## _wp16225

## Source details

**Canonical URL:** [_wp16225](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16225.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16225.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16225.pdf.json)

---

### I. Introduction — recent history and current state of IMF arrears
- Protracted arrears are defined as obligations overdue by six months or more.
- Remaining protracted arrears cases at time of writing started in 1984, 1987, and 2001: Sudan, Somalia, and Zimbabwe.
- No IMF member accumulated protracted arrears since 2001.
- Most recent resolution noted: Liberia repaid SDR 543 million in March 2008 after 22 years of ineligibility due to HIPC debt relief.
- Zimbabwe cleared its arrears on October 20, 2016.
- Short-term arrears: missing payments up to 180 days. About 20 countries incurred short-term arrears per year in the 1980s–1990s; in 2015 only a handful did, and one must go back to 2006 to find ten country cases.
- Objective: re-examine determinants and duration of IMF arrears using expanded 1975–2015 data and political factors; paper structure: literature review (II), stylized facts (III), determinants/logit estimation (IV), duration modeling (V), conclusion (VI).

### II. Literature review — prior empirical findings on IMF arrears
- Few studies focus specifically on IMF arrears; related literature on sovereign default to private or Paris Club creditors.
- Key prior studies:
  - Aylward and Thorne (1998): logit for 138 countries, 1976–93 (1,410 country-year observations; 103 arrears observations); arrears defined as overdue ≥ three months; arrears more likely with larger IMF credit use (quota terms), higher inflation, lower per capita income, or slower growth.
  - Oka (2003): 118 countries, 1980–97 (1,375 country-year observations; 63 arrears observations); arrears defined as overdue ≥ six months; arrears more likely with high IMF debt outstanding relative to quota, high IMF debt service to exports, low reserves-to-imports or low export growth, or rising arrears to other external creditors.
- Contribution of this paper:
  - Joint analysis of occurrence and duration of arrears (duration analysis new relative to Aylward and Thorne (1998) and Oka (2003)).
  - Uses expanded dataset 1975–2015 and models protracted arrears (>6 months) and other duration cutoffs (>1 month, >1 week, any arrears).
  - Thirty instances of protracted arrears since 1945; 189 IMF members of which only 27 incurred arrears in last 75 years — treated as population of events rather than a sample. Small number of protracted events motivates descriptive and qualitative augmentation of econometrics.

### III. Stylized facts — comparative and dynamic patterns
- Country groups: protracted arrears, short-term arrears (over 7 days and less than 6 months), and countries with IMF credit outstanding but not in arrears.
- Comparative indicators (exact figures preserved):
  - Civil unrest (above average magnitude score): protracted arrears ≈ 18 percent; no-arrears group 3 percent; short-term arrears 4 percent (protracted is six times higher than no-arrears, 4.5 times higher than short-term).
  - Institutional development (CPIA < 2.5): protracted ≈ 38 percent; short-term 14 percent; no-arrears 12 percent.
  - IMF credit outstanding > 150 percent of quota: protracted 40 percent; comparator groups about half that percentage.
  - Reserves in months of imports > 3 months: protracted 10 percent; short-term 30 percent; other countries with IMF credit outstanding 50 percent.
  - Total external debt to GDP > 50 percent: protracted ≈ 73 percent; short-term 58 percent; other countries 44 percent.
  - Inflation > 10 percent: protracted 74 percent; short-term 45 percent; no-arrears 41 percent.
  - Real GDP growth positive: protracted 74 percent (26 percent shrank); short-term 81 percent; no-arrears 83 percent.
- Dynamics around occurrence and clearance:
  - Protracted arrears countries had higher IMF credit outstanding than short-term arrears countries at occurrence and for five years before and after.
  - Protracted arrears countries were poorer with worse macroeconomic and political indicators throughout the window.
  - Civil unrest rose steeply before protracted arrears; fell shortly before/after clearance but remained higher than in short-term arrears countries.
  - External debt/GDP rose more steeply before protracted arrears.
  - IMF credit/ quota ratio peaks ~ three years before protracted arrears onset, decreases afterwards, approaching short-term arrears levels at clearance.
  - Reserves (months of imports) persistently lower for protracted arrears countries, converging only after clearance.
  - Growth lower on average for protracted arrears countries across full span; inflation rose immediately before and after protracted arrears onset; exports/GDP behaved oppositely.
- Sample sizes for figure averages: average number of observations is 23 for start and clearing of protracted arrears, and 309 for short-term arrears.

### IV. Triggers and common determinants of arrears
A. Triggers — classification and examples
- Trigger types:
  - Political (wars, political acts, refusal to service obligations).
  - Global (commodity price swings, weather events, debt crises).
  - Country-specific (poor policy implementation, corruption, loss of aid, inappropriate exchange rate policy).
- Notable examples and patterns:
  - Afghanistan and Iraq accrued arrears without IMF credit outstanding due to net SDR holdings below SDR allocations, driven by wars, political turbulence, and sanctions.
  - Nine cases (Cambodia, Democratic Republic of Congo (twice), Honduras, Peru, Vietnam, Zambia (twice), Zimbabwe) involved authorities deciding not to service IMF obligations despite apparent resources — nearly one third of all arrears cases.
  - Political triggers implicated in 21 cases (70 percent), often combined with country-specific or global triggers.
  - Global triggers appear in 13 cases at least partly; only the Dominican Republic is attributed solely to global triggers (commodity price declines common among global triggers).
  - Country-specific triggers involved in 21 cases (majority); five caused solely by country-specific issues and others in combination with global/political triggers.
  - Commodity price troughs/peaks often coincide with arrears start/end dates for affected countries.

