## _wp1103 - Appendix 1. Data Sources

## Source details

**Canonical URL:** [_wp1103 - Appendix 1. Data Sources](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp1103.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp1103.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp1103.pdf.json)

---

### I. Research questions and empirical focus
- Research questions:
  - How has trade evolved following banking and debt crises?
  - Do such crises have lasting effects on a country’s imports and exports?
  - Do crises influence trade only through their effect on output and standard gravity determinants, or do they lead to other impediments to trade?
- Scope:
  - Episodes of banking and debt crises over the period 1970–2009.
  - Sample: 153 advanced, emerging, and developing economies; full sample contains 372,060 bilateral importer*exporter*year observations and 20,003 importer*exporter pairs.
  - Crisis episodes: 129 systemic banking crises since 1970; 60 sovereign debt crisis episodes since 1970; 10 “twin banking and debt crises”; 13 banking crisis episodes during 2007-08 concentrated in advanced economies.

### II. Key empirical findings (preserving reported magnitudes)
- Imports:
  - Gravity estimates: imports fall by 9 percent below gravity-predicted level in the year of the crisis, and by a further 10 percent in the following year.
  - On average, imports are 19 percent lower than predicted by the gravity framework in the year following a crisis (summary statement).
  - Five years after a crisis, imports remain 10 percent below normal (gravity estimates).
  - It takes more than 10 years for imports to return to normal: 11 years until the lag coefficient is statistically indistinguishable from zero.
  - Simple regressions: imports fall by about 11 percent in the crisis year and an additional 13 percent the following year; even 10 years after the crisis, imports are about 5 percent below predicted non-crisis levels.
  - Aggregate (collapsed) specification: aggregate imports fall by 9 percent in the year following the crisis, are 6 percent below predicted 5 years following, and recover to normal in 10 years.
- Exports:
  - On average, exports are 4 percent below predicted in the year of the crisis and return to normal within one year (summary statement).
  - Simple regressions: exports drop by only 3 percent on average at onset and recover within two years.
  - Gravity estimates: exports back to predicted level within one year; medium-run slight increases become insignificant once exchange rates are controlled for.
- Sector/product heterogeneity (imports):
  - Capital and consumer durables: average drop of 23 percent in the year after a crisis; durables imports remain 16 percent below normal after 5 years.
  - Intermediate products: imports remain 21 percent below predicted in the year after crisis, recovering only slowly.
  - Consumer non-durables: drop of 13 percent in the year following the crisis; 8 percent below predicted after 5 years.
  - Primary goods: least affected.
- Crises during global downturns:
  - Global downturn years identified: 1975, 1982, 1991, 2001, and 2008.
  - When crisis coincides with a global downturn, imports fall 28 percent below predicted in the year following the crisis—almost 10 percentage points higher than a crisis not accompanied by a downturn.
  - Imports remain almost 10 percent below normal even ten years after a crisis that coincides with a global downturn.
- Crises concurrent in both trading partners:
  - Five years following the crisis, imports from trading partners that also had a crisis are more than 30 percent below normal; imports from other countries are only 10 percent below normal.

### III. Methodology and estimation framework
- Core model:
  - Standard gravity model of bilateral (log) imports lnM_{i,j,t}, augmented with contemporaneous and lagged crisis indicators in the importer (C_{i,t-k}) and exporter (C_{j,t-k}), importer-exporter pair dummies (ψ_{ij}), time dummies (π_t), and time-varying pair controls X_{ij,t}.
  - Coefficients of interest: α_k (effect of importer crisis on imports k years after onset) and β_k (effect of exporter crisis on exports k years after onset).
  - Estimation in levels due to evidence of cointegration between imports and GDP.
  - Standard errors clustered at the importer*exporter level.
  - Number of lags chosen top-down; importer-crisis coefficients become insignificant after 10 lags.
- Model fit and controls:
  - Gravity model explains about 87 percent of variation in bilateral trade flows.
  - Controls include real GDP, population, distance, colonial ties, common border, language, legal origin, currency union and FTA indicators, exchange rates, domestic absorption, prices, and time-varying multilateral resistance terms.

