## _wp1371

## Source details

**Canonical URL:** [_wp1371](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1371.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1371.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1371.pdf.json)

---

### I. Setting the stage
- Literature context:
  - Traditional approaches estimate average GDP per capita growth over long periods or use dynamic panels (Barro and Sala-i-Martin (1991); Mankiw, Romer and Weil (1992); Islam (1995); Caselli, Esquivel and Lefort (1996)).
  - Real-world growth shows “hills, plateaus, mountains and plains” (Pritchett, 1998); increasing focus on growth slowdowns and the “middle-income trap.”
- Paper contributions:
  - Proposes a novel identification procedure for growth slowdowns grounded in growth theory.
  - Finds slowdowns are disproportionately likely in middle-income countries.
  - Systematically identifies determinants using 42 explanatory variables grouped into seven categories and validates probit results with Bayesian model selection (WALS and BMA).

### II. Stylized facts and decompositions
- Cross-country trajectories and selected thresholds:
  - Charts compare GDP per capita relative to U.S. levels from the year countries reached US$ 3000 (2005 PPP).
  - East Asian examples: Korea and Taiwan Province of China rose from 10‒20 percent of U.S. to 60–70 percent.
  - China: less than a decade above the threshold in sample; Malaysia outperformed Latin American comparators; Thailand and Indonesia mixed/poor performance.
- Growth decomposition (methods preserved):
  - Physical capital: perpetual inventory method from Penn World Tables.
  - Human capital: log h = .134 * pyr + .101 * syr + .068 * hyr (Psacharopoulos (1994)).
  - Capital share assumed to be one-third.
  - TFP (residual) often critical:
    - Steep falls in TFP growth important in past slowdowns (notably Latin America in the 1980s).
    - East Asian successes underpinned by robust TFP growth, often accounting for more than half of GDP per capita growth.

### III. Identification of growth slowdowns
- Data and sample:
  - Annual per capita income in constant 2005 international dollars.
  - Five-year rolling geometric averages; five-year panel of GDP per capita growth rates.
  - Sample: 138 countries; 11 periods (1955–2009); 1125 observations.
- Slowdown definition (residual-based, relative to predicted growth):
  - Residuals = actual growth rate − estimated (predicted) rate (parsimonious conditional convergence framework).
  - Country i experiences a slowdown in period t if both:
    - residual_{i,t} − residual_{i,t−1} ≤ p(0.20)
    - residual_{i,t+1} − residual_{i,t−1} ≤ p(0.20)
  - p(0.20) denotes the 20th percentile of empirical distribution of differences in residuals between successive periods.
  - Empirical count: 123 slowdowns identified (around 11 percent of the sample).

### IV. Middle-income trap operationalization
- Threshold experimentation:
  - Lower threshold a_L ∈ {1000, 2000, 3000} (2005 PPP $).
  - Upper threshold a_H ∈ {12,000, 13,000, 14,000, 15,000, 16,000} (2005 PPP $).
  - Generates 15 classifications (3 × 5).
- Main classification adopted:
  - 2/15 rule: low-income threshold = 2000 (2005 PPP $); high-income threshold = 15,000 (2005 PPP $).
  - This classification closely matches World Bank GNI per capita classification (97 percent overlap noted).
- Main result:
  - Middle-income countries are disproportionately likely to experience growth slowdowns across tested thresholds.

### V. Determinants of growth slowdowns: methodology
- Empirical approach:
  - Probit specifications estimate probability of a slowdown in a given period.
  - 42 candidate explanatory variables grouped into seven categories:
    - (i) Institutions; (ii) Demography; (iii) Infrastructure; (iv) Macroeconomic Environment and Policies; (v) Economic Structure; (vi) Trade structure; (vii) Other.
  - Allow levels (lagged beginning-of-period) and lagged differences to capture threshold and dynamic effects.
- Robustness to model uncertainty and data gaps:
  - Bayesian model averaging techniques used for robustness:
    - Weighted Average Least Squares (WALS).
    - Bayesian Model Averaging (BMA).
- Estimation steps:
  - Step 1: For each category, run probit specifications with forward/backward selection (10 percent inclusion, 10 percent exclusion) to identify robust regressors.
  - Step 2: Apply BMA and WALS to corresponding linear probability models; report PIPs (BMA) and t-ratios (WALS).

### VI. Determinants of growth slowdowns: empirical results (key findings and magnitudes)
- Institutions:
  - Variables: Economic Freedom of the World indices (Size of Government, Rule of Law, Freedom to Trade Internationally, Regulation); Chinn-Ito financial openness.
  - Key coefficients (preserved):
    - L.Strong rule of law: −0.089, P>z = 0.005.
    - D.Small government: −0.173, P>z = 0.003.
    - D.Light regulation: −0.210*** (significance levels as in source).
  - Bayesian checks (WALS/BMA) confirm level of Legal Structure and lagged changes in Size of Government and Regulation as robust.
- Demography:
  - Variables: Dependency Ratio, Sex Ratio (men/women); Fertility omitted in levels due to correlation.
  - Key coefficients (preserved):
    - L.Dependency ratio: 0.008, P>z = 0.003.
    - D.Sex ratio: 0.075, P>z = 0.001.
  - Interpretation: higher dependency ratio and increases in male-to-female sex ratio raise slowdown probability.
- Infrastructure:
  - Variables: Telephone Lines (log per 1000), Power (log gigawatts per 1000), Roads (log road length per sq km).
  - Full-sample result: “No infrastructure variable is significant.”
  - Caveat: infrastructure effects differ and become significant when sample restricted to middle-income countries.
- Macroeconomic environment and policies:
  - Variables include: Gross Capital Inflows/GDP, Investment share, Trade Openness, Public Debt/GDP, Banking Crisis dummy, Terms of Trade shocks.
  - Key coefficients (preserved):
    - L.Gross capital inflows/GDP: 0.028, P>z = 0.001.
    - D.Investment share: 0.059, P>z = 0.000.
    - D.Trade openness: −0.013, P>z = 0.008.
    - D.Public debt: −0.005, P>z = 0.040 (driven by HIPC debt relief episodes; loses significance when HIPC recipients excluded).
    - D.Gross capital inflows: −0.016, P>z = 0.051.
  - Interpretation: high initial capital inflows and rapid increases in investment share are associated with higher subsequent slowdown probability; increases in trade openness reduce slowdown probability.
- Economic structure (output composition):
  - Variables: Agriculture share, Services share (Manufacturing omitted).
  - Key coefficients (preserved):
    - L.Agriculture share: −0.012, P>z = 0.045.
    - L.Services share: −0.015, P>z = 0.035.
    - D.Agriculture share: −0.039, P>z = 0.015.
    - D.Services share: −0.035, P>z = 0.011.
  - Interpretation: economies undergoing structural change (declining agriculture/services shares as industry expands) face higher slowdown risk; sectoral diversification appears protective when examined separately.
- Trade structure:
  - Variables: GDP-weighted Distance, Regional Integration (intra-regional trade share), Export Diversification (Theil index).
  - Key coefficients (preserved):
    - L.Distance: 0.116, P>z = 0.007.
    - L.Regional integration: −0.008, P>z = 0.011.
  - Interpretation: greater geographic distance increases slowdown probability; greater regional integration reduces it. Export diversification shows protective effects in larger samples when estimated separately.
- Other (geography, conflict, etc.):
  - Variables: Tropics (fraction of land area in tropical zone), Wars and civil conflicts, ELF, Spanish colony, Buddhist share, Natural Disasters.
  - Key coefficients (preserved):
    - Tropics: 0.264, P>z = 0.026.
    - War and civil conflicts: 0.476, P>z = 0.003.
  - Interpretation: tropical location and presence of wars/civil conflicts strongly increase slowdown probability; wars and civil conflicts show strong Bayesian signals (WALS t = 2.08, BMA PIP = 0.59).
- Summary magnitudes and policy-relevant impacts (selected entries from Table 10 preserved exactly):
  - L.Strong rule of law: Probit Coeff. −0.089***; Average Marginal Effects −1.7; p(50)-p(25) −3.1; p(75)-p(50) −2.6.
  - D.Small government: Probit Coeff. −0.173***; Average Marginal Effects −3.2; p(50)-p(25) −1.8; p(75)-p(50) −1.9.
  - D.Light regulation: Probit Coeff. −0.210***; Average Marginal Effects −3.9; p(50)-p(25) −2.3; p(75)-p(50) −2.2.
  - L.Gross capital inflows: Probit Coeff. 0.028***; Average Marginal Effects 0.5; p(50)-p(25) 1.4; p(75)-p(50) 2.1.
  - D.Investment share: Probit Coeff. 0.059***; Average Marginal Effects 1.1; p(50)-p(25) 3.4; p(75)-p(50) 4.2.
  - D.Trade openness: Probit Coeff. −0.013***; Average Marginal Effects −0.2; p(50)-p(25) −1.3; p(75)-p(50) −1.5.
  - L.Distance: Probit Coeff. 0.116***; Average Marginal Effects 2.4; p(50)-p(25) 2.9; p(75)-p(50) 1.9.
  - L.Regional integration: Probit Coeff. −0.008***; Average Marginal Effects −0.2; p(50)-p(25) −2.5; p(75)-p(50) −3.4.
  - Tropics: Probit Coeff. 0.264**; Average Marginal Effects 5.0; p(50)-p(25) 3.0; p(75)-p(50) 1.9.
  - War and civil conflicts: Probit Coeff. 0.476***; Average Marginal Effects 9.0.
- Significance notation:
  - Asterisks indicate significance at the 10 percent, 5 percent and 1 percent level as in the source table. Prefix L. = levels; D. = differences.