B. Econometric analysis — model, data, and design choices
- Model: pooled logit estimating probability P(arrears) with logit transformation and maximum likelihood estimation.
- Dependent variable variants: protracted arrears (>6 months), >1 month, >1 week, and any arrears.
- For protracted arrears, only the first country-year observation used as marker of occurrence; years after first-year up to clearance set as missing to avoid state-dependence confounding.
- Dataset: all countries with IMF credit outstanding in a given year: 3,354 country-year observations over 1974–2014; 142 different countries, 114 experienced short-term or protracted arrears (many repeatedly).
- Non-random exclusion: some protracted arrears countries excluded from regressions due to incomplete data; excluded countries on average had more civil conflict and worse institutional quality at time of arrears — likely biases estimated political variable effects downward.
- Explanatory variables (variable names and expected sign preserved):
  - total external debt as percent of GDP — Positive
  - short-term external debt to reserves — Positive
  - debt service on external debt as percent of exports — Positive
  - IMF credit outstanding as percent of quota — Positive
  - multilateral debt as percent of total external debt — Positive
  - arrears to IMF in previous 5 years (dummy) — Positive
  - total arrears (to IMF and other creditors) as percent of external debt — Positive
  - reserves as percent of imports — Negative
  - current account balance as percent of GDP — Ambiguous
  - total foreign assets and liabilities to GDP — Positive
  - exports and imports as percent of GDP — Negative
  - per capita GDP — Negative
  - real GDP growth — Negative
  - Inflation — Positive
  - “civil unrest” magnitude — Positive
  - institutional quality (CPIA) — Negative
  - non-fuel commodity price index — Negative
  - US real interest rate (alternatively global GDP growth rate) — Negative
  - time dummy for 1982-89 — Positive (controls for 1980s debt crisis)
- Methodological notes:
  - CPIA only available from 2005; imputed 2005 value for earlier years treating CPIA as slow-moving.
  - Fiscal position indicators omitted due to data limits.
  - Pooled logit preferred for main results; fixed effects discussed but would exclude countries that never incurred arrears and restrict sample severely.
  - Prior literature found strong state dependence; this paper uses first-year-only observations for protracted arrears to separate entry determinants from duration.

### V. Key empirical and descriptive conclusions
- Countries with protracted arrears perform worse across institutional quality, civil unrest, reserves, debt burdens, inflation, and growth than short-term arrears or no-arrears countries.
- Political factors prominent: political triggers (wars, sanctions, policy refusals) implicated in large share of protracted arrears and interact with country-specific and global triggers.
- Global factors, notably commodity price declines, critical in a notable subset and often coincide with arrears timing.
- Country-specific policy failures remain major triggers, alone or combined with other shocks.
- Econometric findings tempered by small protracted-arrears sample and non-random missingness; qualitative analysis essential to complement statistical results.

### Annex I — Major findings from pooled logit estimations (selected)
- Variables statistically significant across at least four of five regressions:
  - arrears to the IMF in previous five years
  - reserves coverage of imports
  - institutional quality
  - the 1982-89 time dummy
- Variables significant in at least two specifications:
  - total arrears as a share of external debt
  - openness (exports + imports as percent of GDP)
  - real GDP growth
  - inflation
  - non-fuel commodity price index
- Variables significant in one specification include short-term external debt to reserves, IMF credit outstanding in percent of quota, current account balance as percent of GDP, international financial integration, and US interest rate.
- Variables generally insignificant: remaining explanatory variables (signs of estimated coefficients generally as expected).
- Signs and interpretation (exact relationships preserved from regressions):
  - Lower reserves associated with higher likelihood of arrears.
  - Higher growth reduces likelihood; higher inflation increases likelihood (inflation statistically significant for protracted arrears only).
  - Better institutional quality decreases likelihood across specifications.
  - Civil unrest not statistically significant in regressions, caveated by data limitations.
  - Debt burden measures (external debt to GDP; debt service to exports) not statistically significant—possibly reflecting IMF preferred creditor status.
- Summary characterization: history of IMF arrears, higher total arrears burden, lower reserves, higher inflation, lower GDP growth, lower institutional quality, and weak non-fuel commodity prices increase likelihood of IMF arrears; arrears occurrence was more likely during 1982-89.

### Key coefficients and statistics (selected exact values reported)
- From pooled logit full specification (Table 2; column labels preserved):
  - Arrears to Fund in previous 5 years: 0.768; 1.137**; 1.160***; 1.421***; 1.675**
  - Total arrears as percent of external debt: 0.058***; 0.027**; 0.012; 0.002; 0.054***
  - Reserves as percent of imports: -0.916**; -0.783**; -0.382***; -0.167**; -0.597**
  - Real GDP growth: -0.058; -0.094***; -0.054**; -0.047**; -0.024
  - Inflation: 0.006**; 0.001; 0.002; -0.000; 0.008***
  - Institutional quality: -1.646***; -2.293**; -1.012**; -0.312; -1.903***
  - Non-fuel commodity price index: -0.107; -0.037**; -0.028***; -0.026***; -0.026
  - Time dummy 1982-1989: 4.953***; 1.388**; 1.308***; 1.217***; 3.038***
  - Observations: 1098; 1098; 1098; 1098; 1149
  - No. of arrears cases: 1240; 1222; 676; 3 (table shows "No. of arrears cases 124012226763")
  - Pseudo R-squared: 0.50; 0.38; 0.29; 0.23; 0.64
- Parsimonious specification (Table 3 selected):
  - Total external debt as percent of GDP: 0.006; 0.006***; 0.005***; 0.004**; 0.005
  - Arrears to Fund in previous 5 years: 0.290; 1.396***; 1.468***; 1.596***; 2.154***
  - Reserves as percent of imports: -0.546**; -0.280*; -0.231***; -0.179***; -0.612***
  - Real GDP growth: -0.069**; -0.069***; -0.039*; -0.029*; -0.030
  - Institutional quality: -1.250**; -0.838; -0.277; -0.147; -2.485***
  - Non-fuel commodity price index: -0.040; -0.033***; -0.029***; -0.020***; 0.012**
  - Time dummy 1982-1989: 2.374***; 1.791***; 1.388***; 1.175***; 0.633
  - Observations: 1531; 1531; 1531; 1531; 1606
  - No. of arrears cases: 1772; 197; 497; 1992 (table shows "No. of arrears case s177219741992")
  - Pseudo R-squared: 0.29; 0.26; 0.23; 0.22; 0.42

