## wpiea2020041-print-pdf

## Source details

**Canonical URL:** [wpiea2020041-print-pdf](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020041-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020041-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020041-print-pdf.pdf.json)

---

### I. Introduction — scope, motivation, and research questions
- Context:
  - Marked proliferation of violence and conflicts across developing countries over the past two decades, especially after the Arab spring from 2011.
  - World Bank (2017) estimate cited: more than 400,000 death toll and US$200-300 billion loss in GDP in Syria since the conflict started in 2011.
- Research gap and objectives:
  - Literature documents adverse effects of instability but lacks a comprehensive empirical study on the impact of conflict and political instability on systemic banking crises.
  - Study evaluates the impact of conflicts and political instability on the probability of systemic banking crises in 92 developing countries over 1970–2016.
  - Additional objectives:
    - Analyze cross-border spillovers from neighboring countries’ conflicts.
    - Examine whether conflict duration raises banking-crisis probability.
    - Explore transmission channels through which conflict and political instability affect banking-crisis likelihood.

### III. Data, definitions, and sample
- Sample and outcomes:
  - Yearly data for 92 emerging and developing countries during 1970–2016.
  - Sample covers 191 episodes of banking crises.
  - Banking-crisis definition (Laeven and Valencia, 2018): requires (i) significant signs of financial distress and (ii) significant banking policy intervention measures; interventions considered significant if at least three out of six measures are used (a–f as listed in source).
- Conflict and political-instability measures:
  - Civil war data from UCDP:
    - Internal armed conflict intensity-scaled: value of 1 if related death toll in a given year is 25–999, 2 if 1000 or more, 0 otherwise.
    - Binary variable equal to 1 if a civil conflict happens in the country and 0 otherwise.
  - Six political-instability indicators from Banks and Wilson (2019): (i) Government cabinet changes; (ii) Changes in effective executive; (iii) Anti-government demonstrations (≥100 people); (iv) Major government crises; (v) General strikes (≥1,000 workers across employers aimed at national government policies); (vi) Political assassinations (politically motivated murder or attempted murder of a high government official or politician).
  - Additional robustness sources: ICRG, COW, Political Terror Scale, GTD, Major Episodes of Political Violence Database, Coup d’Etat Events Database, State Fragility Index.
- Controls and sources:
  - Real exchange rate, inflation, external debt (% of GDP), GDP per-capita, real GDP growth, terms of trade from IMF WEO.
  - M2/reserves and credit growth from World Bank WDI.
  - Degree of democracy from Polity IV Project.

### Stylized facts and unconditional statistics
- Associations and unconditional probabilities:
  - Positive relationship between number of countries in conflict and number of countries in banking crises; major waves of conflict tend to coincide with higher rates of banking crises.
  - Unconditional probability of a banking crisis: 6.4 percent (also reported as 6.3 percent elsewhere: 191 crises out of 3,026 observations = 6.3 percent).
  - Conditional probabilities (examples preserved exactly):
    - Conditional probability of a banking crisis in a year without a conflict: 5.3 percent.
    - Conditional probability in years of conflict: 10 percent.
    - For general strikes and political assassinations: probability increases from 5.7 percent in years without these events to 12.5 percent in years with them.
  - Joint conditional probability examples:
    - Conditional probability of a banking crisis after a joint occurrence of a conflict and a fiscal crisis: 16.7 percent.
    - Conditional probability of a banking crisis after a conflict when a fiscal crisis does not materialize: 4.2 percent.
  - Statistical tests: Pearson chi-squared and t-tests reported; t-test p-values reported in tables (example: T-test p-value 0.00 for difference in Table 2).