### VII. Are middle-income countries different? (MIC subsample results)
- Strategy: restrict sample to MICs per 2/15 rule and repeat regressions.
- Key differences for MICs (highlights from Table 11):
  - Institutions:
    - Government Size replaces Rule of Law as most significant level variable for MICs.
    - D.Light regulation coefficient about twice as large for MICs than full sample → deregulation particularly important for MICs.
  - Infrastructure:
    - For MICs, insufficient Road Networks and Telephone Lines emerge as significant risk factors (not significant in full sample).
  - Trade:
    - Regional integration reduces slowdown probability for MICs but significance sensitive to outliers (bottom and top deciles excluded).
  - Diversification:
    - Output and trade diversification protective effects in full sample disappear when restricted to MICs.

### VIII. Policy implications and illustrative risk mapping
- Short-term/implementable policy levers identified:
  - Prudential regulation to limit excessive capital inflows and cushion sudden stops.
  - Measures to enhance regional trade integration.
  - Public investment in infrastructure (communications, roads, power).
  - Deregulation to reduce red tape and foster private-sector dynamism.
- Medium- to long-term targets:
  - Strengthening rule of law and institutions.
  - Demographic policies (fertility, gender equality).
- Variables less amenable to short-term policy:
  - Geographic distance and climatic conditions (Tropics).
- Illustrative “growth slowdown risk” mapping:
  - Applied MIC coefficients from Table 11 to latest data for seven Asian MICs and MIC comparators in Latin America and MENA across seven categories.
  - Findings from maps and spider-web charts:
    - Malaysia, the Philippines and China face larger institution-related slowdown risks in this exercise; Vietnam, India, and Indonesia most at risk from infrastructure deficits.
    - Asia stands at higher risk than other regions from infrastructure-related slowdowns (communications in particular).
    - Asia compares favorably on trade integration relative to Latin America and MENA.
  - Figures use latest available observations and projected 2020 dependency ratios per UN baseline for ranking.

### IX. Key quantitative facts (preserved exactly)
- Sample and identification:
  - Sample: 138 countries; 11 periods (1955–2009).
  - Observations: 1125; slowdowns identified: 123 (around 11 percent).
  - Five-year rolling geometric averages used.
- Middle-income threshold choice:
  - Adopt 2/15 definition: low-income threshold = 2000 (2005 PPP $); high-income threshold = 15,000 (2005 PPP $).
- Representative coefficients and statistics preserved exactly:
  - L.Strong rule of law: −0.089, P>z = 0.005.
  - D.Small government: −0.173, P>z = 0.003.
  - L.Gross capital inflows/GDP: 0.028, P>z = 0.001.
  - D.Investment share: 0.059, P>z = 0.000.
  - L.Distance: 0.116, P>z = 0.007.
  - Tropics: 0.264, P>z = 0.026.
  - War and civil conflicts: 0.476, P>z = 0.003.
- Human capital and capital share (preserved):
  - log h = .134 * pyr + .101 * syr + .068 * hyr (Psacharopoulos (1994)).
  - Capital share assumed to be one-third.

### Appendices: model averaging techniques and data inventories
- Appendix I — Bayesian Model Averaging Technique (summary of methods preserved):
  - Two sources of model uncertainty: number of RHS variables and their specification.
  - BMA formal unconditional estimator: β̂_j,BMA = Σ_{i=1}^m ω_i β̂_{j,i}.
  - Partition regressors into Focus and Auxiliary groups; in application Focus = constant term; auxiliary = all other variables.
  - Priors: noninformative for β_F and σ^2; Zellner (1986) form for auxiliary parameters; posterior model weights π_i = P(M_i | y) computed with uniform prior over model space (P(M_i) = 2^{-k_A}).
  - Unconditional BMA estimates: β̂_F = Σ_{i=1}^m π_i β̂_{F,i}; β̂_A = Σ_{i=1}^m π_i Ψ_i β̂_{A,i}.
  - Significance criterion: Posterior Inclusion Probability (PIP); threshold PIP > 0.5 adopted as indicating a robust regressor.
- Appendix I — WALS:
  - WALS uses Laplace prior and an orthogonal transformation of auxiliary regressors to reduce computation.
  - WALS reduces computation from 2^{k_A} estimations to k linear combinations; criterion |t| > 1 for robustness suggested.
- Appendix II — Tables and Charts:
  - Table A.2.1 lists identified Growth Slowdown Episodes by income group (Middle Income, High Income / Low Income) exactly as reported.
  - Table A.2.2 maps slowdown episodes to three classification columns (Conditional Convergence, Absolute Convergence, Eichengreen and others (2011)) and reports aggregate total: Total12312084 (as printed).
  - Table A.2.3 lists independent variables, units, sources and sample coverage (years and frequency) exactly as reported.
  - Table A.2.4 reports sample statistics by category and region (entries preserved as printed).
  - Appendix A.2.5 lists composition of regions exactly as presented.