### Robustness checks and predictive marginal probabilities
- Robustness: alternative specifications (replace US real interest rate with world GDP growth; drop 1982-89 dummy; add commodity-exporter dummy and interaction; drop institutional quality) yield few changes in coefficient size or significance; main conclusions robust.
- Baseline prediction (column (2) in Table 2; arrears ≥ 1 month):
  - Explanatory variables at mean: predicted probability = 0.06 percent
  - At median values: predicted probability = 0.08 percent
  - Unconditional probability (sample-based): around 3.7 percent (40 protracted arrears country-year observations / 1,089 country-year observations for regression data availability)
  - For all 185 protracted arrears country-year observations and 3,354 total country-year observations, unconditional probability = 5.5 percent
- Marginal probability examples (Table 6):
  - Baseline π (in %): 0.06
  - previous arrears = 0 => 0.03
  - previous arrears = 1 => 0.10
  - 10 percentage points higher arrears to debt => 0.07
  - 1 month more reserves (to imports) => 0.03
  - 1% higher growth => 0.05
  - 0.1 point higher institutional quality => 0.05
  - 10 point higher commodity price index => 0.04
  - time dummy = 0 => 0.04; time dummy = 1 => 0.18
- Interpretation: predicted probabilities low on average due to rarity of arrears; worst-recorded covariate values predict near-certain arrears; best-recorded values predict near-zero probability.

### VI. Duration analysis of protracted arrears
- Observed durations:
  - Life-times range from ~8 months (Dominican Republic) to 23 years (Liberia).
  - Zimbabwe, Somalia, and Sudan remained in arrears for 14.9, 28.5, and 31.5–32 years respectively (ongoing at time of writing).
  - Among resolved cases, 22 resolved in under 10 years and five in more than 10 years.
- Econometric approach:
  - Proportional hazard model with exponential baseline hazard (constant baseline hazard): h(t|X) = exp(β0) exp(X'β)
  - Monthly transition rates modeled assuming constant monthly hazards over a year; regressors on annual country-year basis; ongoing arrears treated as right-censored; estimation by maximum likelihood.
- Covariates chosen for duration model:
  - IMF credit outstanding to quota (higher expected to increase duration)
  - Real GDP growth (higher expected to decrease duration)
  - Episodes of civil unrest (expected to increase duration)
  - Share of exports to advanced economies (proxy for advanced-economy support; higher share expected to reduce duration)
- Non-parametric survival findings (Kaplan-Meier and Nelson-Aalen):
  - Non-fragile countries exit protracted arrears faster; fragile states have median life-time about twice that of non-fragile at ~9 years.
  - Non-fragile protracted arrears cease after 18 years; fragile states continue past 20 years with survival probability ≈ 25 percent.
  - Nelson-Aalen: steady exits until about year five, slowing until ~year 23 when ongoing cases remain; hazard rates decline to zero sooner for non-fragile states.
- Parametric hazard results (Table 8 excerpts; signs and hazard ratios preserved):
  - Fragile states (exponential model):
    - IMF credit outstanding as percent of quota: -0.007**  hazard ratio 0.993  (0.003)
    - Real GDP growth: 0.062***  hazard ratio 1.064  (0.019)
    - Civil unrest: -0.165*  hazard ratio 0.848  (0.074)
    - Share of exports to advanced economies: 0.028**  hazard ratio 1.028  (0.011)
    - Constant -2.603***  hazard ratio 0.074  (0.977)
    - Log-likelihood -18.61; Number of observations 130
  - Non-fragile states (exponential model excerpts):
    - IMF credit outstanding as percent of quota: -0.005  hazard ratio 0.995  (0.003)
    - Real GDP growth: 0.113**  hazard ratio 1.119  (0.056)
    - Civil unrest: -0.235*  hazard ratio 0.790  (0.110)
    - Share of exports to advanced economies: 0.044**  hazard ratio 1.045  (0.023)
    - Alternative panels report slightly different coefficients (see source for exact table details); Number of observations 67 / 63 in alternate panels.
- Interpretation:
  - More IMF credit outstanding and more civil unrest lower the hazard of arrears clearance (lengthen duration), particularly in fragile countries.
  - Higher real GDP growth and higher export share to advanced economies increase clearance hazard (shorten duration); export-share effect more pronounced in non-fragile states.
- Predicted durations and sensitivity (Tables 9–10 selected findings):
  - With all variables at sample means predicted mean (median) duration ≈ 10.3 (7.2) years.
  - With 10 percentage points less IMF credit outstanding or a one-point drop in civil unrest score, predicted mean (median) duration drops to ≈ 6.1 (5.0) years.
  - An extra percentage point of growth has relatively small effect on duration.
  - Predicted median durations track actual medians reasonably; predicted average for fragile states higher than actual average due to inclusion of ongoing cases.
  - Ongoing-case observed durations: Somalia 28.5 years; Sudan 31.5 years; Zimbabwe 14.9 years.