### IV. Data sources, construction, and sample statistics
- Primary data sources:
  - Direction of Trade Statistics (DOTS), World Economic Outlook (WEO), International Financial Statistics (IFS), NBER-UN World Trade Flows (Feenstra and others, 2005), Laeven and Valencia (2008, 2010).
- Trade data and deflation:
  - Bilateral import and export flows from DOTS in current U.S. dollars deflated by world import/export price deflators from IFS to obtain real imports/exports.
  - Real GDP in U.S. dollars from WEO.
  - Product-level imports and exports from NBER-UN extended using UN Comtrade; SITC Rev. 2 matched to UN BEC and classified into Capital Goods, Consumer Durables, Consumer Non Durables, Intermediate Goods and Primary Goods.
- Summary statistics (exact values from Table A1):
  - Crisis in Importer: 372060 observations | Mean 0.034 | StDev 0.182 | Min 0 | Max 1
  - Crisis in Exporter: 372060 | Mean 0.034 | StDev 0.182 | Min 0 | Max 1
  - Log Trade: 372060 | Mean 1.009 | StDev 3.736 | Min -32.898 | Max 12.255
  - Log Importer GDP: 372060 | Mean 3.450 | StDev 2.160 | Min -2.800 | Max 9.376
  - Log Exporter GDP: 372060 | Mean 3.564 | StDev 2.112 | Min -2.800 | Max 9.376
  - Log Importer GDP Per Capita: 372060 | Mean -12.871 | StDev 1.642 | Min -16.315 | Max -9.776
  - Log Exporter GDP Per Capita: 372060 | Mean -12.819 | StDev 1.636 | Min -16.315 | Max -9.776
  - Log Distance: 291764 | Mean 4.071 | StDev 0.826 | Min 0.297 | Max 5.661
  - 1 if Common Border: 291764 | Mean 0.029 | StDev 0.168 | Min 0 | Max 1
  - 1 if Island: 372060 | Mean 0.010 | StDev 0.101 | Min 0 | Max 1
  - 1 if Landlock: 372060 | Mean 0.011 | StDev 0.102 | Min 0 | Max 1
  - 1 if Common Legal System: 291764 | Mean 0.359 | StDev 0.480 | Min 0 | Max 1
  - 1 if Common Language: 291764 | Mean 0.303 | StDev 0.460 | Min 0 | Max 1
  - 1 if Colonial Times: 291764 | Mean 0.018 | StDev 0.134 | Min 0 | Max 1
  - 1 if Currency Union: 372060 | Mean 0.014 | StDev 0.116 | Min 0 | Max 1
  - 1 if Free Trade Area: 372060 | Mean 0.048 | StDev 0.214 | Min 0 | Max 1
- Crisis distribution and characteristics:
  - Debt crises distribution: Latin America 34 percent; Sub-Saharan Africa 41 percent; two-thirds occurred during the 1980s.
  - Almost 7 percent of observations involve a contemporaneous crisis in either the importing or exporting country.
  - Of 153 countries in the main sample, about three-quarters (119) had a crisis at some point during the sample period.

### V. Robustness checks and alternative specifications (high-level outcomes)
- Fixed effects specifications:
  - Importer*exporter pair fixed effects and time dummies used; alternative with separate importer and exporter dummies yields similar findings.
- HMR two-step procedure (Helpman, Melitz and Rubinstein (2008)):
  - Crisis-dummy coefficients similar to baseline.
- Allowing GDP elasticity to vary (trend vs. cyclical, crisis vs. non-crisis):
  - Crisis-dummy coefficients almost identical to baseline; total effects evaluated at average sample GDP for crisis episodes similar to baseline.
- Controls and exclusions:
  - Controlling for domestic absorption instead of GDP: main findings robust.
  - Controlling for bilateral exchange rates and relative prices: import effects very similar; export effects smaller and mostly insignificant.
  - Excluding episodes accompanied by currency crises (“pure” financial crises): results similar to baseline (a quarter of financial crisis episodes accompanied by currency crisis in same or previous year; widening window to ±2 years yields roughly half accompanied by currency crisis).
  - Excluding recent crises (omitting 13 most recent banking crises since 2007): short-run import losses smaller but main findings hold (example: imports 7 percent and 17 percent below predicted in crisis year and following year respectively, vs. 9 percent and 19 percent in baseline).
  - Decadal multilateral resistance terms and collapsed aggregate gravity: crisis coefficients fall in magnitude under aggregation (about half) but remain statistically significant.
- Tariff-equivalent cost estimates (using σ = 5 or 10):
  - Tariff-equivalents for importer-crisis: between 1-2 percent in the year of the crisis, and in the range of 2-5 percent in the year following the crisis.
  - Persistent costs remain between 1-3 percent 5 years following the crisis.
- Exchange rates and volatility:
  - Average US-dollar exchange rate depreciates by 72 percent in the year following the crisis; takes about 8 years to get back to normal.
  - Exchange rate volatility (annual std. dev. of monthly changes) increases on average by almost half in the first two years following a crisis, then returns to normal.
- Extensive margin:
  - Number of imported varieties (product-country at 6-digit HS) declines by 7 percent in the year following crises and recovers slowly.

### VI. Endogeneity, feedback, and falsification checks
- Endogeneity concerns acknowledged; simple approaches applied:
  - Drop contemporaneous crisis episodes, augment with lags of GDP as predictors, control for lags of imports.
  - Results (Table 5 and Figure 7 in source) provide some additional light but are not definitive.
- Feedback and falsification:
  - Simple specification testing whether current or previous-year aggregate trade drives crisis occurrence finds little evidence of feedback effects.
  - Falsification replacing crisis variable with recession episodes: recession dummies have little effect; small 3-4 percent negative and statistically significant effect on imports in year of recession and following years, and no significant effect thereafter.
- Heterogeneity and severity:
  - Imports from small trading partners hit more following a crisis; imports from large trading partners more resilient (large defined as top 20 in terms of imports).
  - Import losses occur regardless of severity, with larger initial loss for severe crises (severe defined as output loss in first two years above the median).

