## Appendix IV. Supplementary graphs and tables (wpiea2021176-print-pdf)

## Source details

**Canonical URL:** [Appendix IV. Supplementary graphs and tables (wpiea2021176-print-pdf)](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021176-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021176-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021176-print-pdf.pdf.json)

---

### Introduction — scope and objectives
- Goal: assess trade losses from financial crises from a sectoral perspective for emerging countries.
- Sample: 41 emerging countries over the period 1980–2018 with 38 debt crises, 34 banking crises, and 36 currency crises.
- Novelty: disaggregation by traded category (agricultural, mining, manufactured goods, and services) and causal identification combining local projections and impact assessment with the Augmented Inverse Propensity Weighted estimator.

### Key aggregate findings
- Aggregate five-year cumulative effects on exports (in percentage points of pre-crisis GDP):
  - Debt crises: 5.6 percentage points (pp.)
  - Banking crises: 8.9 pp.
  - Currency crises: 7.7 pp.
- Aggregate five-year cumulative effects on imports (in percentage points of pre-crisis GDP):
  - Debt crises: 11 pp.
  - Banking crises: 14 pp.
  - Currency crises: 9.1 pp.
- Trade balance:
  - Debt and banking crises increase the trade balance for all years in the aftermath by more than 5 pp. of pre-crisis GDP, reflecting larger contractions in imports than exports.
  - Currency crises increase the trade balance for years 1–4 but are not found to significantly affect the trade balance in year 5.

### Sectoral disaggregation — main patterns and key statistics
- Manufacturing
  - Most affected by financial crises; largest contributor to aggregate trade losses.
  - Five-year cumulative losses on manufacturing exports:
    - 6.1 pp. of pre-crisis GDP for debt crises.
    - 4 pp. for currency crises.
    - 2.3 pp. for banking crises.
  - Five-year cumulative losses on manufacturing imports:
    - 4.6 pp. for currency crises.
    - 5.4 pp. for debt crises.
    - 7 pp. for banking crises.
  - Contributions to total export collapse (manufacturing):
    - 107 percent for debt crises.
    - 51 percent for currency crises.
    - 26 percent for banking crises.
  - Manufacturing accounts for around 50 percent of the total collapse of imports for all types of crises.

- Services
  - Significant declines, particularly after banking crises.
  - Five-year cumulative export losses:
    - 3.5 pp. of pre-crisis GDP following banking crises (contribution of 39 percent of the total export collapse).
    - 2.9 pp. following debt crises.
    - 0.9 pp. following currency crises.
  - Five-year cumulative import losses:
    - 2.6 pp. following debt and banking crises.
    - 0.9 pp. following currency crises.

- Mining
  - Trade declines following several types of crises, notably after banking crises.
  - Five years after debt crises, exports of mining goods increase by 1.1 pp. of pre-crisis GDP.
  - After banking and currency crises, exports of mining goods collapse by 1.6 pp. and 2.9 pp., respectively.
  - Imports of mining goods over five years are reduced by:
    - 1.9 pp. for debt crises.
    - 3 pp. for both banking and currency crises.

- Agriculture
  - Least affected; evidence of substitution effect, particularly following debt crises.
  - Agricultural exports: left unchanged after currency crises; decrease by 1.5 pp. over five years after banking crises.
  - Agricultural imports: significantly reduced by between 0.6 and 1.5 pp. for all types of crises.

- Descriptive raw averages (non-counterfactual, year t−1 to t+5, % of pre-crisis GDP):
  - Exports and imports decline respectively by between 11 and 14 percentage points of pre-crisis GDP following all crisis types.

### Identified transmission channels (structural, demand, supply)
- Structural channel
  - Financial crises hurt manufacturing exports more in countries with a relatively lower share of manufacturing exports.
  - Diversification reduces trade losses: countries with diversified export products and trading partners experience significantly lower trade losses.
  - Key channel proxies: share of manufactured exports; export diversification index; trading partners’ diversification index.

- Demand channel
  - Crises that lower demand from trading partners have larger adverse trade impacts, especially for manufacturing trade.
  - Agricultural and mining goods and services exhibit lower income elasticity of demand compared with manufactured goods.
  - Proxy: trading partners’ growth over five years following crises.

- Supply channel
  - Deteriorated domestic and external financial conditions and sudden stops exert significant detrimental effects on trade, mainly through manufacturing and services.
  - Crises associated with large reductions in gross capital inflows and deterioration in investors’ credit rating risks trigger larger trade declines.
  - Proxies: financial development evolution, gross capital inflows, investors’ credit rating risks.

### Methodology — identification and estimation
- Combined approach: local projections (LP) à la Jordà (2005) + impact assessment with the Augmented Inverse Propensity Weighted (AIPW) estimator.
- Local projections setup:
  - Horizon h ∈ ⟦0;5⟧.
  - Dependent variable: Δy_{i,t+h}^k = (y_{i,t+h}^k − y_{i,t−1}^k)/GDP_{t−1} ×100, cumulative change between t−1 and t+h in 100 times trade flows of variable k scaled by pre-crisis GDP.
  - Crisis dummies: D_{i,t}^d, D_{i,t}^b, D_{i,t}^c equal to 1 at crisis start; non-starting crisis years are dropped.
  - Includes lagged changes ∆y_{i,t−1}^k and ∆y_{i,t−2}^k, controls X_{i,t+h}^x, and estimation for each horizon h.
- AIPW estimator and treatment model:
  - ATE: Λ_h = E[y_{i,t+h}(1) − y_{i,t−1} | D_{i,t}=1] − E[y_{i,t+h}(0) − y_{i,t−1} | D_{i,t}=0].
  - Independence assumption conditional on covariates Z_{i,t}.
  - Treatment model: covariate balancing propensity score (CBPS) p̂_{i,t} estimated on Z_{i,t}; CBPS ensures balance and reduces misspecification bias.
  - Outcome model: LP eq. (1) estimated separately for treated and control groups to predict potential outcomes.
  - Final estimator: Λ̂_AIPW_h, doubly robust semi-parametric AIPW; maximum inverse propensity weight truncated at 10 (robustness: 5 and 20).

### Data sources and controls
- Crises definitions and sources:
  - Debt crises: Reinhart and Rogoff (2009).
  - Banking crises: Laeven and Valencia (2018).
  - Currency crises: Frankel and Rose (1996) (nominal depreciation ≥ 25 percent and at least a 10 percent increase in the rate of depreciation).
- Trade data:
  - Goods: UN Comtrade via WITS–World Bank at 3-digit SITC, aggregated into agricultural, mining, and manufactured goods per WTO classification.
  - Services: UNCTAD.
  - Total trade: aggregation of agriculture, mining, manufacturing, and services; deflated by export/import prices.
- Treatment-model covariates (lagged one year) Z_{i,t} used to estimate propensity scores:
  - (i) number of past-5 years financial crises except the one of interest
  - (ii) intensity of conflicts
  - (iii) log of Real GDP
  - (iv) log of public debt to GDP
  - (v) log of domestic credit to GDP
  - (vi) log of liquid liabilities to GDP
  - (vii) net capital inflows to GDP
  - (viii) log of foreign reserves to GDP
  - (ix) current account to GDP
  - (x) financial openness index
  - (xi) log of trade openness to GDP
  - (xii) terms of trade growth
  - (xiii) floating exchange rate regime dummy
  - (xiv) government accountability index
  - (xv) corruption index
  - (xvi) trading partners’ growth
  - (xvii) US interest rate on government’s securities
- Outcome-model controls X_{i,t}: (i) change of trade flows one and two years before crisis onset, (ii) other crises, and (iii) exchange rate regime at horizon h.

