## _wp16117 - Section V concludes.

## Source details

**Canonical URL:** [_wp16117 - Section V concludes.](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16117.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16117.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16117.pdf.json)

---

### II. Literature review and motivation
- Academic approaches:
  - Ex-post assessment using standard regressions on historical data.
  - Ex-ante predictions using computable general equilibrium (CGE) methods based on projections.
- Key points from ex-post literature:
  - Gravity equation explains cross-sectional variation in trade flows by incomes, bilateral distance, and dummies for common language, land borders, and trade agreements.
  - Early ex-post studies show no clear evidence that trade agreements boost trade (Tinbergen, 1962; Bergstrand, 1985).
  - Trade agreement dummy coefficients can be “fragile” (Ghosh and Yamarik, 2004).
  - Trade agreements are endogenous; countries self-select into agreements (Magee, 2003; Baier and Bergstrand, 2004).
  - Methods to address endogeneity: instrumental variables, nonparametric matching, Probit models.
  - Baier and Bergstrand (2009, 2007) find trade agreements increase trade by 100 percentage points after ten years when accounting for endogeneity.
- Trade diversion versus trade creation:
  - Early theories emphasized trade diversion (Viner 1950; Lipsey 1960).
  - Economies with significant pre-agreement trade are “natural trading blocs”; agreements likely lead to more trade creation than diversion (Frankel, Stein, and Wei 1995).
  - Empirical evidence of trade diversion exists (Carrere, 2006).
- Ex-ante (CGE) limitations:
  - CGE approaches face considerable uncertainty, especially quantifying NTBs.
  - Observed trade increases often outstrip CGE predictions (Kehoe, 2003; Corcos et al., 2012).
- Motivation for this paper:
  - Renewed interest from Trans-Pacific Partnership (TPP).
  - Need to complement CGE ex-ante studies with ex-post evidence that addresses endogeneity.
  - Contribution: employ the synthetic control method (SCM) across many trade agreements; SCM allows unobserved confounders to vary with time.

### III. Empirical strategy and SCM methodology
- SCM overview and advantages:
  - Designed for comparative case studies with few treated aggregate units.
  - Constructs synthetic control as a weighted average of donor units based on pre-intervention covariates and outcomes.
  - Allows time-varying unobserved confounders, unlike standard difference-in-differences.
- SCM limitations and inference:
  - Standard statistical inference inappropriate; use placebo/falsification tests.
  - Requires no other events affecting only the treated area; donor units unaffected by the intervention.
  - Requires good pre-intervention fit and a donor pool that can approximate the treated unit.
- Key formal elements (notation and setup presented):
  - Units i = 1,...,J+1 with treated unit i = 1; pre-intervention periods t = 1,...,T0.
  - Observed outcome: Y_it = Y_it^N + α_it D_it, where D_it is the indicator.
  - Counterfactual factor model: Y_it^N = δ_t + X_iβ + Z_iξ_t + ε_it (expressed in source notation).
  - Synthetic control weight vector W = (w2,...,wJ+1)′ with wj ≥ 0 and Σwj = 1.
  - Optimal W* chosen to match pre-intervention outcomes/covariates; treatment effect estimator α̂_t = Y_1t - Σ_{j=2}^{J+1} w_j^* Y_jt for t > T0.
  - Inference via placebo/permutation tests; sensitivity to concurrent events and pre-treatment fit emphasized.