*Source: _wp1371 - IMF staff estimates and analysis as presented in the provided content.*

### References .............................................................................................................

### _wp1371 - References

### Figures
- 1. Cross-Country Comparison ......................................................................................................... 5
- 2. Growth Trajectories ..................................................................................................................... 6
- 3. Low-Income Countries ................................................................................................................ 6
- 4. Slowdown in Latin America: 1970s vs. 1980 .............................................................................. 7
- 5. Growth Success in Asia ............................................................................................................... 8
- 6. Is There a Middle Income Trap? ............................................................................................... 12
- 7. Asian MIC’s Current Strengths and Weaknesses ...................................................................... 36
- 8. Asian MIC’s Current Strengths and Weaknesses Relative to Other Emerging Regions ........... 37

### Tables
- 1. Distribution of Slowdown Episodes by Region ......................................................................... 11
- 2. Distribution of Slowdown Episodes by Time Period ................................................................ 11
- 3. Institutions ................................................................................................................................. 17
- 4. Demography .............................................................................................................................. 19
- 5. Infrastructure .............................................................................................................................. 20
- 6. Macroeconomic Environment and Policies ............................................................................... 23
- 7. Output Composition ................................................................................................................... 24
- 8. Trade .......................................................................................................................................... 27
- 9. Other .......................................................................................................................................... 28
- 10. Summary Table ........................................................................................................................ 29
- 11. Middle-Income Countries vs. Full Sample .............................................................................. 31
- 12. A “Growth Slowdown Risk” Map for Asian Middle-Income Countries ................................. 33
- 13. A “Trap Map” for Middle-Income Countries .......................................................................... 34

### Appendices
- 1. Bayesian Model Averaging Technique ..................................................................................... 38

*Source: _wp1371 - References.*

### 2. Tables and Charts ...................................................................................................

### _wp1371 - 2. Tables and Charts

### I. Setting the stage
- Literature context:
  - Traditional approaches estimate average GDP per capita growth over long periods or use dynamic panels, implying smooth convergence paths (Barro and Sala-i-Martin (1991); Mankiw, Romer and Weil (1992); Islam (1995); Caselli, Esquivel and Lefort (1996)).
  - Real-world growth shows “hills, plateaus, mountains and plains” (Pritchett, 1998); increasing focus on growth slowdowns and the “middle-income trap.”
- Paper contributions:
  - Proposes a novel identification procedure for growth slowdowns grounded in growth theory.
  - Shows slowdowns are disproportionately likely in middle-income countries.
  - Systematically identifies determinants using a comprehensive set of explanatory variables and validates probit results with Bayesian model selection (WALS and BMA).

### II. Stylized facts
- Cross-country trajectories (selected observations and thresholds):
  - Charts compare GDP per capita relative to U.S. levels for countries from the year they reached US$ 3000 (2005 PPP).
  - East Asian successes: Korea and Taiwan Province of China increased per capita income from 10‒20 percent of U.S. to 60–70 percent.
  - Latin American outcomes: Brazil and Mexico stagnated in relative terms; Peru fell behind.
  - Recent middle-income Asia: China outstripped earlier East Asian stories (less than a decade above threshold in sample); Malaysia performed better than Latin American comparators; Thailand and Indonesia mixed/poor performance.
- Decompositions:
  - GDP per capita growth decomposed into contributions from physical capital, human capital, working-age population expansion, and TFP (residual), using:
    - Physical capital via perpetual inventory method from Penn World Tables.
    - Human capital: weighted average of years of primary, secondary and higher schooling from Barro-Lee dataset; log h = .134 * pyr + .101 * syr + .068 * hyr (Psacharopoulos (1994)).
    - Capital share assumed to be one-third (Gollin, 2002; Aiyar and Dalgaard, 2009).
  - Findings:
    - Steep falls in TFP growth played an important role in past slowdowns (notably Latin America in the 1980s).
    - East Asian successes (China, Taiwan Province of China) underpinned by robust TFP growth, often accounting for more than half of GDP per capita growth.

### III. Identifying growth slowdowns
- Data and baseline specification:
  - Annual data on per capita income in constant 2005 international dollars.
  - Five-year rolling geometric averages used to compute a five-year panel of GDP per capita growth rates.
  - Sample covers 138 countries over 11 periods (1955–2009).
  - Specification: per capita GDP growth regressed on lagged income level and standard measures of physical and human capital (parsimonious conditional convergence framework).
- Slowdown definition (residual-based, relative to predicted growth path):
  - Residuals = actual growth rate − estimated (predicted) rate.
  - Country i identified as experiencing a growth slowdown in period t if both hold:
    - (1) residual_{i,t} − residual_{i,t−1} ≤ p(0.20)
    - (2) residual_{i,t+1} − residual_{i,t−1} ≤ p(0.20)
  - p(0.20) denotes the 20th percentile of empirical distribution of differences in residuals between successive periods.
  - Intuition: large sudden and sustained negative deviations from predicted growth path; rules out temporary one-period dips.
- Empirical counts and patterns:
  - Out of 1125 observations, the algorithm selects 123 slowdowns, i.e., around 11 percent of the sample.
  - Regional frequency of slowdowns higher in developing regions, notably Latin America, Middle East and North Africa, sub-Saharan Africa, and East Asia.
  - Frequency varied by time period: higher than average over 1975–85; low during 1960–65.
- Middle-income trap operationalization:
  - Test robustness across a range of lower and upper thresholds for middle income:
    - Lower threshold a_L ∈ {1000, 2000, 3000} (2005 PPP $).
    - Upper threshold a_H ∈ {12,000, 13,000, 14,000, 15,000, 16,000} (2005 PPP $).
    - Generates 15 classifications (3 × 5).
  - Main result:
    - Middle-income countries are disproportionately likely to experience growth slowdowns across the range of thresholds tested.
    - Adopted working definition for the paper: the 2/15 rule — low-income threshold 2000 (2005 PPP $) and high-income threshold 15,000 (2005 PPP $). This classification closely matches World Bank GNI per capita classification (97 percent overlap noted).