### VII. Conclusions and policy-relevant implications
- Main conclusions:
  - Financial crises are associated with large and persistent losses in imports that are over and above import compression from lower output and standard gravity determinants.
  - Exports are less adversely affected; their behavior largely explained by standard gravity determinants.
  - Imports of consumer and capital durables are hit the most.
  - Episodes coinciding with global downturns and episodes concurrent with a crisis in partner countries have more adverse effects on trade.
  - Trade has tended to become more resilient over time (post-1990 episodes show faster recovery).
- Conjectured channels behind persistent import declines:
  - No strong evidence of large increases in tariffs or antidumping measures, though hidden “behind the border” protectionism cannot be ruled out.
  - Significant short-term rise in exchange rate volatility; extensive-margin adjustments (loss in imported varieties).
  - Composition effects left as unresolved hypotheses due to data limitations.
- Implications for the recent crisis:
  - Recent financial crisis concentrated in large advanced economies; findings consistent with sharp drop in import demand in countries with recent financial crises.
  - If past experience guides, imports of these countries are likely to remain below normal for a protracted period.

*Source: _wp1103 - Appendix 1. Data Sources; sections III–V, robustness checks, and Conclusions (IMF working paper content provided in the supplied PDF excerpt).*

### Appendix 1. Data Sources...............................................................................................5

### _wp1103 - Appendix 1. Data Sources...............................................................................................5

### I. Introduction and research questions
- Financial crises have been frequent since the mid-1970s and are associated with large and persistent declines in output, currency crises, asset market collapses, and rising government debt (Reinhart and Rogoff (2008, 2009a, 2009b); Cerra and Saxena (2008); IMF (2009); Kaminsky and Reinhart (1999)).
- This paper investigates how international trade evolves after a country experiences a banking or debt crisis, asking:
  - How has trade evolved following banking and debt crises?
  - Do such crises have lasting effects on a country’s imports and exports?
  - Do crises influence trade only through their effect on output and standard gravity determinants, or do they lead to other impediments to trade?

### II. Literature context and motivation
- Most existing work focuses on trade following global downturns, particularly the “Great Trade Collapse” of the 2008-09 global economic crisis (Baldwin, 2009).
- Freund (2009): decline in world trade following four previous global downturns was almost five times bigger than the corresponding decline in world GDP; world trade growth resumes quickly but pre-downturn levels of trade openness take more than three years to reach.
- Prior gravity-model studies of shocks:
  - Glick and Taylor (2010), Martin, Mayer, and Thoenig (2008): war’s effects on bilateral trade.
  - Blomberg and Hess (2006): violence raises tariff-equivalent costs of trade by between 7 and 17 percent.
- Two previous gravity-based studies of crises and trade:
  - Ma and Cheng (2003): 52 countries, 1981-1998; short-term effects up to two years; find negative impact of banking crises on imports and positive effect on exports (trade values).
  - Berman and Martin (2010): crises in trading partners reduce exports of partners, larger effect for African exporters.

### III. Empirical focus and sample
- Examines episodes of banking and debt crises over the past 40 years and tracks changes in imports and exports following crises.
- Distinguishes between global downturn effects and country-specific crisis effects; compares trade dynamics of countries with and without banking crises during the 2008-09 episode:
  - For countries without a banking crisis, real imports were on average back to their precrisis peak by Q1 2010.
  - For the thirteen countries identified by Laeven and Valencia (2010) as having had a systemic banking crisis from 2007 onwards, real imports were on average well below their precrisis peaks.

### IV. Main empirical findings (preserving reported magnitudes)
- Even after controlling for output and standard gravity controls:
  - Imports: In the year following a crisis, imports of the crisis country are 19 percent lower than the level predicted by the gravity framework; imports recover slowly, taking roughly 10 years to return to normal.
  - Exports: On average, exports are 4 percent below predicted in the year of the crisis and return to normal within one year.
- These patterns are robust across multiple alternative specifications and tests.

### V. Methodology and robustness checks
- Primary framework: gravity model of trade used to assess deviations from “normal” trade behavior following financial crises.
- Robustness and alternative approaches include:
  - Including exporter and importer fixed effects and their interaction to control for bilateral time-invariant characteristics.
  - Addressing omitted-variable and selection biases using the Helpman, Melitz and Rubinstein (2008) framework.
  - Allowing the elasticity of trade with respect to GDP to vary across cyclical and trend components of output and across crisis and non-crisis periods.
  - Introducing additional controls: exchange rates, domestic absorption, prices, and time-varying multilateral resistance terms.
  - Isolating episodes of “pure” financial crises not accompanied by large depreciations.
  - Looking at aggregate rather than bilateral trade patterns.
  - Conducting simple tests to address the potential endogeneity of crises; these tests support the main findings though they are not definitive.

