## Appendix I. Treated and Control Countries by Region (_wp07156)

## Source details

**Canonical URL:** [Appendix I. Treated and Control Countries by Region (_wp07156)](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2007/_wp07156.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2007/_wp07156.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2007/_wp07156.pdf.json)

---

### Periods Covered
- A. Period 1991–2000
- B. Period 1981–90
- C. Period 1971–80
- D. Period 1961–70

### Purpose and Methodology — core points
- Research question:
  - Evaluate the effect of a binary treatment—trade openness or economic liberalization—on changes in per capita income.
- Main methodological approaches used:
  - Cross-country pooled regressions (OLS) and panel methods.
  - Nearest-neighbor matching estimators (matching in covariate space using V diagonal with inverses of variances).
  - Synthetic control method (SCM) in a panel set-up to construct transparent time-dimensioned counterfactuals and account for time-varying country unobservables.
- Identification and diagnostics emphasized:
  - Unconfoundedness / conditional independence assumption required for matching: Y(1), Y(0) ⊥ T | X and 0 < Pr(T = 1 | X) < 1 (weaker form for ATT).
  - Importance of checking common support between treated and control units (investment share and secondary school enrollment highlighted).
  - Use of area dummies (Africa, Asia, Latin America, Middle East, OECD, transition economies) to control for region-specific unobservables.
  - Trade-off: restricting sample to regions with common support improves internal validity but reduces external validity.

### Main empirical conclusions (high-level)
- After removing groups lacking common support, the authors:
  - Confirm a significant effect of openness on growth within selected areas and after 1970.
- Methodological adequacy:
  - Matching results are instructive but cannot fully address unobservable country heterogeneity and reverse causality.
  - Panel methods overcome some OLS weaknesses; SCM addresses time-varying unobservables and provides clear counterfactuals.
- Application summary:
  - For most studied episodes of trade liberalization in IMF MCD countries after 1965, trade liberalization has had a positive effect on growth.

### Key cross-sectional estimates and exact statistics (preserved)
- Table 2 (1961-2000, Matching with area dummies):
  - 1991-2000: Estimate (ATE) 1.318 **; Standard error (0.672)
  - 1981-90: Estimate (ATE) 0.641; Standard error (0.462)
  - 1971-80: Estimate (ATE) 2.399 ***; Standard error (0.825)
  - 1961-70: Estimate (ATE) -0.442; Standard error (0.829)
  - 1991-2000: Estimate (ATT) 1.130; Standard error (0.742)
  - 1981-90: Estimate (ATT) 0.826; Standard error (0.579)
  - 1971-80: Estimate (ATT) 1.895 **; Standard error (0.831)
  - 1961-70: Estimate (ATT) 0.213; Standard error (0.943)
- Table 1 (1950-98) selected entries:
  - OLS without area dummies, 1990-98: Estimate 1.630 **; Standard error (0.772); Adjusted R2 0.36
  - Matching without area dummies, 1990-98: Estimate (ATE) 1.731 **; Standard error (0.808)
  - Matching with area dummies, 1990-98: Estimate (ATE) 1.604 **; Standard error (0.749)
- Table 11 (Refined evidence, restricted to regions with common support; 1961-2000):
  - Matching without area dummies, 1991-2000: Estimate (ATE) 1.921 ***; Standard error (0.647)
  - Matching with area dummies, 1971-80: Estimate (ATE) 2.240 ***; Standard error (0.684)
  - Matching without area dummies, 1971-80: Estimate (ATT) 2.229 ***; Standard error (0.625)
  - Refined-sample estimated effects reported to lie in the 1.5–2.5 percentage points range where significant.
- Table 12 (Matching with common support restrictions):
  - Matching with common support A (investment share): 1991-2000 ATE 1.318 **; Standard error (0.672)
  - Matching with common support B (secondary school): 1991-2000 ATE 1.644 **; Standard error (0.731)
  - Matching with common support C (both): 1991-2000 ATE 1.644 **; Standard error (0.731)
  - ATT reported in Table 12 for some specifications (e.g., ATT 1.407 *; Standard error (0.819)).

### Findings on matching, common support, and robustness
- Matching primarily:
  - Reveals the specific country comparisons underlying cross-sectional estimates.
  - Allows restriction to common-support regions and inspection of plausibility of nearest-neighbor matches.
- Diagnostic findings:
  - Many baseline nearest-neighbor matches appear implausible without area controls; area dummies help but are not sufficient unless overlap exists.
  - OECD and Latin America in the 1990s: lack of common support (e.g., control group limited to Iceland for OECD, Venezuela sole control in Latin America in some samples).
  - Common-support restrictions:
    - Improve plausibility of matches and sometimes increase magnitude and significance of openness coefficients in the 1990s and 1970s.
    - Leave coefficients not significantly different from zero in the 1960s.
  - Example diagnostic (Figure 5, 1971-80): common support for investment share ranges from 0.11 to 0.39; 27 control countries below this region and 1 treated country above—these 28 should be dropped to meet common-support.

