diversification.
- Trade and investment
  - EMDE-CE: greater trade openness increases number of products (extensive) but can increase concentration (intensive), yielding offsetting effects on total diversification.
  - FDI not important among commodity or diversified exporters in BMA.
- Institutions, infrastructure, and public investment
  - Quality of institutions tends to enhance diversification but effects are sometimes statistically insignificant in BMA for EMDE subsamples.
  - Little evidence that phone subscriptions are associated with more diversification among commodity exporters.
  - Higher public investment associated with lower diversification at the extensive margin for commodity exporters.
- Financial development
  - Credit-to-private-sector helps increase diversification in both commodity and diversified exporters, significant mainly for the intensive margin.

### Doing Business indicators
- Data availability and approach
  - Doing Business (DB) indicators available only from 2004; regressions run including DB components (DTF) and excluding them.
- Key DB findings
  - Trading across Borders (DTF): OLS coefficient -0.018**; BMA -0.021. Lower costs of trading across borders (higher DTF) associated with greater diversification, affecting the intensive margin.
  - Enforcing Contracts (DTF): OLS -0.013*; BMA -0.004 (less robust).
  - Including DB indicators reduces statistical significance of some macro variables (e.g., enrollment rates), possibly because DB indicators capture trade-openness effects.

### Selected regression and sample statistics (reported)
- Observations in regressions (examples across tables):
  - 384, 313 (All Countries and selected columns)
  - 108, 205 (EMDE subsamples)
  - 130, 103 (Doing Business sample columns)
- R-squared examples reported:
  - 0.77, 0.62, 0.57, 0.71, 0.61, 0.48 (Table 2 columns)
  - 0.77, 0.78, 0.55, 0.66, 0.44, 0.54 (Table 4 columns)
  - 0.83, 0.72, 0.73, 0.83, 0.73, 0.72 (Table 5 columns)
- Selected coefficient examples (All Countries: Full Sample)
  - Log of Real GDP per Capita: OLS -0.411; BMA -0.059.
  - Log of Population: OLS -0.251***; BMA -0.261.
  - Absolute Latitude: OLS -0.009**; BMA -0.010.
  - Natural Resource Rent: OLS 0.059***; BMA 0.057 (PIP = 1 in BMA).
  - Trade to GDP: OLS -0.003*; BMA -0.004.
  - Capital Account Openness: OLS -0.490***; BMA -0.572.
  - Quality of Institution (ICRG): OLS -0.845**; BMA -0.697.
  - Trading across Borders - DTF (Doing Business): OLS -0.018**; BMA -0.021.

### Policy implications and priorities
- Structural predispositions toward lower diversification
  - Size of the economy, geography (latitude), and natural resource abundance.
  - Even among commodity exporters, higher natural resource rents associated with exporting fewer products after controlling for other factors.
- High-priority actionable policies (statistically robust associations)
  - Human capital accumulation
    - Primary education: robustly associated with export diversification (dominant effect on number of products exported).
    - For commodity exporters: secondary education more important for extensive-margin diversification.
    - Policy emphasis: both quantity and quality of education.
  - Trade openness and lower trade costs
    - Greater trade associated with greater diversification (especially extensive margin); lower costs of trading across borders (Doing Business Trading across Borders) important for intensive margin.
    - For commodity exporters, openness helps extensive-margin diversification but may increase intensive-margin specialization.
  - Quality of institutions
    - Higher quality of governance associated with less concentrated export base and higher overall diversification.
  - Infrastructure
    - Fixed telephone penetration linked to extensive-margin diversification; mobile penetration linked to intensive-margin diversification.
  - Capital account openness and financial sector development
    - Greater openness to capital flows helps diversification via the intensive margin.
    - More developed financial sector helps diversify EMDEs (intensive margin); policies to improve financial access and allocation of credit are beneficial, especially for commodity exporters facing Dutch-disease effects.
- For resource-abundant economies: recommended priorities
  - More open trade regimes.
  - Improve secondary and higher education outcomes.
  - Develop the financial sector to increase access to and improve allocation of credit to overcome export product-space limitations from resource wealth.

### Caveats on causality and avenues for further research
- Causality caveats
  - Results are associations from extensive set of regressors; methodology does not fully address identification.
  - Physical capital and human capital lagged; trade openness proxied using tariffs and Doing Business trading costs; instruments for full set of regressors needed to address causality comprehensively.
- Suggestions for further research
  - Examine the nature of diversification (e.g., within-agriculture vs. agriculture-to-manufacturing) rather than only the level.
  - Evaluate the role of industrial policies, labor and product market reforms, and firm-level constraints using firm-level data where available.