### Conclusions and policy implications
- Main correlates of arrears occurrence:
  - Arrears to the IMF in previous five years, reserves coverage of imports, and institutional quality.
- Main correlates of protracted-arrears duration:
  - IMF credit outstanding to quota, real GDP growth, civil unrest episodes, and share of exports to advanced economies.
- Political factors:
  - Descriptive evidence indicates political triggers matter for both emergence and duration; formal estimation limited by data and non-random missingness.
- Policy recommendations grounded in analysis:
  - Increase scrutiny of repeat arrangements in light of borrowing country challenges.
  - Recognize that a peaceful political environment shortens predicted time to arrears clearance; reducing fragility is key.
  - IMF approval and review procedures (access norms and limits, debt sustainability reviews) are necessary first-line defenses against arrears accumulation.
- Case narratives for remaining protracted arrears (summary):
  - Sudan: prolonged conflicts and lack of creditor financing assurances for debt relief hinder clearance.
  - Somalia: institutions re-established only recently; IMF recognized the FGS in April 2013 following Federal Government formation in August 2012.
  - Zimbabwe: disruptive macroeconomic policies and politically driven development agenda isolated the country internationally.
  - Clearance requires decisive political turnaround and strengthening of macroeconomic policy frameworks.
- Suggested research agenda:
  - Role of IMF program design and conditionality for arrears occurrence.
  - Analysis of arrears magnitude.
  - Dynamics among global/local factors and arrears occurrence.

*Source — _wp16225 - References (PDF) — content unit provided.*

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

### _wp16225 - References

### I. Introduction — recent history and current state of IMF arrears
- Protracted arrears are defined as obligations overdue by six months or more.
- The IMF experienced steady accumulation of protracted arrears by members in the 1980-90s.
- Today there are only three remaining protracted arrears cases that started in 1984, 1987, and 2001: Sudan, Somalia, and Zimbabwe.
- No IMF member has accumulated protracted arrears since 2001.
- The most recent resolution of protracted arrears noted in the text was in March 2008 when Liberia repaid SDR 543 million after 22 years of ineligibility to borrow from the IMF due to qualifying for HIPC debt relief.
- Zimbabwe was still in arrears at the time of writing the paper; Zimbabwe cleared its arrears on October 20, 2016.
- Short-term arrears: members missing payments to the IMF for up to 180 days. On average about 20 countries incurred short-term arrears per year in the 1980s and 1990s; in 2015 only a handful of members accumulated short-term arrears and one has to go back to 2006 to count ten country cases of short-term arrears.
- Objective of the paper: re-examine determinants of arrears to the IMF and analyze their duration, with new data capturing political factors. Structure: literature review (Section II), stylized facts (Section III), determinants/triggers and logit estimation (Section IV), duration modeling (Section V), conclusion (Section VI).

### II. Literature review — prior empirical findings on IMF arrears
- Few studies focus specifically on arrears to the IMF; most literature addresses sovereign default on private or Paris Club debt.
- Aylward and Thorne (1998):
  - Used a logit model for 138 countries over 1976-93 (1,410 country-year observations; 103 arrears observations).
  - Defined arrears as obligations overdue by at least three months.
  - Concluded arrears more likely with larger use of IMF credit (in terms of quota), higher inflation, lower per capita income, or slower economic growth.
- Oka (2003):
  - Analyzed 118 countries over 1980-97 (1,375 country-year observations; 63 arrears observations).
  - Defined arrears as overdue for at least six months.
  - Found arrears more likely when IMF debt outstanding relative to quota or IMF debt service to exports is high, reserves to imports or export growth are low, or when arrears to other external creditors are rising.
  - Early warning indicators issued many false alarms despite identifying arrears-prone countries.
- Contribution of this paper:
  - Study of both occurrence and duration of arrears (duration is new relative to Aylward and Thorne (1998) and Oka (2003)).
  - Uses expanded dataset 1975-2015 and different modeling strategy.
  - Analyzes protracted arrears (longer than six months) and arrears of other durations (more than a day, a week, a month).
- Small number of protracted arrears observations: thirty instances of protracted arrears since IMF founding in 1945; 189 IMF members of which only 27 incurred arrears in the last 75 years — treated as the population of events rather than a sample. Small observation count affects econometric power; authors augment statistical analysis with descriptive and qualitative analyses.

### III. Stylized facts — comparative and dynamic patterns
- Country groups compared: countries with protracted arrears, countries with short-term arrears (short-term arrears defined as overdue for more than 7 days and less than 6 months), and countries with IMF credit outstanding but not in arrears.
- Key comparative findings (exact figures preserved):
  - Civil unrest: a six times higher percentage of countries with protracted arrears had an above average civil unrest magnitude score (around 18 percent) compared to the no-arrears group (3 percent) and 4.5 times higher compared to the short-term arrears group (4 percent).
  - Institutional development (CPIA score below 2.5): around 38 percent of protracted arrears countries versus 14 percent (short-term arrears) and 12 percent (no arrears).
  - IMF credit outstanding as percent of quota above 150 percent: 40 percent of protracted arrears cases versus around half that percentage in the comparator groups.
  - Reserves in months of imports above 3 months: 10 percent in protracted arrears group, 30 percent for short-term arrears, 50 percent for all other countries with IMF credit outstanding.
  - Total external debt to GDP ratio above 50 percent: around 73 percent of protracted arrears cases, 58 percent of short-term arrears cases, and 44 percent of all other countries.
  - Inflation rate above 10 percent: 74 percent (protracted arrears) versus 45 percent (short-term arrears) and 41 percent (no arrears).
  - Real GDP growth: 74 percent of protracted arrears countries experienced positive growth (i.e., 26 percent shrank) compared with 81 percent (short-term arrears) and 83 percent (no arrears).
- Dynamics around arrears occurrence and clearance:
  - Protracted arrears countries had higher IMF credit outstanding than short-term arrears countries at occurrence and for five years before and after.
  - Protracted arrears countries were poorer and had worse macroeconomic and political indicators than short-term arrears countries throughout the window.
  - Civil unrest increased steeply in the years leading up to protracted arrears; civil unrest dropped shortly before and after arrears resolution but remained higher than in short-term arrears countries.
  - External debt to GDP rose more steeply for protracted arrears countries in the years preceding arrears.
  - IMF credit outstanding to quota ratio peaks about three years before occurrence of protracted arrears, decreases afterwards, and approaches short-term arrears country levels at clearance.
  - Reserves in months of imports were persistently lower for protracted arrears countries, converging only after clearance.
  - Growth was lower on average for protracted arrears countries across the full time span. Inflation rose immediately before and after protracted arrears onset. Exports to GDP behaved oppositely. Many indicators align with short-term arrears values at or shortly after clearance.
- Note on sample sizes for figure averages: average number of observations is 23 for start and clearing of protracted arrears, and 309 for short-term arrears.