### III.B Application to bilateral trade pairs (data and implementation)
- Sample and treatment definition:
  - Main period: 1983-1995 (robustness check: 1973-2001).
  - Data from Head, Mayer and Ries (2010); 104 bilateral country pairs in main sample, yielding two treated pairs per agreement (A→B and B→A).
  - Pair types among 104 pairs: 30 EM-AM, 30 AM-EM, 26 AM-AM, 18 EM-EM.
- Donor pool construction:
  - Excludes pairs that had a trade agreement in the same year and excludes all other trade agreements that the exporting country has within the entire sample.
  - If no synthetic found, alternative donor pool excludes only agreements with the same partner.
- Covariates used (gravity-controls and dummies):
  - Distance; GDP of each country; GDP per capita; Population; Bilateral Real Exchange Rate; Remoteness of each country (proxy for multilateral trade resistance).
  - Dummy variables: Colonial history; Col to; Col from; Contiguity (Contig); Common legal origins (Comleg); Common currency (Comcur); Common language.
  - Exports, lagged by three years.
  - Remoteness defined conceptually as rem_i = Σ_k (GDP_k / GDP_world) * distance_ik in the source.
- Data sources and sample size:
  - Gravity dataset: Head, Mayer and Ries (2010) with IMF DOTS trade data.
  - Trade agreements: Baier and Bergstrand (2007), WTO, Frenkel (1997).
  - Additional controls: World Bank Development Indicators, Angus Maddison dataset, CEPII distance database, World Development Indicators for bilateral real exchange rate.
  - Entire sample contains 38124 observations.
- Illustrative example:
  - Belgium exports to Spain (treated) with EC(12) enlargement in 1986.
  - Synthetic control shows close pre-treatment fit and divergence after treatment attributed to Belgium export gains vis-à-vis Spain; anticipatory divergence one year prior noted.

### IV. Results — trade creation, diversion, and robustness
- Pre-intervention fit:
  - Average NRMSD across 104 pairs is 0.56; median NRMSD is 0.31.
- Main trade-creation findings:
  - Average cumulative gross export gain of treated countries relative to synthetic counterparts: 80 percentage points over the next ten years.
  - This translates into annual average export growth of 3.8 percentage points higher than the average export growth in the synthetic.
  - Distribution across pairs: of 104 pairs, 80 had treated > synthetic after ten years, while 24 had treated ≤ synthetic.
  - Income-group heterogeneity:
    - EM-AM pairs: 93 percentage points higher over ten years relative to synthetic; around 3.9 percentage points higher annual export growth.
    - AM-AM pairs also show high gains; AM-EM pairs show lower gains relative to the average.
  - Specific agreement examples:
    - 1986 EC Enlargement: 95 percentage points higher export growth in 10 years; average annual export growth 3.7 percentage points higher.
    - NAFTA: 79 percentage points higher than synthetic over ten years; average annual export growth 4.3 percentage points higher.
    - US due to NAFTA: ~50 percentage points higher over ten years; annual average export growth 3.0 percentage points higher than synthetic counterpart.
- Robustness and alternative calculations:
  - Measuring gains from first divergence (anticipation effects) rather than t=0 yields higher estimates:
    - Average for 104 pairs: 132 percentage points higher than average synthetic.
    - NAFTA = 128 percentage points higher.
    - 1986 EC Enlargement = 234 percentage points higher.
  - Broader sample robustness (216 country pairs, 1973-2001) yields average export gains due to trade agreements of 90 percentage points.
  - Note in source: "3.8 percentage points. The results are thus stronger than using the shorter sample."
- Placebo (falsification) tests and inference:
  - Permutation-style inference: randomly choose ten treated units; for each, randomly select five country pairs where exporting country A exports to a country without an agreement with A, yielding fifty placebo units.
  - Results summary:
    - Of ten randomly chosen treated units, eight cases where treated > synthetic and two cases where treated < synthetic.
    - Of the eight treated>synthetic cases, seven cases where treated-relative-to-synthetic > placebo-relative-to-synthetic.
    - One placebo-test failure where treated>synthetic but treated-relative-to-synthetic < placebo-relative-to-synthetic.
  - Conclusion: placebo tests largely support SCM inference.
- Trade diversion — import and export diversion findings:
  - Import diversion approach: for each agreement between A and B, apply SCM to A’s trade with top partner C outside the agreement to construct counterfactual.
  - Import-diversion sample: 104 country pairs yield 43 country pairs for import-diversion analysis: 31 AM-AM and 12 EM-AM.
  - Import-diversion results:
    - Synthetic unit slightly higher than treated unit; synthetic is 20 percentage points higher than treated after ten years (cumulative), suggesting some import diversion.
    - Import diversion apparent in AM-AM pairs; EM-AM pairs show less import diversion.
    - Caveat: pre-intervention fit weaker in import-diversion analyses.
  - Export diversion results:
    - No evidence of export diversion; treated unit is higher than synthetic.
    - Exports to third parties increase:
      - Export growth to third parties is 8 percentage points higher over ten years due to the trade agreement.
      - For EM-AM exporters, export gains are around 19 percentage points over ten years; average annual export growth for these EM-AM exporters is 0.9 percentage points higher.
    - Overall: import diversion magnitude (~20 percentage points over ten years) is small relative to trade creation gains and would be more than offset by trade creation.

