## Section 6 concludes.

## Source details

**Canonical URL:** [Section 6 concludes.](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021233-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021233-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021233-print-pdf.pdf.json)

---

### Sovereign Debt Re-negotiations in the 1980s: Overview and Context
- Total outstanding debt of Latin American borrowers: from less than $30 billion in 1970 to more than $320 billion in 1982.
- In 1982, outstanding loans to developing countries by the largest nine US banks were nearly three times the capital of these banks; loans to Latin America were close to 180% of the banks’ capital.
- Most loans denominated in US dollars with floating interest rates linked to US short-term rates.
- Latin American debt crisis commonly dated to August 13-15, 1982: Mexico informed IMF and US Treasury it could no longer service its $86 billion of external debt.
- Defaults/reschedulings:
  - 26 countries “defaulted” between 1982 and 1983 and another 29 in the rest of the 1980s.
  - Out of 23 countries in the region with more than one million inhabitants, 22 “defaulted” between 1980 and 1989; Colombia was the only large country that did not “default” in this period.
- Note on terminology: “default” used in quotation marks because many countries continued servicing interest while entering formal debt rescheduling agreements; these reschedulings often implied large NPV losses for creditors, making them de facto defaults.

### The Three Phases of the Latin American Debt Crisis
- Phase 1 (1982-85):
  - Crisis seen as temporary liquidity problem.
  - Policy: fiscal adjustment and coordinated reprofiling of principal repayments conditional on IMF programs; objective to keep countries current on interest.
  - Region current account: from a 5% deficit of region’s GDP in 1982 to a surplus of nearly 1% in 1984.
  - First-round terms: spreads over LIBOR of about 2%-2.5%, maturities of 6-8 years, large upfront cash commissions.
  - Restructuring in this phase increased the present value of developing countries’ debt.
- Phase 2 (1985-89) — Baker Plan era:
  - Commitments to substantial “new” money and greater involvement of World Bank and regional development banks.
  - Terms: spreads over LIBOR lowered to the order of 0.8%-0.9%, extended amortization periods, no commissions.
  - Market-based menu approach (debt buybacks, exit bonds, debt swaps) led to modest reduction in present value of debts.
- Phase 3 (Brady Plan, announced March 1989):
  - Focus on debt reduction with direct official support.
  - Four key elements:
    - Use of zero-coupon US Treasury bonds to collateralize newly issued bonds (purchase financed with IFI and bilateral loans).
    - Continuation of debt buybacks.
    - Waiver by commercial banks of legal clauses hampering restructuring (e.g., “negative pledge” clauses).
    - Change in IMF policy preventing the Fund from lending into arrears.
  - Over 1990-1998, 11 countries implemented Brady exchanges.
  - Brady plan helped end the 1980s debt crisis and aided emerging markets’ return to international markets via global bonds.

### The Strange Case of Colombia’s Non-Default — Positioning and Actions
- Pre-crisis indicators (circa 1981):
  - External debt-to-GDP: Colombia had four countries with lower debt levels and two large countries (Mexico and Brazil) with debt levels only slightly higher (Mexico 6 percentage points of GDP higher; Brazil 8 points higher).
  - Primary fiscal balance over 1978-81: close to region’s median.
  - GDP growth and inflation: close to region’s median.
  - Current account: better than region’s median but not among top four performers.
- Interest spread: Colombia’s spread over the US prime rate slightly higher than Venezuela and Mexico and just below Chile and Ecuador.
- Two stronger dimensions for Colombia entering the crisis:
  - Reserves-to-imports ratio in 1981-82: about 9 months versus regional average of about 3.5 months (about 2.5 times the average of Argentina, Brazil, Chile, Mexico, Peru, and Venezuela).
  - Composition of external debt: lower levels of debt with private creditors and higher levels with official creditors (especially World Bank and Inter-American Development Bank).
- Deterioration in early 1980s:
  - Net capital inflows collapsed during 1982-83, though less than other large economies.
  - From second half of 1982, Colombia was unable to access international capital markets.
  - Balance of payments: surpluses of about 3% of GDP over 1976-80 → zero in 1981 → current account deficit of 10% of GDP in 1982.
  - International reserves fell by 15% in 1982; reserves fell by $1.6 billion in 1983 (from $4.9 to $3.3 billion) and again by $1.6 billion in 1984, for a total loss from $4.9 billion in 1982 to $1.7 billion in 1984 (two thirds of reserves lost over two years).
  - Debt service absorbed nearly 40% of export revenues in 1983 (up from 15% in 1980).
  - Central Bank extended emergency assistance to nine financial institutions in second half of 1982; government monetization increased central bank balance sheet; 5-day state of economic emergency declared at end-1982 (most measures invalidated by Supreme Court).
  - During 1983 Article IV consultation, authorities feared reserve losses beyond $1 billion could cause market panic; reserves fell by $1.6 billion in 1983.

### Colombia’s Renegotiation Strategy and Key Statistics (1985–1991 Arrangements)
- By mid-1984 Colombia was unable to service external debt but avoided suspending payments; negotiated refinancing of majority of payments due between 1985 and 1994 to preserve reputation as a “good debtor.”
- IMF staff view: strategy to avoid formal rescheduling and maintain exposure of commercial banks and multilateral institutions roughly matching amortization payments as they fell due.
- Four negotiated arrangements (1985–1991):
  - “Jumbo” arrangement of 1985: $1 billion
  - “Concorde” arrangement of 1987: $1 billion
  - “Challenger” arrangement of 1989: $1.5 billion
  - “Hercules” arrangement of 1991: $1.8 billion
- Typical terms across arrangements:
  - Maturities between 9 and 12 years.
  - Grace periods between 3.4 and 6.2 years.
  - Stated spreads over LIBOR between 87 and 150 basis points.
  - Effective average costs reached 193 basis points over LIBOR.
  - These conditions were substantially worse than the average interest rate in 1982, which was 1.1% over LIBOR.
- Secondary market evidence of creditor NPV losses:
  - Between 1987 and 1990, Colombian loans traded between 58 and 64 cents on the dollar.
  - Colombian loans were the highest-valued in the region during the 1980s and second only to Chile in the first half of the 1990s.
- Domestic politics:
  - Opposition argued arrangements were “excessively” creditor friendly and suggested outright rescheduling similar to regional peers.
  - Authorities maintained strategy offered better prospects for subsequent voluntary lending.
- Additional measures:
  - 1984 Central Bank Resolution 33 set rescheduling parameters for non-financial private sector firms: minimum 6 years including 3-year grace period and maximum interest spread over LIBOR of 2.5% to obtain liquidity support.

- Selected detailed terms from arrangements (excerpted data):
  - 3.4 years LIBOR +150 bps for the first 4 years, LIBOR+138bps (1+3/8) for the remaining 6 years. Effective average cost: LIBOR+1.93
  - Concorde (Jan. 1988, $1 billion corresponding to 70% of payments due to banks in 1987-88): 8.4 years, quarterly disbursements conditional on IMF monitoring; planned for 1987-88 but signed and disbursed in 1988.
  - Challenger (Jun. 1989): Syndicate Loan of $1.47 billions and facility of $185 millions. 1989-90 (first disbursement October 1989):
    - 6 years for the syndicated loan and 5 years for the facility (average 5 years).
    - LIBOR+87bps (7/8) for the syndicate loan and LIBOR+93bps (15/16) for the facility. Effective average cost: LIBOR+1.11
    - 12 years for the syndicated loan and 10 years for the facility.
  - Hercules (Dec. 1990): Syndicate Loan of $1.575 billion & facility of $200 million (corresponding to 90% of principal payments coming due over 1991-94). 1991-94:
    - 6.2 years LIBOR+100 bps for the syndicate loan & LIBOR+150bps for the facility. Effective average cost: LIBOR+1.24
    - 12.6 years.