### Data patterns and summary statistics (selected exact values)
- Appendix II summary statistics (Obs. 766 unless noted):
  - Cumulative change of total exports over 5 years (% of pre-crisis GDP): Mean 19.690; Sd 20.795; Min -18.277; Max 124.255.
  - Cumulative change of manufacturing exports over 5 years (% of pre-crisis GDP): Mean 8.362; Sd 13.009; Min -11.702; Max 94.118.
  - Cumulative change of services exports over 5 years (% of pre-crisis GDP): Mean 5.392; Sd 7.282; Min -9.869; Max 49.547.
  - Cumulative change of imports over 5 years (% of pre-crisis GDP): Mean 23.338; Sd 27.805; Min -20.333; Max 179.063.
  - Share of manufacturing exports (% of total exports): Mean 34.545; Sd 21.392; Min 0.306; Max 86.902.
  - Share of manufacturing imports (% of total imports): Mean 55.141; Sd 10.704; Min 26.701; Max 80.015.
  - 1 if Debt crises at start: Mean 0.052; Sd 0.223; Min 0.000; Max 1.000.
  - 1 if Banking crises at start: Mean 0.048; Sd 0.215; Min 0.000; Max 1.000.
  - 1 if Currency crises at start: Mean 0.050; Sd 0.217; Min 0.000; Max 1.000.
  - Observations for some variables: Log. of public debt to GDP Obs. 765; Net capital inflows to GDP Obs. 745; Exports diversification index Obs. 728.

### Robustness and diagnostic notes (selected)
- Propensity-score diagnostics:
  - CBPS shows high classification power and considerable overlap between treated and control propensity score distributions; weighting by CBPS eliminates covariate differences per Rubin (2002) criteria more effectively than a pooled probit.
- Weight truncation and sensitivity:
  - Benchmark truncation of maximum inverse propensity weight to 10; robustness checks with maximum weights of 5 and 20.
- Estimator properties:
  - AIPW is doubly robust: unbiased if either treatment or outcome model is correctly specified; includes adjustment term that stabilizes estimates when propensity scores approach 0 or 1.
- Balance and overlap:
  - CBPS model eliminates differences in characteristics between treated and control groups; pooled probit is less effective.
  - Rubin (2002) cut-offs applied: absolute standardized difference 0.25; variance ratio thresholds 0.5–2.0.

### Robustness and alternative specifications — highlights
- Combined vs. individual crises:
  - Combined crises (another crisis within t−2 to t+2) trigger more significant aggregated and sectoral trade losses, except for imports following debt crises.
  - Individual crises yield lower total and sectoral trade losses (more so for exports than imports), except for imports following debt crises.
- Excluding post-GFC period (drop 2008 onwards):
  - Comparable results for total and sectoral export losses from banking and currency crises; losses from debt crises are significantly reduced.
  - Trade losses on imports present a lower magnitude when the post-GFC period is excluded.
- Alternative crisis sources and definitions:
  - Using Cruces and Trebesch (2013) for debt crises and Reinhart and Rogoff (2009) for banking and currency crises yields 41 debt crises, 44 banking crises, and 69 currency crises; main findings hold aside from lower trade losses for some imports.
- Estimators:
  - Main estimator AIPW; alternative IPW yields consistent results though typically lower magnitudes, especially for import trade losses.

### Channels — heterogeneity and illustrative exact contrasts
- Manufacturing share heterogeneity (example, Table A.15a, Debt crises, Total exports ATE-AIPW):
  - High (above median): h=0: -1.621*** ; h=5: -3.693*** (s.e. 0.300, 1.285).
  - Low (below median): h=0: -0.649* ; h=5: -7.610*** (s.e. 0.336, 1.437).
  - P-value High=Low: 0.000 at reported horizons.
- Trading partners’ growth heterogeneity (example, Table A.18a, Debt crises, Total exports ATE-AIPW):
  - High partners’ growth: h=0: -0.172 ; h=3: -4.365*** ; h=5: 0.138.
  - Low partners’ growth: h=0: -2.507*** ; h=3: -8.602*** ; h=5: -13.270***.
  - P-value High=Low: 0.000 at most horizons.
- Financial development / gross capital inflows heterogeneity:
  - Crises associated with contractions in financial development or large reductions in gross capital inflows generate stronger reductions of total and sectoral trade, especially manufacturing and services.
  - Examples (Table A.20a, Debt crises, gross capital inflows):
    - High inflows: h=1: -1.984*** ; h=4: -6.220***.
    - Low inflows: h=3: -11.091*** ; h=5: -13.326***.
    - P-value High=Low: 0.000 at most horizons.

### Conclusions and policy implications
- Conclusions
  - Aggregate exports and imports significantly decrease following most financial crises.
  - Manufactured goods are the most affected sectorally; mining and services also decline; agricultural goods may benefit from substitution effects, especially after debt crises.
  - Combined crises yield larger trade losses than isolated crises.
  - Three channels mediate trade losses: structural (trade composition and diversification), demand-side (trading partners’ growth), and supply-side (availability of domestic and external finance).

- Policy implications (derived from findings)
  - Diversification of exports and trading partners increases resilience of trade to financial crises.
  - Improvements in financial development reduce vulnerability of trade.
  - Policies to reduce the occurrence of financial crises and increase creditors’ confidence can alleviate trade losses; examples include:
    - sound macroeconomic policies,
    - macroprudential policies,
    - multilateral policies coping with contagion of crises.
  - Special attention to countries with lower industrialization where crises impede structural transformation; policies supporting structural transformation and external finance access may mitigate long-term trade losses.
  - Sectoral focus: stabilize trade credit and reconnect global value chains to protect manufacturing and services; coordinate international support to sustain external demand when trading partners’ growth weakens.

*Italic: Source — wpiea2021176-print-pdf (Appendix IV and associated appendices, excerpts provided).*

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

### wpiea2021176-print-pdf - References................................................................................................................

### References
- References....................................................................................................................... 33

### Appendix I. Sample, variables descriptions and sources
- Appendix I. Sample, variables descriptions and sources .................................................... 39
- A. Financial crises since 1980 in all emerging countries ................................................ 39
- B. List of countries included in regressions analyses...................................................... 39
- C. Data description and sources .................................................................................... 40

### Appendix II. Summary statistics
- Appendix II. Summary statistics ....................................................................................... 41

### Appendix III. Balance and overlap checks
- Appendix III. Balance and overlap checks ........................................................................ 42
- A. Balance checks ........................................................................................................ 42
- B. Overlap checks ........................................................................................................ 43

*Source: wpiea2021176-print-pdf - References........................................................................................................................*

### Appendix IV. Supplementary graphs and tables ................................................................ 44

### Appendix IV. Supplementary graphs and tables

### Introduction — scope and objectives
- Goal: assess trade losses from financial crises from a sectoral perspective for emerging countries.
- Sample: 41 emerging countries over the period 1980–2018 with 38 debt crises, 34 banking crises, and 36 currency crises.
- Novelty: disaggregation by traded category (agricultural, mining, manufactured goods, and services) and causal identification combining local projections and impact assessment with the Augmented Inverse Propensity Weighted estimator.