### VI. Comparative notes and scope
- Focus differs from global downturn literature by examining trade of individual economies that experience banking or debt crises, providing a complement to studies of global trade collapses.
- The analysis uses trade volumes rather than trade values where explicitly noted (contrast with Ma and Cheng (2003), who analyze trade values).

*Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp1103.pdf*

### Section III describes the data, Section IV presents the empirical results and Section V

### _wp1103 - Section III describes the data, Section IV presents the empirical results and Section V

### Methodology
- Estimation framework: standard gravity model of international trade augmented with crisis indicators in the importer and exporter.
- Main estimating equation: bilateral (log) imports lnM_{i,j,t} regressed on:
  - importer-exporter pair dummies (ψ_{ij}) to control time-invariant pair characteristics and multilateral resistance,
  - time dummies (π_t),
  - contemporaneous and lagged dummy variables indicating whether a crisis started in period k≤t in the importer (C_{i,t-k}) or exporter (C_{j,t-k}),
  - time-varying importer-exporter controls X_{ij,t} (e.g., currency union, FTA).
- Estimation in levels (not differences) due to strong evidence of cointegration between imports and GDP.
- Coefficients of primary interest:
  - α_k: average effect of a crisis in the importer on its imports from a given exporter k years after crisis onset.
  - β_k: average effect of a crisis in the exporter on its exports to a given importer k years after crisis onset.
- Standard errors clustered at the importer*exporter level.
- Number of lags chosen via a top-down approach; importer-crisis coefficients become insignificant after 10 lags.

### Data
- Sample coverage:
  - 153 advanced, emerging, and developing economies.
  - Period: 1970–2009.
- Trade data:
  - Bilateral import and export flows from IMF DOTS, reported in current U.S. dollars and deflated by world import/export price deflators from IFS to obtain real imports/exports.
- Crisis dates:
  - From Laeven and Valencia (2008, 2010).
  - 129 episodes of systemic banking crises since 1970.
  - 60 episodes of sovereign debt crises since 1970.
  - 10 cases where banking and debt crises coincide (“twin banking and debt crises”).
  - 13 episodes of banking crisis during 2007-08 concentrated in advanced economies.
  - Debt crises distribution: Latin America 34 percent, Sub-Saharan Africa 41 percent; two-thirds occurred during the 1980s.
- Sample statistics:
  - Full sample contains 372,060 bilateral importer*exporter*year observations.
  - 20,003 importer*exporter pairs involving 153 countries.
  - Of 153 countries, about three-quarters (119) had a crisis at some point during the sample period.
  - Almost 7 percent of observations involve a contemporaneous crisis in either the importing or exporting country.
- Other covariates:
  - Real GDP, population, country-pair characteristics (distance, colonial ties, common border, language, legal origin), currency union and FTA indicators.
  - Data sources: WEO for GDP, Glick and Taylor (2010) for pair variables, Glick and Rose (2002) extended to 2009 for currency union/FTA.

### First look at the data (simple regressions)
- Imports:
  - On average, imports fall by about 11 percent in the crisis year.
  - Additional drop of about 13 percent in the following year.
  - Even 10 years after the crisis, imports are about 5 percent below predicted non-crisis levels.
- Exports:
  - Smaller and often statistically indistinguishable from zero.
  - Exports drop by only 3 percent on average at onset and recover quickly within two years.

### Main empirical findings (augmented gravity estimates)
- Model fit and controls:
  - Gravity model explains about 87 percent of variation in bilateral trade flows.
  - Coefficients on GDP, currency union, FTA are plausible and similar to literature.
- Import effects (gravity estimates, equation (1)):
  - Estimated contemporaneous importer-crisis coefficients negative and highly significant (1 percent level; tenth lag significant at 5 percent).
  - On average, imports fall by 9 percent below gravity-predicted level in the year of the crisis, and by a further 10 percent in the following year.
  - Five years after a crisis, imports remain 10 percent below normal.
  - It takes more than 10 years for imports to return to normal (precisely, 11 years until the lag coefficient is statistically indistinguishable from zero).
- Export effects:
  - Effects on exports much smaller and often insignificant.
  - Small drop in exports in crisis year (about 3 percent in simple regressions), rapid recovery—exports back to predicted level within one year in gravity estimates.
  - In some medium-run estimates exports hover slightly above gravity-predicted levels, but this becomes statistically indistinguishable from zero once exchange rates are controlled for.
- Comparative magnitudes:
  - Contemporaneous import decline from crisis comparable to effect of violence on trade: terrorist incident ~5 percent decline; revolutions ~19 percent; interethnic conflicts ~15 percent (Blomberg and Hess, 2006).
  - Trade decline for neutrals at onset of war ~12 percent and persistent up to seven years—magnitude similar to financial crisis effects on imports.
- Tariff-equivalent cost of crisis:
  - Using elasticity of substitution σ = 5 or 10 (Feenstra (2002), Blomberg and Hess (2006)):
    - Tariff-equivalents for importer-crisis: between 1-2 percent in the year of the crisis, and in the range of 2-5 percent in the year following the crisis.
    - Persistent costs remain between 1-3 percent 5 years following the crisis.
- Exchange rate and prices:
  - Average US-dollar exchange rate depreciates by 72 percent in the year following the crisis; takes about 8 years to get back to normal.
  - Rise in exports in medium run is partly explained by accompanying exchange rate changes.
- Trade policy and other channels:
  - No evidence of increase in average tariffs following crises.
  - Small increase in number of antidumping measures: roughly 2 percent increase in the year following the crisis, then slow decrease.
  - Exchange rate volatility (annual std. dev. of monthly changes) increases substantially: on average increases by almost half in the first two years following a crisis, then returns to normal.
  - Extensive margin adjustments: number of imported varieties (product-country at 6-digit HS) declines by 7 percent in the year following crises and recovers slowly.