### Appendix I — Variables and regression outputs (summary)
- Dependent variables
  - Total Theil Index, Between Theil Index (extensive), Within Theil Index (intensive); all computed using UN Comtrade data.
- Key independent variables and data sources (selected)
  - Log of GDP per capita (PWT 9.0); Log of Population (WEO); Absolute latitude (Sala-I-Martin et al. (2004)); Natural Resource Rent (% of GDP) (WDI); Natural Resource Depletion (% of GNI) (WDI); Trade to GDP (WDI); Capital Account Openness (Chinn & Ito); FDI (lagged) (WEO); Primary/Secondary/Tertiary Enrollment (WDI); Public Investment (WEO); Exchange Rate Undervaluation (Rodrik (2008) approach); Credit to Private Sector (WDI); Fixed and Mobile Phone Subscriptions (WDI); Quality of Institution (ICRG QoG index scaled 0-1); Democracy (Polity 2); Doing Business components (World Bank - Doing Business).
- BMA and OLS reporting conventions in tables
  - For OLS regressions standard errors are in italic; for BMA regressions PIPs are in italic.
  - Statistics in bold for BMA regressions have PIPs >= 0.5.
- Sample sizes and time dummies
  - Time dummies included in all specifications.

*Source: wpiea2019105 (Appendix, Tables, and text as provided).*

### References ________________________________________________________________23

### wpiea2019105 - References ________________________________________________________________23

### Introduction
- Strong policy interest in achieving diversification in low and middle-income countries, especially commodity exporters facing commodity price fluctuations and eventual resource depletion.
- Conceptual tension: diversification partially opposes comparative advantage/specialization but can reduce macroeconomic volatility and support growth in low income countries (cites IMF (2014), Haddad et al. (2013), Koren and Tenreyro (2007)).
- Paper builds on Cadot et al. (2011b), IMF (2014), IMF (2017) with three main extensions:
  - Focus on drivers of export diversification and identify key factors from a much larger set of potential drivers.
  - Explicit lens on resource rich developing economies (commodity exporters).
  - Use of Bayesian Model Averaging (BMA) to address model uncertainty and rank variables by explanatory power.
- BMA highlights factors predisposing countries to lower export diversification:
  - Size of the economy.
  - Natural resource abundance.
- Resource abundance reduces diversification at:
  - Intensive margin (distribution of export earnings across a fixed set of products).
  - Extensive margin (net addition of exported products). Even among commodity exporters, higher natural resource rents associate with exporting fewer products after controlling for other factors.
- Most robust actionable policy areas associated with higher export diversification (in order of importance by explanatory power):
  - Higher levels of human capital:
    - Primary education: key via the extensive margin.
    - Secondary education: more important in commodity exporters mainly through the extensive margin.
  - Greater openness to trade, especially through the extensive margin.
  - Higher quality institutions, proxied by quality of governance.
- Additional associations:
  - More developed financial sector helps diversification, especially among commodity exporters, at the intensive margin.
  - Better infrastructure (phone connectivity) associated with higher diversification through the extensive margin (association weak among commodity exporters).
  - Greater openness to capital flows helps diversification by positively affecting the intensive margin, but not among commodity exporters.
- Doing Business indicators (smaller sample due to shorter time series) show: lower costs of trading across borders associated with greater diversification.
- Paper organization overview: Section II literature review; III data and methodology; IV stylized facts; V–VI drivers for all countries and commodity exporters; VII Doing Business; VIII policy lessons; conclusion.

### Related literature
- Structural change in development: reallocation across activities/sectors; traditional pattern agriculture → industry → services (Herrendorf et al. (2014)).
- Imbs and Wacziarg (2003) (IW): “stages of diversification” — U-shaped relationship between specialization and income (countries first diversify with rising income, then re-concentrate at higher income).
- Klinger and Lederman (2006) and Cadot et al. (2011a): diversification across export products using highly disaggregated export data; extensive margin (introduction of new export products) drives diversification in poor countries.
- IMF (2014): LICs’ diversification largely via extensive margin; move from agriculture to manufacturing exports; divergence in export quality trends across regions.
- Theoretical interpretations:
  - IW: Ricardian trade model with interaction of rising productivity and declining trade costs yields U-shaped pattern.
  - Klinger and Lederman: market failures in new product discoveries (Hausman and Rodrik (2003)) leading to underinvestment in experimentation.
  - Samaniego and Sun (2016): multi-sector closed economy model with sectoral productivity growth differences.
- IMF (2014)/(2017): empirical case for causal effect of diversification on growth in LICs; use of BMA in presence of many potential mechanisms and model uncertainty.
- This paper expands determinants beyond prior work to include: 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, quality of institutions, and Doing Business indicators.
- Focus on resource dependent economies and determinants relevant for commodity exporters: size of resource rents, rate of resource depletion, institutional quality, change in terms of trade, exchange rate overvaluation.
- Policy debate references: Callen et al. (2014) (incentive structures), Cherif et al. (2016) (industrial policy). Paper does not analyze labor market reforms or industrial policy due to data limitations.