### IV. Empirical strategy — model, estimation, and controls
- Binary outcome model:
  - Pr(Y_iit = 1 | X_iit-1, C_i) = F(X_iit-1 β + C_i) = X_iit-1 β + C_i + ε_iit
  - Y_iit: binary indicator equal to 1 if a banking crisis occurs for country i at time t; X_iit-1: vector of explanatory variables lagged by one year.
- Addressing unobserved heterogeneity:
  - Fixed effects exclude countries that never experienced banking crises (40 percent of countries, 37 out of 92, had no banking crises 1970-2017).
  - Correlated random effects (Mundlak, 1978) specification used: C_i = ω + X̄_i δ + a_i; implemented by including time means of time-varying covariates. Advantages: controls time-constant unobservables while retaining countries with no crises.
  - Robustness: traditional probit and logit; probit fixed-effects bias correction (Fernández-Val and Weidner, 2016) — caveat that method excludes countries without banking crises.
- Lagged control variables (lagged by one year) and expected signs (preserved):
  - Real effective exchange rate, M2/reserves, inflation, credit growth, external debt, GDP per capita, GDP growth, terms of trade, degree of democracy.

### V. Empirical results — baseline estimates and key magnitudes
- Estimation framework and sample sizes:
  - Random effects-Mundlak binary-response (logit) estimates reported across alternative conflict and political-instability indicators.
  - Observations: 3,026 in columns with conflict measures; 2,972 or 2,991 in columns with some political-instability measures.
  - Number of countries: 92 in all columns.
- Model performance statistics (AUROC and related):
  - AUROC: 0.75; 0.744; 0.742; 0.739; 0.734; 0.734; 0.740; 0.737 across columns.
  - seAUROC: 0.0174; 0.0174; 0.0175; 0.0183; 0.0181; 0.0178; 0.0177; 0.0177.
  - Rho(LR): 0.482; 0.484; 0.460; 0.457; 0.471; 0.469; 0.465; 0.476. P-value(Rho): 0 in all columns.
  - Wald chi2: 326.9; 323.8; 335.6; 333.5; 327.5; 328.1; 335; 323.8.
  - Log likelihoods: -598.7; -602.1; -599.5; -600.6; -603.3; -602.4; -602.2; -603.1.
- Key conflict and political instability coefficients (selected exact values from Table 3):
  - Conflict (binary), t-1: 0.9097*** (standard error 0.245). Text interpretation: odds of a banking crisis are 2.5 times greater when a country is affected by a conflict; estimated conditional probability rises from 6.3 percent to 13.5 percent when in conflict.
  - Conflict (intensity), t-1: 0.4483*** (0.167).
  - Cabinet changes, t-1: 0.3620*** (0.126). Example: increasing cabinet changes from zero to four raises probability to 21.17 percent from unconditional 6.3 percent.
  - Changes in Effective Executive, t-1: 0.3277** (0.150).
  - Anti-Government Demonstrations, t-1: 0.0295* (0.017).
  - Government Crises, t-1: 0.2894** (0.134).
  - General Strikes, t-1: 0.2211** (0.094).
  - Assassinations, t-1: 0.0788* (0.042).
- Control-variable coefficients (reported exactly as in Table 3):
  - Exchange rate, t-1: -0.0903**; -0.0892**; -0.0931**; -0.0959**; -0.0910**; -0.1024**; -0.0987**; -0.0899** (standard errors ≈ 0.039).
  - M2/reserves, t-1: 0.5660***; 0.5531***; 0.5605***; 0.5658***; 0.5691***; 0.5675***; 0.5666***; 0.5675*** (standard errors ≈ 0.123–0.124).
  - Inflation, t-1: 1.0096***; 1.0140***; 1.0163***; 1.0224***; 1.0509***; 1.0638***; 1.0106***; 1.0148*** (standard errors ≈ 0.169–0.173).
  - Credit growth, t-1: -0.4867*; 0.3878; 0.3468; 0.2923; 0.3007; 0.3571; 0.2995; 0.3088 (standard errors ≈ 0.281–0.293).
  - External debt, t-1: 0.7992***; 0.7981***; 0.8284***; 0.8410***; 0.8267***; 0.8388***; 0.8137***; 0.8282*** (standard errors ≈ 0.144–0.145).
  - GDP per capita, t-1: -0.3966; -0.3913; -0.3530; -0.3983; -0.4916; -0.3989; -0.4452; -0.4245 (standard errors ≈ 0.313–0.318).
  - Economic growth, t-1: -0.0324*; -0.0312; -0.0257; -0.0277; -0.0324*; -0.0289; -0.0331*; -0.0340* (standard errors 0.019–0.020).
  - Terms of trade, t-1: -0.0286; 0.0910; -0.1673; -0.0626; 0.0235; -0.0540; 0.0422; -0.0172 (standard errors ≈ 1.204–1.218).
  - Degree of democracy, t-1: -0.0460**; -0.0469**; -0.0524***; -0.0540***; -0.0466**; -0.0514***; -0.0518***; -0.0483*** (standard errors 0.019).
- Duration effects (selected magnitudes preserved):
  - Duration matters: conflict lasting only one year has no significant effect; when lasting longer the impact becomes apparent and stronger.
  - Probability of experiencing a banking crisis is 25 percent when the conflict lasts 10 years.
  - Probability is 16.4 percent when conflict lasts two years.
- Spillovers from neighboring countries:
  - Conflicts and political instability in neighboring countries increase the likelihood of banking crises domestically but with smaller magnitude than direct domestic effects.
  - Example: a rise in the number of bordering countries affected by conflict from 0 to 3 increases the likelihood of banking crises from 6.3 percent to 11.2 percent (column example preserved).