### Key aggregate findings
- Aggregate five-year cumulative effects on exports (in percentage points of pre-crisis GDP):
  - Debt crises: 5.6 percentage points (pp.)
  - Banking crises: 8.9 pp.
  - Currency crises: 7.7 pp.
- Aggregate five-year cumulative effects on imports (in percentage points of pre-crisis GDP):
  - Debt crises: 11 pp.
  - Banking crises: 14 pp.
  - Currency crises: 9.1 pp.
- Trade balance:
  - Debt and banking crises increase the trade balance for all years in the aftermath by more than 5 pp. of pre-crisis GDP, reflecting larger contractions in imports than exports.
  - Currency crises increase the trade balance for years 1–4 but are not found to significantly affect the trade balance in year 5.

### Sectoral disaggregation — main patterns
- Manufacturing:
  - Most affected by financial crises; largest contributor to aggregate trade losses.
- Services:
  - Significant declines, particularly after banking crises; services trade is not negligible in losses.
- Mining:
  - Trade declines following several types of crises, notably after banking crises.
- Agriculture:
  - Least affected; evidence that agricultural trade may benefit from a substitution effect, particularly following debt crises.
- Overall ordering of trade shares in the sample: manufactured goods dominate, followed by services, mining goods, and agricultural goods.

### Identified transmission channels
- Structural channel:
  - Financial crises hinder structural transformation by hurting manufacturing exports more in countries with a relatively lower share of manufacturing exports.
  - Diversification reduces trade losses: countries with diversified export products and trading partners experience significantly lower trade losses from crises.
- Demand channel:
  - Crises that lower demand from trading partners have larger adverse trade impacts, especially for manufacturing trade.
  - Agricultural and mining goods and services exhibit lower income elasticity of demand compared with manufactured goods, contributing to their relative resilience.
- Supply channel:
  - When crises are associated with deteriorated domestic and external financial conditions and sudden stops, they exert significant detrimental effects on trade, mainly through manufacturing and services.
  - Manufacturing and services are more reliant on external finance and global value chains; credit crunches and disruptions of trade credit and production chains amplify trade declines in these sectors.

### Methodology — identification and estimation
- Combined approach: local projections (LP) à la Jordà (2005) + impact assessment with the Augmented Inverse Propensity Weighted (AIPW) estimator.
- Local projection setup:
  - Horizon h ∈ ⟦0;5⟧.
  - Dependent variable: Δy_{i,t+h}^k = (y_{i,t+h}^k − y_{i,t−1}^k)/GDP_{t−1} ×100, cumulative change between t−1 and t+h in 100 times trade flows of variable k scaled by pre-crisis GDP.
  - Crisis dummies: D_{i,t}^d, D_{i,t}^b, D_{i,t}^c equal to 1 at crisis start, 0 otherwise; non-starting crisis years are dropped.
  - Includes lagged changes ∆y_{i,t−1}^k and ∆y_{i,t−2}^k, controls X_{i,t+h}^x, and estimation for each horizon h.
- AIPW estimator and treatment model:
  - ATE defined as Λ_h = E[y_{i,t+h}(1) − y_{i,t−1} | D_{i,t}=1] − E[y_{i,t+h}(0) − y_{i,t−1} | D_{i,t}=0].
  - Independence assumption conditional on covariates Z_{i,t}.
  - Treatment model: covariate balancing propensity score (CBPS) p̂_{i,t} estimated on Z_{i,t}; CBPS ensures balance and reduces misspecification bias.
  - Outcome model: LP eq. (1) estimated separately for treated and control groups to predict potential outcomes.
  - Final estimator: Λ̂_AIPW_h as the doubly robust semi-parametric AIPW, with truncation of maximum inverse propensity weight to 10 (robustness checks use 5 and 20).

### Data sources and controls
- Crises definitions and sources:
  - Debt crises: Reinhart and Rogoff (2009).
  - Banking crises: Laeven and Valencia (2018).
  - Currency crises: Frankel and Rose (1996) (nominal depreciation ≥ 25 percent and at least a 10 percent increase in the rate of depreciation).
- Trade data:
  - Goods: UN Comtrade via WITS–World Bank at 3-digit SITC, aggregated into agricultural, mining, and manufactured goods per WTO classification.
  - Services: UNCTAD.
  - Total trade: aggregation of agriculture, mining, manufacturing, and services; deflated by export/import prices.
- Treatment-model covariates (lagged one year) used to estimate propensity scores (Z_{i,t}):
  - (i) number of past-5 years financial crises except the one of interest
  - (ii) intensity of conflicts
  - (iii) log of Real GDP
  - (iv) log of public debt to GDP
  - (v) log of domestic credit to GDP
  - (vi) log of liquid liabilities to GDP
  - (vii) net capital inflows to GDP
  - (viii) log of foreign reserves to GDP
  - (ix) current account to GDP
  - (x) financial openness index
  - (xi) log of trade openness to GDP
  - (xii) terms of trade growth
  - (xiii) floating exchange rate regime dummy
  - (xiv) government accountability index
  - (xv) corruption index
  - (xvi) trading partners’ growth
  - (xvii) US interest rate on government’s securities
- Outcome-model controls X_{i,t}: (i) change of trade flows one and two years before crisis onset, (ii) other crises, and (iii) exchange rate regime at horizon h.

### Data patterns and preliminaries
- Crises occurrence and clustering:
  - Financial crises are recurrent in emerging markets and occur by clusters (e.g., debt crises in Latin America and Africa in the 1980s; banking and currency crises in Asia, Latin America, and Eastern Europe in the 1990s).
  - Debt crises have been more frequent in emerging markets (especially in the 1980s and 1990s) than banking and currency crises.
  - Regional incidence: Africa, Latin America, and Middle-East more affected by debt crises; Latin America, Europe & Central Asia, East Asia & Pacific more affected by banking crises; Europe & Central Asia and Latin America more affected by currency crises.
- Inter-crisis linkages (Kaplan–Meier survival estimates):
  - After a debt crisis: a banking crisis follows in one-quarter of cases within five years; a currency crisis in one year.
  - After a banking crisis: a debt or currency crisis follows in one-quarter of cases within two years.
  - After a currency crisis: a debt crisis follows in one-quarter of cases within two years; a banking crisis in four years.
  - Implication: empirical estimation must control for concurrent and sequential crises to avoid overestimating trade losses.
- Descriptive (non-counterfactual) cumulative changes (year t−1 to t+5, scaled by pre-crisis GDP):
  - Total exports and imports decline sharply during all types of financial crises (e.g., exports and imports decline respectively by between 11 and 14 percentage points of pre-crisis GDP following all crisis types in the raw averages).
  - Manufacturing trade drives the aggregate contraction, followed by services, mining, and agricultural goods.

### Robustness and diagnostic notes
- Propensity-score diagnostics:
  - CBPS shows high classification power and considerable overlap between treated and control propensity score distributions; weighting by CBPS eliminates covariate differences per Rubin (2002) criteria more effectively than a pooled probit.
- Weight truncation and sensitivity:
  - Primary truncation of maximum inverse propensity weight to 10; robustness checks conducted with maximum weights of 5 and 20.
- Estimator properties:
  - AIPW is doubly robust: unbiased if either treatment or outcome model is correctly specified; includes an adjustment term that stabilizes estimates when propensity scores approach 0 or 1.

### Implications and policy-relevant findings
- Diversification policy:
  - Export-product and partner diversification increase resilience to trade losses from financial crises.
- Sectoral focus for stabilization:
  - Manufacturing is highly vulnerable to crises due to external-finance dependence and global value-chain linkages; policies to stabilize trade credit and reconnect GVC links can mitigate losses.
- Demand- and supply-side interventions:
  - When crises reduce external demand, policies that support demand from trading partners (e.g., coordinated international support) help limit manufacturing export declines.
  - When crises cause credit crunches or sudden stops, measures to restore domestic and external financial conditions, and targeted support to firms that rely on external finance, can reduce trade contraction—especially in manufacturing and services.