### Methodology and data
- Diversification index: Theil index following Cadot et al. (2011a), constructed for each year as:
  - 퐷퐷푖 = 1 푛 �푥푖푖푛 휇 .푙푛(푥푖푖휇) 푛 푖=1  , where 푥푖푖 is value of product k exported by country i, 푛 is total number of export products, and 휇 is average exports defined as 휇 = 1 푛 �푥푖 푛 푖=1 .
  - The Theil index decomposes into between-group (퐷퐷푖,푏푏푏푏푏푏푏푏푏푏푏푏푏푛) and within-group (퐷퐷푖,푏푏푖푖푏푏ℎ푖푖푛) components:
    - 퐷퐷푖 = �푛 푗 푛 휇 푖푗 휇 ln(휇 푖푗 휇) 퐽 푗=0 ������������ 퐷퐷푖,푏푏푏푏푏푏푏푏푏푏푏푏푏푏 + �푛 푗 푛 휇 푖푗 휇 � 1 푛 푗 �푥푖푖푖 휇푗 .푙푛(푥푖푖푖 휇푗) 푁 푖𝑘 퐺 푗 퐽 푗=1 �������������������� 퐷퐷푖,푏푏푖푖푏푏ℎ푖푖푏푏
  - Between-group component captures extensive margin (net addition of exported products); within-group captures intensive margin (evenness across existing exported products).
  - Lower index values correspond to higher diversification.
- Data construction:
  - Follows IMF (2014): use 6-digit HS product level data on exports, define groups at the 4-digit level.
- Empirical specification (following Cadot et al. (2011b)), estimated for data spanning 1990 to 2015:
  - 퐷퐷푖,푏 = 훼 + 훽1 푌푖,푏 + 훽2 푌푖,푏2 + 훽3 푃푖,푏 + 훾푿풊풊 + 휆푏 + 흂풊 + 푒푖,푏   (1)
  - 퐷퐷푖,푏 is average diversification index for country i over a five-year period t.
  - 푌푖,푏 is average log of real GDP per capita; 푃푖,푏 is average log of population for country i in period t.
  - 휆푏 are time fixed effects (period specific dummies).
  - 흂풊 is set of country specific variables that do not vary over time: includes latitude and 1970 GDP per capita to control for geography and convergence.
- Enhancements relative to Cadot et al. (2011b):
  - Much larger number of potential determinants in 푿풊,풊.
  - Use of Bayesian Model Averaging (BMA) instead of OLS to address model uncertainty and rank determinants via posterior inclusion probability.
- Vector of potential determinants (푿풊,풊) includes: resource rents and resource depletion, openness (goods and capital markets), market access, FDI, human capital, public expenditure and investment, exchange rate misalignment, terms of trade, financial market development, infrastructure, quality of institutions.
- Margin importance note: both intensive and extensive margins matter for policy — e.g., many exported products with concentration in one product still exposes country to sectoral shocks.

### Figures and tables (listed)
- Figures:
  - 1. Diversification and Development
  - 2. Diversification and Potential Determinants
  - 3. Diversification and Potential Determinants
- Tables:
  - 1. Distribution Statistics of Diversification Index
  - 2. Baseline Specification, All Countries and EMDEs
  - 3. Baseline Specification, EMDEs - Commodity and Diversified Exporters
  - 4. Baseline Specification with Doing Business Indicator

*Source: wpiea2019105 - References ________________________________________________________________23*

### Appendix 1). Physical and human capital variables are lagged one period in (1).

### Appendix 1). Physical and human capital variables are lagged one period in (1).

### Methodology and data
- Model uncertainty addressed using Bayesian Model Averaging (BMA) following Magnus, Powell, and Prüfer (2010) and the bma command in stata.
- In the BMA exercise, 푌푌푖,푏, 푌푌푖,푏2, 푃푃푖,푏 and 흂흂풊 are kept fixed (always included) while the algorithm iterates over potential determinants in 푿푿풊풊,풊풊. If there are 푵푵 variables in 푿푿풊풊,풊풊 there are 2푁푁 possible models.
- Estimates of 휸휸 are weighted averages across models with weights given by posterior inclusion probability (PIP). Variables with PIP above 0.5 are considered robustly correlated with export diversification.
- Also estimate (1) using OLS.
- Periods used in baseline sample for regressions: 1990-1994, 1995-1999, 2000-2004, 2005-2009, 2010-2014.
- Stylized facts presented for 1975-2015 due to broader data availability for drivers and exports.
- Sample breakdowns reported: All countries (92), Emerging markets and developing economies (EMDE) (73), EMDE commodity exporters (EMDE-CE) (28), and EMDE diversified exporters (EMDE-DE) (45).
- Doing Business indicators available only from 2004; run two sets of regressions (with and without Doing Business indicators).