### VI. Transmission channels, subsamples, and robustness
- Transmission channel evidence:
  - Fiscal crises identified as the primary transmission channel linking conflict/political instability to banking crises.
  - Conditional results (Table 2 and Table 7): when fiscal crisis and interaction included, coefficients on conflict/political instability become insignificant while fiscal crisis is positive and significant in all columns.
  - Joint probabilities:
    - Conditional probability of a banking crisis after a joint occurrence of a conflict and a fiscal crisis: 16.7 percent.
    - Conditional probability of a banking crisis after a conflict when a fiscal crisis does not materialize: 4.2 percent.
- Subsample results by income group:
  - Emerging markets: conflict raises probability of a banking crisis to 14.12 percent.
  - Low-income developing countries: conflict raises probability of a banking crisis to 17.15 percent.
  - Coefficient significance varies by subgroup and instability measure (detailed in source).
- Robustness checks — alternative data sources and specifications:
  - Alternative conflict/political-instability sources produce consistent positive and significant estimates: ICRG internal/external conflict indices; Marshall (2017) MEPV civil war and violence scores; COW civil war; GTD terrorism incidents; Political Terror Scale; State Fragility Index; assassination of ruling executive.
  - COW conflict example: being in conflict raises probability of a banking crisis from 6.3 percent to 17.7 percent.
  - Including additional covariates (global conditions, commodity prices, natural resource rents, portfolio inflows, real interest rate, control of corruption, export diversification, financial development) does not change baseline qualitative findings.
  - Alternative estimators (probit, logit, probit fixed effects bias-corrected) yield positive and significant coefficients on conflict and political instability (with sample caveats for fixed-effects estimators).

### Concluding findings and policy implications
- Main conclusions:
  - Conflicts and political instability are associated with higher probability of systemic banking crises.
  - Duration and persistence amplify the probability of banking crises (e.g., 16.4 percent at 2 years; 25 percent at 10 years).
  - Cross-border spillovers exist: neighboring-country conflicts and political instability raise domestic banking-crisis probability, but with lower magnitude than direct effects.
  - Fiscal crises are the primary transmission channel from conflict/political instability to banking crises.
- Policy implications (as stated in source):
  - Governments facing conflict and/or political instability should address root causes and mitigate negative effects through appropriate economic policy design and implementation.
  - Creating adequate fiscal space in normal times can reduce the likelihood of fiscal crises and thereby lower the probability of systemic banking crises.
  - Policy makers should monitor conflicts in neighboring countries because their banking systems may suffer negative spillovers even if not directly conflict-afflicted.