### IV. Determinants of growth slowdowns: methodology
- Empirical strategy:
  - Estimate probit specifications for the probability of a slowdown in a given period.
  - Consider broad set of candidate determinants: 42 explanatory variables grouped into seven categories:
    - (i) Institutions; (ii) Demography; (iii) Infrastructure; (iv) Macroeconomic Environment and Policies; (v) Economic Structure; (vi) Trade structure; (vii) Other.
  - Allow both levels (lagged beginning-of-period) and lagged differences for many variables to capture threshold and dynamic effects.
- Addressing model uncertainty and data gaps:
  - Use Bayesian model averaging techniques for robustness:
    - Weighted Average Least Squares (WALS).
    - Bayesian Model Averaging (BMA).
  - Variables grouped into seven categories and estimated separately to preserve sample size and coverage.
- Estimation steps:
  - Step 1: For each category, run probit specifications with lagged levels and differences of candidate variables; use forward and backward selection (10 percent inclusion, 10 percent exclusion thresholds) to identify a restricted set of robust regressors.
  - Step 2: Apply Bayesian averaging (BMA and WALS) on corresponding linear probability models for robustness checks; report PIPs (BMA) and t-ratios (WALS).

### V. Determinants of growth slowdowns: empirical results
- Overview:
  - Results presented by module; each module reports final probit specification and Bayesian robustness (WALS and BMA).
  - Key reported sample sizes, pseudo R-squared, and significance levels retained as in the source.

- A. Institutions
  - Variables used:
    - Four indices from Economic Freedom of the World (EFW): Size of Government, Rule of Law, Freedom to Trade Internationally, Regulation (higher = more growth-friendly).
    - Chinn-Ito index of financial openness.
  - Probit and Bayesian findings (Table 3):
    - Level of Rule of Law significant at the 1 percent level (L.Strong rule of law coefficient −0.089, P>z = 0.005).
    - Differences: D.Small government and D.Light regulation significant (D.Small government coefficient −0.173, P>z = 0.003).
    - WALS and BMA confirm: level of Legal Structure and lagged change in Size of Government and Regulation robust (WALS t and BMA PIP reported; e.g., WALS t for Small government = 0.67, BMA PIP = 0.06; WALS t for Light regulation = −0.78, BMA PIP = 0.08; Gross pattern: Bayesian methods support key variables identified).

- B. Demography
  - Variables used: Fertility Rate (omitted in levels due to high correlation with dependency ratio), Dependency Ratio, Sex Ratio (men/women).
  - Findings (Table 4):
    - L.Dependency ratio coefficient 0.008, P>z = 0.003 (significant).
    - D.Sex ratio coefficient 0.075, P>z = 0.001 (significant).
    - Interpretation: higher dependency ratio and increases in male-to-female sex ratio increase slowdown probability.
    - WALS and BMA support identification (e.g., WALS t for Dependency ratio = 2.74, BMA PIP = 0.7; WALS t for Sex ratio = 0.45, BMA PIP = 0.032).

- C. Infrastructure
  - Variables used (Calderon and Serven, 2004): Telephone Lines (log per 1000 people), Power (log gigawatts generating capacity per 1000 people), Roads (log road length per square km).
  - Findings (Table 5):
    - No significant results in the full-sample probit module ("No infrastructure variable is significant").
    - Caveat: effects differ when restricting to middle-income countries (see Section VII).

- D. Macroeconomic Environment and Policies
  - Variables include: Gross Capital Inflows / GDP, Investment share, Trade Openness, Public Debt / GDP, Banking Crisis dummy (Laeven and Valencia, 2012), Terms of Trade shocks and commodity price shock interactions.
  - Findings (Table 6):
    - L.Gross capital inflows / GDP coefficient 0.028, P>z = 0.001 (higher initial inflows associated with higher slowdown probability).
    - D.Investment share coefficient 0.059, P>z = 0.000 (rapid increase in investment share associated with higher subsequent slowdown probability).
    - D.Trade openness coefficient −0.013, P>z = 0.008 (increases in trade openness reduce slowdown probability).
    - D.Public debt coefficient −0.005, P>z = 0.040 (increase in public debt associated with smaller slowdown probability), but this result is driven by HIPC debt relief episodes and loses significance when HIPC recipients excluded.
    - D.Gross capital inflows coefficient −0.016, P>z = 0.051 (reduction in inflows associated with higher subsequent slowdown probability).
    - Bayesian results: Gross capital inflows show strong WALS/BMA signals (e.g., WALS t = 1.44, BMA PIP = 0.62 for levels).

- E. Economic Structure (Output Composition)
  - Variables: Agriculture share, Services share (Manufacturing share omitted as residual).
  - Findings (Table 7):
    - L.Agriculture share coefficient −0.012, P>z = 0.045.
    - L.Services share coefficient −0.015, P>z = 0.035.
    - D.Agriculture share coefficient −0.039, P>z = 0.015.
    - D.Services share coefficient −0.035, P>z = 0.011.
    - Interpretation: economies undergoing structural change (declining agriculture/services shares as industry expands) face higher slowdown risk; sectoral diversification appears to lower slowdown probability when examined separately (Papageorgiou and Spatafora index).

- F. Trade structure
  - Variables: GDP-weighted Distance, Regional Integration (intra-regional trade share), Export Diversification (Theil index).
  - Findings (Table 8):
    - L.Distance coefficient 0.116, P>z = 0.007 (greater distance increases slowdown probability).
    - L.Regional integration coefficient −0.008, P>z = 0.011 (greater regional integration reduces slowdown probability).
    - Export diversification not selected jointly due to coverage constraints but shows protective effect in larger samples when estimated separately.

- G. Other (geography, conflict, etc.)
  - Variables: ELF (ethno-linguistic fractionalization), Tropics (fraction of land area in tropical zone), Spanish colony, Buddhist population share, Wars and Civil Conflicts, Natural Disasters.
  - Findings (Table 9):
    - Tropics coefficient 0.264, P>z = 0.026 (greater tropical area increases slowdown probability).
    - Wars and civil conflicts coefficient 0.476, P>z = 0.003 (presence of wars/civil conflicts strongly associated with increased slowdown probability).
    - Bayesian results: Wars and civil conflicts strong signal (WALS t = 2.08, BMA PIP = 0.59).

- H. Summary of magnitudes and policy-relevant impacts (Table 10)
  - Selected average marginal effects and percentile impacts (p(50)-p(25), p(75)-p(50)) reported exactly as in source for key regressors. Examples:
    - L.Strong rule of law: Probit Coeff. −0.089***; Average Marginal Effects −1.7; p(50)-p(25) −3.1; p(75)-p(50) −2.6.
    - D.Small government: Probit Coeff. −0.173***; Average Marginal Effects −3.2; p(50)-p(25) −1.8; p(75)-p(50) −1.9.
    - D.Light regulation: Probit Coeff. −0.210***; Average Marginal Effects −3.9; p(50)-p(25) −2.3; p(75)-p(50) −2.2.
    - L.Gross capital inflows: Probit Coeff. 0.028***; Average Marginal Effects 0.5; p(50)-p(25) 1.4; p(75)-p(50) 2.1.
    - D.Investment share: Probit Coeff. 0.059***; Average Marginal Effects 1.1; p(50)-p(25) 3.4; p(75)-p(50) 4.2.
    - D.Trade openness: Probit Coeff. −0.013***; Average Marginal Effects −0.2; p(50)-p(25) −1.3; p(75)-p(50) −1.5.
    - L.Distance: Probit Coeff. 0.116***; Average Marginal Effects 2.4; p(50)-p(25) 2.9; p(75)-p(50) 1.9.
    - L.Regional integration: Probit Coeff. −0.008***; Average Marginal Effects −0.2; p(50)-p(25) −2.5; p(75)-p(50) −3.4.
    - Tropics: Probit Coeff. 0.264**; Average Marginal Effects 5.0; p(50)-p(25) 3.0; p(75)-p(50) 1.9.
    - War and civil conflicts: Probit Coeff. 0.476***; Average Marginal Effects 9.0.
  - Note on significance: asterisks indicate significance at the 10 percent, 5 percent and 1 percent level as in the source table. Prefix L. = levels; D. = differences. Bracketed variables significant only when regressed alone.