### Political Economy and IMF Involvement
- Authorities emphasized maintaining reputation as a “good debtor” (Garay, 1991).
- Colombia considered an IMF program politically unviable due to past disagreements; in 1967 Colombia decided to remain independent from the Fund (Junguito, 1985, p. 75).
- IMF involvement despite political constraints:
  - World Bank lending monitored in part by IMF via a “shadow program.”
  - Private creditors’ “advisory” committees requested access to IMF Article IV staff reports and enhanced surveillance (IMF “enhanced” surveillance) to certify and monitor Colombia’s adjustment program as if a stand-by arrangement were in place.
- Enhanced surveillance:
  - Met reservations on IMF Executive Board but approved with strong support from US Federal Reserve, IMF Managing Director, and US Executive Director Charles Dallara.
  - Declassified IMF Board minutes (July 26, 1985) justify Fund accepting proposed monitoring role.
  - US political support partly linked to cooperation on drug traffic control and broader geopolitical considerations; Paul Volcker advocated tailored approaches.
- Hercules arrangement lasted until 1994. By 1992 Colombia regained market access and began issuing dollar bonds to repay higher-rate World Bank loans. By mid-1993 secondary market discounts dropped sharply. Colombia imposed controls to slow excessive private inflows in September 1993 and tightened restrictions twice in 1994.

### Estimating Colombia’s Default Probability — Data and Methodology
- Raw data: 87 countries over 1976-2017, including 77 default episodes.
- Data sources: vintages of World Bank’s World Development Indicators and International Debt Statistics; default episodes from updated Asonuma and Trebesch (2016) dataset; only first episode retained when restructurings were consecutive.
- Candidate predictors: 22 variables covering solvency, liquidity, volatility, macroeconomic performance, political and institutional quality (see Appendix Table A6).
- Two-step estimation:
  - Step 1: Variable selection via LASSO (logistic LASSO) with 5-fold cross-validation.
    - Cross-validation selects hyper-parameter λ = 0.356.
    - LASSO selects 17 variables (listed in Appendix Table A7).
    - High interest payments on external debt and large current account deficits emerge as best predictors of sovereign defaults.
  - Step 2: Logit estimation using selected variables to predict default probability and compare Colombia in the 1980s to other Latin American defaulters.
- LASSO-logit estimator specified as:
  - β_L = arg max_β ∑_{i=1}^N [y_i X_i β − ln(1 + e^{X_i β})] − λ ∑_{j=1}^p |β_j|

### Predicted Default Probabilities — Summary Statistics for Actual Defaulters (Table 3)
- All Defaulters:
  - Mean: 16.58
  - Median: 6.61
  - St. Dev.: 22.76
  - Min.: 0.16
  - Max.: 99.47
  - p10: 1.01
  - p90: 35.04
- LAC Defaulters:
  - Mean: 20.74
  - Median: 7.38
  - St. Dev.: 27.98
  - Min.: 0.67
  - Max.: 99.47
  - p10: 0.93
  - p90: 81.78
- LAC Defaulters 1980-85:
  - Mean: 26.45
  - Median: 19.59
  - St. Dev.: 31.01
  - Min.: 0.93
  - Max.: 99.47
  - p10: 3.58
  - p90: 81.78
- Note: Probabilities based on Logit model of Table A7; summary statistics refer to estimated probability in the year of actual default for those countries.

### Colombia’s Predicted Default Probability (1979–86) — Key Results
- Predicted probability for Colombia:
  - 3% in 1982
  - 14% in 1983
  - Ranged between 14% and 20% over 1984-86
- Comparisons:
  - Between 1983 and 1986 Colombia’s estimated probability was never significantly lower than mean and median values for actual defaulters.
  - In 1983, 1984, and 1986 Colombia’s estimated probability was significantly higher than the median for all defaulting countries in their default year.
  - Cross-country (1983-86): Panama, Bolivia, Ecuador, Uruguay, Costa Rica, and Dominican Republic had significantly lower predicted probabilities than Colombia; Venezuela, Jamaica, Mexico, Argentina, and Peru were not significantly different; Brazil and Chile displayed significantly higher predicted probabilities than Colombia in 1983-86.

### Empirical Implications and Next-Step Analysis
- LASSO selection and logit prediction indicate Colombia’s 1980s fundamentals were similar to Latin American countries that defaulted.
- Authors proceed to counterfactual analysis using:
  - Synthetic control method (SCM) (Abadie and Gardeazabal, 2003).
  - Synthetic difference-in-differences method (SDID) (Arkhangelsky, Athey, Hirshberg, Imbens, and Wager, 2020).
- SDID generalizes SCM by accounting for unobservable time-invariant factors and common shocks to improve identification.

### Synthetic Control (Methodology and Implementation)
- Treatment: a “non-event” — Colombia’s non-default. Treatment date: 1981. Estimation horizon up to 1985.
- Potential outcomes notation: Yj,t(NDj) = Yj,t(0) if NDj = 0; Yj,t(1) if NDj = 1.
- Synthetic control estimator:
  - ˆY1,t(0) = sum_{j=2}^{n+1} wj Yj,t
  - ˆτ1,t = Y1,t(1) − ˆY1,t(0)
- Inference: placebo/permutation test framework; test statistic is ratio of mean squared prediction errors (RMSPE):
  - RMSPE_j = (∑_{t=T0+1}^{T} (Yj,t − ˆYj,t(0))^2 /(T−T0)) / (∑_{t=0}^{T0} (Yj,t − ˆYj,t(0))^2 / T0)
- Donor pool and predictors:
  - Variables with counterfactuals: real GDP, inflation, exports, imports.
  - Bolivia and Jamaica excluded (defaulted before 1981).
  - Baseline matching: match pre-treatment outcomes on their lagged values only.
- Goodness-of-fit metrics:
  - Pre-treatment RMSPE for models:
    - Log of GDP: 0.006
    - Inflation: 2.194
    - Log of Export: 0.094
    - Log of Import: 0.058
  - Ratio to benchmark RMSPE:
    - Log of GDP: 0.000
    - Inflation: 0.013
    - Log of Export: 0.003
    - Log of Import: 0.004
- Donor-pool weights (Log of GDP | Inflation | Log of Export | Log of Import):
  - ARG 0.038 0.008 0.000 0.000
  - BRA 0.126 0.000 0.149 0.000
  - CHL 0.036 0.000 0.000 0.278
  - CRI 0.231 0.000 0.482 0.000
  - DOM 0.000 0.181 0.000 0.000
  - ECU 0.000 0.000 0.004 0.000
  - MEX 0.276 0.711 0.365 0.032
  - PAN 0.138 0.000 0.000 0.391
  - PER 0.000 0.000 0.000 0.000
  - URY 0.000 0.099 0.000 0.169
  - VEN 0.155 0.000 0.000 0.130
- Note: Mexico has the highest weight for GDP, inflation and exports; Panama highest for imports. Weights are relatively sparse.

### Main SCM Estimated Effects
- Real GDP:
  - Counterfactual under default implies a slowdown in GDP growth of approximately 1 percentage point per year over 1981-85.
- Inflation:
  - SCM counterfactual implies inflation peaking at nearly 80% versus actual approximately 20%.
- Exports and Imports:
  - Exports: about 26% higher over a 4-year period in the default counterfactual (SCM).
  - Imports: about 36% lower over a 4-year period in the default counterfactual (SCM).
  - Combined trade balance effect: default counterfactual implies a positive trade balance of about 10% over 1982-85, instead of actual small trade deficit of about 1%.
- Statistical significance:
  - Estimated effects statistically significant at most horizons for GDP, inflation, and exports.
  - Import effect large but not statistically significant (very wide confidence bands).