### Synthetic Control Method (SCM) — MCD case studies and match summaries
- Treated liberalizations and treatment years (SCM, 1960–2000):
  - Armenia (1996), Azerbaijan (1996), Egypt (1995), Georgia (1996), Morocco (1984), Mauritania (1995), Tajikistan (1996), Tunisia (1989).
- Selected pre-treatment predictor match statistics (Table 13, exact reported numbers preserved for selected cases):
  - Georgia: Secondary school 75.850 vs. Synth. Control A 72.840 and Synth. Control B 92.529; Population growth -0.233 vs. 0.957 and 1.401; Investment share 0.040 vs. 0.124 and 0.115; Pre-treatment GDP 583.067 vs. 678.535 and 594.333.
  - Armenia: Pre-treatment GDP 441.634 vs. Synth. Control A 511.739 and B 594.333.
  - Tajikistan: Pre-treatment GDP 239.526 vs. Synth. Control A 342.553 (noted as a poor match).
  - Tunisia: Pre-treatment GDP 1,089.462 vs. Synth. Control A 1,070.248 and B 1,088.473.
  - Egypt: Pre-treatment GDP 764.775 vs. Synth. Control A 837.891 and B 981.882.
- SCM outcome patterns (qualitative with numeric context preserved where given):
  - Georgia and Armenia: treated countries fare much better than synthetic controls after liberalization (both broad and region-restricted control sets).
  - Azerbaijan: trade liberalization has not had a positive effect in SCM results.
  - Tajikistan: SCM matching unsatisfactory due to per capita GDP being too low to be matched.
  - Uzbekistan (reverse treatment / missed liberalization): synthetic counterfactual constructed from Armenia, Kyrgyz Republic, Tajikistan performed substantially better in per capita GDP, indicating a missed opportunity.
  - Morocco: outperforms synthetic controls; in Case B (controls restricted to Middle East) synthetic control per capita GDP about 40 percent below Morocco’s 10 years after treatment (noting 32 percent of synthetic control is Algeria).
  - Tunisia: evidence similar to Morocco under some specifications.
  - Mauritania: no clear effect associated with trade liberalization.
  - Egypt: mixed results; algorithm assigns low weight to poorer nearby African countries and sometimes picks Syria as main control.

### Main conclusions and implications (preserved)
- Cross-sectional estimates without common-support checks or area-specific controls rely on implausible country comparisons and extrapolation beyond observed overlap.
- After restricting samples to ensure common support and controlling for area heterogeneity:
  - Matching estimates broadly confirm an openness–growth association found in the literature, with stronger impacts in some periods (notably 1970s, 1980s, 1990s) and insignificant effects in the 1960s.
- SCM case studies for MCD countries:
  - Find trade liberalization has had a positive effect on per capita income growth in most treated cases.
- Trade-off emphasized:
  - Restricting samples to satisfy common support improves internal validity but reduces external validity (results should not be extrapolated to developed countries, Africa in 1960s/1970s, or Latin America in the 1990s where common support is lacking).
- Recommended further work (exact intent preserved):
  - Apply the inference procedure proposed by Abadie, Diamond, and Hainmueller (2007) — placebo tests through permutation methods — for robustness in MCD countries.
  - Extend SCM analysis to other regions with relevant liberalization episodes over 1960-2000 (e.g., Latin America, East Asia, Africa).

### Appendix I — Treated and Control Countries by Region (lists by period)

- A. Period 1991–2000
  - AFRICA
    - Treated: Benin, Botswana, Cameroon, Cape Verde, Ethiopia, Gambia,Ghana, Guinea, Ivory Coast, Kenya, Madagascar, Mali, Mauritius, Mozambique, Niger, South Africa, Tanzania, Uganda, Zambia.
    - Control: Angola, Burkina Faso, Burundi, Central African Republic,Chad, Congo, Gabon, Lesotho, Malawi, Nigeria, Rwanda, Senegal, Togo, Zimbabwe.
  - ASIA
    - Treated: Hong Kong, Indonesia, Malaysia, Nepal, Philippines, Singapore, South Korea, Sri Lanka, Thailand.
    - Control: China, India, Pakistan, Papua New Guinea.
  - LATIN AMERICA
    - Treated: Argentina, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador, El Salvador, Guatemala, Guyana, Honduras, Jamaica, Mexico, Nicaragua, Panama, Paraguay, Peru, Trinidad & Tobago, Uruguay.
    - Control: Venezuela.
  - MIDDLE EAST
    - Treated: Egypt, Jordan, Mauritania, Morocco, Tunisia, Yemen.
    - Control: Algeria, Iran, Syria.
  - OECD
    - Treated: Australia, Austria, Belgium, Canada, Cyprus, Denmark, Finland, France, Germany, Greece, Ireland, Israel, Italy, Japan, Luxembourg, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, Turkey, United Kingdom, United States.
    - Control: Iceland.
  - TRANSITION ECONOMIES
    - Treated: Albania, Bulgaria, Czech Republic, Hungary, Latvia, Poland, Romania, Slovak Republic, Slovenia.
    - Control: Belarus, Russia, Ukraine.

- B. Period 1981–90
  - AFRICA
    - Treated: Botswana, Gambia, Ghana, Guinea, Mauritius.
    - Control: Angola, Benin, Burkina Faso, Burundi, Cameroon, Cape Verde, Central African Republic, Chad, Congo, Ethiopia, Gabon, Guinea-Bissau, Ivory Coast, Kenya, Lesotho, Madagascar, Malawi, Mali, Mozambique, Niger, Nigeria, Rwanda, Senegal, Sierra Leone, South Africa, Tanzania, Togo, Uganda, Zambia, Zimbabwe.
  - ASIA
    - Treated: Hong Kong, Indonesia, Malaysia, Singapore, South Korea, Thailand.
    - Control: Bangladesh, China, India, Nepal, Pakistan, Papua New Guinea, Philippines, Sri Lanka.
  - LATIN AMERICA
    - Treated: Barbados, Bolivia, Chile, Colombia, Costa Rica, Mexico.
    - Control: Argentina, Brazil, Dominican Republic, Ecuador, El Salvador, Guatemala, Guyana, Haiti, Honduras, Jamaica, Nicaragua, Panama, Paraguay, Peru, Trinidad & Tobago, Uruguay, Venezuela.
  - MIDDLE EAST
    - Treated: Jordan, Morocco.
    - Control: Algeria, Egypt, Iran, Mauritania, Syria, Tunisia.
  - OECD
    - Treated: Australia, Austria, Belgium, Canada, Cyprus, Denmark, Finland, France, Germany, Greece, Ireland, Israel, Italy, Japan, Luxembourg, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom, United States.
    - Control: Iceland, Turkey.
  - TRANSITION ECONOMIES
    - Treated: None.
    - Control: Hungary, Poland, Romania.