### VI. Are middle-income countries different?
- Strategy: Restrict sample to middle-income countries (MICs, per 2/15 rule) and repeat regressions.
- Key differences (Table 11 summary highlights):
  - Institutions:
    - In MIC subsample, Government Size replaces Rule of Law as most significant level variable.
    - D.Light regulation coefficient about twice as large for MICs than full sample → deregulation particularly important for MICs.
  - Infrastructure:
    - For MICs, insufficient Road Networks and Telephone Lines emerge as significant risk factors (these were not significant in full sample).
  - Trade:
    - Regional Integration reduces slowdown probability for MICs but significance sensitive to outliers (bottom and top deciles excluded).
  - Diversification:
    - Output and trade diversification protective effects in full sample disappear when restricted to MICs, consistent with literature suggesting diversification matters more at low-income stages.

### VII. Policy implications and illustrative risk mapping
- Policy levers identified with relatively short-term implementability:
  - Prudential regulation to limit excessive capital inflows and cushion sudden stops.
  - Measures to enhance regional trade integration.
  - Public investment in infrastructure (communications, roads, power).
  - Deregulation to reduce red tape and foster private sector dynamism.
- Variables less amenable to short-term policy:
  - Geographic distance and climatic conditions (Tropics).
- Medium- to long-term policy targets:
  - Demographic policies (fertility, gender equality).
  - Strengthening rule of law and institutions.
- Illustrative “growth slowdown risk” maps (Tables 12 and 13; Figures 7 and 8):
  - Applied MIC coefficients from Table 11 to latest data for seven Asian MICs and broader MIC comparators in Latin America and MENA to rank relative risk across seven categories.
  - Findings from illustrative maps and spider-web charts:
    - Compared with other Asian economies, Malaysia, the Philippines and China face larger institutional-related slowdown risks in this exercise; Vietnam, India, and Indonesia most at risk from infrastructure deficits.
    - Asia stands at higher risk than other regions from infrastructure-related slowdowns (communications in particular).
    - Asia compares favorably on trade integration relative to Latin America and MENA.
  - Figures 7 and 8 use latest available observations (and projected 2020 dependency ratios per UN baseline) to rank countries on Institutions, Dependency ratio, Infrastructure, Macroeconomic factors (including 2008–2012 changes for capital inflows, trade openness, and investment-to-GDP), and Trade structure.

### VIII. Key quantitative facts (preserved exactly)
- Sample and identification:
  - Sample: 138 countries; 11 periods (1955–2009).
  - Observations: 1125; slowdowns identified: 123 (around 11 percent).
  - Five-year rolling geometric averages used.
- Middle-income threshold choice:
  - Adopt 2/15 definition: low-income threshold = 2000 (2005 PPP $); high-income threshold = 15,000 (2005 PPP $).
- Notable coefficients and significance reported (examples preserved exactly from tables):
  - L.Strong rule of law: −0.089, P>z = 0.005.
  - D.Small government: −0.173, P>z = 0.003.
  - L.Gross capital inflows/GDP: 0.028, P>z = 0.001.
  - D.Investment share: 0.059, P>z = 0.000.
  - L.Distance: 0.116, P>z = 0.007.
  - Tropics: 0.264, P>z = 0.026.
  - War and civil conflicts: 0.476, P>z = 0.003.
- Human capital specification (preserved):
  - log h = .134 * pyr + .101 * syr + .068 * hyr (Psacharopoulos (1994)).
- Capital share assumption:
  - A capital share of one-third is assumed.

*Source: IMF staff estimates and analysis as presented in the provided content.*

### Appendix I. Bayesian Model Averaging Technique

### Appendix I. Bayesian Model Averaging Technique

### Model uncertainty and objective of Model Averaging
- Growth econometrics generates two sources of model uncertainty: (i) uncertainty about the number of variables on the RHS and (ii) uncertainty in the way they have to be specified (lags, differences...).
- Practical consequences:
  - If researchers select k regressors and the “true” model is a combination of them, one could test 2^k specifications; infeasible in practice, leading researchers to limit specifications and tend to report “favorable” results.
  - Limited sample sizes make even favorable results sensitive to small specification changes; contradictory results are common when different covariates are considered.
- Objective of Model Averaging:
  - Run the maximum combination of models.
  - Provide estimates and inference that account for variable performance over the whole set of possible specifications.
- Formal unconditional Model Averaging estimator for coefficient β_j across m models:
  - β̂_j,BMA = Σ_{i=1}^m ω_i β̂_{j,i}
  - ω_i is the weight associated with the estimate of β_j using model i.
- Key design choices in model averaging:
  - Construct conditional estimators β̂_{j,i}.
  - Choose weights ω_i for each model-specific estimator.
  - Specify inference for the final estimator.

### Statistical framework
- Partition regressors into Focus and Auxiliary groups to test robustness of Focus regressors (always included) to inclusion/permutation of Auxiliary regressors.
- In the paper’s application: Focus group contains only the constant term; all remaining variables are Auxiliary.
- Linear regression model:
  - y = X_F β_F + X_A β_A + ε
  - y is n×1, X_F is n×k_F, X_A is n×k_A, β_F and β_A are parameter vectors, ε is n×1 with i.i.d. N(0, σ^2_ε) elements.
- If model uncertainty is limited to auxiliary regressors, number of possible permutations is 2^{k_A}.
- Denote model M_i as:
  - y = X_F β_F + X_{A,i} β_{A,i} + ε_i, for i = 1,...,2^{k_A}
  - X_{A,i} is n×k_{A,i}, β_{A,i} the associated parameters, ε_i disturbances after excluding k_A − k_{A,i} auxiliary regressors.