### SDID and SCM Comparative Macroeconomic Effect Estimates
- Inflation:
  - Accumulated effect: nearly 27% (SDID estimations) versus 41% (SCM estimations).
- Exports:
  - SCM: 36% increase; SDID: 32% increase.
- Imports:
  - Negative effect on imports: 26% in both SCM and SDID estimations.
- Notes on figures:
  - Figure 5 uses SDID model including all pre-treatment outcomes as covariates and country and year fixed effects. Treatment year is 1981.

### Reputation, Literature, and Colombia’s 1980s Strategy
- Theoretical and empirical context:
  - Eaton and Gersovitz (1981): preserving reputation drives repayment.
  - Empirical evidence mixed: reputational effects found short lived or small in several studies (Eichengreen and Portes, 1986; Borensztein and Panizza, 2009; Gelos, Sahay, and Sandleris, 2011).
  - Cruces and Trebesch (2013): reputational costs increase with creditor haircut.
  - Catão and Mano (2017): all defaulters pay a large and long-lived default premium irrespective of haircut size.
  - Endogeneity concerns: unobserved variables (e.g., political instability) may bias estimates.
- Colombia-specific evidence:
  - Authorities believed avoiding outright rescheduling would preserve reputational access (Garay, 1989).
  - First credit rating documented in 1993: investment grade by Standard & Poor’s and one notch below investment grade by Moody’s (Cline, 1995).
  - Comparative evidence: Chile (a defaulter) received higher credit ratings than Colombia; Mexico rated one notch below Colombia; primary market yield spreads in first half of 1990s show Colombian bond yields lower than Argentina and Venezuela but close to Mexico and Uruguay and higher than Chile.
  - Caveat: counterfactual rating if Colombia had defaulted is unobserved; endogeneity in fundamentals may explain rating and yield differences.

### Event Study: Russian Default (August 1998) and Colombian Resilience
- Context:
  - Early 1990s large capital inflows to Latin America; by mid-1998 about one-quarter of total investment (nearly 6% of GDP) of region’s seven largest economies financed by foreign capital.
  - Sudden stop after Russian default: flows to largest seven economies fell from $100 billion over 1997Q3-1998Q2 to $37 billion in 1998Q3-1999Q2; average sovereign yield spreads tripled.
  - Russian crisis dates: August 13, 1998 (stock market crash) and August 17, 1998 (devaluation, domestic default, moratorium on foreign payments).
- Visual spread dynamics:
  - Pre-Russian default: Colombian spread ~450 basis points (150-200 points lower than Argentina, Mexico, Peru; ~400 points lower than Brazil).
  - After August 17, spreads rose rapidly; early September Colombian, Mexican, Peruvian spreads ~900 basis points and peaked over 1000 basis points on September 11, 1998.
  - Mid-October 1998: Argentinean, Colombian, Mexican, Peruvian spreads in 850-900 basis points range; Brazilian spreads above 1300 basis points.
  - One year after Russian default (August 1999): Colombian spread still above 650 basis points, slightly higher than Mexican and Peruvian spreads and about 100 basis points lower than Argentinean spread; nearly 400 basis points higher than Chilean spread.
- Event-study methodology (two-step, MacKinlay, 1997):
  1. Estimate ∆S_Ct = α + β ∆S_Mt + εt over pre-event estimation window.
  2. Predict Colombian spreads during event window and compute accumulated abnormal spreads.
- Baseline choices:
  - Estimation window: 90-day window ending 4 days before event (robust to 60- and 120-day windows).
  - Event window: baseline 6-day window (day before to 4 days after); robust to 12-day window.
  - “Market” spread: first principal factor of changes in Argentinean, Brazilian, Mexican, and Peruvian spreads.
- Event-study results (Table 7):
  - Average accumulated abnormal spreads positive in all specifications — Colombia did worse than comparator countries during crisis.
  - 6-day event window average abnormal spreads (examples): 6.43** (t = 2.38), 9.13*** (t = 3.89), 8.69*** (t = 3.61), 9.98*** (t = 4.52), 5.56** (t = 2.39), 1.83 (t = 0.79).
  - 12-day event window average abnormal spreads (examples): 8.40*** (t = 6.20), 10.57*** (t = 9.01), 10.23*** (t = 8.51), 11.21*** (t = 10.16), 9.23*** (t = 7.94), 7.45*** (t = 6.44).
  - Positive accumulated abnormal spreads statistically significant in 11 out of 12 reported specifications; only non-significant case when using 2 factors and 6-day window.
- Interpretation: reputational advantage Colombia might have gained by not defaulting in the 1980s had evaporated by the Russian crisis; reputational gains were not long-lived.

### Conclusions and Implications
- Research design: combines archival research with econometric analysis to study benefits of repaying during a period of widespread sovereign default by focusing on Colombia.
- Key findings:
  - Early 1980s: Colombia’s fundamentals similar to neighboring defaulting countries; estimated default probability not significantly different.
  - Political economy: authorities sought to maintain creditor relations and avoided IMF-program rescheduling due to strong US political support.
  - Colombia engaged in four rounds of rescheduling with conditions only slightly more favorable to creditors than those applied to other defaulters; syndicated bank loans traded at large discounts in secondary markets.
  - Econometric estimates show significant short-term benefits from avoiding formal default: higher GDP growth, lower inflation, and smoother current account adjustment.
  - No evidence of long-term benefits in terms of superior access to international capital markets during subsequent crises (e.g., Russian shock of August 1998).
- Broader messages:
  - Default episodes are not binary; differences in rescheduling strategies produce distinct short- and long-term economic effects.
  - Further research needed to understand short- and long-term economic effects of different debt rescheduling strategies.

*Source: wpiea2021233-print-pdf - Section 6 concludes.*

### Section 6 concludes.

### Section 6 concludes.

### Sovereign Debt Re-negotiations in the 1980s: Colombia and the Rest of Latin America — Overview and Context
- Total outstanding debt of Latin American borrowers went from less than $30 billion in 1970 to more than $320 billion in 1982.
- In 1982, outstanding loans to developing countries by the largest nine US banks were nearly three times the capital of these banks, and loans to Latin America were close to 180% of the banks’ capital.
- Most loans were in US dollars with a floating interest rate linked to US short-term rates.
- The Latin American debt crisis is usually associated with the weekend of August 13-15, 1982, when Mexico informed the IMF and the US Treasury it was no longer able to service its $86 billion of external debt.
- In total, 26 countries “defaulted” between 1982 and 1983 and another 29 in the rest of the 1980s.
- Out of the region’s 23 countries with more than one million inhabitants, 22 “defaulted” between 1980 and 1989; Colombia was the only large Latin American country that did not “default” in this period.
- The term “default” is used in quotation marks because many countries continued servicing interest while entering formal debt rescheduling agreements; these reschedulings often implied large NPV losses for creditors, making them de facto defaults.