### IV. Triggers and common determinants of arrears
A. Triggers — classification and examples
- Triggers are classified into three types:
  - Political (e.g., wars, political acts).
  - Global (affecting many economies beyond individual control; e.g., commodity price swings, weather events, debt crisis).
  - Country-specific (structural or macroeconomic issues; e.g., corruption, poor policy implementation).
- Notable observations:
  - Afghanistan and Iraq accrued arrears without IMF credit outstanding (due to net SDR holdings below SDR allocations) owing to wars, political turbulence, and international sanctions, illustrating political trigger risk even when no IMF credit is outstanding.
  - Nine cases (Cambodia, Democratic Republic of Congo (twice), Honduras, Peru, Vietnam, Zambia (twice), and Zimbabwe) were characterized by authorities deciding not to service IMF obligations even when resources appeared sufficient; these account for nearly one third of all arrears cases.
  - Political triggers contributed to arrears in 21 cases (70 percent of total), often in combination with country-specific or global triggers.
- Global triggers:
  - Figure 5 lists 13 arrears cases at least partly owing to global triggers. Only the Dominican Republic is ascribed to global triggers alone.
  - Most global factors involved declines in commodity prices for commodity-exporting countries.
- Country-specific triggers:
  - Country-specific triggers played a part in 21 arrears cases (majority). They alone caused five cases; paired with global triggers in three cases; paired with political triggers in four cases; and combined with both global and political triggers in nine cases.
  - Country-specific triggers include poor policy implementation, corruption, loss of aid, and inappropriate exchange rate policy, among others.
- Mapping to commodity prices:
  - For several cases, local troughs and peaks in commodity price time series coincide with the beginning and end of arrears (Figure 7 maps commodity series and protracted arrears start/end dates for affected countries).

B. Econometric analysis — model, data, and key design choices
- Model: pooled logit model estimating probability of arrears occurrence.
  - Arrears accumulation formalized as P* = F(X'β + ε) (equation (1) in text).
  - Observed dependent variable is a binary indicator: 1 for instances of arrears occurrence, 0 otherwise.
  - Logit transformation applied to odds ratio; maximum likelihood estimation used (equation (2) in text).
- Dependent variable specifications:
  - Multiple measures used depending on arrears duration: >6 months (protracted arrears), >1 month, >1 week, and all arrears irrespective of duration.
  - For each arrears case only the first country-year observation is used as the marker of arrears occurrence; years after the first year of occurrence up to clearance are set as missing for protracted arrears to avoid state-dependence confounding determinants of entry vs duration.
  - For completeness, column (5) in Table 2 (not reproduced here) shows results where dependent variable takes value one for all years a country was in protracted arrears.
- Data and coverage:
  - Dataset includes all countries with IMF credit outstanding in a given year: 3,354 country-year observations over 1974-2014.
  - This comprises 142 different countries, of which 114 experienced short-term or protracted arrears (many repeatedly).
  - Incompleteness of country or variable coverage leads to exclusion of some protracted arrears countries from regressions. Excluded countries on average experienced more civil conflict and worse institutional quality at time of arrears occurrence (Figure 8). Exclusion is non-random and likely biases estimated coefficients toward a lower bound for political variables’ impacts.
  - The small number of protracted arrears observations motivates supplementing econometric results with qualitative analysis.
- Explanatory variables (Table 1) across seven categories and expected impacts (exact variable names and expected sign preserved):
  - total external debt as percent of GDP — Positive
  - short-term external debt to reserves — Positive
  - debt service on external debt as percent of exports — Positive
  - IMF credit outstanding as percent of quota — Positive
  - multilateral debt as percent of total external debt — Positive
  - arrears to IMF in previous 5 years (dummy variable) — Positive
  - total arrears (to IMF and other creditors) as percent of external debt — Positive
  - reserves as percent of imports — Negative
  - current account balance as percent of GDP — Ambiguous
  - total foreign assets and liabilities to GDP (measure of international financial integration) — Positive
  - exports and imports as percent of GDP (openness) — Negative
  - per capita GDP — Negative
  - real GDP growth — Negative
  - Inflation — Positive
  - “civil unrest” magnitude — Positive
  - institutional quality (CPIA) — Negative
  - non-fuel commodity price index — Negative
  - US real interest rate (alternatively global GDP growth rate) — Negative
  - time dummy for 1982-89 — Positive (controls for 1980s debt crisis effects)
- Methodological notes:
  - CPIA index only available from 2005 onwards; CPIA treated as slow-moving and 2005 value imputed for all years considered.
  - Fiscal position indicators could not be included due to data limitations.
  - Fixed effects estimation is discussed as illustrative (Annex II) but pooled logit used for main results because fixed effects would exclude countries that never incurred arrears and would severely restrict the data set.
  - Prior literature (Aylward and Thorne 1998) found state dependence: being in arrears in previous year strongly predicts arrears next year; this motivates using only first-year observations for protracted arrears spells.

### V. Key empirical and descriptive conclusions (synthesized from text)
- Countries that accumulated protracted arrears performed worse across a wide range of indicators (institutional quality, civil unrest, reserves, debt burdens, inflation, growth) than countries with short-term arrears or no arrears.
- Political factors are prominent: political triggers (wars, sanctions, policy refusals to service debt) are implicated in a large share of protracted arrears cases and often interact with country-specific and global triggers.
- Global factors, especially commodity price declines, played a critical role in a notable subset of cases—sometimes coinciding with arrears start/end dates.
- Country-specific policy failures remain a major category of triggers and often are the reason why countries seek IMF programs; such triggers can occur alone or combined with political/global shocks.
- Econometric analysis uses a pooled logit with rich explanatory variables; small sample of protracted arrears and non-random missingness (especially of politically troubled countries) reduce statistical power and likely bias estimates for political variables downward, underscoring the importance of complementary qualitative analysis.