### Methodological advantages, limitations, and future research directions
- Advantages of SCM:
  - Allows unobserved confounders to vary with time.
  - Data-driven synthetic control construction is intuitive and appealing.
- Limitations and caveats:
  - Standard statistical inference inappropriate; rely on placebo/permutation tests.
  - Requires no other contemporaneous interventions; results can be contaminated by other non-trade policy changes.
  - Assumes intervention affects outcomes only in intervention area; may not always hold.
  - Fit quality matters—poorer pre-intervention fit reduces confidence (particularly in import-diversion analyses).
- Suggested future research using SCM (from source):
  - Expand to sectoral responses to trade agreements.
  - Extend trade-diversion analysis to more country pairs and multiple sectors to identify which sectors suffer most when excluded.
  - Study impact on formation of global value chains.
  - Apply SCM to other outcome variables beyond exports (welfare, labor adjustments, inequality, costs of liberalization).

### Key statistics (preserved exactly as in source)
- 3.8 percentage points.
- 100 percentage points (Baier and Bergstrand ten-year estimate).
- 80 percentage points (average cumulative export gain over ten years).
- 3.8 percentage points (annual average export growth higher).
- NAFTA: 79 percentage points higher exports over ten years.
- NAFTA: 4.3 percentage points average export growth.
- Placebo tests: ten treated units; eight cases treated>synthetic; seven cases where treated-relative-to-synthetic > placebo-relative-to-synthetic; one placebo failure.
- Import-diversion sample: 104 country pairs results in 43 country pairs: 31 AM-AM and 12 EM-AM.
- Synthetic unit is 20 percentage points higher than treated after ten years (import diversion, cumulative).
- Export diversion: export growth to third parties is 8 percentage points higher over ten years.
- EM-AM export gains: around 19 percentage points over ten years; average annual export growth 0.9 percentage points higher.