*Italic: Authors’ calculations and analysis as presented in Appendix IV of the referenced working paper.*

### 2.2 pp. of pre-crisis GDP over five years, compared with countries unaffected by crises. In

### wpiea2021176-print-pdf - 2.2 pp. of pre-crisis GDP over five years, compared with countries unaffected by crises. In

### Sectoral findings: agricultural, mining, manufacturing, and services trade
- Agricultural trade
  - Exports of agricultural goods are left unchanged in the aftermath of currency crises or significantly decrease by 1.5 pp. over five years after the banking crises.
  - Agricultural imports are significantly reduced by between 0.6 and 1.5 pp. for all types of crises.
  - Interpretation: trade in agricultural goods exhibits greater resilience during financial crises and can even intensify, which may signal a substitution effect in favor of agricultural goods; supported by relatively lower income elasticity and lower external financing needs.

- Mining trade
  - Five years after debt crises, exports of mining goods increase by 1.1 pp. of pre-crisis GDP.
  - In the aftermath of banking and currency crises, exports of mining goods decrease both on impact and over the five years to reach a collapse of 1.6 pp. and 2.9 pp., respectively.
  - Imports of mining goods over five years are significantly reduced by 1.9 pp. for debt crises and 3 pp. for both banking and currency crises.
  - Overall: collapse of mining goods is higher than that of agriculture goods and sometimes lower or higher than services, but more often largely lower than the collapse of manufactured goods.

- Manufacturing trade
  - Five-year cumulative losses on manufacturing exports:
    - 6.1 pp. of pre-crisis GDP for debt crises.
    - 4 pp. for currency crises.
    - 2.3 pp. for banking crises.
  - Five-year cumulative losses on manufacturing imports:
    - 4.6 pp. for currency crises.
    - 5.4 pp. for debt crises.
    - 7 pp. for banking crises.
  - Contributions to total export collapse (manufacturing):
    - 107 percent for debt crises.
    - 51 percent for currency crises.
    - 26 percent for banking crises.
  - Manufacturing accounts for around 50 percent of the total collapse of imports for all types of crises.
  - As manufacturing imports fall more than manufacturing exports, adverse effects on manufactured trade are driven by the increase in overall trade balance for debt and banking crises.
  - Despite large effects, manufactured trade losses leave unexplained a fairly large proportion of aggregated trade losses.

- Services trade
  - Five-year cumulative export losses:
    - 3.5 pp. of pre-crisis GDP over five years following banking crises (contribution of 39 percent of the total export collapse).
    - 2.9 pp. following debt crises.
    - 0.9 pp. following currency crises.
  - Five-year cumulative import losses:
    - 2.6 pp. following debt and banking crises.
    - 0.9 pp. following currency crises.
  - Services trade is significantly reduced by all types of financial crises but shows greater resilience than manufactured goods except in the aftermath of banking crises.

- Summary of sectoral hierarchy (magnitude of declines)
  - Manufactured goods: most affected.
  - Mining and services: significant declines for several crisis types.
  - Agricultural goods: relatively resilient; possible substitution effect, especially following debt crises.

### Robustness checks and alternative specifications
- Combined vs. individual crises
  - Combined crises (another crisis within t−2 to t+2) trigger more significant and higher-magnitude aggregated and sectoral trade losses, except for imports following debt crises.
  - Individual crises yield lower total and sectoral trade losses (more so for exports than imports), except for imports following debt crises.
  - Implication: studies ignoring simultaneity of crisis types may overestimate or misattribute effects.

- Excluding post-GFC period
  - Dropping the post-GFC period (from 2008 onwards) leads to comparable results for total and sectoral export losses from banking and currency crises; losses from debt crises are significantly reduced.
  - Trade losses on imports present a lower magnitude when the post-GFC period is excluded.

- Maximum weights in treatment models
  - Benchmark maximum weight: 10.
  - Alternative maximum weights tested: 20 and 5.
  - Using maximum weight of 20 associated with somewhat lower magnitudes for some goods/crises; main findings remain robust.

- Alternative crisis sources and definitions
  - Using Cruces and Trebesch (2013) definition for debt crises (private creditor restructurings only), Reinhart and Rogoff (2009) for banking and currency crises (currency crisis = at least a 20 percent nominal depreciation vs. US dollar).
  - Data under these definitions: 41 debt crises, 44 banking crises, and 69 currency crises.
  - Aside from lower trade losses for total and sectoral imports following debt and currency crises, main findings hold.

- Estimators
  - Main estimator: Augmented Inverse Propensity Weighted (AIPW).
  - Alternative: Inverse Propensity Weighted (IPW) — results consistent with main findings though lower magnitudes, especially for import trade losses.

### Channels: structural, demand-side, and supply-side evidence
- Method: split crises by median of channel-proxy variables; compute 5-year cumulated ATE-AIPW estimates above/below median.

- Structural channel (trade structure, diversification)
  - Proxy variables: share of manufactured exports in total exports; export diversification index; trading partners diversification index (year before crises).
  - Findings:
    - Total, manufacturing, and services exports and imports fall more after crises in countries with a lower pre-crisis share of manufacturing exports.
    - Trade in agricultural and mining goods shows similar patterns across high- and low-manufacturing-share countries.
    - Countries with higher share of manufacturing exports are more resilient: resilience effect dominates vulnerability effect.
    - Export diversification: crises lead to higher contractions of total, manufacturing, and services trade in countries with lower export diversification; diversification strengthens resilience.
    - Trading-partner diversification: except for banking crises, countries with more trading partners experience lower adverse effects after debt and currency crises.

- Demand-side channel (trading partners’ growth)
  - Proxy: evolution of trading partners’ growth over the five years following crises.
  - Findings:
    - Financial crises associated with higher trading partners’ growth have milder or no adverse effect on trade, except for banking crises.
    - For debt crises with higher partners’ growth: no aggregate effect because increases in agriculture and mining exports offset collapses in manufacturing and services exports.
    - Currency crises: larger reductions in exports and imports in countries with lower trading partners’ growth, driven by falls across goods and services except agriculture exports.
    - Banking crises: similar or slightly higher trade losses when associated with higher trading partners’ growth.
    - Implication: lower external demand amplifies trade losses; global downturns (e.g., post-GFC) can cause unprecedented international trade collapses.

- Supply-side channel (external and domestic finance)
  - Proxies: evolution of financial development, gross capital inflows, investors’ credit rating risks over five years following crises.
  - Findings:
    - Crises associated with a low increase (or contraction) of financial development lead to higher declines in trade.
    - Crises associated with large reductions in gross capital inflows generate stronger reductions of total and sectoral trade (driven by fall of manufacturing exports for debt and currency crises, service exports for banking crises, and manufacturing imports for all crises).
    - Crises associated with higher deterioration of investors’ sentiment (credit rating risks) trigger larger detrimental effects on total and sectoral trade.
    - Overall: deterioration in domestic financial development and external financial conditions (sudden stops) intensify trade losses, mostly via manufacturing and services.