_Italic: Source — _wp16225 - References (PDF) — content unit provided._

### Annex I.

### Annex I.

### Major findings from pooled logit estimations
- Variables statistically significant across at least four out of five regressions:
  - arrears to the IMF in previous five years
  - reserves coverage of imports
  - institutional quality
  - the 1982-89 time dummy
- Variables statistically significant in at least two out of five specifications:
  - total arrears as a share of external debt
  - openness (exports + imports as percent of GDP)
  - real GDP growth
  - inflation
  - non-fuel commodity price index
- Variables significant in one specification:
  - short-term external debt as percent of reserves
  - IMF credit outstanding in percent of quota
  - current account balance as percent of GDP
  - international financial integration
  - US interest rate
- Variables generally statistically insignificant: remaining explanatory variables (signs of estimated coefficients are as expected where estimated).
- Signs and interpretation:
  - Lower reserves associated with higher likelihood of arrears.
  - Higher growth reduces likelihood of arrears; higher inflation increases likelihood of arrears (inflation coefficient statistically significant for protracted arrears cases only).
  - Better institutional quality decreases likelihood of arrears across all specifications.
  - Political environment measure (civil unrest) not statistically significant in regressions, with caveat that data limitations may explain insignificance.
  - Debt burden measures (external debt to GDP; debt service to exports) not statistically significant—possibly reflecting the IMF’s preferred creditor status.
- Summary conclusion from Table 2:
  - A country with a history of arrears to the IMF, higher total arrears burden, lower international reserves, higher inflation, lower GDP growth, lower institutional quality, and weak non-fuel commodity prices is more likely to accumulate arrears to the IMF than its opposite.
  - Arrears occurrence was more likely during 1982-89 than at other times.

### Key coefficients and statistics (selected, as reported)
- From Table 2 (pooled logit full specification; column labels include (1) all protracted arrears, (2) arrears longer than 1 month, (3) arrears longer than 1 week, (4) all arrears (short-term and protracted), (5) beginning and ongoing spells for all protracted):
  - Arrears to Fund in previous 5 years: 0.768; 1.137**; 1.160***; 1.421***; 1.675** (clustered standard errors shown in table)
  - Total arrears as percent of external debt: 0.058***; 0.027**; 0.012; 0.002; 0.054*** 
  - Reserves as percent of imports: -0.916**; -0.783**; -0.382***; -0.167**; -0.597** 
  - Real GDP growth: -0.058; -0.094***; -0.054**; -0.047**; -0.024 
  - Inflation: 0.006**; 0.001; 0.002; -0.000; 0.008*** 
  - Institutional quality: -1.646***; -2.293**; -1.012**; -0.312; -1.903*** 
  - Non-fuel commodity price index: -0.107; -0.037**; -0.028***; -0.026***; -0.026 
  - Time dummy 1982-1989: 4.953***; 1.388**; 1.308***; 1.217***; 3.038*** 
  - Observations: 1098; 1098; 1098; 1098; 1149
  - No. of arrears cases: 1240; 1222; 676; 3 (table shows multiple counts) — note: table lists "No. of arrears cases 124012226763" (preserve as in source)
  - Pseudo R-squared: 0.50; 0.38; 0.29; 0.23; 0.64

- From Table 3 (parsimonious specification):
  - Total external debt as percent of GDP: 0.006; 0.006***; 0.005***; 0.004**; 0.005
  - Arrears to Fund in previous 5 years: 0.290; 1.396***; 1.468***; 1.596***; 2.154*** 
  - Reserves as percent of imports: -0.546**; -0.280*; -0.231***; -0.179***; -0.612*** 
  - Real GDP growth: -0.069**; -0.069***; -0.039*; -0.029*; -0.030 
  - Institutional quality: -1.250**; -0.838; -0.277; -0.147; -2.485*** 
  - Non-fuel commodity price index: -0.040; -0.033***; -0.029***; -0.020***; 0.012** 
  - Time dummy 1982-1989: 2.374***; 1.791***; 1.388***; 1.175***; 0.633
  - Observations: 1531; 1531; 1531; 1531; 1606
  - No. of arrears cases: 1772; 197; 497; 1992 (table shows "No. of arrears case s177219741992")
  - Pseudo R-squared: 0.29; 0.26; 0.23; 0.22; 0.42

### Robustness checks
- Alternative specifications tested in Table 4:
  - Replacing US real interest rate with world (real) GDP growth.
  - Dropping the time dummy 1982-89.
  - Adding a dummy for commodity exporters and its interaction with non-fuel commodity price index.
  - Dropping the control for institutional quality.
- Robustness results summary:
  - Few coefficients change in significance or size across alternative specifications.
  - Main conclusions remain fairly robust to alternative specifications.
  - Example coefficients from Table 4 (selected):
    - Reserves as percent of imports: -0.854**; -0.158**; -1.068*; -0.178**; -0.989**; -0.166**; -0.799**; -0.145*** (across columns)
    - Total arrears as percent of external debt: 0.056***; 0.002; 0.036**; 0.000; 0.061***; 0.002; -0.004; -0.001
    - Arrears to Fund in previous 5 years: 0.777; 1.445***; -0.095; 1.331***; 0.817; 1.430***; 0.323; 1.503*** 
  - Observations vary across robustness specifications (e.g., 1098; 1506; 1531) and No. of arrears cases vary (e.g., 1226; 712; 1435 etc. as shown in table).