- C. Period 1971–80
  - AFRICA
    - Treated: Mauritius.
    - Control: Angola, Benin, Botswana, Burkina Faso, Burundi, Cameroon, Cape Verde, Central African Republic, Chad, Congo, Ethiopia, Gabon, Gambia, Ghana, Guinea, Guinea-Bissau, Ivory Coast, Kenya, Lesotho, Madagascar, Malawi, Mali, Mozambique, Niger, Nigeria, Rwanda, Senegal, Sierra Leone, Tanzania, Togo, Uganda, Zambia, Zimbabwe.
  - ASIA
    - Treated: Hong Kong, Indonesia, Malaysia, Singapore, South Korea, Thailand.
    - Control: Bangladesh, China, India, Nepal, Pakistan, Papua New Guinea, Philippines, Sri Lanka.
  - LATIN AMERICA
    - Treated: Barbados, Bolivia, Chile, Ecuador.
    - Control: Argentina, Brazil, Colombia, Costa Rica, Dominican Republic, El Salvador, Guatemala, Guyana, Haiti, Honduras, Jamaica, Mexico, Nicaragua, Panama, Paraguay, Peru, Trinidad & Tobago, Uruguay, Venezuela.
  - MIDDLE EAST
    - Treated: Jordan.
    - Control: Algeria, Egypt, Iran, Mauritania, Morocco, Syria, Tunisia.
  - OECD
    - Treated: Australia, Austria, Belgium, Canada, Cyprus, Denmark, Finland, France, Greece, Ireland, Italy, Japan, Luxembourg, Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom, United States.
    - Control: Iceland, Israel, New Zealand, Turkey.
  - TRANSITION ECONOMIES
    - Treated: None.
    - Control: Hungary, Poland, Romania.

- D. Period 1961–70
  - AFRICA
    - Treated: None.
    - Control: Angola, Benin, Botswana, Burkina Faso, Burundi, Cameroon, Cape Verde, Central African Republic, Chad, Congo, Ethiopia, Gabon, Gambia, Ghana, Guinea, Guinea-Bissau, Ivory Coast, Kenya, Lesotho, Madagascar, Malawi, Mali, Mauritius, Mozambique, Niger, Nigeria, Rwanda, Senegal, Sierra Leone, South Africa, Tanzania, Togo, Uganda, Zambia, Zimbabwe.
  - ASIA
    - Treated: Hong Kong, Malaysia, Singapore, Thailand.
    - Control: Bangladesh, China, India, Indonesia, Nepal, Pakistan, Papua New Guinea, Philippines, South Korea, Sri Lanka.
  - LATIN AMERICA
    - Treated: Barbados, Bolivia, Ecuador, Jamaica, Peru.
    - Control: Argentina, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, El Salvador, Guatemala, Guyana, Haiti, Honduras, Mexico, Nicaragua, Panama, Paraguay, Trinidad & Tobago, Uruguay, Venezuela.
  - MIDDLE EAST
    - Treated: Jordan.
    - Control: Algeria, Egypt, Iran, Mauritania, Morocco, Syria, Tunisia.
  - OECD
    - Treated: Australia, Austria, Belgium, Canada, Cyprus, Denmark, Finland, France, Greece, Ireland, Italy, Japan, Luxembourg, Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom, United States.
    - Control: Iceland, Israel, New Zealand, Turkey.
  - TRANSITION ECONOMIES
    - Treated: None.
    - Control: Romania.

*Italic: Source — _wp07156 (Appendix I. Treated and Control Countries by Region).*