### Robustness checks and alternative specifications
- Alternative gravity specification with separate importer and exporter dummies (instead of importer*exporter fixed effects):
  - Main findings robust: imports fall substantially and persistently; exports hurt less and recover quickly.
- HMR two-step procedure (addressing zero trade flows and asymmetries; probit first stage with common religion as exclusion):
  - Crisis-dummy coefficients similar to baseline.
- Allowing elasticity of imports w.r.t. GDP to vary (trend vs. cyclical components via HP filter):
  - Crisis-dummy coefficients almost identical to baseline.
- Allowing GDP elasticity to differ during crisis vs. non-crisis periods:
  - Total effect of crisis on imports and exports similar to baseline when evaluated at average sample GDP for crisis episodes.
- Controlling for domestic absorption (consumption + investment) in importer instead of GDP:
  - Main findings robust.
- Controlling for bilateral exchange rates and relative prices (GDP deflator):
  - Import effects very similar to baseline.
  - Export effects smaller and mostly insignificant apart from crisis year.
- Excluding episodes accompanied by currency crises (pure financial crises):
  - Quarter of financial crisis episodes accompanied by currency crisis in same or previous year; widening window to ±2 years yields roughly half accompanied by currency crisis.
  - Results for “pure” financial crises similar to baseline.
- Excluding recent crises (restricting dataset to 2006 to omit 13 most recent banking crises):
  - Short-run import losses smaller when omitting recent episodes (example: imports 7 percent and 17 percent below predicted in crisis year and following year respectively, vs. 9 percent and 19 percent in baseline), but main findings still hold.
- Allowing multilateral resistance terms to vary by decade (ψ_{ij} by decade):
  - Results qualitatively similar to baseline.
- Collapsed (aggregate) gravity specification (country-level aggregated imports/exports across partners, partner-weighted covariates):
  - Crisis coefficients fall to about half the magnitude relative to baseline but remain statistically significant.
  - Aggregate imports fall by 9 percent in the year following the crisis, 6 percent below predicted 5 years following, and recover to normal in 10 years.
  - Aggregate exports do not deviate significantly from normal in short or medium term.
- Summary: multiple robustness tests confirm persistent decline in imports following financial crises and muted effects on exports.

### Endogeneity and reverse causality concerns
- Potential biases acknowledged: occurrence of a crisis may be affected by trade behavior; underlying factors may affect both crisis likelihood and trade.
- Approaches to explore endogeneity (following Cerra and Saxena (2008)):
  - Drop contemporaneous crisis episodes more likely endogenous to trade.
  - Augment main equation with lags of GDP as predictors of a crisis.
  - Control for lags of imports to address reverse causality.
- Results of these checks presented in Table 5 (columns 2-4) and Figure 7 (not reproduced here); authors state results provide some additional light but are not definitive with respect to endogeneity.

*Source: IMF working paper content provided in the supplied PDF excerpt.*

### 8. The estimated coefficients on the crisis indicators are almost identical to the baseline

### 8. The estimated coefficients on the crisis indicators are almost identical to the baseline

### Robustness and endogeneity checks
- The total effect of a crisis in the specification with lags of imports includes both direct effects (estimated coefficients in Table 5) and indirect effects through the dynamics of the model.
- Tests for heterogeneity in partner-size response:
  - Imports from small trading partners are hit more following a crisis.
  - Imports from large trading partners seem to be more resilient to a crisis (large trading partners defined as top 20 in terms of imports).
- Severity test:
  - Import losses occur regardless of whether a financial crisis is severe or moderate, although the initial import loss is larger for severe crises (severe crises defined as episodes where the size of the output loss in the first two years of the crisis is above the median).
- Bias from fixed effects and lagged dependent variables:
  - Nickell (1981) shows that the order of the bias is 1/T.
  - Judson and Owne (1999) show that the bias of the fixed effect estimator is approximately 2-3 percent on the lagged dependent variable and less than 1 percent on other regressors for a panel of size N=100, T=30, and low persistence.
- Feedback test:
  - A simple specification testing whether current or previous-year aggregate trade drives crisis occurrence finds little evidence of feedback effects.
- Falsification test replacing crisis variable with recession episodes:
  - Recession dummies generally have little effect on imports and exports.
  - There is a small (3-4 percent) negative and statistically significant effect on imports in the year of the recession and the following years, and no significant effect thereafter.
- Summary conclusion:
  - The simple controls for endogeneity presented (Table 5 and Figure 8) are far from definite but show a picture consistent with baseline results.
  - Estimates could capture the effect of the crisis on imports and exports or any underlying factor that may have led to it.