*Source: wpiea2020041-print-pdf — https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020041-print-pdf.pdf*

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

### wpiea2020041-print-pdf - References .............................................................................................................

### I. Introduction — scope, motivation, and research questions
- Context:
  - Marked proliferation of violence and conflicts across developing countries over the past two decades, especially after the Arab spring from 2011.
  - World Bank (2017) estimate cited: more than 400,000 death toll and US$200-300 billion loss in GDP in Syria since the conflict started in 2011.
- Research gap:
  - Literature documents adverse effects of instability on growth, public investment, trade, tourism, and fiscal outcomes, but lacks a comprehensive empirical study on the impact of conflict and political instability on systemic banking crises.
- Research objectives:
  - Evaluate the impact of conflicts and political instability on the probability of systemic banking crises in 92 developing countries over 1970–2016.
  - Analyze cross-border spillovers from neighboring countries’ conflicts.
  - Examine whether conflict duration raises banking-crisis probability.
  - Explore transmission channels through which conflict and political instability affect banking-crisis likelihood.

### III. Data, definitions, and sample
- Sample:
  - Yearly data for 92 emerging and developing countries during the period 1970–2016.
  - Sample covers 191 episodes of banking crises.
- Banking crisis definition (Laeven and Valencia, 2018):
  - Banking crisis defined as an event that meets two conditions:
    - (i) significant signs of financial distress in the banking system (bank runs, losses, and/or bank liquidations);
    - (ii) significant banking policy intervention measures in response to significant losses in the banking system.
  - Policy interventions are considered significant if at least three out of six measures are used:
    - (a) deposit freezes and/or bank holidays;
    - (b) significant bank nationalizations;
    - (c) bank restructuring fiscal costs (at least 3 percent of GDP);
    - (d) extensive liquidity support (at least 5 percent of deposits and liabilities to nonresidents);
    - (e) significant guarantees put in place;
    - (f) significant asset purchases (at least 5 percent of GDP).
- Conflict and political instability measures:
  - Civil war data from Uppsala Conflict Data Program (UCDP):
    - Internal armed conflict intensity-scaled measure: value of 1 if related death toll in a given year is 25–999, 2 if 1000 or more, 0 otherwise.
    - Also constructed a binary variable equal to 1 if a civil conflict happens in the country and 0 otherwise.
  - Six political-instability indicators from Banks and Wilson (2019)’s Cross-National Time-Series Data Archive:
    - (i) Government cabinet changes;
    - (ii) Changes in effective executive;
    - (iii) Anti-government demonstrations (peaceful gatherings of at least 100 people);
    - (iv) Major government crises;
    - (v) General strikes (any strike of 1,000 or more industrial or service workers involving more than one employer and aimed at national government policies);
    - (vi) Political assassinations (politically motivated murder or attempted murder of a high government official or politician).
  - Additional robustness sources (used later): ICRG, COW, Political Terror Scale, GTD, Major Episodes of Political Violence Database, Coup d’Etat Events Database, State Fragility Index.
- Other controls and sources:
  - Real exchange rate, inflation, external debt (% of GDP), GDP per-capita, real GDP growth, terms of trade from IMF WEO.
  - M2/reserves and credit growth from World Bank WDI.
  - Degree of democracy from Polity IV Project.