### 1.   Appendix I. Treated and Control Countries by Region . . . . .. . . . . . . . . . . . . .  44

### Appendix I. Treated and Control Countries by Region

### Periods Covered
- A. Period 1991–2000
- B. Period 1981–90
- C. Period 1971–80
- D. Period 1961–70

### Tables and Figures Inventory (as listed)
- Tables:
  - 1. Openness and Growth, Cross-Country Evidence (I), 1950-98
  - 2. Openness and Growth, Cross-Country Evidence (II), 1961-2000
  - 3. Cross-Country Matches, Treated Countries, 1991-2000
  - 4. Cross-Country Matches, Control Countries, 1991-2000
  - 5. Cross-Country Matches, Treated Countries, 1981-90
  - 6. Cross-Country Matches, Control Countries, 1981-90
  - 7. Cross-Country Matches, Treated Countries, 1971-80
  - 8. Cross-Country Matches, Control Countries, 1971-80
  - 9. Cross-Country Matches, Treated Countries, 1961-70
  - 10. Cross-Country Matches, Control Countries, 1961-70
  - 11. Openness and Growth, Refined Evidence (I), 1961-2000
  - 12. Openness and Growth, Refined Evidence (II), 1961-2000
  - 13. Economic Growth Predictor Means in the Pre-Treatment Period
- Figures:
  - 1. Common support for investment share, 1991-2000
  - 2. Common support for secondary school enrollment, 1991-2000
  - 3. Common support for investment share, 1981-90
  - 4. Common support for secondary school enrollment, 1981-90
  - 5. Common support for investment share, 1971-80
  - 6. Common support for secondary school enrollment, 1971-80
  - 7. Common support for investment share, 1961-70
  - 8. Common support for secondary school enrollment, 1961-70
  - 9. Trends in Real GDP Per Capita, Georgia vs. Synthetic Control - Case A
  - 10. Trends in Real GDP Per Capita, Georgia vs. Synthetic Control - Case B
  - 11. Trends in Real GDP Per Capita, Armenia vs. Synthetic Control - Case A
  - 12. Trends in Real GDP Per Capita, Armenia vs. Synthetic Control - Case B
  - 13. Trends in Real GDP Per Capita, Azerbaijan vs. Synthetic Control - Case A
  - 14. Trends in Real GDP Per Capita, Azerbaijan vs. Synthetic Control - Case B
  - 15. Trends in Real GDP Per Capita, Tajikistan vs. Synthetic Control - Case A
  - 16. Trends in Real GDP Per Capita, Uzbekistan vs. Synthetic Control - Case A
  - 17. Trends in Real GDP Per Capita, Morocco vs. Synthetic Control - Case A
  - 18. Trends in Real GDP Per Capita, Morocco vs. Synthetic Control - Case B
  - 19. Trends in Real GDP Per Capita, Tunisia vs. Synthetic Control - Case A
  - 20. Trends in Real GDP Per Capita, Tunisia vs. Synthetic Control - Case B
  - 21. Trends in Real GDP Per Capita, Mauritania vs. Synthetic Control - Case A
  - 22. Trends in Real GDP Per Capita, Mauritania vs. Synthetic Control - Case B
  - 23. Trends in Real GDP Per Capita, Egypt vs. Synthetic Control - Case A
  - 24. Trends in Real GDP Per Capita, Egypt vs. Synthetic Control - Case B

### Introduction — Purpose and Methodology (selected excerpts)
- Research question:
  - Evaluate the effect of a binary treatment—trade openness or economic liberalization—on changes in per capita income.
- Methodological concerns highlighted:
  - Cross-country regression estimates suffer from specification problems, endogeneity, and measurement errors.
  - Matching estimators (microeconometric) are applied to make estimation more transparent while relying on the conditional independence assumption.
  - Importance of controlling for continent or macro-region dummies to improve cross-country comparisons.
  - Need to ensure common support between treated and control units; developed countries may provide no within-group variation for some openness indicators.
- Alternative approach adopted:
  - Synthetic control method in a panel set-up to account for time-varying country unobservables and construct transparent counterfactuals.
  - Treatment is time-dimensioned (trade liberalization episodes after 1965 for IMF’s Middle East and Central Asia Department (MCD) countries).
- Key methodological trade-off noted:
  - Cleaning sample for regions without common support improves internal validity but reduces external validity.

### Main Empirical Conclusions (as stated)
- After removing groups lacking common support, the authors:
  - Confirm a significant effect of openness on growth within selected areas and after 1970.
- On methodological adequacy:
  - Matching results are instructive but cannot fully address unobservable country heterogeneity and reverse causality.
  - Panel methods can overcome some OLS weaknesses, but transparency issues remain; synthetic control method addresses time-varying unobservables and provides a clear counterfactual construction.
- Application result summary:
  - For most studied episodes of trade liberalization in IMF MCD countries after 1965, trade liberalization has had a positive effect on growth.

*Source: _wp07156 - 1.   Appendix I. Treated and Control Countries by Region (PDF chapter/section).*

### Section III, we briefly present the data sources and variables of interest. In Section IV, we

### _wp07156 - Section III, we briefly present the data sources and variables of interest. In Section IV, we

### II. Literature Review — Empirical Studies on Trade Openness and Growth
- Cross-country evidence on the causal effect of trade on growth is challenging due to measurement of openness and econometric modeling.
- Seminal cross-country studies finding a positive link between trade openness and growth: Dollar (1992), Sachs and Warner (1995), Edwards (1992, 1998).
- Vamvakidis (2002): evidence that trade is associated with growth after 1970 but not before.
- Critiques and limitations:
  - Rodriguez and Rodrik (2001): choice of openness measure and weak econometric strategies; find little evidence of a significant association after corrections.
  - Harrison (1996): most of explanatory power of the Sachs and Warner (1995) composite openness dummy comes from non-trade components.
  - Bhagwati and Srinivasan (2002, p.181): “cross-country regressions are a poor way to approach this question”; sensitivity to period, sample, and proxies.
  - Levine and Renelt (1992) and Temple (2000): extreme-bounds analysis shows lack of robustness of cross-country growth regressions.
- Identification issues in cross-country studies:
  - Reverse causality and endogeneity (country-specific omitted characteristics affecting both openness and growth).
  - Instrumental variables (IV) approaches:
    - Frankel and Romer (1999) and Irwin and Tervio (2002): use gravity-model based instruments (population, land area, borders, distances) to isolate geographical components of openness; find positive effect from trade to growth, but instruments may have indirect effects on growth.
    - Dollar and Kraay (2003): suggest estimating regressions in differences and using lagged openness as instrument; simultaneity bias may persist.
    - Lee, Ricci, and Rigobon (2004): identification through heteroskedasticity in a panel framework; find small, positive, but not robust effect; rely on non-testable assumption of uncorrelated structural shocks.
- Panel methods and difference-in-differences:
  - Harrison (1996): fixed-effect estimators find stronger impact of openness indicators in panel versus cross-section.
  - Wacziarg and Welch (2003): update SW openness indicator; show positive effect in panel even in the 1990s when focusing on within-country effects.
  - Slaughter (2001): difference-in-differences on four trade liberalization events finds no systematic link between trade liberalization and per capita income convergence.
  - Giavazzi and Tabellini (2005): difference-in-differences finds positive effect of economic liberalization on growth, but not wholly attributable to international trade since liberalizations coincide with other policy improvements.