*Source: _wp16117 - Section V concludes.*

### Section V concludes.

### _wp16117 - Section V concludes.

### II. LITERATURE REVIEW AND MOTIVATION
- Academic literature on the impact of trade agreements is grouped into:
  - ex-post assessment using standard regressions on historical data;
  - ex-ante predictions using computable general equilibrium (CGE) methods based on projections.
- Ex-post studies:
  - The gravity equation has been the empirical workhorse in international trade for ~40 years, explaining cross-sectional variation in trade flows by incomes, bilateral distance, and dummies for common language, land borders, and trade agreements (Baier and Bergstrand (2007); UNCTAD/WTO (2012)).
  - Earlier ex-post studies show no clear evidence that trade agreements boost trade (Tinbergen, 1962; Bergstrand, 1985).
  - Ghosh and Yamarik (2004) find trade agreement dummy coefficients are “fragile”.
  - Trade agreements are endogenous: countries self-select into agreements for reasons correlated with trade (Magee, 2003; Baier and Bergstrand, 2004).
  - Methods to address endogeneity include instrumental variables (Baier and Bergstrand (2007); Magee (2003)), nonparametric matching (Baier and Bergstrand (2009)), and Probit models (Mansfield and Reinhardt, 2003 and 2008; Baier and Bergstrand, 2004).
  - Baier and Bergstrand (2009, 2007) find trade agreements increase trade by 100 percent after ten years when accounting for endogeneity.
- Trade diversion versus trade creation:
  - Early theories emphasized trade diversion (Viner 1950; Lipsey 1960).
  - Economies with significant pre-agreement trade are “natural trading blocs”; agreements likely lead to more trade creation than diversion (Frankel, Stein, and Wei 1995).
  - Empirical evidence of trade diversion exists (Carrere, 2006).
- Ex-ante studies:
  - Partial-equilibrium (PE) and general-equilibrium (GE/CGE) models simulate baseline versus policy-change scenarios to estimate impacts (UNCTAD/WTO (2012)).
  - CGE approaches face considerable uncertainty due to difficulty quantifying many channels, especially NTBs (Petri and Plummer, 2016).
  - Kehoe (2003) finds observed NAFTA trade increases in many sectors outstripped three ex-ante CGE predictions by more than ten-fold; Corcos et al. (2012) find similar results for the EU.
- Motivation:
  - Renewed interest generated by the Trans-Pacific Partnership (TPP).
  - Need to complement CGE ex-ante studies with ex-post evidence addressing endogeneity.
  - This paper contributes by employing the synthetic control method (SCM) across a large number of trade agreements to ask: do trade agreements matter?
  - SCM advantage: allows unobserved confounders to vary with time; first paper to employ SCM across many trade agreements for this question.

### III. EMPIRICAL STRATEGY
- SCM overview:
  - SCM is designed for comparative case studies when few aggregate units are treated (Abadie, Diamond and Hainmueller (2010)).
  - SCM constructs a synthetic control as a weighted average of donor units based on similarity in pre-intervention covariates and outcomes.
  - SCM allows time-varying unobserved confounders, unlike standard difference-in-differences.
  - Popular applications include policy interventions, natural disasters, reunification, inflation targeting (Abadie, Diamond and Hainmueller (2014); Lee (2011); Craig (2015)).
  - Limitations of SCM (Craig (2015)):
    - Standard statistical inference is inappropriate; placebo/falsification tests used instead.
    - Requires no other events affecting only the treated area.
    - Assumes donor units are unaffected by the intervention.
    - Requires good pre-intervention fit between treated and synthetic.
    - Requires donor pool contains weighted combinations similar to treated unit; extremes may be problematic.
- SCM technical setup (following Abadie, Diamond and Hainmueller (2010)):
  - Notation and assumptions:
    - ܻ௜௧ே is outcome without intervention for unit i at time t; J+1 regions with first exposed.
    - T0 is number of pre-intervention periods.
    - Observed outcome ܻ௜௧ = ܻ௜௧ே + ܙ௜௧ܦ௜௧ where ܙ௜௧ is effect when exposed and ܦ is indicator.
    - Aim to estimate ߙ (treatment effects) for t > T0 via ܻ௜௧ே counterfactual.
  - Factor model for counterfactual:
    - ܻ௜௧ே = ߜ௧ߠ + ܼ௜ߣ௧ + ߤ௜ߝ௜௧ (equation (3)): common factors, observed covariates ܼ, unobserved common factors ߣ, loadings ߤ, transitory shocks ߝ.
  - Synthetic control weights W = (w2,...,wJ+1)′ satisfy wj ≥ 0 and sum wj = 1.
  - Synthetic outcome constructed as weighted sum of control outcomes (equations (5)-(8)).
  - Weights chosen to minimize pseudo-distance between treated pre-intervention characteristics ܺ௵ and weighted controls ܺ௴.
  - Inference via placebo/falsification tests; SCM sensitive to other concurrent events and pre-treatment fit.