### Stylized facts on diversification (1975–2015)
- Advanced economies (AEs) are more diversified than EMDEs; EMDE-DEs are more diversified than EMDE-CEs.
- Over time (1975 to 2015), median and mean diversification for All countries increased (Theil index decreased), but trends differ:
  - AEs experienced a decrease in diversification.
  - EMDEs experienced an increase in diversification.
  - EMDE-CE have become less diversified.
  - EMDE-DE have become more diversified.
- Dispersion (IQR) increased between 1975 and 2015 except for AEs.
- Relationship between diversification and GDP per capita:
  - For All countries diversification increases as income increases; non-linearity is clear when splitting into AEs and EMDEs.

### Unconditional correlations (Figures 2 and 3 summary)
- Negative correlation with Theil index (i.e., positive with diversification): human capital, population size, quality of institutions (ICRG Quality of Government), FDI, credit-to-GDP.
  - FDI exhibits weak association with diversification, particularly for EMDEs.
- Weak positive relationship: trade (% of GDP), public investment (% of GDP).
- Exchange rate undervaluation:
  - For All countries undervaluation inversely correlated with diversification.
  - For EMDEs undervaluation positively correlated with diversification.
  - Undervaluation constructed following Rodrik (2008): positive values indicate undervalued currency, negative values overvalued.
- Resource dependence (total natural resource rents % of GDP) associated with lower diversification.

### Drivers of export diversification — key empirical findings
- General note: GDP per capita, GDP per capita squared, population size, latitude, and GDP per capita in 1970 are always included in models (PIP = 1 by construction).
- Resource endowment:
  - Natural resource rents are an important determinant of export diversification.
  - Higher natural resource rent associated with lower diversification at both intensive and extensive margins.
  - In BMA the variable has PIP = 1 across specifications and is highly significant in OLS.
  - Quantitative impact larger on the intensive margin.
  - Faster depletion of natural resources not robustly associated with overall diversification (opposing effects on intensive vs extensive margins cancel out).
- Human capital:
  - Primary education positively associated with overall diversification and affects both extensive and intensive margins.
  - Secondary and tertiary education not significant in BMA for the All countries and EMDE samples, though they become important in some subsamples (see EMDE-CE vs EMDE-DE).
- Cross-border flows: trade and investment
  - Trade (trade-to-GDP) strongly associated with greater diversification mainly through the extensive margin; posterior inclusion probability near 1 for extensive-margin specification.
  - Preferential trade agreements not statistically significant.
  - Higher FDI associated with greater specialization (lower diversification), mainly at the extensive margin; correlation weaker in EMDE sample (PIP < 0.5).
  - Capital account openness (Shin and Ito index) exhibits robust negative association with Theil index (i.e., less restricted capital flows associated with higher diversification) operating through the intensive margin.
- International relative prices and volatility
  - Change in terms of trade and exchange rate undervaluation not significantly associated with diversification once trade flows are accounted for.
  - Exchange rate volatility, terms-of-trade volatility, and interactions with credit-to-GDP are not statistically significant for diversification (not reported in detail).
- Quality of institutions
  - Higher quality of governance (ICRG QoG index average of Corruption, Law and Order, Quality of Bureaucracy normalized to one) associated with higher diversification largely at the extensive margin.
  - Democracy not robustly associated with diversification overall.
- Infrastructure
  - Fixed telephone subscriptions associated with the extensive margin.
  - Mobile phone subscriptions associated with the intensive margin.
  - These infrastructure associations do not hold among EMDEs; road density and ICT access do not show robust relationships.
  - Public investment (% of GDP) has a negative correlation with diversification via the extensive margin.
- Financial sector development
  - Credit-to-GDP not associated with diversification for All countries, but associated with increased diversification among EMDEs along the intensive margin.
  - More developed financial sector helps diversify economies via intensive margin.