### II. Literature review — prior findings relevant to channels
- Established findings on conflict impacts:
  - Conflicts reduce long-run economic performance, public investment, trade, tourism, and worsen fiscal outcomes.
  - Examples: Alesina et al. (1996); Alesina and Perotti (1996); Jong-A-Pin (2009); Aisen and Veiga (2013); Rother et al. (2016); Murdoch and Sandler (2002).
- Fiscal channels:
  - Conflicts undermine government revenue by disrupting activity, destroying tax base, lowering tax administration efficiency.
  - Examples cited:
    - Afghanistan conflict led to total revenue loss of about $3 billion between 2005 and 2016 (Barrett, 2018).
    - Central government revenue collapsed by about 60 percent following the outbreak of the conflict in Yemen in 2015 (Rother et al., 2016).
  - IMF (2019) finding for sub-Saharan Africa: conflicts entail, on average, a loss of tax revenue by about 2 percent of GDP.
- Links from fiscal stress to banking crises (literature):
  - Worsened fiscal positions can trigger banking crises via balance-sheet linkages and banks’ portfolio exposures, accumulation of arrears, higher non-performing loans, and reduced collateral eligibility for funding.
  - References: Von Hagen and Ho (2007); Reinhart and Kaminsky (1999); Dornbusch et al. (1995); Caprio and Honohan (2008); Caruana and Avdjiev (2012); Popov and Van Horen (2013); Gertler and Kiyotaki (2010).

### Stylized facts and unconditional statistics
- Association between conflict and banking crises:
  - Figure evidence (described): positive relationship between number of countries in conflict and number of countries in banking crises; major waves of conflict tend to coincide with higher rates of banking crises.
- Table 1 summary (unconditional and conditional probabilities):
  - Unconditional probability of a banking crisis (sample of country-year observations): reported proportion identified with a start of a banking crisis.
  - Conditional probabilities:
    - Example values reported:
      - Conditional probability of a banking crisis in a year without a conflict: 5.3 percent.
      - Conditional probability in years of conflict: 10 percent.
      - For general strikes and political assassinations: probability increases from 5.7 percent in years without these events to 12.5 percent in years with them.
  - Statistical tests:
    - Pearson chi-squared statistic shows a statistically significant relationship between banking crises and conflict or political instability.
    - T-tests on differences in conditional probabilities reported with p-values (column (7) in table).

### Main empirical results (summary of key findings)
- Effect of conflict and political instability on systemic banking-crisis probability:
  - Conflicts and political instability are associated with higher probability of systemic banking crises.
  - Estimated magnitude:
    - "Odds of a banking crisis are 2.5 times greater when a country is affected by a conflict."
- Spillovers:
  - Conflicts and political instability in neighboring countries increase the likelihood of banking crises in a given country, but spillover effects are less impactful than primary (domestic) channels.
- Duration effects:
  - Duration of conflict is positively associated with increased probability of a banking crisis.
  - Reported magnitudes:
    - Probability of experiencing a banking crisis is 25 percent when the conflict lasts 10 years.
    - Probability is 16.4 percent when conflict lasts two years.
- Transmission channel identified:
  - Evidence points to fiscal crises occurring after conflict or political instability as the likely transmission channel to the banking sector.
- Robustness:
  - Findings robust to:
    - Use of alternative conflict and political instability indicators from 10 different sources.
    - Alternative empirical strategies.
    - Inclusion of additional covariates.

### IV–VI (empirical strategy, results, robustness testing) — methodological notes and structure
- Empirical strategy:
  - Multiple measures of conflict and political instability used in baseline estimates.
  - Analysis of spillovers from neighboring countries.
  - Examination of conflict duration and channel analysis focusing on fiscal crises.
- Robustness checks:
  - Alternative indicators (ICRG, COW, Political Terror Scale, GTD, Major Episodes of Political Violence Database, Coup d’Etat Events Database, State Fragility Index).
  - Alternative empirical strategies and additional covariates (details referenced in later sections and appendices).
- Additional sample/statistics notes:
  - Appendix Table A1: correlations between conflict and political variables; majority show low degrees of correlation (less than 0.3), suggesting variables provide additional information across equations.

*Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020041-print-pdf.pdf*

### 11.7 percent, respectively, in years of general strikes and political assassinations. The t-test in

### IV. EMPIRICAL STRATEGY

### Banking-crisis incidence and conditional probabilities
- Unconditional probability of a banking crisis: 6.4 percent (also reported as 6.3 percent elsewhere in the sample description: 191 crises out of 3,026 observations = 6.3 percent).
- Conditional probability highlights (from Tables 1 and 2):
  - Conditional probability of a banking crisis after a joint occurrence of a conflict and a fiscal crisis: 16.7 percent.
  - Conditional probability of a banking crisis after a conflict when a fiscal crisis does not materialize: 4.2 percent.
  - Table 1 reported probabilities across groups including "No conflict or political instability" and "Conflict or political instability", with examples:
    - Overall probability of a banking crisis: 6.3; 5.3; 10.0; 4.6; 8.0; 0.00 (table entries presented verbatim in the source).
  - Table 2 reports conditional statistics by fiscal crisis status; examples:
    - All: Observations 6413; Number of banking crises 66; Probability of a banking crisis 10.3; No fiscal crisis: Observations 3031; Number of banking crises 4; Probability 4.2; Fiscal crisis: Observations 1628; Number of banking crises 16; Probability 16.7; Difference ((5)-(4)): 12.4; T-test p-value 0.00; Pearson chi2 unspecified here but reported in table rows.

### Empirical specification (binary outcome model)
- Binary response model (equation (1)):
  - Pr(Y_iit = 1 | X_iit-1, C_i) = F(X_iit-1 β + C_i) = X_iit-1 β + C_i + ε_iit
  - Y_iit: binary indicator equal to 1 if a banking crisis occurs for country i at time t.
  - X_iit-1: vector of observed explanatory variables lagged by one year.
  - β: parameter vector; C_i: unobserved time-invariant country fixed effect; ε_iit: zero-mean error term uncorrelated with RHS terms.
- Lagging controls: All control variables are lagged by one year to avoid simultaneity and endogeneity. Footnote: results remain consistent if lags up to five years are used.

### Unobserved heterogeneity and estimation approach
- Concern: C_i likely correlated with explanatory variables (e.g., religion, language, regulatory framework, ethnic diversity) violating E(ε|X,C)=0 and creating incidental-parameters issues when treated as parameters.
- Fixed effects approach:
  - Pros: controls time-invariant omitted variables.
  - Cons: excludes countries that never experienced banking crises. In the sample, 40 percent of countries (37 out of 92) have not experienced banking crises (1970-2017), raising selection bias concerns.
- Correlated random effects (Mundlak, 1978) / random effects-Mundlak model:
  - Assumes C_i = ω + X̄_i δ + a_i, where X̄_i is the time-average of X_i, and a_i is uncorrelated with X̄_i and normally distributed.
  - Implemented by including means of all time-varying covariates for each country in the estimates.
  - Advantages: controls for time-constant unobserved heterogeneity while allowing measurement of time-constant independent variables; includes countries with no crises, avoiding sample exclusion.
- Robust checks:
  - Traditional probit and logit models.
  - Probit fixed-effects model of Fernández-Val and Weidner (2016) that corrects incidental-parameter bias by subtracting a plug-in estimator of the bias; caveat: this method excludes countries without banking crises.

### Control variables included (lagged by one year)
- Real effective exchange rate: depreciation expected to be positively associated with the likelihood of a banking crisis.
- M2/reserves: measures banks’ exposure to foreign exchange risk; expected positive correlation with banking crises.
- Inflation: high inflation expected to be positively correlated with banking crises.
- Credit growth: high credit growth expected to be positively correlated with banking crises (literature mixed).
- External debt: higher external debt-to-GDP expected to be positively correlated with banking crises.
- GDP per capita: captures development level; literature finds negative association with banking crises.
- GDP growth: deteriorating growth expected to be negatively associated with banking crises (i.e., higher growth reduces crisis probability).
- Terms of trade: deterioration expected to be negatively correlated with banking stability (worse terms increase crisis risk).
- Degree of democracy: expected negative association with banking crises (better institutions reduce probability).