### Conclusions and policy implications
- Key conclusions
  - Aggregate exports and imports significantly decrease following most financial crises.
  - Manufacturing goods are the most affected sectorally; mining and services also decline; agricultural goods may benefit from substitution effects, especially after debt crises.
  - Combined crises yield larger trade losses than isolated crises.
  - Three channels mediate trade losses: structural (trade composition and diversification), demand-side (trading partners’ growth), and supply-side (availability of domestic and external finance).

- Policy implications (derived from findings)
  - Diversification of exports and trading partners increases resilience of trade to financial crises.
  - Improvements in financial development reduce vulnerability of trade.
  - Policies to reduce the occurrence of financial crises and increase creditors’ confidence can alleviate trade losses; these include:
    - sound macroeconomic policies,
    - macroprudential policies,
    - multilateral policies coping with contagion of crises.
  - Special attention to countries with lower industrialization where crises impede structural transformation; policies supporting structural transformation and external finance access may mitigate long-term trade losses.

*Italic: Source — wpiea2021176-print-pdf (excerpts provided) — sectoral trade losses and channels after debt, banking, and currency crises.*

### REFERENCES

### REFERENCES

### Major themes and representative citations
- Trade dynamics and the Great Trade Collapse: Baldwin (2011); Bems, R., R. C. Johnson, And K. M. Yi (2011); Levchenko, A. A., L. T. Lewis, And L. L. Tesar (2010); Eaton, J., S. Kortum, B. Neiman, And J. Romalis (2016).
- Financial crises, sudden stops, and sovereign defaults: Bordo, M. D. (2006); Mendoza, E. G. (2010); Cruces, J. And C. Trebesch (2013); Forni, L., C. Richmond, G. Palomba, And J. Pereira (2016); Laeven, L. And F. Valencia (2018); Kuvshinov, D. And K. Zimmermann (2019).
- Trade resilience and sectoral patterns (goods and services): Ariu, A. (2016); Borchert, I. And A. Mattoo (2010); Miroudot, S., A. Ragoussis, And R. Lanz (2009); Amiti, M. And D. Weinstein (2011).
- Export diversification and country vulnerability: Abreha, K. G., V. Smeets, And F. Warzynski (2020); Romeu, R. And N. C. Da Costa Neto (2011); Martinez, J. And G. Sandleris (2011).
- Empirical methods for causal inference and treatment effects: Rosenbaum, P. And D. Rubin (1983); Imbens, G. (2004); Cole, S. And M. Hernan (2008); Imai, K. And M. Ratkovic (2014); Glynn, A. And K. Quinn (2009); Jorda, O. (2005).
- Financial openness, development, and macro links to trade: Beck, T. (2002); Beck, T. And A. Demirguc-Kunt (2009); Chinn, M. D. And H. Ito (2008); Mbaye, S., M. Moreno Badia, And K. Chae (2018).
- Governance, political violence, and external conditions: Sachs, J. And A. Warner (1995); Glick, R. And A. K. Rose (1999); Ilzetzki, E., C. Reinhart, And K. Rogoff (2017).

*Selected individual references are listed in the source document.*

### Appendix I. Sample, variables descriptions and sources

- Figure A.1: Reports the starting date of the various crises since 1980 in all emerging countries.
- B. List of countries included in regressions analyses:
  - Albania, Algeria, Argentina, Armenia, Bolivia, Brazil, Chile, China, Colombia, Costa Rica, Croatia, Dominican Republic, Ecuador, Egypt, Arab Rep., Gabon, Georgia, Guatemala, Hungary, India, Indonesia, Jamaica, Jordan, Lebanon, Mexico, Morocco, Pakistan, Panama, Paraguay, Peru, Philippines, Poland, Romania, Russian Federation, South Africa, Sri Lanka, Thailand, Tunisia, Turkey, Ukraine, Uruguay, Venezuela.
- C. Data description and sources (variables and sources as listed):
  - Trade variables
    - International trade in goods (agricultural, mining): World Integrated Trade Solution (WITS) code at 3-digit SITC classification
    - International trade in services: United Nations Conference on Trade and Development (UNCTAD)
    - Cumulative change of trade variables from the onset of financial crises to years 1-5 after crises: Authors’ calculation based in WITS and UNCTAD
  - Financial crises
    - Debt crises: Reinhart and Rogoff (2009) and Cruces and Trebesch (2013)
    - Banking crises: Laeven and Valencia (2018) and Reinhart and Rogoff (2009)
    - Currency crises: Authors’ calculation based on exchange rate taken from Penn World Tables 9.0 and Bruegel datasets, and using the definition in Frankel and Rose (1996)
  - Other variables and sources
    - Intensity of conflicts: Major episodes of political violence (MEPV)
    - Log of real GDP: World Development Indicators
    - Log. of public debt to GDP: Global Debt Database of the IMF (Mbaye et al. (2018))
    - Log. of domestic credit to GDP: World Economic Outlook, IMF, and World Development Indicators, WB
    - Log. of liquid liabilities to GDP: Beck and Demirguc-Kunt (2009)
    - Net capital inflows to GDP: Authors' calculations based on the Balance of Payments and International Investment Position dataset from the IMF
    - Gross Capital inflows (% of GDP): Authors' calculations based on the Balance of Payments and International Investment Position dataset from the IMF
    - Log. of foreign reserves to GDP: World Economic Outlook, IMF, and World Development Indicators, WB
    - Current account to GDP: World Economic Outlook, IMF, and World Development Indicators, WB
    - Financial openness index: Chinn and Ito (2008)
    - Log of trade openness to GDP: World Development Indicators
    - Terms of trade growth: Penn World Tables 9.1
    - Floating exchange rate regime: Authors’ calculation based on Ilzetzki et al. (2017)
    - Government accountability index: Varieties of Democracy (V-Dem) project
    - Corruption index: Varieties of Democracy (V-Dem) project
    - Trading partners' growth: Global Economic Environment, IMF
    - US interest rate on gov.'s debt securities: Bank of International Settlements
    - Exports diversification index: Export Diversification and Quality database, IMF
    - Trading partners diversification index: Export Diversification and Quality database, IMF
    - Financial development index: Financial development index from IMF
    - Investors' credit ratings risks: Country Credit Ratings from the IMF