### Heterogeneity: EMDE commodity exporters (EMDE-CE) vs EMDE diversified exporters (EMDE-DE)
- Resource endowment
  - EMDE-CE: natural resource rent highly significant and robustly associated with lower diversification at both intensive and extensive margins (PIP for intensive margin marginally < 0.5).
  - EMDE-DE: natural resource rents depress total and intensive margin diversification but help facilitate diversification at the extensive margin (extensive-margin channel has sufficiently high PIP; intensive PIP < 0.5).
  - Depletion rates not robustly associated with higher diversification for commodity exporters (OLS positive but not robust in BMA).
  - Diversified exporters experience decreases in total and extensive-margin diversification when depletion rates rise.
- Human capital
  - Primary education’s positive role in EMDEs driven mainly by diversified exporters via intensive margin.
  - Secondary enrollment aids extensive-margin diversification in EMDE-CEs.
  - Tertiary enrollment helps intensive-margin diversification among EMDE-DEs.
- Trade and investment
  - For EMDE-CE, greater trade openness increases number of products (extensive) but also increases concentration (intensive), offsetting effects on total diversification.
  - FDI not important among commodity or diversified exporters.
- International prices
  - No change in lack of association between diversification and terms of trade change or exchange rate undervaluation when splitting EMDE sample.
- Infrastructure
  - Little evidence that fixed or mobile phone subscriptions associated with more diversification among commodity exporters.
  - Higher public investment associated with lower diversification at the extensive margin for commodity exporters.
- Quality of institutions
  - Higher quality institutions tend to enhance diversification but effect statistically insignificant in BMA; OLS significance not robust.
  - Democracy associated positively with total diversification for EMDE-CEs with PIP almost 0.5, driven by intensive margin.
- Financial development
  - Credit-to-private-sector helps increase diversification in both commodity and diversified exporters, significant only for intensive margin.

### Doing Business indicators
- Augment baseline with World Bank Doing Business (DB) indicators (enterprise-survey based); data available only from 2004.
- Among DB indicators, cost of trading across borders (DTF values) is robustly correlated with diversification: lower costs of trading across borders (higher DTF) associated with greater diversification, affecting the intensive margin.
- Inclusion of DB indicators reduces statistical significance of some macro variables (e.g., enrollment rates lose significance; trade-to-GDP loses power except for extensive margin in All countries), possibly because cost of trading captures trade openness.

### Policy implications (as stated)
- Key structural predispositions toward lower export diversification: size of the economy, geography, and natural resource abundance.
- Even among commodity exporters, higher natural resource rents associated with exporting fewer products after controlling for other factors.
- Areas where policy actions are most likely to bear fruit (statistically robust associations):
  - Human capital accumulation
    - Primary education robustly associated with export diversification (dominant effect on number of products exported).
    - For commodity exporters, secondary education is more important for diversification (robustly associated with number of products).
    - Policies should focus on both quantity and quality of education.
  - Trade openness
    - Greater trade associated with greater diversification (intensive and extensive margins); lower trade costs (Doing Business) important.
    - For commodity exporters openness helps extensive-margin diversification but may increase intensive-margin specialization.
  - Quality of institutions
    - Higher quality of governance associated with less concentrated export base and higher overall diversification.
  - Infrastructure
    - Fixed telephone penetration important for extensive margin; mobile phone penetration associated with intensive margin diversification.
  - Capital account openness
    - Greater openness to capital flows helps diversification via intensive margin.
  - Financial sector development
    - More developed financial sector helps diversify EMDEs at the intensive margin, including among commodity exporters.
    - Policies to improve financial access and allocation of credit across sectors and firms would be beneficial, especially for commodity exporters facing Dutch disease effects.
- For resource-abundant economies, recommended focus: more open trade regimes, improve secondary and higher education outcomes, and develop the financial sector to increase access to and improve allocation of credit to overcome export product-space limitations from resource wealth.

*Source: Appendix 1 of the referenced IMF working paper (content as provided).*

### CONCLUSION

### CONCLUSION

### Methodology and approach
- Identifies 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.
- Methodology allows ranking 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.

### Main empirical findings
- Natural resource abundance and smaller size (in terms of population) predispose countries towards being less diversified.
- 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.

### Policy implications and priorities
- To diversify, policy makers should prioritize:
  - Human capital accumulation.
  - Reduce barriers to trade.
- Other policy areas (in order of importance):
  - Improving quality of institutions.
  - Quality of infrastructure.
  - Deepening financial markets.
- For commodity exporters, specific findings:
  - Secondary education is the most important driver of diversification.
  - Greater trade is associated with greater number of products exported.
  - Improving access as well as allocation of credit.

### Caveats on causality and identification
- One must be careful in interpreting the findings as evidence of causality.
- Although the study significantly expands the set of regressors and controls for important time-invariant features of economies (initial GDP and geography), the specification implicitly views the regressors as the causal factors, but the methodology does not address the issue of identification completely.
- 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, which is challenging given the large set of regressors considered.