### The Three Phases of the Latin American Debt Crisis
- Phase 1 (1982-85): Predominant view that the crisis was due to temporary liquidity problems; fiscal adjustment and coordinated reprofiling of principal repayments (conditional on IMF programs) aimed to keep countries current on interest payments.
  - The region’s current account turned from a 5% deficit of the region’s GDP in 1982 to a surplus of nearly 1% in 1984.
  - First round terms: spreads over LIBOR of about 2%-2.5%, maturities of 6-8 years, and large upfront cash commissions.
  - Restructuring in this phase led to an increase in the present value of developing countries’ debt.
- Phase 2 (1985-89): Opened by the “Baker Plan” (1985), featuring commitments to substantial amounts of “new” money from private and official sectors and greater involvement of the World Bank and regional development banks.
  - Restructurings under the Baker Plan had spreads over LIBOR lowered to the order of 0.8%-0.9%, extended amortization periods, and no commissions.
  - Through debt buybacks, exit bonds, and debt swaps (the “market-based menu approach”), the second phase led to a modest reduction in the present value of developing countries’ debts.
- Phase 3 (announced March 1989 by US Treasury Secretary Nicholas Brady): Focused on debt reduction with direct official financial support via the Brady plan.
  - Four key elements of the Brady plan:
    - (i) Use of zero-coupon US Treasury bonds to collateralize newly issued bonds (purchase financed with loans from international financial institutions and bilateral lenders).
    - (ii) Continuation of debt buybacks.
    - (iii) Waiving by commercial banks of legal clauses that hampered debt restructuring (e.g., “negative pledge” clauses).
    - (iv) Change in IMF policy that prevented the Fund from lending into arrears.
  - Over 1990-1998, 11 countries implemented Brady exchanges.
  - The Brady plan contributed to the end of the 1980s debt crisis and to emerging markets’ regained access to international financial markets via global bonds.

### The Strange Case of Colombia’s Non-Default — Comparative Positioning and Actions
- Colombia was not the best performer on standard macroeconomic and debt indicators in the run-up to the crisis, though it was better positioned than the regional median.
  - External debt-to-GDP ratio in 1981: Colombia had four countries with lower debt levels and two large countries (Mexico and Brazil) with debt levels only slightly higher (Mexico 6 percentage points of GDP higher; Brazil 8 points higher).
  - Primary fiscal balance over 1978-81: Colombia was close to the region’s median level.
  - Rate of GDP growth and inflation: Colombia was close to the region’s median.
  - Current account: Colombia was in a better position relative to the region’s median but not among the top four performers.
- Colombia’s interest spread over the US prime rate was a bit higher than those of Venezuela and Mexico and just below those of Chile and Ecuador.
- Two dimensions where Colombia was stronger entering the crisis:
  - Reserves-to-imports ratio in 1981-82: about 9 months versus the regional average of about 3.5 months (about 2.5 times the average of Argentina, Brazil, Chile, Mexico, Peru, and Venezuela).
  - Composition of external debt: lower levels of debt with private creditors and higher levels of debt with official creditors (especially the World Bank and the Inter-American Development Bank).
- Deterioration in early 1980s:
  - Net capital inflows collapsed during 1982-83, though to a lesser extent than other large Latin American economies.
  - From the second half of 1982, Colombia was unable to access the international capital market.
  - Balance of payments: surpluses of about 3% of GDP over 1976-80 deteriorated to zero in 1981 and to a current account deficit of 10% of GDP in 1982.
  - International reserves fell by 15% in 1982; reserves fell by $1.6 billion in 1983 (from $4.9 to $3.3 billion) and again by $1.6 billion in 1984, for a total loss from $4.9 billion in 1982 to $1.7 billion in 1984 (two thirds of reserves lost over two years).
  - Debt service absorbed nearly 40% of export revenues in 1983 (up from 15% in 1980).
  - Central Bank extended emergency assistance to nine financial institutions in the second half of 1982; government monetization increased the central bank balance sheet, and a 5-day state of economic emergency was declared at end-1982 (most measures invalidated by the Supreme Court).
  - During the 1983 Article IV consultation, authorities feared that reserve losses beyond $1 billion could cause market panic; reserves fell by $1.6 billion in 1983.

### Colombia’s Renegotiation Strategy and Key Statistics (1985–1991 Arrangements)
- By mid-1984 Colombia was unable to service its external debt but avoided suspending payments; instead it negotiated arrangements to refinance the majority of payments due between 1985 and 1994, aiming to preserve a reputation as a “good debtor.”
- IMF staff characterization: Colombia’s strategy was to avoid a formal debt rescheduling and maintain exposure of commercial banks and multilateral institutions roughly matching amortization payments as they fell due.
- Four negotiated arrangements over 1985-1991:
  - “Jumbo” arrangement of 1985: $1 billion
  - “Concorde” arrangement of 1987: $1 billion
  - “Challenger” arrangement of 1989: $1.5 billion
  - “Hercules” arrangement of 1991: $1.8 billion
- Typical terms across these arrangements:
  - Maturities ranging between 9 and 12 years.
  - Grace periods ranging between 3.4 and 6.2 years.
  - Stated spreads over LIBOR ranging between 87 and 150 basis points.
  - Effective average costs reached 193 basis points over LIBOR.
  - These conditions were substantially worse than the average interest rate in 1982, which was 1.1% over LIBOR.
- Market evidence of NPV losses for creditors:
  - Between 1987 and 1990, Colombian loans traded in the secondary market at prices that ranged between 58 and 64 cents on the dollar, indicating significant NPV losses for lenders.
  - Nevertheless, secondary market valuations of Colombian loans were the highest in the region during the 1980s and second only to Chile in the first half of the 1990s, outperforming valuations of Argentine, Brazilian, Venezuelan, and Peruvian loans.
- Domestic political debate:
  - Opposition argued arrangements were “excessively” creditor friendly and suggested outright rescheduling with conditions similar to other regional countries.
  - Authorities maintained that the strategy offered better prospects for a subsequent return to voluntary lending.
- Additional measures:
  - In 1984, rescheduling parameters for non-financial private sector firms unable to service external debt were set (Central Bank’s Resolution 33 required a minimum of 6 years including a 3-year grace period and a maximum interest spread over LIBOR of 2.5% to obtain liquidity support).
- Secondary-market performance and policy context:
  - Chile’s secondary market valuations benefited from debt reduction via debt-for-equity swaps; Chile’s low spreads in the early 1990s likely reflected reduced post-swap external debt and reform reputation.

_Italic line: Source: wpiea2021233-print-pdf - Section 6 concludes._

### 3.4 yearsLIBOR   +150   bps

### wpiea2021233-print-pdf - 3.4 yearsLIBOR   +150   bps

### Debt arrangements and terms (excerpted data)
- 3.4 years LIBOR +150 bps for the first 4 years, LIBOR+138bps (1+3/8) for the remaining 6 years. Effective average cost: LIBOR+1.93
- 8.4 years Quarterly disbursements conditional on IMF monitoring. Jan. 1988, Concorde $1 billion (corresponding to 70% of payments due to banks in 1987-88). Planned for 1987-88, but only signed and disbursed in 1988
- 5.0 years LIBOR+93bps (15/16). Effective average cost: LIBOR+1.42
- 10 years. Authorities had to share copies of Article IV reports with creditors. Jun. 1989, Challenger Syndicate Loan of $1.47 billions and facility of $185 millions. 1989-90 (first disbursement October 1989)
  - 6 years for the syndicated loan and 5 years for the facility (average 5 years)
  - LIBOR+87bps (7/8) for the syndicate loan and LIBOR+93bps (15/16) for the facility. Effective average cost: LIBOR+1.11
  - 12 years for the syndicated loan and 10 years for the facility
- Dec. 1990, Hercules Syndicate Loan of $1.575 billion & facility of $200 million (corresponding to 90% of principal payments coming due over 1991-94). 1991-94 6.2 years LIBOR+100 bps for the syndicate loan & LIBOR+150bps for the facility. Effective average cost: LIBOR+1.24
- 12.6 years