### Bayesian Model Averaging (BMA) methodology
- BMA ingredients: likelihood, prior on regression parameters, prior on model space.
- Conditional on model M_i, sample likelihood implied by the linear model (equation (4) in source).
- Priors used:
  - Noninformative priors for β_F and σ^2.
  - Uninformative Gaussian prior for auxiliary parameters β_{A,i}. Joint prior given by equation (5) with V_{A,i} the prior variance-covariance matrix taking Zellner (1986) form.
- Conditional posterior conditional estimates (as shown in source):
  - β̂_{F,i} = (X_F'X_F)^{-1} X_F'(y − X_{A,i} β̂_{A,i})
  - β̂_{A,i} = (1/(1+g)) (X_{A,i}'M_F X_{A,i})^{-1} X_{A,i}'M_F y  (expressions as presented in the source, equations (7) and (8)).
- Model posterior weights:
  - π_i = P(M_i | y) = [P(y | M_i) P(M_i)] / Σ_j P(y | M_j) P(M_j)
  - With uniform prior over model space, P(M_i) = 2^{-k_A}.
- Unconditional BMA estimates:
  - β̂_F = Σ_{i=1}^m π_i β̂_{F,i}
  - β̂_A = Σ_{i=1}^m π_i Ψ_i β̂_{A,i}, where Ψ_i are k_A×k_{A,i} matrices that set excluded coefficients to zero (equation (10)).
- Posterior variance-covariance matrices of unconditional estimators are computed as weighted sums of conditional variances plus between-model variability (expressions provided in source).

### Weighted Average Least Squares (WALS)
- WALS builds on Magnus and Durbin (1999) and Danilov and Magnus (2004); conceptually similar to BMA with important differences.
- Key differences:
  - Prior for model-specific parameters: Laplace distribution in WALS versus Normal in standard BMA. Using a Laplace distribution implies bounded risk (Magnus, Powell, and Prüfer (2010)).
  - WALS uses a preliminary orthogonal transformation of auxiliary regressors and their parameters, greatly reducing computational burden.
- Orthogonal transformation details:
  - Compute orthogonal k_A×k_A matrix Q and diagonal k_A×k_A matrix Δ such that Q' X_A' M_F X_A Q = Δ.
  - Define X̃_A = X_A Q Δ^{-1/2} and ζ_A = Δ^{1/2} Q' β_A so that X̃_A' M_F X̃_A = I_{k_A} and X̃_A ζ_A = X_A β_A.
  - Advantage: all models that use variable X as regressor will have the same estimator for β_j and the same t-ratio for β_j; computation reduces from 2^{k_A} estimations (BMA) to k linear combinations (WALS).
- WALS estimator expressions (equation (11) in source):
  - β̂_F,WALS = (X_F' X_F)^{-1} X_F'(y − X_A β̂_{A,WALS})
  - β̂_{A,WALS} = Δ^{-1/2} Q ζ̂, where ζ̂ is the Laplace estimator of the vector of theoretical t-ratios τ = (τ_1,...,τ_{k_A}).
- Variance-covariance expressions for WALS estimators are given in the source, involving Δ, Q, and the diagonal variance matrix of ζ̂.

### Criteria for significance and interpretation
- BMA:
  - Use Posterior Inclusion Probability (PIP) to assess robustness.
  - Adopt threshold: PIP > 0.5 indicates a robust regressor (Masanjala and Papageorgiou (2008)); this corresponds approximately to a t-ratio of 1.
  - Rationale: with a uniform prior over model space, the prior probability of including any regressor ex-ante is 0.5; a posterior inclusion probability above the prior indicates support for inclusion.
  - Note: no universally adopted PIP threshold; thresholds differ across studies because prior inclusion probabilities differ.
  - Example: Sala-i-Martin, Dopplehofer, and Miller (2004) use a lower threshold 0.104 due to penalization of large models.
- WALS:
  - PIPs cannot be computed for WALS estimators (WALS estimators are biased and their distribution is not Gaussian).
  - Magnus, Powell, and Prüfer (2010) suggest absolute t-ratio > 1 as criterion for robustness.
  - Motivation:
    - In a simple regression with one auxiliary regressor, the adjusted R^2 will decrease when removing the auxiliary variable if and only if the t-ratio of the auxiliary parameter is smaller than 1 in absolute value.
    - Define theoretical t-ratio τ = β √(n)/[σ/√(X' M_F X)] (source notation); if |τ| ≤ 1 then inclusion does not improve fit or estimator precision.
    - WALS sets priors so that the benchmark theoretical t-ratio is one; thereafter |t| > 1 qualifies the variable as robust.

*Source: Appendix I. Bayesian Model Averaging Technique*

### Appendix II. Tables and Charts

### Appendix II. Tables and Charts

### Table A.2.1. Growth Slowdowns Episodes (By income group)
- Middle Income: list of country-period slowdown episodes as reported:
  - Algeria 1980-1985
  - Haiti 1980-1985
  - Papua New Guinea 1995-2000
  - Algeria 1985-1990
  - Honduras 1960-1965
  - Paraguay 1980-1985
  - Argentina 1980-1985
  - Honduras 1980-1985
  - Peru 1975-1980
  - Argentina 1995-2000
  - Indonesia 1995-2000
  - Peru 1980-1985
  - Belize 1990-1995
  - Iran 1970-1975
  - Poland 1980-1985
  - Bolivia 1975-1980
  - Iran 1975-1980
  - Portugal 1970-1975
  - Botswana 1975-1980
  - Iraq 1980-1985
  - Romania 1975-1980
  - Botswana 2000-2005
  - Jamaica 1970-1975
  - Romania 1980-1985
  - Brazil 1975-1980
  - Jamaica 1990-1995
  - South Africa 1980-1985
  - Brazil 1980-1985
  - Jordan 1965-1970
  - Spain 1965-1970
  - Bulgaria 1980-1985
  - Jordan 1980-1985
  - Swaziland 1990-1995
  - Chile 1995-2000
  - Korea, Republic of 1970-1975
  - Syria 1975-1980
  - Congo, Republic of 1985-1990
  - Malaysia 1980-1985
  - Syria 1980-1985
  - Cyprus 1980-1985
  - Malaysia 1995-2000
  - Syria 1995-2000
  - Dominican Republic 1975-1980
  - Maldives 1985-1990
  - Thailand 1995-2000
  - Ecuador 1975-1980
  - Malta 1980-1985
  - Tonga 1985-1990
  - Ecuador 1980-1985
  - Mauritius 1975-1980
  - Trinidad & Tobago 1960-1965
  - Egypt 1995-2000
  - Mexico 1980-1985
  - Trinidad & Tobago 1980-1985
  - El Salvador 1975-1980
  - Namibia 1970-1975
  - Tunisia 1975-1980
  - El Salvador 1995-2000
  - Nicaragua 1965-1970
  - Uruguay 1995-2000
  - Gabon 1975-1980
  - Nicaragua 1985-1990
  - Venezuela 1975-1980
  - Guatemala 1980-1985
  - Panama 1980-1985
  - Yemen 2000-2005
  - Guyana 2000-2005
  - Papua New Guinea 1980-1985
  - Zambia 1970-1975
- High Income / Low Income: list of country-period slowdown episodes as reported:
  - Japan 1970-1975
  - Afghanistan 1985-1990
  - Pakistan 1965-1970
  - Japan 1990-1995
  - Benin 1985-1990
  - Sierra Leone 1990-1995
  - Finland 2000-2005
  - Burundi 1970-1975
  - Sudan 2000-2005
  - Ireland 2000-2005
  - Burundi 2000-2005
  - Togo 1990-1995
  - Malta 2000-2005
  - Cameroon 1985-1990
  - Uganda 1970-1975
  - Portugal 1990-1995
  - Republic of Congo 1970-1975
  - Zambia 1975-1980
  - Portugal 2000-2005
  - Cote d`Ivoire 1970-1975
  - Zimbabwe 1975-1980
  - Spain 1975-1980
  - Egypt 1965-1970
  - Zimbabwe 1990-1995
  - Spain 2000-2005
  - Ghana 1970-1975
  - Zimbabwe 2000-2005
  - Barbados 1970-1975
  - Indonesia 1975-1980
  - Barbados 1980-1985
  - Kenya 1990-1995
  - Barbados 2000-2005
  - Lao P.D.R. 1985-1990
  - Bahrain 1980-1985
  - Liberia 1980-1985
  - Cyprus 1990-1995
  - Liberia 1985-1990
  - Israel 1975-1980
  - Liberia 2000-2005
  - Kuwait 1995-2000
  - Malawi 1970-1975
  - Brunei 1980-1985
  - Malawi 1975-1980
  - Hong Kong SAR 1980-1985
  - Malawi 1980-1985
  - Hong Kong SAR 1990-1995
  - Mauritania 1975-1980
  - Korea 1990-1995
  - Mongolia 1990-1995
  - Korea 1995-2000
  - Morocco 1965-1970
  - Singapore 1995-2000
  - Mozambique 1975-1980
  - Niger 1980-1985
- Note: the table presents these slowdown episodes by income group and period exactly as listed.