### II. Literature Review — Empirical Studies Applying Matching Estimators to Macro Data
- Growing strand applies microeconometric treatment-evaluation estimators to cross-country macro data to overcome OLS weaknesses.
- Examples:
  - Persson and Tabellini (2005): propensity-score matching for political institutions.
  - Edwards and Magendzo (2003): matching for macroeconomic record of dollarized economies.
  - Atoyan and Conway (2006): matching to evaluate IMF program impacts.
- Advantages of matching estimators:
  - Relax OLS linearity assumption without imposing parametric functional form.
  - Allow checking for common support in distributions of treated and control units across covariates.
  - Transparency: in small country samples, matched treated and control units can be easily identified and inspected.
- Limitations:
  - Unconfoundedness (selection on observables) is unlikely to fully hold in cross-country context; matching relies on same identification condition as regression.
  - Matching useful to transparently reveal country comparisons behind cross-sectional results and to restrict estimates to the region of common support.

### III. Data and Variables of Interest
- Outcome variable used: difference of (log) per capita GDP (dynamic impact of trade openness over time).
- Two groups of openness indicators:
  - Simple trade-volume measures (subject to endogeneity, often used with IV).
  - Trade-policy measures capturing barriers and policy regime; primary binary indicator used is the SWWW (short SWWW) trade openness policy dummy (Sachs and Warner (1995) extended by Wacziarg and Welch (2003)).
- SWWW binary indicator: country considered closed in a given year if at least one condition holds:
  - (i) average tariffs exceed 40 percent;
  - (ii) non-tariff barriers cover more than 40 percent of its imports;
  - (iii) it has a socialist economic system;
  - (iv) the black market premium on the exchange rate exceeds 20 percent;
  - (v) much of its exports are controlled by a state monopoly.
- Definition of treatment in synthetic control context (Section V): “treatment” = event of becoming open after being closed in the preceding year according to the SWWW dummy (capturing policy changes that reduce constraints below a critical threshold along these five dimensions).
- Data sets used:
  - Vamvakidis (2002) data (focus on post-1950): repeated country cross-sections for periods 1950-70, 1970-90, and 1990-98. Variables: average GDP per capita growth (outcome), SWWW openness dummy, initial GDP, investment share, population growth, secondary school enrollment, inflation, black market premium.
  - Persson and Tabellini (PT) / Giavazzi and Tabellini (2005) panel (about 180 countries, 1960–2000): annual observations for updated SWWW, log change in per capita GDP, and control variables as above with two exceptions—inflation is not reported and a democracy dummy is present. Outcome extended to 2005 where available using IMF World Economic Outlook database to gain additional observations for synthetic control comparisons (important for recent liberalizers, e.g., transition economies).

### IV. Cross-Country Analysis and Matching Estimators — Methodology
- Framework: Rubin’s (1974) potential-outcome causal inference.
  - Potential outcomes: Y(0) = GDP growth if T = 0 (closed), Y(1) = GDP growth if T = 1 (open).
  - Causal effect defined as Y(1) − Y(0) (counterfactual problem: one potential outcome is unobserved).
- Parameters of interest:
  - Average Treatment Effect (ATE): E[Y(1) − Y(0)].
  - Average effect of Treatment on the Treated (ATT): E[Y(1) − Y(0) | T = 1].
- Identification condition: Unconfoundedness / conditional independence assumption:
  - Y(1), Y(0) ⊥ T | X
  - 0 < Pr(T = 1 | X) < 1
  - Weaker condition for ATT: Y(0) ⊥ T | X and Pr(T = 1 | X) < 1.
- Matching versus regression:
  - Both rely on unconfoundedness.
  - Matching allows checking common support and transparency in country matches; nearest-neighbor covariate matching implemented (rather than propensity-score matching).
- Nearest-neighbor matching formalism (as presented):
  - Vector norm ||x||_V = (x' V x)^{1/2} with positive definite weight matrix V.
  - Distance ||x − z||_V between covariate vectors.
  - d(i) = smallest distance from covariates of country i to covariates of countries with opposite treatment.
  - J(i) = {k = 1, ..., N | T_k = 1 − T_i, ||X_k − X_i||_V = d(i)}.
  - Estimated potential outcomes:
    - ˆY_i(l) = Y_i if T_i = l.
    - ˆY_i(l) = (1 / #J(i)) ∑_{k∈J(i)} Y_k if T_i = 1 − l.
  - ATE estimator: τ_ATE = (1 / I) ∑_{i=1}^I [ˆY_i(1) − ˆY_i(0)].
  - ATT estimator: τ_ATT = (1 / I_T) ∑_{i: T_i = 1} [Y_i − ˆY_i(0)].
  - V used is diagonal matrix with inverses of variances of covariates on diagonal (Abadie and others (2004)); estimates robust to alternative distance metrics (e.g., Mahalanobis distance).
- Role of matching in this context:
  - Not a panacea — unconfoundedness still required.
  - Advantage: transparency of matches, ability to check common support, restrict estimates to common-support region, inspect plausibility via country-level matches.