*Appendix I contents are drawn directly from the source document.*

### Appendix II. Summary statistics (Table B.2)
- Obs. 766; Mean 19.690; Sd 20.795; Min -18.277; Max 124.255 — Cumulative change of total exports over 5 years (% of pre-crisis GDP)
- Obs. 766; Mean 2.646; Sd 3.923; Min -2.543; Max 44.719 — Cumulative change of agricultural mining exports over 5 years (% of pre-crisis GDP)
- Obs. 766; Mean 3.291; Sd 8.324; Min -18.363; Max 72.770 — Cumulative change of mining exports over 5 years (% of pre-crisis GDP)
- Obs. 766; Mean 8.362; Sd 13.009; Min -11.702; Max 94.118 — Cumulative change of manufacturing exports over 5 years (% of pre-crisis GDP)
- Obs. 766; Mean 5.392; Sd 7.282; Min -9.869; Max 49.547 — Cumulative change of services exports over 5 years (% of pre-crisis GDP)
- Obs. 766; Mean 23.338; Sd 27.805; Min -20.333; Max 179.063 — Cumulative change of imports over 5 years (% of pre-crisis GDP)
- Obs. 766; Mean 2.228; Sd 3.039; Min -4.678; Max 25.207 — Cumulative change of agricultural mining imports over 5 years (% of pre-crisis GDP)
- Obs. 766; Mean 3.819; Sd 5.958; Min -10.939; Max 34.175 — Cumulative change of mining imports over 5 years (% of pre-crisis GDP)
- Obs. 766; Mean 13.219; Sd 16.641; Min -20.028; Max 102.713 — Cumulative change of manufacturing imports over 5 years (% of pre-crisis GDP)
- Obs. 766; Mean 4.072; Sd 5.516; Min -4.676; Max 40.976 — Cumulative change of services imports over 5 years (% of pre-crisis GDP)
- Obs. 766; Mean 19.128; Sd 15.711; Min 0.160; Max 71.509 — Share of agricultural exports (% of total exports)
- Obs. 766; Mean 17.572; Sd 21.172; Min 0.001; Max 95.987 — Share of mining exports (% of total exports)
- Obs. 766; Mean 34.545; Sd 21.392; Min 0.306; Max 86.902 — Share of manufacturing exports (% of total exports)
- Obs. 766; Mean 28.755; Sd 19.331; Min 2.558; Max 90.934 — Share of services exports (% of total exports)
- Obs. 766; Mean 10.474; Sd 4.338; Min 2.797; Max 30.705 — Share of agricultural imports (% of total imports)
- Obs. 766; Mean 13.775; Sd 7.875; Min 0.804; Max 46.405 — Share of mining imports (% of total imports)
- Obs. 766; Mean 55.141; Sd 10.704; Min 26.701; Max 80.015 — Share of manufacturing imports (% of total imports)
- Obs. 766; Mean 20.610; Sd 7.227; Min 5.017; Max 49.055 — Share of services imports (% of total imports)
- Obs. 766; Mean 0.052; Sd 0.223; Min 0.000; Max 1.000 — 1 if Debt crises at start
- Obs. 766; Mean 0.048; Sd 0.215; Min 0.000; Max 1.000 — 1 if Banking crises at start
- Obs. 766; Mean 0.050; Sd 0.217; Min 0.000; Max 1.000 — 1 if Currency crises at start
- Obs. 766; Mean 1.219; Sd 2.227; Min 0.000; Max 10.000 — Intensity of conflicts
- Obs. 766; Mean 7.212; Sd 3.687; Min -8.147; Max 15.742 — Log of real GDP
- Obs. 765; Mean 3.661; Sd 0.696; Min 0.991; Max 5.355 — Log. of public debt to GDP
- Obs. 757; Mean 3.484; Sd 0.653; Min 1.380; Max 5.076 — Log. of domestic credit to GDP
- Obs. 760; Mean 3.635; Sd 0.573; Min 1.687; Max 5.482 — Log. of liquid liabilities to GDP
- Obs. 745; Mean 2.531; Sd 5.075; Min -22.404; Max 24.586 — Net capital inflows to GDP
- Obs. 766; Mean 2.201; Sd 0.863; Min -1.138; Max 4.677 — Log. of foreign reserves to GDP
- Obs. 766; Mean -2.588; Sd 5.050; Min -29.363; Max 22.671 — Current account to GDP
- Obs. 765; Mean 0.469; Sd 0.336; Min 0.000; Max 1.000 — Financial openness index
- Obs. 766; Mean 4.021; Sd 0.493; Min 2.446; Max 5.116 — Log of trade openness to GDP
- Obs. 766; Mean 0.005; Sd 0.053; Min -0.323; Max 0.360 — Terms of trade growth (%)
- Obs. 766; Mean 0.110; Sd 0.313; Min 0.000; Max 1.000 — Floating exchange rate regime
- Obs. 766; Mean 0.647; Sd 0.827; Min -1.529; Max 1.986 — Government accountability index
- Obs. 766; Mean -0.589; Sd 0.250; Min -0.960; Max -0.037 — Corruption index
- Obs. 751; Mean 3.607; Sd 2.034; Min -4.810; Max 13.262 — Trading partners' growth (%)
- Obs. 766; Mean 4.847; Sd 4.159; Min 0.125; Max 22.000 — US interest rate on gov.'s debt securities (%)
- Obs. 728; Mean 2.840; Sd 0.793; Min 1.498; Max 5.558 — Exports diversification index
- Obs. 728; Mean 2.657; Sd 0.550; Min 1.651; Max 4.614 — Trading partners diversification index
- Obs. 766; Mean 0.279; Sd 0.121; Min 0.061; Max 0.632 — Financial development index
- Obs. 761; Mean 42.797; Sd 14.103; Min 7.650; Max 81.050 — Investors' credit ratings risks

*Appendix II reproduces the summary statistics table as presented in the source document.*

*Source: wpiea2021176-print-pdf - REFERENCES (as provided).*

### APPENDIX III. BALANCE AND OVERLAP CHECKS

### APPENDIX III. BALANCE AND OVERLAP CHECKS

### A. Balance checks
- Methodological note: Rubin (2002) suggests a cut-off of 0.25 on the absolute value of the standardized difference to define imbalance. Variables with absolute standardized difference higher than 0.25 indicate a significant difference between treated and control groups. Rubin (2002) also proposes using the ratio of treated and control variances as a balance measure of the second moment; balance is defined by values close to 1.0 and variables are out of balance if the variance ratio is greater than 2.0 or less than 0.5.
- Model performance summary:
  - The CBPS model eliminates all the differences in characteristics between treated and control groups.
  - The pooled probit model is less effective at eliminating differences in characteristics between treated and control groups.
- Selected balance diagnostics (examples from tables for Debt, Banking, and Currency crises):
  - Banking crises (non-weighted): 
    - # of Banking crises over past-5 years (t-5,t): Treated 0.316, Control 0.147, Std. mean 0.405, Var. ratio 1.767
    - # of Currency crises over past-5 years (t-5,t): Treated 0.474, Control 0.148, Std. mean 0.687, Var. ratio 2.245
    - Log. of foreign reserves to GDP (t-1): Treated 1.269, Control 2.276, Std. mean -1.147, Var. ratio 1.294
    - Floating exchange rate regime (t-1): Treated 0.395, Control 0.091, Std. mean 0.751, Var. ratio 2.969
    - US interest rate on gov.'s securities (t-1): Treated 8.409, Control 4.442, Std. mean 0.916, Var. ratio 1.570
  - Debt crises (non-weighted):
    - # of Debt crises over past-5 years (t-5,t): Treated 0.147, Control 0.149, Std. mean -0.005, Var. ratio 1.266
    - Log. of Real GDP per capita (t-1): Treated 6.890, Control 7.277, Std. mean -0.099, Var. ratio 1.258
    - Log. of public debt to GDP (t-1): Treated 3.551, Control 3.680, Std. mean -0.161, Var. ratio 1.835
    - Log. of foreign reserves to GDP (t-1): Treated 1.809, Control 2.244, Std. mean -0.503, Var. ratio 1.073
    - US interest rate on gov.'s securities (t-1): Treated 6.973, Control 4.535, Std. mean 0.567, Var. ratio 1.434
  - Currency crises (non-weighted):
    - # of Debt crises over past-5 years (t-5,t): Treated 0.500, Control 0.131, Std. mean 0.735, Var. ratio 2.773
    - # of Banking crises over past-5 years (t-5,t): Treated 0.472, Control 0.139, Std. mean 0.767, Var. ratio 2.134
    - Log. of public debt to GDP (t-1): Treated 3.749, Control 3.670, Std. mean 0.114, Var. ratio 1.039
    - Log. of foreign reserves to GDP (t-1): Treated 1.252, Control 2.274, Std. mean -1.314, Var. ratio 0.750
    - Floating exchange rate regime (t-1): Treated 0.389, Control 0.092, Std. mean 0.733, Var. ratio 2.923
- Interpretation guidance provided in notes:
  - Variables with variance ratios greater than 2.0 or less than 0.5 are considered out of balance.
  - The tables report (1) Non weighted, (2) Weighted, (21) Covariate Balancing propensity score (CBPS), and (22) Pooled probit diagnostics.