### Political economy and IMF involvement
- Colombian authorities emphasized maintaining a reputation as a “good debtor” (Garay, 1991), expecting differentiated, more favorable treatment from creditors.
- Colombia considered an IMF program not politically viable due to past disagreement on IMF conditionality; in 1967 Colombia had decided to remain independent from the Fund (Junguito, 1985, p. 75).
- The Fund was nevertheless involved:
  - World Bank lending to Colombia was monitored in part by the IMF with a “shadow program”.
  - Private creditors’ “advisory” committees requested access to IMF Article IV staff reports and enhanced surveillance (IMF “enhanced” surveillance) to certify and monitor Colombia’s adjustment program as if a stand-by arrangement were in place.
- Enhanced surveillance arrangement:
  - Met reservations on the IMF Executive Board but was approved with strong support from the US Federal Reserve, the IMF Managing Director, and the US Executive Director Charles Dallara.
  - Declassified IMF Board minutes (July 26, 1985) justify the Fund accepting the proposed monitoring role.
  - US political support was partly linked to cooperation on drug traffic control and broader geopolitical considerations in Latin America; Paul Volcker (Fed Chairman) played a key role advocating tailored approaches.
- The “Hercules” arrangement lasted until 1994. By 1992 Colombia had regained market access and began issuing dollar bonds to repay higher-rate World Bank loans. By mid-1993 secondary market discounts dropped sharply and Colombia imposed controls to slow excessive private inflows in September 1993 and tightened restrictions twice in 1994.

### Estimating Colombia’s default probability — data and methodology
- Raw data: 87 countries over 1976-2017, including 77 default episodes. Macroeconomic and debt data from vintages of the World Bank’s World Development Indicators and International Debt Statics. Default episodes from updated Asonuma and Trebesch (2016) dataset; only first episode retained when restructurings were consecutive.
- Candidate predictors: 22 variables covering solvency, liquidity, volatility, macroeconomic performance, political and institutional quality (see Appendix Table A6).
- Two-step estimation:
  - Step 1: Variable selection via LASSO (logistic LASSO) with 5-fold cross-validation.
    - Cross-validation selects hyper-parameter λ = 0.356.
    - LASSO selects 17 variables (listed in Appendix Table A7).
    - High interest payments on external debt and large current account deficits emerge as best predictors of sovereign defaults.
  - Step 2: Logit estimation using selected variables to predict default probability and compare Colombia in the 1980s to other Latin American defaulters.
- LASSO-logit estimator specified as:
  β_L = arg max_β ∑_{i=1}^N [y_i X_i β − ln(1 + e^{X_i β})] − λ ∑_{j=1}^p |β_j|

### Predicted default probabilities — summary statistics for actual defaulters
- Table 3: Estimated Probability of Defaults for Actual Defaulters
  - All Defaulters
    - Mean: 16.58
    - Median: 6.61
    - St. Dev.: 22.76
    - Min.: 0.16
    - Max.: 99.47
    - p10: 1.01
    - p90: 35.04
  - LAC Defaulters
    - Mean: 20.74
    - Median: 7.38
    - St. Dev.: 27.98
    - Min.: 0.67
    - Max.: 99.47
    - p10: 0.93
    - p90: 81.78
  - LAC Defaulters 1980-85
    - Mean: 26.45
    - Median: 19.59
    - St. Dev.: 31.01
    - Min.: 0.93
    - Max.: 99.47
    - p10: 3.58
    - p90: 81.78
- Note: Estimated probabilities based on the Logit model of Table A7; summary statistics refer to the estimated probability in the year of actual default for those countries.

### Colombia’s predicted default probability (1979–86) — key results
- Predicted probability of default for Colombia:
  - 3% in 1982
  - 14% in 1983 (implying predictors of default increased in 1982, since lagged explanatory variables are used)
  - Ranged between 14% and 20% over 1984-86
- Comparison with defaulters:
  - Between 1983 and 1986, Colombia’s estimated probability of default was never significantly lower than the mean and median values reported for actual defaulters.
  - In 1983, 1984, and 1986 Colombia’s estimated probability was significantly higher than the median for all defaulting countries in their default year.
  - Cross-country comparisons (1983-86): predicted default probabilities in Panama, Bolivia, Ecuador, Uruguay, Costa Rica, and Dominican Republic were significantly lower than Colombia’s; Venezuela, Jamaica, Mexico, Argentina, and Peru were not significantly different from Colombia; Brazil and Chile displayed significantly higher predicted default probabilities than Colombia in 1983-86.

### Empirical implications and next-step analysis
- LASSO selection and logit prediction indicate that Colombia’s fundamentals in the 1980s were similar to those of Latin American countries that defaulted.
- The authors proceed to counterfactual analysis using:
  - Synthetic control method (SCM) (Abadie and Gardeazabal, 2003)
  - Synthetic difference-in-differences method (SDID) (Arkhangelsky, Athey, Hirshberg, Imbens, and Wager, 2020)
- SDID is described as a generalization of SCM that accounts for unobservable time-invariant factors and common shocks to improve identification.

*Source: wpiea2021233-print-pdf*

### 4.1    Synthetic Control

### 4.1    Synthetic Control

### Methodology: synthetic control construction
- Treatment definition and context
  - The treatment is a “non-event”, namely the non-default of Colombia.
  - In a set of Latin American countries j = 1,...,N over T periods, Colombia (country 1) is the only country that receives the treatment in period T0, since all other countries in the region did experience a default event.
  - Treatment date chosen: 1981. Estimation horizon: up to 1985 to limit confounding shocks.

- Potential outcomes notation
  - Yj,t(NDj) =
    - Yj,t(0) if NDj = 0
    - Yj,t(1) if NDj = 1

- Synthetic control estimator
  - Given weights W = (w2,...,wn+1):
    - ˆY1,t(0) = sum_{j=2}^{j+1} wj Yj,t
    - ˆτ1,t = Y1,t(1) − ˆY1,t(0)

- Inference approach
  - Placebo/permutation test framework following Firpo and Possebom (2018), building on Imbens and Rubin (2015) and Abadie, Diamond, and Hainmueller (2010, 2015).
  - Test statistic used: ratio of mean squared prediction errors (RMSPE):
    - RMSPE_j = (∑_{t=T0+1}^{T} (Yj,t − ˆYj,t(0))^2 /(T−T0)) / (∑_{t=0}^{T0} (Yj,t − ˆYj,t(0))^2 / T0)
  - The RMSPE ratio (post- to pre-treatment) avoids discarding countries with poor pre-treatment fit.
  - Confidence sets constructed by inverting the RMSPE-based test statistic.

- Donor pool and predictors
  - Variables with counterfactuals built: real GDP, inflation, exports, imports.
  - Bolivia and Jamaica excluded from the donor pool as they defaulted before 1981.
  - Baseline matching: match pre-treatment outcomes on their lagged values only (no additional controls) to avoid ‘cherry-picking’ predictors.

### Implementation details and goodness-of-fit metrics
- Pre-treatment fit assessment
  - Goodness of fit assessed with ratio of pre-treatment RMSPE to RMSPE obtained with a zero-fit benchmark (Adhikari and Alm, 2016).
  - Benchmark RMSPE defined as RMSPE_benchmark = √(1/11 ∑_{t=0}^{1980} (Y1,t)^2) where 11 corresponds to the number of pre-treatment years.
  - Interpretation: RMSPE_j = 0 → ratio = 0 (perfect fit); ratio = 1 → RMSPE_j identical to zero-fit model.

- Donor-pool weight structure
  - Weights tend to be relatively sparse, consistent with synthetic control estimators.