### IV. Cross-Country Analysis and Matching Estimators — The Unbearable Lightness of Cross-Country Estimates (Empirical Findings)
- Vamvakidis (2002) data set results (Table 1 summary):
  - Confirmed: openness (Sachs-Warner dummy) has a significant effect on growth after 1970, but not before.
  - Coefficients indicate open country grows, on average, by 1.5–2.0 percentage points per year faster than a closed economy.
  - Matching estimates (ATE and ATT) qualitatively and quantitatively similar to standard OLS results.
  - Estimates robust to inclusion of regional dummies.
  - Data set drawbacks:
    - (i) data pooled for 20-year intervals;
    - (ii) information stops in 1998 (too early for former Soviet Union countries);
    - (iii) small sample size in the 1950s and 1960s (mainly OECD and Latin American countries).
- Persson and Tabellini (PT) data set results (Table 2 summary):
  - Using Wacziarg and Welch (2003) updated SWWW, pooled estimates by decades for whole data set.
  - Matching results very similar to OLS; significant effect of trade on growth in the 1990s and 1970s.
  - In the 1990s and 1970s, open countries grew on average by 1.5–2.0 percentage points faster than closed countries (with or without regional dummies).
  - Growth effect of openness not significantly different from zero in the 1980s and 1960s.
- Transparency analysis via nearest-neighbor matching:
  - Nearest-neighbor matching used to display full lists of treated countries and their nearest neighbors (ATE estimation) for decades: 1990s, 1980s, 1970s, 1960s (Tables 3, 5, 7, 9).
  - Corresponding tables for control (closed) countries and their nearest neighbors (Tables 4, 6, 8, 10).
  - Tables report:
    - Column 1: Country under consideration.
    - Column 2 (Baseline): nearest neighbor used to estimate counterfactual in ATE without area dummies.
    - Column 3 (Area): nearest neighbor used to estimate counterfactual in ATE with area dummies.
  - Example: Baseline matches in Tables 3 and 4 for the 1990s underlie the 1.505 coefficient reported in Table 2 (effect of openness on growth without area dummies); Area matches underlie the counterpart with area dummies.
- Overall synthesis:
  - Matching does not necessarily change OLS point estimates in these data sets; it reveals the specific country comparisons underlying cross-sectional estimates and permits restricting analysis to common-support regions for plausibility checks.
  - Matching’s transparency allows inspection of whether treated and control countries are comparable across covariates before attributing causal interpretation.

*Italic: Source — _wp07156 (excerpt: Sections II–IV).*

### 1.318 coefficient in Table 2 (i.e., the effect of openness aftercontrolling for area dummies).

### _wp07156 - 1.318 coefficient in Table 2 (i.e., the effect of openness after controlling for area dummies)

### Methodology and identification concerns
- Uses cross-country pooled regressions, nearest-neighbor matching (matching estimators), and panel synthetic control methods (SCM).
- Treatment: trade openness dummy (Sachs and Warner, 1995; Wacziarg and Welch, 2003 / Vamvakidis (2002) in other tables).
- Control variables: initial GDP per capita, secondary school enrollment, population growth, investment share, black market premium, inflation (depending on decade and dataset).
- Area dummies used for Africa, Asia, Latin America, Middle East, OECD, and transition economies (OECD participation takes precedence over geographic region).
- Emphasis on checking common support before estimating effects; matching requires overlap between treated and control units across covariates (investment share, secondary school enrollment highlighted).

### Key cross-sectional estimates and exact statistics (preserving reported values)
- Table 2 (1961-2000, Matching with area dummies):
  - 1991-2000: Estimate (ATE) 1.318 **; Standard error (0.672)
  - 1981-90: Estimate (ATE) 0.641; Standard error (0.462)
  - 1971-80: Estimate (ATE) 2.399 ***; Standard error (0.825)
  - 1961-70: Estimate (ATE) -0.442; Standard error (0.829)
  - 1991-2000: Estimate (ATT) 1.130; Standard error (0.742)
  - 1981-90: Estimate (ATT) 0.826; Standard error (0.579)
  - 1971-80: Estimate (ATT) 1.895 **; Standard error (0.831)
  - 1961-70: Estimate (ATT) 0.213; Standard error (0.943)
- Table 1 (1950-98) selected entries (preserving exact values):
  - OLS without area dummies, 1990-98: Estimate 1.630 **; Standard error (0.772); Adjusted R2 0.36
  - Matching without area dummies, 1990-98: Estimate (ATE) 1.731 **; Standard error (0.808)
  - Matching with area dummies, 1990-98: Estimate (ATE) 1.604 **; Standard error (0.749)
- Table 11 (Refined evidence, restricted to regions with common support; 1961-2000):
  - Matching without area dummies, 1991-2000: Estimate (ATE) 1.921 ***; Standard error (0.647)
  - Matching with area dummies, 1971-80: Estimate (ATE) 2.240 ***; Standard error (0.684)
  - Matching without area dummies, 1971-80: Estimate (ATT) 2.229 ***; Standard error (0.625)
  - All estimated effects (refined samples) reported to lie in the 1.5–2.5 percentage points range for several decades where significant effects are found.
- Table 12 (Matching with common support restrictions):
  - Matching with common support A (investment share): 1991-2000 ATE 1.318 **; Standard error (0.672)
  - Matching with common support B (secondary school): 1991-2000 ATE 1.644 **; Standard error (0.731)
  - Matching with common support C (both): 1991-2000 ATE 1.644 **; Standard error (0.731)
  - Comparable ATT estimates are reported in Table 12 with exact values preserved (e.g., ATT 1.407 *; Standard error (0.819) for some specifications).

### Findings on country matches, common support, and robustness
- Many baseline nearest-neighbor matches appear far-fetched; matching often pairs implausible country pairs unless area-specific unobservables are controlled for.
- OECD and Latin America in the 1990s: no common support between treated and control countries (resulting in matches like "all OECD members are matched with Iceland" and Venezuela being the only match in Latin America for some samples).
- Introducing area dummies alone is insufficient unless there is sufficient overlap of treated and untreated units in all areas.
- Restricting samples to regions with common support (dropping regions lacking overlap):
  - Improves the intuitiveness and plausibility of matches (counter-intuitive matches reduced).
  - Yields slightly larger and sometimes more statistically significant openness coefficients in the 1990s and 1970s.
  - Produces marginally significant positive effect for 1980s (especially ATT).
  - Continues to find coefficients not significantly different from zero in the 1960s.
- Common-support diagnostics (Figures 1–8) show frequent lack of overlap for investment share and secondary school enrollment across decades:
  - Example: Figure 5 (1971-80) common support for investment share ranges from 0.11 to 0.39; 27 control countries below this region and 1 treated country above—these 28 should be dropped to meet common-support.