### Table A.2.2. Growth Slowdowns Episodes (By criteria)
- The table maps slowdown episodes to three classification columns: Conditional Convergence, Absolute Convergence, Eichengreen and others (2011). Selected rows (economy-year and codes) include:
  - Honduras 1960-65 110
  - Trinidad and Tobago 1960-65 110
  - Jordan 1960-65 000
  - China 1960-65 000
  - Austria 1960-65 001
  - New Zealand 1960-65 001
  - Spain 1965-70 110
  - Nicaragua 1965-70 110
  - Jordan 1965-70 110
  - Egypt 1965-70 110
  - Hong Kong SAR 1965-70 000
  - Pakistan 1965-70 110
  - Mauritania 1965-70 000
  - Morocco 1965-70 110
  - Niger 1965-70 000
  - Namibia 1965-70 000
  - Togo 1965-70 000
  - Papua New Guinea 1965-70 000
  - Australia 1965-70 001
  - Denmark 1965-70 001
  - Japan 1965-70 001
  - New Zealand 1965-70 001
  - United States 1965-70 001
  - Japan 1970-75 110
  - Greece 1970-75 001
  - Portugal 1970-75 111
  - Barbados 1970-75 110
  - Jamaica 1970-75 110
  - Iran 1970-75 111
  - Korea, Republic of 1970-75 110
  - Burundi 1970-75 110
  - Congo, Republic of 1970-75 110
  - Ghana 1970-75 110
  - Cote d`Ivoire 1970-75 110
  - Malawi 1970-75 110
  - Mauritania 1970-75 010
  - Namibia 1970-75 110
  - Uganda 1970-75 110
  - Zambia 1970-75 110
  - Argentina 1970-75 001
  - Australia 1970-75 001
  - Belgium 1970-75 001
  - Denmark 1970-75 001
  - Finland 1970-75 001
  - France 1970-75 001
  - Ireland 1970-75 001
  - Israel 1970-75 001
  - Netherlands 1970-75 001
  - Puerto Rico 1970-75 001
  - Singapore 1970-75 001
  - (Table continues with many additional economy-year rows through 2000-05)
- Aggregate total at end of table: Total12312084
- Source for table: IMF staff estimates.

### Note on matching Eichengreen, Park, and Shin (2012) slowdowns to five-year panels
- Rule for imputing a slowdown year t into five-year panel periods with panel index s = 1...12:
  - if s−2 ≤ t ≤ s+2 then the slowdown is imputed to the period s−2 … s+2
  - if s−3 ≤ t ≤ s−2 then the slowdown is imputed to the period s−3 … s−2
  - if s+3 ≥ t ≥ s−2 then the slowdown is imputed to the period s+2 … s+3
- Example: slowdowns over the period 2000–05 reported in the right column include slowdowns that started in 1998, 1999, 2000, 2001 and 2002 according to Eichengreen, Park, and Shin (2012).
- Additional mapping notes:
  - Slowdowns that started between 1993 and 1997 are counted as 1995–2000 slowdowns.
  - Slowdowns that started between 2003 and 2007 are imputed to 2005–10.
  - A string of consecutive years can be imputed to 2 consecutive periods in the five-year panel dimension.

### Table A.2.3. Independent Variables: Unit and Sources
- Variable categories, descriptions, sources, and sample coverage (start, end, frequency) as listed:
  - Fertility rate, total (births per woman) — WDI — Demography — 1960 to 2009 — Annual
  - Dependency ratio — United Nations — Demography — 1950 to 2005 — 5-year
  - Sex ratio — United Nations — Demography — 1950 to 2005 — 5-year
  - Agriculture share of value added (percent of GDP) — WDI — Economic Structure — 1970 to 2011 — Annual
  - Services share of value added (percent of GDP) — WDI — Economic Structure — 1970 to 2011 — Annual
  - Industry share value added (percent of GDP) — WDI — Economic Structure — 1970 to 2011 — Annual
  - Output diversification — Papageorgiou and Spatafora (2012) — Economic Structure — 2000 to 2010 — Annual
  - Telephone lines — Calderon and Serven (2004); WDI — Infrastructure — 1960 to 2010 — 5-year
  - Power (generating capacity) — Calderon and Serven (2004); WDI — Infrastructure — 1960 to 2010 — 5-year
  - Roads — Calderon and Serven (2004); WDI — Infrastructure — 1960 to 2010 — 5-year
  - Size of government — Economic Freedom dataset — Institutions — 1960 to 2010 — 5-year
  - Rule of law — Economic Freedom dataset — Institutions — 1960 to 2010 — 5-year
  - Freedom to trade internationally — Economic Freedom dataset — Institutions — 1960 to 2010 — 5-year
  - Regulation (note: refers to credit market, labor market, and business regulations; of six subindices covering labor market regulations, only the three that are not taken from the Employing Workers Index of the World Bank’s Doing Business database are considered) — Economic Freedom dataset — Institutions — 1960 to 2010 — 5-year
  - Financial openness — Chinn and Ito (2006) — Institutions — 1970 to 2009 — Annual
  - Gross capital inflows as percentage of GDP — World Economic Outlook — MACRO — 1970 to 2009 — Annual
  - Gross capital outflows as percentage of GDP — World Economic Outlook — MACRO — 1970 to 2009 — Annual
  - Banking crisis dummy — Laeven and Valencia (2012) — MACRO — 1975 to 2008 — Annual
  - Real exchange rate — IMF staff calculations — MACRO — 1950 to 2009 — Annual
  - Trade openness at 2005 constant prices (percent) — PWT — MACRO — 1950 to 2009 — Annual
  - CPI inflation — WDI — MACRO — 1970 to 2010 — Annual
  - Price level of investment — PWT — MACRO — 1950 to 2009 — Annual
  - External debt (net) to GDP ratio — Lane and Milesi Ferretti — MACRO — 1970 to 2010 — Annual
  - Public debt to GDP ratio — Abbas and others (2010) — MACRO — 1950 to 2010 — Annual
  - Terms of trade — World Economic Outlook — MACRO — 1970 to 2009 — Annual
  - Reserves/GDP ratio — World Economic Outlook — MACRO — 1970 to 2010 — Annual
  - Investment share of PPP GDP per capita at 2005 constant — PWT — MACRO — 1960 to 2010 — Annual
  - Oil exporters' price shock — IMF staff calculations — MACRO — 1950 to 2010 — Annual
  - Food exporters' price shock — IMF staff calculations — MACRO — 1950 to 2010 — Annual
  - Oil importers' price shock — IMF staff calculations — MACRO — 1950 to 2010 — Annual
  - Food importers' price shock — IMF staff calculations — MACRO — 1950 to 2010 — Annual
  - Fraction of country in tropics — Sala-i-martin and others (2004) — Other — 1950 to 2010 — Annual
  - Spanish colony — Sala-i-martin and others (2004) — Other — 1950 to 2010 — Annual
  - Fraction Buddhist — Sala-i-martin and others (2004) — Other — 1950 to 2010 — Annual
  - Ethno linguistic fractionalization — Sala-i-martin and others (2004) — Other — 1950 to 2010 — Annual
  - War and civil conflicts — Correlates of War Project — Other — 1950 to 2010 — Annual
  - Natural disaster — International Disaster Database — Other — 1950 to 2010 — Annual
  - Distance (GDP weighted) — World Bank — TRADE — 1950 to 2010 — Annual
  - Regional integration — IMF staff calculations — TRADE — 1960 to 2010 — Annual
  - Trade diversification - Theil Index — Papageorgiou and Spatafora (2012) — TRADE — 1960 to 2010 — Annual