### Predicting protracted arrears (marginal probabilities)
- Baseline prediction (column (2) in Table 2; arrears of 1 month and longer):
  - With all explanatory variables at their mean, predicted probability of arrears occurrence: 0.06 percent
  - At median values: 0.08 percent
  - Unconditional probability (sample-based): around 3.7 percent (calculated as ratio of 40 country-year observations of protracted arrears and 1,089 total country-year observations based on data availability in the regression model)
  - For all 185 country-year observations of protracted arrears (irrespective of data availability) and 3,354 total country-year observations, unconditional probability would be 5.5 percent.
- Table 5 marginal probabilities (selected table entries preserved):
  - Variable coefficients and sample statistics used (column labels in source): 
    - Arrears to Fund in previous 5 years: coefficient 1.137; Mean 0.511; Median 1; Worst 10; Best 0
    - Total arrears as percent of external debt: coefficient 0.027; Mean 7.47; Median 0.001; Worst 09.55; Best 0.00
    - Reserves to imports: coefficient -0.783; Mean 3.18; Median 2.90; Worst 0.00; Best 14.55
    - Real GDP growth: coefficient -0.094; Mean 4.11; Median 4.47; Worst -41.89; Best 36.94
    - Institutional quality: coefficient -2.293; Mean 3.08; Median 3.20; Worst 2.20; Best 3.90
    - Non-fuel commodity price index: coefficient -0.037; Mean 102.31; Median 94.05; Worst 76.03; Best 190.04
    - Time dummy 1982-1989: coefficient 1.388; Mean 0.18; Median 1; Constant 5.211
  - π (in %) values reported: 0.06; 0.08; 99.89; 0.00 (for mean, median, worst, best respectively)
- Table 6: marginal probabilities for changes (selected):
  - Baseline π (in %): 0.06
  - previous arrears = 0 => 0.03
  - previous arrears = 1 => 0.10
  - 10 percentage points higher arrears to debt => 0.07
  - 1 month in reserves (to imports) extra => 0.03
  - 1% growth extra => 0.05
  - 0.1 point higher institutional quality index => 0.05
  - 10 point increase in commodity price index => 0.04
  - time dummy = 0 => 0.04
  - time dummy = 1 => 0.18
- Interpretation:
  - Predicted probabilities are very small on average because arrears accumulation is a rare occurrence over the sample period.
  - If explanatory variables were at their worst recorded values in the sample, the model predicts arrears emergence as nearly certain; conversely, virtually nil at best values.

### Duration analysis of protracted arrears
- Focus: protracted arrears (short-term arrears typically resolve within six months or less).
- Observed durations:
  - Life-times of protracted arrears cases vary from around 8 months (Dominican Republic) to 23 years (Liberia).
  - Zimbabwe, Somalia, and Sudan remain in arrears for 15, 29, and 32 years respectively.
  - Among resolved cases, 22 resolved in under 10 years and five in more than 10 years.
- Econometric approach:
  - Proportional hazard model with exponential hazard function assumed (baseline hazard constant).
  - Hazard specification (as reported): h(t|X) = h0(t) exp(X'β) = exp(β0) exp(X'β)
  - Monthly transition rates from protracted arrears to clearance modeled, assuming monthly hazard rates constant over one year; regressors computed on an annual, country-year basis; ongoing arrears treated as right-censored; estimation by maximum likelihood.
- Choice of explanatory variables for duration model:
  - IMF credit outstanding to quota (higher outstanding credit expected to increase duration)
  - Real GDP growth (higher growth expected to decrease duration)
  - Episodes of civil unrest (expected to increase duration)
  - Percentage share of exports to advanced economies (proxy for support by advanced IMF members; higher share expected to reduce duration)
- Non-parametric survival analysis:
  - Kaplan-Meier unconditional survival functions estimated for all protracted arrears countries and separately for “fragile states” and “non-fragile” states.
  - Fragile states account for about half of protracted arrears cases; among three continuing arrears cases the longest two are fragile states.

*Source: Annex I. (from the provided IMF content unit).*

### Section IV. B), but found the one reported here the best intuitive and statistical fit.

### _wp16225 - Section IV. B), but found the one reported here the best intuitive and statistical fit.

### Survival dynamics of protracted arrears
- The Kaplan-Meier graph for all protracted arrears countries begins at 1 (or 100 percent probability of remaining in arrears) at spell start.
- After year one, once arrears cases are protracted, the probability of survival as a protracted arrears case:
  - Is much lower for non-fragile countries than for fragile countries.
  - At about 9 years, the median life-time as a protracted arrears case for fragile states is about twice the median life-time for non-fragile states.
  - Non-fragile protracted arrears cases cease after eighteen years.
  - Fragile protracted arrears cases continue past the twenty-year mark with a survival probability of about 25 percent.
- Nelson-Aalen cumulative hazard estimates:
  - For all arrears cases suggest a steady exit from arrears until about year five, which slows progressively until about year 23 when only ongoing cases remain.
  - For non-fragile states hazard rates change more rapidly and decline to zero sooner than for fragile states.
  - For fragile states cumulative hazard stabilizes at about year 23 as fragile arrears cases become the longest surviving ones.

### Parametric hazard model results and interpretation
- Table 7 / duration-model overview:
  - Positive estimated coefficients (hazard ratios > 1) imply a one-unit increase raises the probability arrears clearance occurs in a given time interval.
  - Negative estimated coefficients (hazard ratios < 1) imply a one-unit increase lowers the probability arrears clearance occurs in a given time interval.
  - Intuition and main correlations:
    - More IMF credit outstanding and more civil unrest are associated with lower probability of arrears clearance.
    - Higher real GDP growth and a higher export share to advanced economies raise the probability of arrears clearance.
  - Robustness: exponential hazard model compared to semiparametric Cox model — both models broadly similar.