### Synthetic Control Methods (SCM) evidence — MCD case studies and numerical match summaries
- SCM applied to MCD (Middle East, North Africa, Central Asia/Caucasus) liberalization episodes 1960–2000.
- Treated liberalizations and years (as used in SCM):
  - Armenia (1996), Azerbaijan (1996), Egypt (1995), Georgia (1996), Morocco (1984), Mauritania (1995), Tajikistan (1996), Tunisia (1989).
- SCM pre-treatment predictor match statistics (Table 13, preserving exact numbers for selected cases):
  - Georgia: Secondary school 75.850 vs. Synth. Control A 72.840 and Synth. Control B 92.529; Population growth -0.233 vs. 0.957 and 1.401; Investment share 0.040 vs. 0.124 and 0.115; Pre-treatment GDP 583.067 vs. 678.535 and 594.333.
  - Armenia: Pre-treatment GDP 441.634 vs. Synth. Control A 511.739 and B 594.333.
  - Tajikistan: Pre-treatment GDP 239.526 vs. Synth. Control A 342.553 (noted as a poor match).
  - Tunisia: Pre-treatment GDP 1,089.462 vs. Synth. Control A 1,070.248 and B 1,088.473.
  - Egypt: Pre-treatment GDP 764.775 vs. Synth. Control A 837.891 and B 981.882.
- SCM outcomes (qualitative summary of exact reported patterns):
  - Georgia and Armenia: treated countries fare much better than synthetic controls after liberalization (both broad and region-restricted control sets).
  - Azerbaijan: trade liberalization has not had a positive effect in SCM results.
  - Tajikistan: matching procedure unsatisfactory due to per capita GDP being too low to be matched.
  - Uzbekistan (reverse treatment / missed liberalization): counterfactual constructed from Armenia, Kyrgyz Republic, Tajikistan performed substantially better in per capita GDP, indicating a missed opportunity.
  - Morocco: outperforms synthetic controls; in Case B (restricting controls to Middle East) synthetic control per capita GDP about 40 percent below Morocco’s 10 years after treatment (noting that 32 percent of synthetic control is Algeria, which experienced a costly civil war).
  - Tunisia: evidence similar to Morocco; treated country outperforms synthetic control under some specifications.
  - Mauritania: no clear effect associated with trade liberalization.
  - Egypt: mixed—broad set shows treated below synthetic control once restricting controls; algorithm assigns low weight to poorer nearby African countries (e.g., Sudan) and sometimes picks Syria as main control.

### Main conclusions and implications (as stated)
- Cross-sectional estimates that do not check for common support or that do not control for area-specific unobservables rely on implausible country comparisons and parametric extrapolation beyond the region of common support.
- After restricting samples to ensure common support and controlling for area heterogeneity, matching estimates broadly confirm an openness–growth association found in the literature, with somewhat stronger impacts in some periods (notably 1970s, 1980s, 1990s) and insignificant effects in the 1960s.
- SCM case studies for MCD countries find that trade liberalization, in most treated cases, has had a positive effect on per capita income growth.
- Trade-offs: restricting samples to satisfy common support improves internal validity of estimated effects but reduces external validity (e.g., results should not be extrapolated to developed countries, Africa in 1960s/1970s, and Latin America in the 1990s where common support is lacking).
- Recommended further work (exact wording preserved in intent):
  - Apply the inference procedure proposed by Abadie, Diamond, and Hainmueller (2007) — placebo tests through permutation methods — to assess robustness in MCD countries.
  - Extend the SCM analysis to other regions with relevant liberalization episodes over 1960-2000 (e.g., Latin America, East Asia, Africa).

*Source: _wp07156 - 1.318 coefficient in Table 2 (i.e., the effect of openness aftercontrolling for area dummies).*

### Appendix I. Treated and Control Countries by Region

### Appendix I. Treated and Control Countries by Region

### A. Period 1991–2000

- AFRICA  
  - Treated: Benin, Botswana, Cameroon, Cape Verde, Ethiopia, Gambia,Ghana, Guinea, Ivory Coast, Kenya, Madagascar, Mali, Mauritius, Mozambique, Niger, South Africa, Tanzania, Uganda, Zambia.  
  - Control: Angola, Burkina Faso, Burundi, Central African Republic,Chad, Congo, Gabon, Lesotho, Malawi, Nigeria, Rwanda, Senegal, Togo, Zimbabwe.

- ASIA  
  - Treated: Hong Kong, Indonesia, Malaysia, Nepal, Philippines, Singapore, South Korea, Sri Lanka, Thailand.  
  - Control: China, India, Pakistan, Papua New Guinea.

- LATIN AMERICA  
  - Treated: Argentina, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador, El Salvador, Guatemala, Guyana, Honduras, Jamaica, Mexico, Nicaragua, Panama, Paraguay, Peru, Trinidad & Tobago, Uruguay.  
  - Control: Venezuela.

- MIDDLE EAST  
  - Treated: Egypt, Jordan, Mauritania, Morocco, Tunisia, Yemen.  
  - Control: Algeria, Iran, Syria.

- OECD  
  - Treated: Australia, Austria, Belgium, Canada, Cyprus, Denmark, Finland, France, Germany, Greece, Ireland, Israel, Italy, Japan, Luxembourg, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, Turkey, United Kingdom, United States.  
  - Control: Iceland.