### Differences in trade dynamics across products
- Method:
  - Product-level gravity regression with four product categories: consumer nondurables, capital and consumer durables, intermediate goods, and primary goods; product fixed effects and standard gravity controls varying by product category.
- Key findings on imports:
  - Capital and consumer durables:
    - Average drop of 23 percent in the year after a crisis.
    - Durables imports remain 16 percent below normal after 5 years.
  - Intermediate products:
    - Imports remain 21 percent below predicted in the year after crisis, recovering only slowly in the medium-term.
  - Consumer non-durables:
    - Drop of 13 percent in the year following the crisis.
    - 8 percent below predicted after 5 years.
  - Primary goods:
    - Least affected by a crisis.
- Key findings on exports:
  - Exports of all product categories are much less affected than imports, with all categories except primary goods recovering in the year after the crisis.
  - Capital and consumer durables experience a sharp decline in the year of the crisis (5 percent), but recover quickly.
  - Decline in exports of primary goods is marginally smaller on impact but more persistent—recovering to normal 3 years after the crisis.

### Are postcrisis trade dynamics different during global downturns?
- Design:
  - Interact contemporaneous and lagged crisis dummies with indicators of whether the year of the episode coincided with a global downturn (global downturn years defined as in Freund (2009)).
  - About a fifth of the crisis episodes in the sample occurred during global downturns.
  - Global downturn years identified by the procedure: 1975, 1982, 1991, 2001, and 2008.
- Findings:
  - Crises during global downturns produce deeper import and export losses, conditional on standard gravity controls.
  - Imports of a crisis country fall 28 percent below predicted in the year following the crisis when the crisis coincides with a global downturn—almost 10 percentage points higher than a crisis episode not accompanied by a downturn.
  - The decline is much more persistent if the crisis is accompanied by a global downturn; imports remain almost 10 percent below normal even ten years after a crisis.
  - Exports also fall more and take longer to recover for crises accompanied by global downturns, but exports remain much less dramatic compared to imports and frequently are statistically indistinguishable from zero.
- Interpretation:
  - Recent financial crises that coincided with global downturns may result in deeper trade losses than historical episodes that did not coincide with global downturns.

### Crises concurrent in both trading partners
- Specification adds an indicator equal to one if the crisis occurs in both the importer and exporter.
- Findings:
  - Trade decline between two countries that both had a crisis is disproportionately more severe and more persistent.
  - Five years following the crisis, imports from trading partners that also had a crisis are more than 30 percent below normal.
  - In comparison, imports from other countries are only 10 percent below normal (as shown in Figure 4).

### Do postcrisis trade dynamics vary over time?
- Design:
  - Interact crisis dummies with an indicator for whether the episode occurred after 1990 to compare pre- and post-1990 crises.
- Findings:
  - Post-1990 episodes:
    - Imports are 16 percent below normal in the year following a crisis.
    - Imports are almost back to normal levels just three years after the crisis.
    - Export behavior in post-1990 episodes seems to be completely explained by standard gravity controls.
  - Pre-1990 episodes:
    - Maximum decline in imports as much as 21 percent.
    - Imports still 6 percent below predicted after 10 years.
    - Export losses are typically higher for pre-1990 episodes.
- Interpretation:
  - Evidence suggests trade has become more resilient over time, consistent with greater production sharing and strengthened production linkages (a “beachhead effect”).

*Source: 8. The estimated coefficients on the crisis indicators are almost identical to the baseline*

### CONCLUSIONS

### CONCLUSIONS

### Main findings on trade dynamics following financial crises
- Using bilateral trade data for a large set of countries over the period 1970-2009, the paper provides evidence that financial crises are associated with large and persistent losses in imports.
- The import losses are over and above any import compression due to lower output and changes in other standard determinants of trade flows as a result of a crisis.
- Exports are not as adversely affected; their behavior can be explained by standard gravity determinants.
- Imports of consumer and capital durables are hit the most.
- Episodes coinciding with global downturns and episodes concurrent with a crisis in partner countries have more adverse effects on trade.
- Trade has tended to become more resilient over time.