- Table of country weights and fit (exact values)
  - Country-specific weights for each model (Log of GDP | Inflation | Log of Export | Log of Import)
    - ARG 0.038 0.008 0.000 0.000
    - BRA 0.126 0.000 0.149 0.000
    - CHL 0.036 0.000 0.000 0.278
    - CRI 0.231 0.000 0.482 0.000
    - DOM 0.000 0.181 0.000 0.000
    - ECU 0.000 0.000 0.004 0.000
    - MEX 0.276 0.711 0.365 0.032
    - PAN 0.138 0.000 0.000 0.391
    - PER 0.000 0.000 0.000 0.000
    - URY 0.000 0.099 0.000 0.169
    - VEN 0.155 0.000 0.000 0.130
  - Pre-treatment RMSPE (exact values)
    - Log of GDP: 0.006
    - Inflation: 2.194
    - Log of Export: 0.094
    - Log of Import: 0.058
  - Ratio to benchmark RMSPE (exact values)
    - Log of GDP: 0.000
    - Inflation: 0.013
    - Log of Export: 0.003
    - Log of Import: 0.004
  - Note: Mexico has the highest weight in the models for GDP, inflation and exports; Panama has the highest weight in the model for imports. Brazil, Costa Rica, and Venezuela also carry relatively high weights in more than one model.

### Main estimated effects (SCM results)
- Real GDP
  - Counterfactual indicates that, under default, Colombia would have observed a slow-down in GDP growth of approximately 1 percentage point per year over the period 1981-85.
  - Top-left panel of Figure 3 and accumulated effect discussed further in SDID section.

- Inflation
  - Default counterfactual implies a spike in inflation, with inflation peaking at nearly 80% in the counterfactual, against an actual value of approximately 20%.

- Exports and imports (current account components)
  - Exports: about 26% higher over a 4-year period in the default counterfactual.
  - Imports: about 36% lower over a 4-year period in the default counterfactual.
  - Combined effect on trade balance: default counterfactual implies a positive trade balance of about 10% over 1982-85, instead of an actual small trade deficit of about 1%.

- Statistical significance
  - Figure 4 (inference) shows estimated effects with 90% confidence bands.
  - Estimated effects are statistically significant at most horizons for GDP, inflation, and exports.
  - Estimated effect for imports is large but not statistically significant (imprecisely estimated with very wide confidence bands).

### Interpretation and contextual findings
- Comparison to other literature
  - The GDP slowdown effect (~1 percentage point per year) is about half the size of what is normally found in cross-country regressions (Borensztein and Panizza, 2009).

- Mechanisms and robustness notes
  - The import effect is unlikely driven by reduced trade credit because access to trade credit for Colombia was not different from access for countries that defaulted and had an IMF program (Alvarez and Flores Zendejas, 2014).
  - The absence of a major devaluation for Colombia, unlike other countries in the sample, may have played an important role in its trade and macro dynamics.

*Italic: Source: 4.1 Synthetic Control, wpiea2021233-print-pdf*

### 3.  With respect to inflation (top right panel) we find an accumulated effect of nearly 27%

### wpiea2021233-print-pdf - 3.  With respect to inflation (top right panel) we find an accumulated effect of nearly 27%

### Macroeconomic effect estimates (SDID and SCM)
- Inflation:
  - Accumulated effect of nearly 27% (SDID estimations), which is lower than the 41% effect found in the SCM estimations.
- Exports:
  - Positive effect of exports: 36% in the SCM estimations and 32% in the SDID estimations.
- Imports:
  - Negative effect on imports: 26% in both the SCM and the SDID estimations.
- Figure context:
  - Figure 5 reports Colombia’s synthetic controls (grey solid line) and the actual series (black solid line) for: (a) Log of real GDP; (b) Inflation; (c) Log of Export; (d) Log of Import.
  - The black dashed line corresponds to the trend in the observed series; the blue dashed line reports the trend we would have observed under a default.
  - The pink areas represent the time weights λt. The treatment year is 1981.
  - The synthetic control estimates are based on a SDID model that includes all the pre-treatment outcomes as covariates and country and year fixed effects.

### Reputation, empirical literature, and Colombia’s strategy in the 1980s
- Literature overview:
  - Eaton and Gersovitz (1981) framework: desire to preserve reputation drives willingness to repay.
  - Empirical evidence mixed: reputational effects of sovereign defaults found to be either short lived or small by several studies (Eichengreen and Portes, 1986; Borensztein and Panizza, 2009; Gelos, Sahay, and Sandleris, 2011).
  - Cruces and Trebesch (2013): reputational costs increase with the haircut imposed on creditors.
  - Catão and Mano (2017): all defaulters pay a large and long-lived default premium, irrespective of haircut size.
  - Endogeneity concerns: unobserved variables (e.g., political instability) may correlate with both sovereign spreads and default likelihood, biasing estimates.
- Colombia in the 1980s:
  - Colombian authorities believed avoiding outright rescheduling would preserve reputational access to capital markets (Garay, 1989).
  - Cline (1995) documents Colombia received its first credit rating in 1993: investment grade by Standard & Poor’s and only one notch below investment grade by Moody’s (Table 5).
  - Comparative evidence: Chile (a defaulter) received higher credit ratings than Colombia; Mexico was rated just one notch below Colombia; primary market yield spreads in the first half of the 1990s show Colombian bond yields lower than Argentina and Venezuela but close to Mexico and Uruguay and higher than Chile (Table 6).
  - Caveat: these comparisons do not prove causality because the counterfactual rating for Colombia had it defaulted is unobserved; endogeneity in fundamentals may explain rating and yield differences.

### Event study: Russian default (August 1998) and Colombian resilience
- Context:
  - Early 1990s: large capital inflows to several Latin American countries; by mid-1998 about one-quarter of total investment (or nearly 6% of GDP) of the region’s seven largest economies was financed by foreign capital (Calvo and Talvi, 2005).
  - Sudden stop after Russian default: flows to the largest seven Latin American economies fell from $100 billion over 1997Q3-1998Q2 to $37 billion in 1998Q3-1999Q2; average sovereign yield spreads in the region tripled (Calvo and Talvi, 2005).
  - Russian crisis dates: August 13, 1998 (stock market crash) and August 17, 1998 (devaluation, domestic default, moratorium on foreign payments).
- Visual evidence (Figure 6):
  - Before the Russian default, the Colombian spread stood at about 450 basis points, 150-200 points lower than Argentina, Mexico, and Peru (600–650 range), and about 400 points lower than the Brazilian spread.
  - After August 17, spreads climbed rapidly; by early September Colombian, Mexican and Peruvian spreads hovered around 900 basis points and peaked at over 1000 basis points on September 11, 1998.
  - By mid-October 1998, Argentinean, Colombian, Mexican, and Peruvian spreads were in the 850-900 basis points range; Brazilian spreads remained above 1300 basis points.
  - One year after the Russian default (August 1999), the Colombian spread was still above 650 basis points, slightly higher than Mexican and Peruvian spreads and about 100 basis points lower than the Argentinean spread; nearly 400 basis points higher than the Chilean spread.
- Event-study methodology:
  - Two-step procedure (MacKinlay, 1997):
    1. Estimate relationship between daily changes in Colombian EMBI spreads (∆S_Ct) and daily changes in “market” spreads (∆S_Mt) over a pre-event estimation window: ∆S_Ct = α + β ∆S_Mt + εt.
    2. Use estimated parameters to predict Colombian spreads during the event window and compute excess (“abnormal”) changes as out-of-sample forecast errors; accumulate abnormal spreads over the event window.
  - Baseline choices:
    - Estimation window: 90-day window ending 4 days before the event (robust to 60- and 120-day windows).
    - Event window: baseline 6-day window (starting the day before the event and ending 4 days after); robust to a 12-day window (day before to 10 days after).
    - “Market” spread: first principal factor of changes in Argentinean, Brazilian, Mexican, and Peruvian spreads (robust to using first and second principal factors of seven EMBI countries).
  - Statistical test:
    - Average daily excess spread = CA∆S / W with variance σ^2_A∆S / W, t statistic = CA∆S / (σ_A∆S √W).
- Event-study results (Table 7):
  - Average accumulated abnormal spreads are positive in all specifications, implying Colombia did worse than comparator countries during the crisis.
  - Statistical significance:
    - For the 6-day event window, average abnormal spreads reported include: 6.43**, 9.13***, 8.69***, 9.98***, 5.56**, 1.83 (t-statistics in parentheses: (2.38), (3.89), (3.61), (4.52), (2.39), (0.79)).
    - For the 12-day event window, average abnormal spreads reported include: 8.40***, 10.57***, 10.23***, 11.21***, 9.23***, 7.45*** (t-statistics in parentheses: (6.20), (9.01), (8.51), (10.16), (7.94), (6.44)).
    - Estimation windows and specifications in Table 7: N. Days = 90, 90, 120, 60, 90, 90; Ending on = Aug. 9, Aug. 13, Aug. 9, Aug. 9, Aug. 9, Aug. 9; N. of princ. factors = 1, 1, 1, 1, 1, 2; Countries used for market = ARG, BRA, MEX, PER (with alternatives PAN, PER, VEN noted).
  - Significance summary:
    - Positive average accumulated abnormal spreads are statistically significant in 11 out of the 12 reported specifications.
    - The only non-statistically-significant case is when using 2 factors and a 6-day window.
- Interpretation:
  - Formal tests confirm the visual impression from Figure 6: at the time of the Russian default the reputational advantage Colombia might have gained by not defaulting in the 1980s had evaporated.
  - The empirical evidence suggests reputational gains were not long-lived.