- TRANSITION ECONOMIES  
  - Treated: Albania, Bulgaria, Czech Republic, Hungary, Latvia, Poland, Romania, Slovak Republic, Slovenia.  
  - Control: Belarus, Russia, Ukraine.

### B. Period 1981–90

- AFRICA  
  - Treated: Botswana, Gambia, Ghana, Guinea, Mauritius.  
  - Control: Angola, Benin, Burkina Faso, Burundi, Cameroon, Cape Verde, Central African Republic, Chad, Congo, Ethiopia, Gabon, Guinea-Bissau, Ivory Coast, Kenya, Lesotho, Madagascar, Malawi, Mali, Mozambique, Niger, Nigeria, Rwanda, Senegal, Sierra Leone, South Africa, Tanzania, Togo, Uganda, Zambia, Zimbabwe.

- ASIA  
  - Treated: Hong Kong, Indonesia, Malaysia, Singapore, South Korea, Thailand.  
  - Control: Bangladesh, China, India, Nepal, Pakistan, Papua New Guinea, Philippines, Sri Lanka.

- LATIN AMERICA  
  - Treated: Barbados, Bolivia, Chile, Colombia, Costa Rica, Mexico.  
  - Control: Argentina, Brazil, Dominican Republic, Ecuador, El Salvador, Guatemala, Guyana, Haiti, Honduras, Jamaica, Nicaragua, Panama, Paraguay, Peru, Trinidad & Tobago, Uruguay, Venezuela.

- MIDDLE EAST  
  - Treated: Jordan, Morocco.  
  - Control: Algeria, Egypt, Iran, Mauritania, Syria, Tunisia.

- OECD  
  - Treated: Australia, Austria, Belgium, Canada, Cyprus, Denmark, Finland, France, Germany, Greece, Ireland, Israel, Italy, Japan, Luxembourg, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom, United States.  
  - Control: Iceland, Turkey.

- TRANSITION ECONOMIES  
  - Treated: None.  
  - Control: Hungary, Poland, Romania.

### C. Period 1971–80

- AFRICA  
  - Treated: Mauritius.  
  - Control: Angola, Benin, Botswana, Burkina Faso, Burundi, Cameroon, Cape Verde, Central African Republic, Chad, Congo, Ethiopia, Gabon, Gambia, Ghana, Guinea, Guinea-Bissau, Ivory Coast, Kenya, Lesotho, Madagascar, Malawi, Mali, Mozambique, Niger, Nigeria, Rwanda, Senegal, Sierra Leone, Tanzania, Togo, Uganda, Zambia, Zimbabwe.

- ASIA  
  - Treated: Hong Kong, Indonesia, Malaysia, Singapore, South Korea, Thailand.  
  - Control: Bangladesh, China, India, Nepal, Pakistan, Papua New Guinea, Philippines, Sri Lanka.

- LATIN AMERICA  
  - Treated: Barbados, Bolivia, Chile, Ecuador.  
  - Control: Argentina, Brazil, Colombia, Costa Rica, Dominican Republic, El Salvador, Guatemala, Guyana, Haiti, Honduras, Jamaica, Mexico, Nicaragua, Panama, Paraguay, Peru, Trinidad & Tobago, Uruguay, Venezuela.

- MIDDLE EAST  
  - Treated: Jordan.  
  - Control: Algeria, Egypt, Iran, Mauritania, Morocco, Syria, Tunisia.

- OECD  
  - Treated: Australia, Austria, Belgium, Canada, Cyprus, Denmark, Finland, France, Greece, Ireland, Italy, Japan, Luxembourg, Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom, United States.  
  - Control: Iceland, Israel, New Zealand, Turkey.

- TRANSITION ECONOMIES  
  - Treated: None.  
  - Control: Hungary, Poland, Romania.

### D. Period 1961–70

- AFRICA  
  - Treated: None.  
  - Control: Angola, Benin, Botswana, Burkina Faso, Burundi, Cameroon, Cape Verde, Central African Republic, Chad, Congo, Ethiopia, Gabon, Gambia, Ghana, Guinea, Guinea-Bissau, Ivory Coast, Kenya, Lesotho, Madagascar, Malawi, Mali, Mauritius, Mozambique, Niger, Nigeria, Rwanda, Senegal, Sierra Leone, South Africa, Tanzania, Togo, Uganda, Zambia, Zimbabwe.

- ASIA  
  - Treated: Hong Kong, Malaysia, Singapore, Thailand.  
  - Control: Bangladesh, China, India, Indonesia, Nepal, Pakistan, Papua New Guinea, Philippines, South Korea, Sri Lanka.

- LATIN AMERICA  
  - Treated: Barbados, Bolivia, Ecuador, Jamaica, Peru.  
  - Control: Argentina, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, El Salvador, Guatemala, Guyana, Haiti, Honduras, Mexico, Nicaragua, Panama, Paraguay, Trinidad & Tobago, Uruguay, Venezuela.

- MIDDLE EAST  
  - Treated: Jordan.  
  - Control: Algeria, Egypt, Iran, Mauritania, Morocco, Syria, Tunisia.

- OECD  
  - Treated: Australia, Austria, Belgium, Canada, Cyprus, Denmark, Finland, France, Greece, Ireland, Italy, Japan, Luxembourg, Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom, United States.  
  - Control: Iceland, Israel, New Zealand, Turkey.

- TRANSITION ECONOMIES  
  - Treated: None.  
  - Control: Romania.

*Source: Appendix I. Treated and Control Countries by Region (content unit)._wp07156 - Appendix I. Treated and Control Countries by Region*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2007/_wp07156.pdf_