### III.A SCM METHODOLOGY — Key formal elements (as presented)
- Model assumptions and notation:
  - Units i = 1,...,J+1; treated unit is i = 1.
  - Pre-intervention periods t = 1,...,T0.
  - Observed outcome: ܻ௜௧ = ܻ௜௧ே + ߙ௜௧ܦ௜௧ with ܦ௜௧ indicator.
  - Counterfactual factor model: ܻ௜௧ே = ߜ௧ߠ + ܼ௜ߣ௧ + ߤ௜ߝ௜௧.
- Synthetic control construction:
  - Weight vector W = (w2,...,wJ+1)′ with nonnegative weights summing to 1.
  - Optimal W* chosen to closely match pre-intervention outcomes and covariates (equation (6) approximated in practice).
  - Estimator of treatment effect: ߙൌܻ௵ - Σ wj* ܻj,t for t > T0 (equation (7)).
  - Minimization of ||X1 - X0W|| with restrictions wj ≥ 0 and Σwj = 1 (equation (8)).

### III.B APPLICATION OF SCM TO BILATERAL TRADE PAIRS
- Sample and treatment definition:
  - Paper examines regional trade agreements in period 1983-1995 (robustness check: broader sample 1973-2001).
  - Trade agreement data from Head, Mayer and Ries (2010).
  - Each bilateral agreement yields two treated pairs (A→B and B→A), yielding 104 bilateral country pairs in main sample.
  - Pair types among 104 pairs: 30 EM-AM, 30 AM-EM, 26 AM-AM, 18 EM-EM.
- Donor pool construction:
  - Donor pool excludes pairs that had a trade agreement in the same year and excludes all other trade agreements that the exporting country has within the entire sample.
  - If no synthetic found, alternative donor pool excludes only agreements with the same partner.
- Covariates for SCM drawn from gravity model literature:
  - Covariates used:
    - Distance between bilateral pairs
    - GDP of each country in the bilateral pair
    - GDP per capita of each country in the bilateral pair
    - Population of each country in the bilateral pair
    - Bilateral Real Exchange Rate
    - Remoteness of each country (proxy for multilateral trade resistance)
    - Dummy variables:
      - Colonial history = 1 if pair ever in colonial relationship
      - Col to = 1 if export from hegemon to colony
      - Col from = 1 if export from colony to hegemon
      - Contig = 1 for contiguity
      - Comleg = 1 for common legal origins
      - Comcur = 1 for common currency
      - Common language = 1 for common official language
    - Exports, lagged by three years
  - Gravity equation representation for exports x_ijt: x_ijt = µ_t F_{it}^{α} F_{jt}^{β} φ_{ijt} (equation (9) notation).
  - Remoteness (MTR proxy) defined as: rem_i = Σ_k (GDP_k / GDP_world) * distance_ik (formula (10) conceptual).
- Data sources:
  - Underlying gravity dataset: Head, Mayer and Ries (2010) (trade data from IMF DOTS).
  - Trade agreements from Baier and Bergstrand (2007), WTO, Frenkel (1997).
  - Gravity controls: World Bank Development Indicators, national sources, Angus Maddison dataset.
  - Legal origins, currency: Shleifer/Glick and Glick and Rose (2002).
  - Distances and common language: CEPII distance database.
  - Bilateral real exchange rate: World Development Indicators.
  - Entire sample contains 38124 observations.

### III.B Illustrative example
- Treated unit: Belgium exports to Spain (B-S); treatment: EC(12) enlargement in 1986.
- Synthetic B-S constructed as weighted average of control bilateral pairs that did not experience the same agreement.
- Figure 2 (described): close pre-treatment fit; divergence after treatment attributed to Belgium export gains vis-à-vis Spain; anticipatory divergence one year prior noted.