### B. Overlap checks
- Propensity score estimation:
  - The predicted propensity scores reported are obtained after applying the CBPS estimator.
  - The dependent variable equals 1 at the start of financial crises, 0 in country-years observations without crises (unfolding).
  - The largest set of control variables described in the data section are included with lags to reduce potential issues of endogeneity.
- Kernel-density findings:
  - Figures display a high probability of financial crises for the treated groups (especially for debt and currency crises) and a low probability for their counterparts.
  - The figures show a significant overlap between the treated and control groups.
- Estimation practicality:
  - Since some observations receive a large weight, the maximum weight was set to 10 for the ATE-AIPW estimates.

*Source: APPENDIX III. BALANCE AND OVERLAP CHECKS (tables and notes as provided).*

### APPENDIX IV. SUPPLEMENTARY GRAPHS AND TABLES

### APPENDIX IV. SUPPLEMENTARY GRAPHS AND TABLES

### A. Graphs — Benchmark and robustness results
- Figures D.3–D.9 present robustness checks for cumulative trade losses over five years after financial crises:
  - Notes across figures: "Robust standard errors clustered at the country-level in parentheses. AIPW estimates. The dependent variables are 100 times the cumulative change of agricultural, mining, manufacturing, and services exports and imports relative to the year prior to the onset of the crisis for years 1–5 after the onset of the crisis, scaled pre-crisis GDP. Accumulated losses over five years."
  - Additional robustness variations listed in figure captions include: "Drop the period 2008 onwards", "Maximum weight set to 20", "Maximum weight set to 5", "Alternative sources and definitions of crises", and "Inverse Probability Weighted (IPW) estimator."
- Figures D.10–D.16 examine channels (manufacturing share, export diversification, trading partners diversification, partners’ growth, financial development, gross capital inflows, investors’ credit rating risks) with identical estimation notes: AIPW estimates, dependent variables as above, accumulated over five years, maximum weights truncated at 10.

### B. Tables — First-stage treatment model
- Table D.6: Treatment models predicting the likelihood of financial crises, pooled probit, average marginal effects (selected coefficients and significance)
  - "# of Debt crises over past-5 years (t-5,t)": Debt: -0.012 ; Banking: 0.034**
  - "# of Banking crises over past-5 years (t-5,t)": Debt: 0.026 ; Banking: 0.060***
  - "Log of real GDP (t-1)": Currency: -0.005**
  - "Log. of public debt to GDP (t-1)": Debt: 0.042***
  - "Log. of domestic credit to GDP (t-1)": Banking: 0.049***
  - "Log. of liquid liabilities to GDP (t-1)": Banking: -0.051**
  - "Log. of foreign reserves to GDP (t-1)": Debt: -0.038*** ; Currency: -0.034***
  - "Floating exchange rate regime (t-1)": Debt: 0.054*** ; Currency: 0.047**
  - "Trading partners' growth (t-1)": Debt: -0.010** ; Banking: 0.010**
  - "US interest rate on gov.'s debt securities (t-1)": Debt: 0.004**
  - Observations: 732; # of crises: Debt: 38, Banking: 34, Currency: 36 (as reported in columns); Classification: Debt 94.262, Banking 95.219, Currency 95.082; Model AUC: Debt 0.9220, Banking 0.769, Currency 0.915 (s.e. AUC: 0.016, 0.038, 0.019); pseudo R2: 0.338, 0.121, 0.343.
  - Notes: "Robust standard errors clustered at the country-level in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01. Pooled probit model. The coefficients are the average marginal effects at the mean. AUC denotes Area Under Receiver Operating Characteristic curve."

### B.2 Benchmark and robustness results — Cumulative trade losses over five years after financial crises
- Table D.7 (Benchmark results, AIPW estimates; dependent variables = 100 times cumulative change relative to year prior to onset, scaled pre-crisis GDP; accumulated over five years; maximum weights truncated at 10)
  - Panel I: Exports (percent of pre-crisis GDP), Panel II: Imports (percent of pre-crisis GDP). Panels A–C cover Debt, Banking, Currency crises; Panels A'–C' appear as complementary panels.
  - Selected benchmark estimates for Total (exports) — ATE-AIPW (examples across horizons h=0..h=5):
    - One reported block: h=0: -0.134 ; h=1: -3.415*** ; h=2: -3.348*** ; h=3: -5.925*** ; h=4: -6.413*** ; h=5: -5.639*** (standard errors in parentheses 0.342, 0.466, 0.597, 0.860, 1.093, 1.514 respectively).
  - Corresponding imports block (examples):
    - h=0: -4.397*** ; h=1: -8.365*** ; h=2: -9.358*** ; h=3: -10.403*** ; h=4: -11.188*** ; h=5: -11.012*** (s.e. 0.471, 0.649, 1.004, 1.322, 1.668, 2.246).
  - Sectoral ATE-AIPW highlights (exports panels, example values and significance)
    - Agriculture: h=2: 0.382** ; h=5: 2.248*** (s.e. 0.184, 0.623)
    - Mining: h=0: 1.420*** ; h=5: 1.068* (s.e. 0.215, 0.590)
    - Manufacturing: h=0: -0.862*** ; h=5: -6.065*** (s.e. 0.145, 0.736)
    - Services: h=0: -0.546*** ; h=5: -2.890*** (s.e. 0.082, 0.380)
  - Observations: 732; # of Crises (varies by panel e.g., 38) ; # of Countries: 41.
  - P-value tests reported for "Total Exports = Total Imports" at horizons (examples): 0.000 for many horizons, and 0.231 in one block (indicating some horizon-specific equality tests).

- Robustness tables (selected highlights)
  - Table D.8: Robustness, cumulative trade losses after combined financial crises (observations 689 in many specifications; # of crises 24; # of countries 40). Example totals:
    - Exports total (ATE-AIPW): h=0: -2.398*** ; h=4: -9.691*** ; h=5: -9.805*** (s.e. 0.308, 1.077, 1.328).
    - Imports total (ATE-AIPW): h=0: -4.958*** ; h=5: -5.690*** (s.e. 0.481, 1.819).
  - Table D.9: Robustness, individual financial crises (observations 670; # of crises often 13 or 17 depending on panel). Example:
    - Exports total (ATE-AIPW): h=0: 2.186*** ; h=1: -3.367*** ; h=5: -3.521* (s.e. 0.433, 0.601, 1.899).
    - Imports total (ATE-AIPW): h=0: -4.211*** ; h=1: -10.576*** ; h=5: -16.934*** (s.e. 0.464, 0.662, 2.776).
  - Table D.10: Robustness, Drop the period 2008 onwards (observations 409; # of crises 31–33; # of countries 39). Example:
    - Exports total (ATE-AIPW): h=1: -1.142** ; h=3: -2.795** ; h=4: -2.855** (s.e. 0.567, 1.128, 1.374).
    - Imports total (ATE-AIPW): h=0: -4.044*** ; h=5: -8.638*** (s.e. 0.649, 2.300).
  - Table D.11, D.12: Robustness with maximum weight set to 20 and 5 respectively (observations 732; # of crises vary). Example:
    - Table D.11 exports total (ATE-AIPW) block: h=1: -3.186*** ; h=3: -4.965*** ; h=5: -4.224* (s.e. 0.646, 1.291, 2.526).
    - Table D.12 exports total (ATE-AIPW) block: h=1: -3.766*** ; h=5: -6.020*** (s.e. 0.364, 1.058).
  - Table D.13: Robustness, alternative sources and definitions of crises (observations 590; # of countries 31). Example:
    - Exports total (ATE-AIPW): h=0: -1.408*** ; h=5: -5.063*** (s.e. 0.384, 1.611).
    - Imports total (ATE-AIPW): h=0: -3.102*** ; h=3: -4.693*** (s.e. 0.464, 1.385).
  - Table D.14: Robustness, Inverse Probability Weighted (IPW) estimator (observations 732; # of crises 38; # of countries 41). Example:
    - ATE-IPW Total (exports): h=0: -1.217*** ; h=3: -5.509*** ; h=5: -4.068*** (s.e. 0.284, 1.164, 0.850).
    - ATE-IPW Total (imports): h=0: -5.012*** ; h=5: -3.581*** (s.e. 0.838, 0.280).