### Conclusions and implications
- Research approach:
  - Study focuses on benefits of repaying during a period of widespread sovereign default by analyzing Colombia, the only large Latin American country that did not default in the 1980s, combining archival research with econometric analysis.
- Key findings:
  - In early 1980s Colombia’s economic fundamentals were similar to neighboring defaulting countries and estimated default probability was not significantly different from those countries.
  - Political factors mattered: Colombian authorities sought to maintain creditor relations and secured avoidance of an IMF-program rescheduling due to strong US political support.
  - Colombia engaged in four rounds of debt rescheduling with conditions only slightly more favorable to creditors than those applied to other Latin American defaulters; syndicated bank loans traded at a large discount in secondary markets.
  - Econometric estimates indicate significant short-term benefits from avoiding formal default: higher GDP growth, lower inflation, and smoother current account adjustment.
  - No evidence of long-term benefits in terms of better access to international capital markets during subsequent crises (e.g., Russian shock of August 1998).
- Broader messages:
  - Default episodes should not be treated as binary events; differences in rescheduling strategies produce distinct short- and long-term economic effects.
  - More research is needed to understand the short and long-term economic effects of different debt rescheduling strategies.

*Source: wpiea2021233-print-pdf.*

### References

### References and Appendix (wpiea2021233-print-pdf)

### Major referenced literature and thematic coverage
- Literature spans methods and applications relevant to sovereign debt, sovereign default costs, debt restructurings, contagion, and causal inference.
- Methodological sources include Synthetic Control Methods, LASSO, Bayesian Lasso, Regression Shrinkage (Lasso), Synthetic Difference in Differences, and causal inference foundations.
- Empirical and historical studies cover Latin America debt crises, sovereign debt restructurings, banking crisis risk, output costs of sovereign default, and long-run sovereign bond history.

### Key empirical lists and inventories (Appendix Table A1)
- Table A1: List of Sovereign Defaults in the 1980s by region (selected entries shown in source):
  - East Asia: Philippines 1983; Vietnam 1985; Thailand entries not listed.
  - Europe: Poland 1981; Romania 1981; Turkey 1982; Yugoslavia 1983.
  - Latin America & Caribbean: Argentina 1982; Bolivia 1980; Brazil 1983; Chile 1983; Costa Rica 1983; Cuba 1983; Dominican Republic 1982; Ecuador 1982; El Salvador 1983; Guatemala 1985; Haiti 1982; Honduras 1981; Jamaica 1981; Mexico 1982; Nicaragua 1980; Panama 1983; Paraguay 1986; Peru 1983; Trinidad and Tobago 1988; Uruguay 1983; Venezuela 1983.
  - Middle East North Africa: Iraq 1987; Jordan 1989; Morocco 1983; Yemen 1985.
  - Sub-Saharan Africa: Angola 1985; Burkina Faso 1983; Cabo Verde 1981; Cameroon 1985; Centra African Rep. 1981; Cote d’Ivoire 1983; Gabon 1986; Gambia 1986; Ghana 1987; Guinea 1986; Guinea-Bissau 1983; Liberia 1987; Madagascar 1981; Malawi 1982; Mozambique 1980; Niger 1983; Nigeria 1982; Sao Tome & Principe 1987; Senegal 1981; Sierra Leone 1983; South Africa 1985; Tanzania 1984; Togo 1982; Uganda 1980; Zambia 1983.
- Source attribution in table: Borensztein and Panizza (2009) and Das, Papaioannou, and Trebesch (2012). Note: In the case of multiple restructurings, only the first default is listed.

### Selected bank debt restructuring terms in Latin America and the Caribbean (Appendix Table A2 — selected entries)
- Argentina:
  - 1983 New financing: Grace Period (3)4.5, Maturity (21?)3? — source lists "34.5 213-225" (as formatted in PDF).
  - 1984 Restructuring: Grace Period 3, Maturity 10-12, Spread over LIBOR 137.
  - 1984 New financing: Grace Period 3, Maturity 10, Spread 125-163.
- Brazil:
  - 1983 Restructuring: Grace Period 2.5, Maturity 8, Spread 200-225.
  - 1983 New financing: Grace Period 2.5, Maturity 8, Spread 188-212.
  - 1984 Restructuring: Grace Period 5, Maturity 9, Spread 175-200.
- Chile:
  - 1983 New financing: Grace Period 4, Maturity 7, Spread 212-225.
  - 1983 Restructuring: Grace Period 4, Maturity 8, Spread 200-212.
  - 1984 New financing: Grace Period 5, Maturity 9, Spread 150-175.
- Mexico:
  - 1983 Restructuring: Grace Period 4, Maturity 8, Spread 175-188.
  - 1983 New financing: Grace Period 3, Maturity 6, Spread 213-225.
  - 1984 New financing: Grace Period 5.5, Maturity 10, Spread 113-125.
  - 1984 Restructuring: Grace Period 0-1, Maturity 14, Spread 88-125.
  - 1986 Restructuring and New money entries with spreads and maturities as listed in source.
- Uruguay:
  - 1983 Restructuring/New financing: Grace Period 2, Maturity 6, Spread 213-225.
  - 1986 Restructuring: Grace Period 3, Maturity 12, Spread 138 and 163 in separate entries.
- Venezuela:
  - 1984 Restructuring: Grace Period –, Maturity 12.5, Spread 113.
- Source: IMF (1987a)