### Avenues for further research
- The paper focuses on the level of diversification and not on the nature of diversification; it does not differentiate, for example, between diversification within agriculture and diversification from agriculture into manufacturing when the theil index changes by the same amount — this qualitative difference may be important from a growth standpoint and should be examined.
- The role of industrial policies in enhancing diversification is not addressed due to lack of comparable data on different policy instruments across countries and over time.
  - Recent literature (Aghion et al. (2011); Cherif et al. (2016); Cherif and Hasanov (2019)) debates the state’s role in development, but this paper does not engage that issue empirically.
- 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 an interesting area for future research; this could shed light on key constraints to desirable firm level dynamics and efficient allocation of resources (see Hsieh and Klenow (2014, 2009)).
- Labor and product market reforms were not explored because data availability is poor especially among LICs, yet these reforms could also be important for diversification.

*Source: CONCLUSION (wpiea2019105)*

### Appendix I

### Appendix I — Summary of Variables and Regression Outputs (wpiea2019105 - Appendix I)

### Variables (Definitions and Sources)
- Dependent Variables
  - Total Theil Index — Measure of total diversification; Computed based on UN Contrade data
  - Between Theil Index — Extensive margin of diversification; Computed based on UN Contrade data
  - Within Theil Index — Intensive margin of diversification; Computed based on UN Contrade data

- Independent Variables (definition — data source)
  - Log of GDP per capita — Expenditure-side real GDP chained PPPs in million 2011 US$ per capita, in natural logs; PWT 9.0
  - Log of GDP per capita square — Square of Log GDP per capita; Computed
  - Log of Population — Population, in natural logs; WEO
  - Absolute latitude — Absolute latitude; Sala-I-Martin et al. (2004)
  - Real GDP per capita (PPP) in 1970 — Real GDP per capita (PPP) in 1970; Computed using data from PWT 9.0
  - Inflation — Inflation rate (percent change in CPI); WEO
  - Natural Resource Rent — Total natural resources rents (% of GDP); WDI
  - Natural Resource Depletion (% of GNI) — Adjusted savings: natural resources depletion (% of GNI); WDI
  - Trade to GDP — Trade (% of GDP); WDI
  - Capital Account Openness — Capital account openness index; Chin & Ito database
  - Preferential Trade Agreement — Market access measure: weighted sum of preferential trade agreements (PTAs); Computed: GDP data from WDI; PTA data Jeffrey Bergstrand's EIA database
  - Foreign Direct Investment (lagged) — Lagged: Foreign direct investment, (BPM6), percent of GDP in U.S. dollars; WEO
  - Primary School Enrollment (Lagged) — Lagged: Gross enrollment ratio, primary, both sexes (%); WDI
  - Secondary School Enrollment (Lagged) — Lagged: Gross enrollment ratio, secondary, both sexes (%); WDI
  - Tertiary Education (Lagged) — Lagged: Gross enrollment ratio, tertiary, both sexes (%); WDI
  - Public Investment (lagged) — Lagged: Public Investment (% of GDP); WEO
  - Exchange Rate Undervaluation Index — Exchange rate undervaluation index (following Rodrick (2008)); Computed using data from WEO, WDI and
  - Terms of Trade — Terms of trade, total, US Dollars; WEO
  - Credit to Private Sector — Domestic credit to private sector (% of GDP); WDI
  - Fixed Phone Subscription (per 100 people) — Fixed telephone subscriptions (per 100 people); WDI
  - Mobile Phone Subscription (per 100 people) — Mobile cellular subscriptions (per 100 people); WDI
  - Quality of Institution (ICRG) — ICRG: Quality of Government; mean of ”Corruption”, ”Law and Order” and ”Bureaucracy Quality”, scaled 0-1; QoG institute
  - Democracy (Polity 2) — Combined policy score: p_democ − p_autoc, range +10 to -10; QoG institute
  - Doing Business components (DTF) — Starting a Business; Dealing with Construction Permits; Registering Property; Getting Credit; Protecting Minority Investors; Paying Taxes; Trading across Borders; Enforcing Contracts; Resolving Insolvency; World Bank - Doing Business

### Table 1: Key variable list (as provided)
- All definitions and sources are listed above; no additional numeric summary provided in the source beyond variable definitions.

### Table 2: Baseline Specification, All Countries and EMDEs (selected coefficients and statistics)
- Notes on table:
  - Columns present OLS and BMA results for different dependent variables (Total Theil Index, Between Theil Index, Within Theil Index) across samples (All Countries: Full Sample; EMDE: Full Sample).
  - For OLS regressions, standard errors are in Italic; for BMA regressions, PIPs are in italic. Statistics in bold for BMA regressions have PIPs >= 0.5.
  - Significance notation: * 0.05<p<0.1; **0.01<p<0.05; ***p<0.01