### B.3 Channels — Heterogeneity by country/channel characteristics
- Tables A.15a–A.21c present channel analyses across crises types (Debt, Banking, Currency) and channels:
  - Channel variables include: higher vs lower manufacturing exports share; higher vs lower export diversification; higher vs lower trading partners’ diversification; higher vs lower trading partners’ growth; higher vs lower financial development; higher vs lower gross capital inflows; higher vs lower investors’ credit ratings risks.
  - Each panel reports ATE-AIPW or ATE-IPW for High and Low groups across horizons h=0..h=5, for Total, Agriculture, Mining, Manufacturing, Services; also reports P-value High=Low and average trade in countries above and below cutoff.
- Selected illustrative channel findings (exact reported ATEs and tests):
  - Table A.15a (Debt crises, manufacturing exports channel, Panel a: Total, Panel I Exports):
    - ATE-AIPW High: h=0: -1.621*** ; h=1: -1.750*** ; h=2: -2.203*** ; h=3: -3.785*** ; h=4: -4.775*** ; h=5: -3.693*** (s.e. 0.300, 0.480, 0.620, 0.812, 1.002, 1.285).
    - ATE-AIPW Low: h=0: -0.649* ; h=1: -3.375*** ; h=2: -4.900*** ; h=3: -7.863*** ; h=4: -7.811*** ; h=5: -7.610*** (s.e. 0.336, 0.455, 0.620, 0.873, 1.091, 1.437).
    - P-value High=Low across horizons: 0.000 at all reported horizons.
    - Avg.(trade) in countries above cutoff (exports) at horizons: 3.030, 6.212, 9.802, 13.553, 17.101, 21.167.
  - Table A.15b (Banking crises, manufacturing exports channel, Panel a: Total, exports):
    - ATE-AIPW High: h=0: -0.693** ; h=1: -5.293*** ; h=2: -5.197*** ; h=3: -3.396*** ; h=4: -5.414*** ; h=5: -5.917***.
    - ATE-AIPW Low: h=0: -1.800*** ; h=1: -5.228*** ; h=2: -6.804*** ; h=3: -9.202*** ; h=4: -12.355*** ; h=5: -13.474***.
    - P-value High=Low at h=0: 0.000 ; at h=1: 0.787 ; at h=2: 0.000 ; at later horizons many P-values 0.000.
  - Table A.15c (Currency crises, manufacturing exports channel, Panel a: Total, exports and imports):
    - Exports ATE-AIPW High: h=0: -2.029*** ; h=1: -3.126*** ; h=2: -4.221*** ; h=3: -3.771*** ; h=4: -3.391*** ; h=5: -2.809***.
    - Imports ATE-AIPW High: h=0: -5.643*** ; h=1: -7.081*** ; h=2: -7.551*** ; h=3: -8.797*** ; h=4: -8.196*** ; h=5: -6.686***.
    - ATE-AIPW Low and P-values reported for high vs low contrasts.
- Additional channel examples (selected exact values):
  - Export diversification (Table A.16a, Debt crises, Panel a Total exports):
    - ATE-AIPW High: h=0: -1.600*** ; h=5: -5.678***.
    - ATE-AIPW Low: h=0: -1.104*** ; h=5: -8.603***.
    - P-value High=Low: 0.051 at h=0; 0.000 at many horizons.
  - Trading partners’ diversification (Table A.17a, Debt crises, Panel a Total exports):
    - ATE-AIPW High: h=0: -2.122*** ; h=5: -5.571***.
    - ATE-AIPW Low: h=0: -0.819*** ; h=5: -11.499***.
    - P-value High=Low: 0.000 at many horizons, some horizons with higher P-values (e.g., 0.067 at h=1).
  - Trading partners’ growth (Table A.18a, Debt crises, Panel a Total exports):
    - ATE-AIPW High: h=0: -0.172 ; h=3: -4.365*** ; h=5: 0.138.
    - ATE-AIPW Low: h=0: -2.507*** ; h=3: -8.602*** ; h=5: -13.270***.
    - P-value High=Low: 0.000 at most horizons.
  - Financial development evolution (Table A.19a, Debt crises, Panel a Total exports):
    - ATE-AIPW High: h=0: -1.330*** ; h=1: -3.509*** ; h=3: -6.209***.
    - ATE-AIPW Low: h=0: -1.688*** ; h=1: -1.924*** ; h=5: -7.029***.
    - P-values show heterogeneity across horizons (examples: P-value High=Low at h=1: 0.000; at h=0: 0.153).
  - Gross capital inflows (Table A.20a, Debt crises, Panel a Total exports):
    - ATE-AIPW High: h=0: -0.284 ; h=1: -1.984*** ; h=4: -6.220***.
    - ATE-AIPW Low: h=0: -2.969*** ; h=3: -11.091*** ; h=5: -13.326***.
    - P-value High=Low: 0.000 at most horizons.
  - Investors’ credit rating risks (Table A.21a, Debt crises, Panel a Total exports):
    - ATE-AIPW High: h=0: -0.921*** ; h=3: -3.828*** ; h=5: -2.395*.
    - ATE-AIPW Low: h=0: -0.812*** ; h=3: -8.587*** ; h=5: -10.174***.
    - P-value High=Low: 0.000 at many horizons; heterogeneity notable across panels and sectors.

- General channel-table notes (repeated across all channel tables):
  - "Robust standard errors clustered at the country-level in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01. IPW estimates or AIPW estimates as indicated. The dependent variables are 100 times the cumulative change of agricultural, mining, manufacturing, services, and total exports and imports relative to the year prior to the onset of the crisis for years 1-5 after the onset of the crisis, scaled pre-crisis GDP. Accumulated costs over five years. Observations in the treated and control groups are weighted by the propensity scores predicted in the treatment model. Maximum weights truncated at 10 (or as specified). Panel I: Exports (% of pre-crisis GDP). Panel II: Imports (% of pre-crisis GDP). Cutoffs reported (median) for channel splits (examples: manufacturing share median 17.031; banking crises manufacturing cutoff 27.083; export diversification cutoff -3.200; partners’ diversification cutoff -2.647; partners’ growth cutoff 7.041; financial development cutoff -0.008; gross capital inflows cutoff -1.017; investors' credit rating risks cutoff -5.567 or other medians as reported per table)."

*Italic: Source — APPENDIX IV. SUPPLEMENTARY GRAPHS AND TABLES (wpiea2021176-print-pdf).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021176-print-pdf.pdf_