### Debt levels and macroeconomic indicators in the run-up to the Latin American Debt Crisis (Appendix Table A3 — summary statistics preserved)
- Table A3 (selected country-level extremes and sample statistics as reported):
  - Mean: External debt to GDP 48.85; Primary Bal. to GDP -0.4; Current Acc. Bal. to GDP -5.76; Real GDP growth 3.3; Inflation 28.15; Spread (bps) 104.
  - Median: External debt to GDP 36.74; Primary Bal. to GDP -0.5; Current Acc. Bal. to GDP -5.97; Real GDP growth 4.31; Inflation 17.5; Spread (bps) 86.
  - Min: External debt to GDP 15.2; Primary Bal. to GDP -9.27; Current Acc. Bal. to GDP -23.24; Real GDP growth -7.3; Inflation 4.14; Spread (bps) 57.
  - Max: External debt to GDP 146.5; Primary Bal. to GDP 5.31; Current Acc. Bal. to GDP 3.1; Real GDP growth 10.33; Inflation 130.75; Spread (bps) 222.
  - St. Dev.: External debt to GDP 33.99; Primary Bal. to GDP 4.07; Current Acc. Bal. to GDP 5.71; Real GDP growth 3.97; Inflation 30.23; Spread (bps) 55.
  - Colombia minus regional average difference (and in parentheses regional St. Dev.):
    - Difference: -25.9; -0.53; 5.25; 1.74; -5.59; -41.45
    - (35.3) (4.54) (5.85) (4.15) (31.75) (58.65)
- Source: IMF (1987a)

### Evolution of selected economic variables in Colombia (Appendix Table A4 — selected series)
- Debt Service (% of exp.) by year (1979–1990): 15.1 15.3 28 29.9 38 40.3 46 38.5 43.1 46.9 47.3 43.8
- Govt bal. (% of GDP) by year (1979–1990): -1.2 -2.5 -6.4 -7.6 -7.6 -6.3 -3.5 -0.3 -0.5 -1.5 -1.6 -0.9
- Intl. res. (Bill. USD) by year (1979–1990): 4.1 5.4 5.6 4.9 3.3 1.7 1.3 3.6 3.6 3.8 3.7 4.5
- Curr. acc. bal. (% of GDP) by year (1979–1990): 20.4 -6.1 -10.1 -9.8 -7.6 1.6 -0.2 -0.7 -0.4 1.4
- Sources: Colombia Recent Economic Developments reports 1983, 1985, 1987, 1989, 1990, 1993; Colombia Article IV reports 1983, 1985, 1987, 1989.

### Colombian EMBI Spreads and “Market” Spreads regression results (Appendix Table A5)
- Coefficient estimates (standard errors in parenthesis):
  - F1: 11.38*** (1.93) in column (1); 13.06*** (1.87) in column (2); 11.84*** (1.95) in column (3); 12.47*** (2.12) in column (4); 11.73*** (1.89) in column (5); 10.71*** (1.96) in column (6).
  - F2 shown in one column as 3.468 (4.50).
  - Constant values across columns: 0.827 (0.61); 1.108 (0.71); 0.791 (0.60); 0.768 (0.61); 0.769 (0.54); 0.936 (0.70).
- Sample sizes and model diagnostics:
  - N. Obs: 90 90 90 90 120 60 (by column).
  - R2: 0.28 0.36 0.30 0.30 0.25 0.34.
  - Window length: 90 days 90 days 90 days 90 days 120 days 60 days.
  - Window ends on: Aug. 9; Aug. 13; Aug. 9; Aug. 9; Aug. 9; Aug. 9.
  - Countries used for market returns vary by column (examples: ARG, BRA; ARG, BRA, ECU, MEX; ARG, BRA, MEX, PER; PAN, PER, VEN).
- Note: Standard errors in parenthesis. *** statistically significant at the 1% confidence level. F1 and F2 represent the first and second principal factors of the changes in EMBI spreads of the countries listed.

### Variable summary statistics used in LASSO-Logit (Appendix Table A6 — selected variables preserved exactly)
- Broad money growth (%) — Mean 46.46; Median 16.41; St. Dev 63.20; Min -88.79; Max 12,513; p10 3.73; p90 46.76
- Inflation, GDP defl. (%) — Mean 50.91; Median 7.86; St. Dev 605.36; Min -29.69; Max 26,766; p10 0.80; p90 35.50
- GDP growth (%) — Mean 3.73; Median 4.16; St. Dev 5.55; Min -51.03; Max 106.28; p10 -1.88; p90 8.77
- GDP per capita growth (%) — Mean 1.74; Median 2.10; St. Dev 5.38; Min -50.23; Max 91.65; p10 -3.89; p90 6.96
- Short-term debt (% of ext. debt) — Mean 12.06; Median 9.52; St. Dev 11.10; Min 0.00; Max 83.15; p10 0.65; p90 26.02
- Domestic Absorption (% of GDP) — Mean 107.72; Median 105.30; St. Dev 15.68; Min 55.86; Max 161.43; p10 94.26; p90 122.66
- Int. on ext. debt (% of Exp.) — Mean 6.92; Median 4.92; St. Dev 6.77; Min 0.08; Max 69.81; p10 1.04; p90 15.19
- Total debt service (% of exports) — Mean 17.93; Median 14.46; St. Dev 14.49; Min 0.12; Max 156.86; p10 3.08; p90 36.49
- Ext. debt stock, PPG (% of GDP) — Mean 47.89; Median 32.57; St. Dev 59.85; Min 0.00; Max 862.11; p10 9.97; p90 95.75
- Debt forgiveness (% of GDP) — Mean 0.64; Median 0.00; St. Dev 3.77; Min -0.98; Max 93.51; p10 0.00; p90 0.77
- (Table contains additional variable summary statistics in the source.)

### Drivers of the Probability of Default — Logit model coefficients (Appendix Table A7)
- Variables (all lagged) with coefficient estimates and standard errors (standard errors in parenthesis):
  - Broad money growth (%) — -0.008 (0.008)
  - GDP growth (%) — 0.101 (0.197)
  - GDP per capita growth (%) — -0.193 (0.203)
  - Short-term debt (% of external debt) — -0.022 (0.023)
  - Domestic Absorption (% of GDP) — -0.049*** (0.021)
  - Interest on external debt (% of exports) — 0.133*** (0.039)
  - Interest on external debt (% of GDP) — 0.732 (0.456)
  - Interest on PNG external debt (% of GDP) — -0.084 (1.009)
  - Interest on PPG external debt (% of GDP) — -0.968 (0.585)
  - Interest on short-term debt (% of GDP) — 0.203 (0.605)
  - Total debt service (% of GDP) — 0.042 (0.044)
  - Current account balance (% of GDP) — -0.091*** (0.028)
  - Exports (% of GDP) — 0.031* (0.017)
  - Foreign direct investment, net inflows (% of GDP) — -0.027 (0.038)
  - External debt stocks, PPG (% of GDP) — 0.007* (0.004)
  - External debt stocks, PNG (% of GDP) — -0.096 (0.067)
  - Debt forgiveness grants (% of GDP) — -0.644* (0.343)
  - Constant — -0.759 (2.547)
- N. Obs. 2,098
- Pseudo R2 0.269
- Statistical significance notation: *** 1% confidence level; ** 5% confidence level; * 10% confidence level.

### Sources of variables for the counterfactual exercise (Appendix Table A8)
- Real GDP: GDP at constant 2010 USD — World Development Indicators
- Inflation: CPI annual percent change — World Development Indicators
- Exports: Exports of goods and services (current USD) — World Development Indicators
- Imports: Imports of goods and services (current USD) — World Development Indicators
- Notes: Missing CPI observations for Argentina, Brazil, Chile, and Venezuela are filled with annual inflation calculated using the GDP deflator obtained from the World Development Indicators.

*Content unit: wpiea2021233-print-pdf - References and Appendix (as provided).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021233-print-pdf.pdf_