- Selected coefficients (All Countries: Full Sample)
  - Log of Real GDP per Capita
    - OLS: -0.411
    - BMA: -0.059
    - Alternate specifications: -0.293, -0.488, -0.118, -0.311, 0.487, 0.557, -0.522, -0.687, 1.009, 1.035
    - Accompanying PIP/SE entries: 0.6131.000.3651.000.6871.000.7141.000.4401.000.8071.00
  - Log of Real GDP per Capita (squared)
    - OLS: 0.036
    - BMA: 0.005
    - Alternate: 0.030, 0.042, 0.005, 0.009, -0.016, -0.030, 0.044, 0.053, -0.060, -0.071
    - PIP/SE row: 0.0381.000.0231.000.0431.000.0441.000.0271.000.0501.00
  - Log of Population
    - OLS: -0.251***
    - BMA: -0.261
    - Alternate: -0.076***, -0.078, -0.175***, -0.174, -0.238***, -0.224, -0.081***, -0.070, -0.157***, -0.185
    - PIP/SE row: 0.0301.000.0181.000.0331.000.0351.000.0211.000.0391.00
  - Absolute Latitude
    - OLS: -0.009**
    - BMA: -0.010
    - Alternate: -0.004*, -0.004, -0.005, -0.005, -0.010*, -0.011, -0.005*, -0.006, -0.005, -0.004
    - PIP/SE row: 0.0031.000.0021.000.0041.000.0041.000.0021.000.0041.00
  - Natural Resource Rent
    - OLS: 0.059***
    - BMA: 0.057
    - Alternate: 0.012**, 0.009, 0.047***, 0.042, 0.063***, 0.054, 0.009, 0.002, 0.053***, 0.055
    - PIP/SE row: 0.0081.000.0050.700.0091.000.0091.000.0050.240.0101.00
  - Natural Resource Depletion (% of GNI)
    - OLS: -0.012
    - BMA: -0.001
    - Alternate: 0.020***, 0.023, -0.032**, -0.021, -0.020, -0.003, 0.027***, 0.034, -0.048***, -0.041
    - PIP/SE row: 0.0100.080.0060.960.0110.690.0110.170.0071.000.0120.97
  - Trade to GDP
    - OLS: -0.003*
    - BMA: -0.004
    - Alternate: -0.003***, -0.003, 0.000, 0.000, -0.003, -0.001, -0.005***, -0.004, 0.002, 0.000
    - PIP/SE row: 0.0010.910.0011.000.0010.070.0020.270.0011.000.0020.08
  - Capital Account Openness
    - OLS: -0.490***
    - BMA: -0.572
    - Alternate: -0.045, -0.005, -0.445**, -0.490, -0.485***, -0.562, 0.001, -0.003, -0.487**, -0.423
    - PIP/SE row: 0.1271.000.0750.080.1420.980.1440.990.0890.070.1620.87
  - Foreign Direct Investment (% of GDP) (lagged)
    - OLS: 0.029*
    - BMA: 0.010
    - Alternate: 0.024**, 0.017, 0.005, 0.001, 0.029, 0.019, 0.030*, 0.011, 0.000, 0.000
    - PIP/SE row: 0.0140.330.0090.720.0160.060.0200.440.0120.420.0220.06
  - Quality of Institution (ICRG)
    - OLS: -0.845**
    - BMA: -0.697
    - Alternate: -0.547**, -0.572, -0.299, -0.030, -0.880*,-0.300, -0.535*,-0.299, -0.346, -0.033
    - PIP/SE row: 0.2930.810.1750.960.3290.090.3550.390.2190.570.4010.08
  - Credit to Private Sector
    - OLS: -0.003
    - BMA: -0.001
    - Alternate: 0.000, 0.000, -0.002, 0.000, -0.007**, -0.008, 0.001, 0.000, -0.008**, -0.007
    - PIP/SE row: 0.0010.230.0010.050.0020.120.0020.960.0010.080.0020.97
  - R-squared values (reported)
    - 0.77, 0.62, 0.57, 0.71, 0.61, 0.48 (across columns/specifications)
  - Observations
    - 384 384 384 384 384 384 313 313 313 313 313 313
  - Time Dummies — Yes (in all specifications)

### Table 4: Baseline Specification, EMDEs — Commodity and Diversified Exporters (selected coefficients)
- Notes:
  - Columns show OLS and BMA for EMDE subsamples: Commodity Exporters and Diversified Exporters, across Total, Between, Within Theil Indices.
  - Same notation and significance markers as Table 2.