### Table A.2.4. Sample Statistics by Category
- Regional columns: Adv. (Advanced), East Asia Pacific, Europe and Central Asia, Latin America and the Caribbean, Middle East and North Africa, South Asia, Sub-Saharan Africa, Total
- Category 1: Institutions
  - Subsample size 128 51 9 100 58 23 99 468 42 (as printed)
  - Full sample size 215 147 83 214 1296 1276 1125 (as printed)
  - Subsample regional coverage (in percent) 27 11 22 11 25 21 (as printed)
  - Full sample regional coverage (in percent) 19 13 71 91 15 25 (as printed)
- Category 2: Demography
  - Subsample size 184 111 74 192 1205 6240 97787 (as printed)
  - Full sample size 215 147 83 214 1296 1276 1125 (as printed)
  - Subsample regional coverage (in percent) 19 11 82 01 26 25 (as printed)
  - Full sample regional coverage (in percent) 19 13 71 91 15 25 (as printed)
- Category 3: Infrastructure
  - Subsample size 154 71 31 133 390 311 066 1655 (as printed)
  - Full sample size 215 147 83 214 1296 1276 1125 (as printed)
  - Subsample regional coverage (in percent) 25 12 52 21 55 17 (as printed)
  - Full sample regional coverage (in percent) 19 13 71 91 15 25 (as printed)
- Category 4: Macroeconomic environment and policy
  - Subsample size 108 33 14 9348 1947 36232 (as printed)
  - Full sample size 215 147 83 214 1296 1276 1125 (as printed)
  - Subsample regional coverage (in percent) 39 42 61 35 13 (as printed)
  - Full sample regional coverage (in percent) 19 13 71 91 15 25 (as printed)
- Category 5: Output Composition
  - Subsample size 1337 241 996 129 171 60654 (as printed)
  - Full sample size 215 147 83 214 1296 1276 1125 (as printed)
  - Subsample regional coverage (in percent) 22 12 71 61 05 28 (as printed)
  - Full sample regional coverage (in percent) 19 13 71 91 15 25 (as printed)
- Category 6: Trade
  - Subsample size 1266 436 12559 3256 49844 (as printed)
  - Full sample size 215 147 83 214 1296 1276 1125 (as printed)
  - Subsample regional coverage (in percent) 25 13 72 51 26 11 (as printed)
  - Full sample regional coverage (in percent) 19 13 71 91 15 25 (as printed)
- Category 7: Other
  - Subsample size 2158 116 18 967 532 598 8078 (as printed)
  - Full sample size 215 147 83 214 1296 1276 1125 (as printed)
  - Subsample regional coverage (in percent) 24 92 21 86 29 (as printed)
  - Full sample regional coverage (in percent) 19 13 71 91 15 25 (as printed)
- Source: IMF staff estimates.

*Source: IMF staff estimates.*

### Appendix A.2.5. Composition of Regions

### Appendix A.2.5. Composition of Regions

### Regions and country composition

- Advanced: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, 
Greece, Iceland, Ireland, Italy, Japan, Luxembourg, Netherlands, New Zealand, Norway, 
Portugal, Spain, Sweden, Switzerland, Turkey, United Kingdom, and the United States.

- East Asia and Pacific: Brunei, Cambodia, China, Fiji, Hong Kong SAR, Indonesia, 
Republic of Korea, Lao P.D.R., Malaysia, Mongolia, Papua New Guinea, Philippines, 
Singapore, Taiwan Province of China, Thailand, Tonga, and Vietnam.

- Europe and Central Asia: Albania, Armenia, Bulgaria, Croatia, Czech Republic, 
Estonia, Hungary, Kazakhstan, Kyrgyzstan, Lithuania, Poland, Romania, Russia, Slovak 
Republic, Slovenia, Tajikistan, and Ukraine.

- Latin America and the Caribbean: Argentina, Barbados, Belize, Bolivia, Brazil, Chile, 
Colombia, Costa Rica, Dominican Republic, Ecuador, El Salvador, Guatemala, Guyana, 
Haiti, Honduras, Jamaica, Mexico, Nicaragua, Panama, Paraguay, Peru, Trinidad 
&Tobago, Uruguay, and Venezuela

- Middle East and North Africa: Algeria, Bahrain, Cyprus, Egypt, Iran, Iraq, Israel, 
Jordan, Kuwait, Libya, Malta, Morocco, Qatar, Saudi Arabia, Syria, Tunisia, United Arab 
Emirates, and Yemen.

- South Asia: Afghanistan, Bangladesh, India, Maldives, Nepal, Pakistan, Sri Lanka

- Sub-Saharan Africa: Benin, Botswana, Burundi, Cameroon, Central African Republic, 
Congo, Republic of, Cote d`Ivoire, Gabon, Gambia, Ghana, Kenya, Lesotho, Liberia, 
Malawi, Mali, Mauritania, Mauritius, Mozambique, Namibia, Niger, Rwanda, Senegal, 
Sierra Leone, South Africa, Sudan, Swaziland, Tanzania, Togo, Uganda, Zambia, and 
Zimbabwe.

*Appendix A.2.5. Composition of Regions*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2013/_wp1371.pdf_