### IV. RESULTS — Overview and Trade Creation
- Empirical setup for outcome plotting:
  - Outcome: gross exports plotted against time to trade agreement with t=0 at enactment; t=-1,...,-10 pre-intervention; t=+1,...,+10 post-intervention.
  - Treated series = actual gross exports; synthetic series = SCM counterpart.
  - Fit assessed via normalized root-mean-square deviation (NRMSD) computed over ten pre-intervention years: NRMSD = RMSD / average exports over ten pre-intervention years.
- Pre-intervention fit:
  - Average NRMSD across 104 pairs is 0.56; median NRMSD is 0.31, suggesting generally good pre-intervention fit.
- Main trade-creation findings:
  - Average gross exports of treated countries with trade agreement was 80 percentage points higher over the next ten years on a cumulative basis compared to average synthetic counterparts (figure 6).
  - This translates into annual average export growth of 3.8 percentage points higher than the average export growth in the synthetic (figure 7).
  - Distribution across pairs:
    - Of 104 pairs, 80 had treated > synthetic after ten years, while 24 had treated ≤ synthetic.
  - Income-group heterogeneity:
    - Highest cumulative gains: EM-AM pairs with 93 percentage points increase over ten years relative to synthetic, translating to around 3.9 percentage points higher annual export growth.
    - AM-AM pairs also show high gains.
    - Lower gains for AM-EM pairs (advanced exporters to emerging market importers) relative to the average.
  - Specific agreements:
    - 1986 EC Enlargement: 95 percentage points higher export growth in 10 years; average annual export growth 3.7 percentage points higher.
    - NAFTA: 79 percentage points higher than synthetic over ten years; average annual export growth 4.3 percentage points higher.
    - US due to NAFTA: ~50 percentage points higher over ten years; annual average export growth 3.0 percentage points higher than synthetic counterpart (US exports to Canada and Mexico; figure 8).
- Robustness and alternate calculations:
  - If gains are measured from first divergence (anticipation effects) rather than t=0 through ten years post-agreement, average export growth for 104 pairs is 132 percentage points higher than average synthetic; NAFTA = 128 percentage points higher; 1986 EC Enlargement = 234 percentage points higher.
  - Robustness check with broader sample (216 country pairs, 1973-2001) yields average export gains due to trade agreements of 90 percentage points (note: longer sample not baseline due to data gaps).

*Source: _wp16117 - Section V concludes.*

### 3.8 percentage points. The results are thus stronger than using the shorter sample.

### _wp16117 - 3.8 percentage points. The results are thus stronger than using the shorter sample.

### Major findings on trade agreement impacts
- Trade agreements can substantially boost exports:
  - Average cumulative export gain of 80 percentage points over ten years.
  - Annual average export growth 3.8 percentage points higher.
- Gains are higher for emerging markets in agreements with advanced markets.
- NAFTA-specific results:
  - All bilateral country pairs in NAFTA show increased exports relative to synthetic controls.
  - Exports 79 percentage points higher over ten years.
  - Average export growth 4.3 percentage points.
- Comparison to prior literature:
  - Baier and Bergstrand (2007, 2009) find trade gains 100 percentage points higher over ten years (using their methodology).
  - When export gains are calculated using t=0 results are lower; results are higher when anticipation effects are accounted for.
  - Krueger (1999) finds impact of NAFTA over its first three years around a 3 percent increase in trade and not statistically significant.
- Differentiation across income groups:
  - Trade gains may depend on income group pairing (AM-AM vs EM-AM).
  - Possible reasons: heterogeneity in depth/nature of agreements; exposure of emerging markets to large advanced economy markets.

### Placebo (Falsification) tests and inference
- Methodology:
  - Follow permutation-style inference as in Abadie, Diamond and Hainmueller (2010).
  - Procedure: randomly choose ten treated units; for each, randomly select five country pairs where exporting country A exports to a country without an agreement with A, yielding fifty placebo units. Apply SCM to placebos and compare treated-relative-to-synthetic to placebo-relative-to-synthetic.
- Results summary (Figure 9 narration):
  - Of ten randomly chosen treated units, eight cases where treated > synthetic and two cases where treated < synthetic.
  - Of the eight treated>synthetic cases, seven cases where treated-relative-to-synthetic > placebo-relative-to-synthetic.
  - One placebo-test failure (chart in grey): treated>synthetic but treated-relative-to-synthetic < placebo-relative-to-synthetic.
  - Conclusion: placebo tests largely support the SCM inference; inferential technique considered strong given these results.