### Conjectured channels behind persistent import declines
- Rise in protectionism following past crises:
  - The authors do not find much evidence for a large increase in tariffs or antidumping measures following past episodes, but cannot rule out a rise in more hidden forms of protectionism that are difficult to quantify (see Gregory et al. (2010) for discussion of “behind the border” measures).
- Rise in exchange rate volatility:
  - There is a significant increase in the volatility of the exchange rate in the short term that declines over the medium term.
- Adjustment on the extensive margin:
  - Evidence suggests the fall in imports after a crisis reflects a reduction in the number of imported varieties.

### Limits of the analysis and unresolved hypotheses
- The paper cannot test whether “composition effects” (large drops in demand for goods that constitute a larger share of trade than of output) explain trade dynamics in past financial crises because of the lack of comprehensive historical data on the composition of demand for various categories of goods.
  - Eaton et al (2010) and Levchenko et al. (2010) find that composition effects can account for the bulk of the disproportionate decline in trade relative to output in the 2008-09 global downturn.
- Establishing the full mechanisms behind the persistent import losses is left for future research.

### Implications for the recent crisis (as discussed in the paper)
- Although caution is advised when extrapolating from historical episodes to the recent crisis given its severity and global nature, the paper’s findings shed light on likely trade developments.
- The recent financial crisis has been concentrated in many large, advanced economies that account or almost of global demand.
- The findings are consistent with the sharp and substantial drop in import demand that occurred in countries that recently went through a financial crisis.
- If past experience is any guide, imports of these countries are likely to remain below normal—i.e., below where they would have been in the absence of a crisis—for a protracted period.

*Source: _wp1103 - CONCLUSIONS (PDF chapter/section) — content covering conclusions, tables and figures, 1970-2009 analysis.*

### Appendix 1. Data Sources

### Appendix 1. Data Sources

### Primary data sources
- The primary data sources for the chapter are the IMF’s Direction of Trade Statistics (DOTS), World Economic Outlook (WEO) and International Financial Statistics (IFS) databases, the NBER-UN World Trade Flows database (2005), and Laeven and Valencia (2008, 2010).
- Additional data sources are listed in Table A3.

### Crisis indicators and episodes
- Crisis indicators are from Laeven and Valencia (2008, 2010).
- Laeven and Valencia (2010) present new and comprehensive data on the starting dates and characteristics of systemic banking crises over the period 1970-2009, building on earlier work by Caprio et al. (2005), Laeven and Valencia (2008), and Reinhart and Rogoff (2009a).
- They update the Laeven and Valencia (2008) database on systemic banking crises to include the recent episodes following the U.S. mortgage crisis of 2007, and identify 129 episodes since 1970.
- Laeven and Valencia (2008) also identify debt crisis episodes based on sovereign debt default and restructuring by relying on information from Beim and Calomiris (2001), World Bank (2002), Sturzenegger and Zettelmeyer (2006), and IMF Staff reports.
- Using this approach, they identify 60 episodes of sovereign debt defaults and restructurings since 1970.

### Trade data, deflation, and partner-weighted measures
- Data on bilateral and aggregate imports and exports from the DOTS database are reported in current U.S. dollars.
- These are deflated using the world import and export price deflators, respectively, from the IFS database, to get each economy’s real imports and exports.
- The series on real GDP in U.S. dollars is from the WEO database.
- Import- and export-weighted partner GDP and GDP per capita are constructed using real GDP in U.S. dollars and import and export weights from the DOTS database.
- These weights vary each year based on the actual import and export flows between economies.

### Product-level imports and classification
- Data on imports and exports by product category are constructed from the NBER-UN World Trade Flows database (see Feenstra and others, 2005).
- The database is first extended using the UN Comtrade database.
- The Standard International Trade Classification, Revision 2 (SITC Rev. 2) codes that identify products in the NBER-UN trade data are matched to the UN Broad Economic Classification (BEC) codes.
- These are then classified into Capital Goods, Consumer Durables, Consumer Non Durables, Intermediate Goods and Primary Goods, following Pula and Peltonen (2009).

### Trade liberalization and other measures
- The measure of trade liberalization is from the IMF Structural Reforms Database and is described in IMF (2008).

### Time-series properties of imports and GDP (Appendix 2)
- Objective: examine time-series properties of real imports and real GDP using techniques for non-stationary macro panel data allowing for country fixed effects, heterogeneous trends, and common time effects.
- Unit root tests applied:
  - (i) Im, Pesaran and Shin augmented Dickey-Fuller test, adjusted for small sample size distortions using bootstrapping techniques (Im, Pesaran, and Shin, 2003, and Pedroni and Yao, 2006).
  - (ii) Bai and Ng (2004) test.
  - (iii) Pesaran (2007) cross-sectionally augmented Dickey-Fuller (CADF) test.
- Test findings:
  - The results reported in Table A2 show that most of the tests fail to reject the null of unit root in both real imports and real GDP.
- Cointegration tests:
  - Tests of cointegration between real imports and real GDP as suggested by Pedroni (1999, 2004) were conducted.
  - The tests treat all parameters as heterogeneous across members of the panel, and allow for both heterogeneous dynamics and heterogeneous cointegrating vectors, as well as complete endogeneity.
  - Four test statistics reported (bottom panel of Table A2): (i) pooled Phillips-Perron t-statistics, (ii) pooled ADF t-statistics, (iii) group-mean Phillips-Perron t-statistics and (iv) group mean ADF t-statistics.
  - All four test statistics reject the null of no cointegration.
- Implication:
  - Given the evidence for cointegration, the empirical specification is estimated in levels.
  - Under cointegration, standard panel techniques produce superconsistent estimates of slope coefficients; precise estimates are attainable even in small panels and even if regressors are endogenous.