### Duration analysis split by fragility (Table 8)
- Fragile states (exponential model / Cox model column excerpts):
  - IMF credit outstanding as percent of quota -0.007**  hazard ratio 0.993  (0.003)
  - Real GDP growth 0.062***  hazard ratio 1.064  (0.019)
  - Civil unrest -0.165*  hazard ratio 0.848  (0.074)
  - Share of exports to advanced economies 0.028**  hazard ratio 1.028  (0.011)
  - Constant -2.603***  hazard ratio 0.074  (0.977)
  - Log-likelihood -18.61
  - Number of observations 130
- Non-fragile states (exponential model / Cox model column excerpts):
  - IMF credit outstanding as percent of quota -0.005  hazard ratio 0.995  (0.003)
  - Real GDP growth 0.113**  hazard ratio 1.119  (0.056)
  - Civil unrest -0.235*  hazard ratio 0.790  (0.110)
  - Share of exports to advanced economies 0.044**  hazard ratio 1.045  (0.023)
  - (Alternate panel in table)
    - IMF credit outstanding as percent of quota -0.014** hazard ratio 0.986 (0.007)
    - Real GDP growth 0.074*** hazard ratio 1.076 (0.018)
    - Civil unrest -0.277* hazard ratio 0.758 (0.111)
    - Share of exports to advanced economies -0.010 hazard ratio 0.990 (0.012)
    - Constant 1.243  hazard ratio 3.464 (3.730)
    - Log-likelihood -5.17 / -9.40 (two panels)
    - Number of observations 67 / 63 (total 6763 shown in table header)
- Interpretation:
  - IMF credit outstanding to quota and civil unrest are relevant determinants of arrears duration in fragile countries.
  - Export share to advanced economies is significant only in the model for non-fragile states.
  - Slightly different forces operate in fragile versus non-fragile countries.

### Predicted durations and sensitivity analysis (Tables 9 and 10)
- Actual and predicted (high-level findings):
  - Predicted mean and median duration of protracted arrears is much longer for fragile states than for non-fragile states.
  - Predicted median durations do not deviate far from actual medians; predicted average duration for fragile states is four time higher than the actual average (bias due to inclusion of ongoing cases).
  - Ongoing arrears cases observed durations: Somalia 28.5 years, Sudan 31.5 years, Zimbabwe 14.9 years (actual averages shorter than predicted).
- Predicted failure time by covariates (upper panel sample means and variations; lower panel predicted durations):
  - Variables (sample means / variation example):
    - IMF credit outstanding as percent of quota 147.09 137.09
    - Real GDP growth 2.12 3.12
    - Civil unrest 2.76 1.76
    - Share of exports to advanced economies 60.24 70.24
  - Predicted durations (Duration in years):
    - Predicted mean failure time 10.3 6.1 9.7 6.6 7.8
    - Predicted median failure time 7.2 5.0 6.7 4.6 5.4
  - Sensitivity highlights:
    - With all variables at their means the predicted mean (median) duration is about 10 (7) years.
    - With 10 percentage points less IMF credit outstanding or a one point drop in the civil unrest score, the predicted mean (median) duration drops to 6 (5) years.
    - An extra percentage point of growth has relatively small effects on duration.
    - A 10 percentage point higher share of exports to advanced economies slightly reduces predicted duration.
  - Policy-relevant inference: large improvements in duration of arrears come from lower accumulation of credit to the IMF and from reduction of civil unrest incidences.

### Conclusions and policy implications
- Main correlates of occurrence of arrears:
  - Arrears to the IMF in previous five years, reserves coverage of imports, and institutional quality.
- Main correlates of duration of protracted arrears:
  - IMF credit outstanding to quota, real GDP growth, episodes of civil unrest, and percentage share of exports to advanced economies.
- Political factors:
  - Descriptive evidence suggests political factors matter for both emergence and duration of arrears; formal estimation limited by data.
- Policy recommendations grounded in analysis:
  - Increase scrutiny of repeat arrangements in line with challenges faced by the borrowing country.
  - Recognize the importance of a peaceful political environment for arrears clearance; reducing fragility shortens predicted times to arrears clearance.
  - IMF’s approval and review procedures (access norms and limits, debt sustainability reviews) are necessary first-line defenses against arrears accumulation.
- Case narratives for remaining protracted arrears:
  - Sudan: prolonged conflicts and lack of financing assurances from creditors for debt relief have hindered arrears clearance.
  - Somalia: institutions only recently re-established; IMF recognized the FGS in April 2013 (recognition following formation of the Federal Government of Somalia in August 2012).
  - Zimbabwe: disruptive macroeconomic policies and political-driven development agenda cut it off from the international community.
  - Clearance in these cases will require a decisive turnaround in political fortunes and strengthening of the macroeconomic policy framework.
- Research agenda items remaining:
  - Implications of IMF program design and conditionality for occurrence of arrears.
  - Analysis of the magnitude of arrears.
  - Dynamic interaction among global/local factors and arrears occurrence.

### Annex and data notes (selected)
- Annex I: Overview of Protracted Arrears Cases and Inclusion in Econometric Model (lists country-level start dates, periods, and inclusion flags for econometric analysis; examples include Somalia Jul-87 ongoing; Sudan Jul-84 ongoing yes; Zimbabwe Feb-01 ongoing yes).
- Annex II: Fixed Effects Logit Model — selected coefficients and significance (examples):
  - Arrears to Fund in previous 5 years 0.359  (1)  / 1.158**  (2) / 0.919***  (3)
  - Reserves as percent of imports -1.732*** (1) / -0.437*** (2) / -0.214** (3)
  - Total arrears as percent of external debt 0.105** (1) / 0.007 (2) / 0.005 (3)
  - Observations 366 / 763 / 1103; No. of arrears cases 40 / 122 / 267; Pseudo R-squared 0.47 / 0.24 / 0.19
- Annex IV: Data and data sources (variable-by-variable sources include IMF Finance Department, IMF WEO, IMF IFS, Center for Systemic Peace, World Bank Group CPIA database, Lane and Milesi-Ferretti (2007), International Debt Statistics, Direction of Trade Statistics, IMF GAS, World Development Indicators).

*Source: _wp16225 - Section IV. B).*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2016/_wp16225.pdf_