### Trade diversion: approach and key results
- Approach:
  - For each country with a trade agreement (A and B), apply SCM to A’s trade with its top trading partner C outside the agreement to construct a counterfactual of trade with C absent the agreement with B.
  - Control group excludes the same country pairs as in trade creation analysis; in some cases control group extended due to data constraints.
- Sample composition for import diversion exercise:
  - 104 country pairs considered yield 43 country pairs for the import-diversion analysis: 31 AM-AM and 12 EM-AM.
- Import diversion (Figure 10, Figures 11–12):
  - Synthetic unit slightly higher than treated unit, suggesting imports from third countries would have been slightly higher if no trade agreement existed.
  - Synthetic unit is 20 percentage points higher than the treated unit after ten years (cumulative basis).
  - By income groups: import diversion apparent in AM-AM pairs; EM-AM pairs show positive results (i.e., less import diversion).
  - Caveat: pre-intervention fit between treated and synthetic is not as good as in trade creation/export diversion analyses; interpret with caution.
- Export diversion (Figures 13–15):
  - Treated unit is higher than synthetic counterpart — no evidence of export diversion.
  - Exports to third parties not in the agreement increase:
    - Export growth to third parties is 8 percentage points higher over ten years due to the trade agreement.
    - For emerging markets exporting to advanced economies, export gains are around 19 percentage points over ten years.
    - Average annual export growth for these EM-AM exporters is 0.9 percentage points higher.
  - Overall: import diversion magnitude (~20 percentage points over ten years) is small relative to trade creation gains and would be more than offset by trade creation.

### Methodological advantages and limitations
- Advantages of SCM:
  - Allows unobserved confounders to vary with time (unlike methods that only allow time-invariant unobservables).
  - Data-driven construction of a synthetic control group is intuitive and appealing.
- Limitations and assumptions to bear in mind:
  - Standard statistical inference is inappropriate; placebo/permutation tests used for inference.
  - Assumes no other contemporaneous interventions that affect outcomes; results could be contaminated by other non-trade policy changes.
  - Assumes the intervention affects outcomes only in the intervention area—may not always hold.
  - Fit quality matters: weaker pre-intervention fit reduces confidence (noted especially for import-diversion analyses).

### Policy implications and directions for future research
- Policy relevance:
  - After accounting for endogeneity, trade agreements can meaningfully increase trade — relevant amid observed trade slowdown.
  - Results support the view that trade agreements “matter” for trade flows, particularly for emerging markets engaging with advanced economies.
- Suggested future research using SCM:
  - Expand analysis to sectoral responses to trade agreements.
  - Extend trade-diversion analysis to more country pairs and multiple sectors to identify which sectors suffer most when excluded.
  - Study the impact of trade agreements on formation of global value chains.
  - Apply SCM to other outcome variables beyond exports (welfare, labor adjustments, inequality, costs of liberalization).

### Key statistics (preserved exactly as in source)
- 3.8 percentage points.
- 100 percentage points (Baier and Bergstrand ten-year estimate).
- 80 percentage points (average cumulative export gain over ten years).
- 3.8 percentage points (annual average export growth higher).
- NAFTA: 79 percentage points higher exports over ten years.
- NAFTA: 4.3 percentage points average export growth.
- Placebo tests: ten treated units; eight cases treated>synthetic; seven cases where treated-relative-to-synthetic > placebo-relative-to-synthetic; one placebo failure.
- Import-diversion sample: 104 country pairs results in 43 country pairs: 31 AM-AM and 12 EM-AM.
- Synthetic unit is 20 percentage points higher than treated after ten years (import diversion, cumulative).
- Export diversion: export growth to third parties is 8 percentage points higher over ten years.
- EM-AM export gains: around 19 percentage points over ten years; average annual export growth 0.9 percentage points higher.

*Source: _wp16117 - 3.8 percentage points. The results are thus stronger than using the shorter sample.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16117.pdf_