- Selected coefficients (EMDE subsamples)
  - Log of Real GDP per Capita
    - Examples: 1.226, 1.065, -1.071, -0.994, 2.297, 1.730, -1.946, -1.374, -0.276, -0.684, -1.670, -1.451
    - Accompanying PIP/SE entries: 1.0721.000.7641.001.3921.001.0941.000.6101.001.1821.00
  - Natural Resource Rent
    - Examples: 0.078***, 0.053, 0.033***, 0.029, 0.045**, 0.011, -0.010, 0.000, -0.017*,-0.021, 0.007, 0.004
    - PIP/SE row: 0.0131.000.0090.990.0170.470.0150.070.0080.850.0160.22
  - Natural Resource Depletion (% of GNI)
    - Examples: -0.052**, -0.017, -0.004, 0.000, -0.048*, -0.005, 0.054**, 0.050, 0.053***, 0.055, 0.001, 0.002
    - PIP/SE row: 0.0180.460.0130.070.0230.190.0180.990.0101.000.0190.12
  - Trade to GDP
    - Examples: 0.003, 0.000, -0.011***, -0.009, 0.014**, 0.008, -0.003, -0.001, 0.000, 0.000, -0.003, -0.001
    - PIP/SE row: 0.0040.070.0031.000.0050.710.0020.270.0010.070.0020.23
  - Public Investment (% of GDP) (lagged)
    - Examples: 0.025, 0.009, 0.025, 0.017, 0.000, 0.000, 0.017, 0.005, 0.001, 0.000, 0.017, 0.004
    - PIP/SE row: 0.0180.270.0130.530.0240.060.0100.240.0060.070.0110.22
  - R-squared values (reported)
    - 0.77, 0.78, 0.55, 0.66, 0.44, 0.54
  - Observations
    - 108 108 108 108 108 108 205 205 205 205 205 205
  - Time Dummies — Yes (in all specifications)

### Table 5: Baseline Specification with Doing Business Indicators (selected coefficients)
- Notes:
  - OLS and BMA regressions including Doing Business (DTF) component indicators; columns cover All Countries: Full Sample and EMDE: Full Sample across Total, Between, Within Theil Indices.
  - Significance notation: * 0.05<p<0.1; **0.01<p<0.05; ***p<0.01

- Selected coefficients (All Countries and EMDEs with Doing Business indicators)
  - Log of Real GDP per Capita
    - Examples: 1.708, -0.428, -1.405*, -0.936, 3.113*, 0.594, 1.826, 0.922, -1.735*,-1.629, 3.560*, 2.441
    - PIP/SE row: 1.2031.000.5921.001.3471.001.3501.000.7101.001.5451.00
  - Log of Real GDP per Capita (squared)
    - Examples: -0.081, 0.038, 0.096**, 0.065, -0.178*, -0.027, -0.089, -0.036, 0.116**, 0.108, -0.205*, -0.134
    - PIP/SE row: 0.0731.000.0361.000.0811.000.0821.000.0431.000.0941.00
  - Natural Resource Rent
    - Examples: 0.087***, 0.083, 0.031***, 0.027, 0.057**, 0.008, 0.080***, 0.084, 0.028**, 0.024, 0.052*, 0.014
    - PIP/SE row: 0.0161.000.0081.000.0180.240.0171.000.0091.000.0200.30
  - Natural Resource Depletion (% of GNI)
    - Examples: -0.072***, -0.066, -0.008, -0.001, -0.064**, -0.008, -0.070**, -0.073, -0.001, 0.000, -0.069*, -0.015
    - PIP/SE row: 0.0200.970.0100.100.0230.170.0230.980.0120.050.0260.26
  - Trade to GDP
    - Examples: 0.001, 0.000, -0.003*,-0.002, 0.004, 0.000, 0.000, 0.000, -0.005**,-0.001, 0.005, 0.000
    - PIP/SE row: 0.0020.040.0010.580.0030.050.0040.060.0020.260.0040.04
  - Trading across Borders - DTF
    - OLS: -0.018**
    - BMA: -0.021
    - Alternate columns: 0.005, 0.002, -0.023***, -0.028, -0.020***, -0.023, 0.005, 0.001, -0.025***, -0.031
    - PIP/SE row: 0.0061.000.0030.360.0061.000.0061.000.0030.230.0071.00
  - Enforcing Contracts - DTF
    - OLS: -0.013*
    - BMA: -0.004
    - Alternate: 0.005, 0.000, -0.018*, -0.002, -0.016, -0.006, 0.006, 0.000, -0.021*, -0.011
    - PIP/SE row: 0.0070.300.0030.050.0070.190.0080.360.0040.050.0090.49
  - R-squared values (reported)
    - 0.83, 0.72, 0.73, 0.83, 0.73, 0.72
  - Observations
    - 130 130 130 130 130 130 103 103 103 103 103 103
  - Time Dummies — Yes (in all specifications)

*Note: For OLS regressions, standard errors are in Italic; for BMA regressions, PIPs are in italic. Statistics in bold for BMA regressions have PIPs >= 0.5.*

*Appendix I, wpiea2019105 - Appendix I (source PDF: wpiea2019105 - Appendix I).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2019/wpiea2019105.pdf_