### Empirical tests and sample details (from Table A2)
- Unit Root Tests (Null=Unit Root, Large negative values imply rejection):
  - Bootstrapped Im, Pesaran & Shin ADF statistic: Ln (imports) 0.10; Ln (real GDP) 0.19; 1% critical values -1.28; 5% critical values -1.64.
  - Bai-Ng (2004) with orthogonalized data: Im, Pesaran & Shin ADF statistic 5.73 for Ln (imports) and -1.86 for Ln (real GDP) (table layout indicates these values).
  - Pesaran (2007) cross-sectionally augmented ADF statistic: Ln (imports) -2.26; Ln (real GDP) -2.12; 1% critical values -5.01; 5% critical values -4.01 (table presents these critical values in context).
- Cointegration Tests (Null=No Cointegration, Large negative values imply rejection):
  - Pooled Phillips-Perron statistic: -5.98.
  - Pooled ADF statistic: -1.93.
  - Group mean Phillips-Perron statistic: -7.60.
  - Group mean ADF statistic: -3.84.
  - Critical values listed as 1% -1.28 and 5% -1.64 in the table context.
- Panel dimensions:
  - Number of countries 179.
  - Number of periods 40.
- Notes:
  - Countries with less than 10 years of data are dropped.
  - All reported test statistics (except for Pesaran (2007) test statistic) are distributed N(0,1) under null of unit root or no cointegration.
  - The reported values of test statistics shown in the table are calculated assuming 3 lags. The values are similar if 4, 5 or 6 lags are used.

### Summary statistics (from Table A1; variables preserved exactly)
- Table A1. Summary Statistics to Main Variables (columns: Variable | Nobs | Mean | StDev | Min | Max)
  - Crisis in Importer: 372060 | 0.034 | 0.182 | 0 | 1
  - Crisis in Exporter: 372060 | 0.034 | 0.182 | 0 | 1
  - Log Trade: 372060 | 1.009 | 3.736 | -32.898 | 12.255
  - Log Importer GDP: 372060 | 3.450 | 2.160 | -2.800 | 9.376
  - Log Exporter GDP: 372060 | 3.564 | 2.112 | -2.800 | 9.376
  - Log Importer GDP Per Capita: 372060 | -12.871 | 1.642 | -16.315 | -9.776
  - Log Exporter GDP Per Capita: 372060 | -12.819 | 1.636 | -16.315 | -9.776
  - Log Distance: 291764 | 4.071 | 0.826 | 0.297 | 5.661
  - 1 if Common Border: 291764 | 0.029 | 0.168 | 0 | 1
  - 1 if Island: 372060 | 0.010 | 0.101 | 0 | 1
  - 1 if Landlock: 372060 | 0.011 | 0.102 | 0 | 1
  - 1 if Common Legal System: 291764 | 0.359 | 0.480 | 0 | 1
  - 1 if Common Language: 291764 | 0.303 | 0.460 | 0 | 1
  - 1 if Colonial Times: 291764 | 0.018 | 0.134 | 0 | 1
  - 1 if Currency Union: 372060 | 0.014 | 0.116 | 0 | 1
  - 1 if Free Trade Area: 372060 | 0.048 | 0.214 | 0 | 1

### Data sources listing (from Table A3)
- Annual Data:
  - Real Exports and Imports: Direction of Trade Statistics (DOTS) Database
  - Real GDP in U.S. Dollars: World Economic Outlook (WEO) Database
  - Real GDP per Capita in U.S. Dollars: WEO Database
  - World Import/Export Price Deflator: International Financial Statistics (IFS) Database
  - Bilateral exchange rate: IMF
  - Real Effective Exchange Rate: IMF
  - Product-Level Imports and Exports: Feenstra and others (2005), COMTRADE, Pula and Peltonen (2009)
  - Trade Liberalization: IMF
  - Debt Crisis Indicators: Laeven and Valencia (2008)
  - Banking Crisis Indicators: Laeven and Valencia (2010)
- High-Frequency Data:
  - Real Exports and Imports: CPB Netherlands Bureau of Economic Policy Analysis, DOTS Database, Global Trade Atlas, Haver Analytics
  - Antidumping Data: Bown (2010)
  - World Import/Export Price Deflator: IFS Database, CPB Netherlands Bureau of Economic Policy Analysis

*Source: Appendix 1. Data Sources (content unit).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2011/_wp1103.pdf_