*Italicized empirical-method and data notes are retained exactly as presented in the source.*

---

### V. EMPIRICAL RESULTS

### Baseline results (random effects-Mundlak estimates summarized from Table 3)
- Model: random effects-Mundlak binary-response (logit) estimated for different conflict and political-instability indicators.
- Key summary statistics reported across columns (each column corresponds to a different conflict/instability measure):
  - Observations: 3,026 in columns with conflict measures; 2,972 or 2,991 in columns with some political-instability measures.
  - Number of countries: 92 in all columns.
  - AUROC (goodness-of-fit): 0.75; 0.744; 0.742; 0.739; 0.734; 0.734; 0.740; 0.737 across columns.
  - seAUROC: 0.0174; 0.0174; 0.0175; 0.0183; 0.0181; 0.0178; 0.0177; 0.0177.
  - Rho(LR): 0.482; 0.484; 0.460; 0.457; 0.471; 0.469; 0.465; 0.476. P-value(Rho): 0 in all columns.
  - Wald chi2: 326.9; 323.8; 335.6; 333.5; 327.5; 328.1; 335; 323.8.
  - Log likelihoods reported: -598.7; -602.1; -599.5; -600.6; -603.3; -602.4; -602.2; -603.1.

### Conflict and political instability effects (coefficients and significance from Table 3)
- Conflict (binary), variable in column (1), t-1:
  - Coefficient (exponentiated odds interpretation discussed in text): 0.9097*** (standard error 0.245). Text interpretation: odds of a banking crisis are 2.5 times greater when a country is affected by a conflict; estimated conditional probability rises from 6.3 percent to 13.5 percent when in conflict.
- Conflict (intensity), column (2): 0.4483*** (0.167).
- Cabinet changes, column (3): 0.3620*** (0.126). Text example: increasing cabinet changes from zero to four (maximum observed) raises probability to 21.17 percent from unconditional 6.3 percent.
- Changes in Effective Executive, column (4): 0.3277** (0.150).
- Anti-Government Demonstrations, column (5): 0.0295* (0.017).
- Government Crises, column (6): 0.2894** (0.134).
- General Strikes, column (7): 0.2211** (0.094).
- Assassinations, column (8): 0.0788* (0.042).

### Control variables – estimated coefficients (reported exactly as in Table 3)
- Exchange rate, t-1: -0.0903**; -0.0892**; -0.0931**; -0.0959**; -0.0910**; -0.1024**; -0.0987**; -0.0899** (standard errors ≈ 0.039).
- M2/reserves, t-1: 0.5660***; 0.5531***; 0.5605***; 0.5658***; 0.5691***; 0.5675***; 0.5666***; 0.5675*** (standard errors ≈ 0.123–0.124).
- Inflation, t-1: 1.0096***; 1.0140***; 1.0163***; 1.0224***; 1.0509***; 1.0638***; 1.0106***; 1.0148*** (standard errors ≈ 0.169–0.173).
- Credit growth, t-1: -0.4867*; 0.3878; 0.3468; 0.2923; 0.3007; 0.3571; 0.2995; 0.3088 (standard errors ≈ 0.281–0.293).
- External debt, t-1: 0.7992***; 0.7981***; 0.8284***; 0.8410***; 0.8267***; 0.8388***; 0.8137***; 0.8282*** (standard errors ≈ 0.144–0.145).
- GDP per capita, t-1: -0.3966; -0.3913; -0.3530; -0.3983; -0.4916; -0.3989; -0.4452; -0.4245 (standard errors ≈ 0.313–0.318).
- Economic growth, t-1: -0.0324*; -0.0312; -0.0257; -0.0277; -0.0324*; -0.0289; -0.0331*; -0.0340* (standard errors 0.019–0.020).
- Terms of trade, t-1: -0.0286; 0.0910; -0.1673; -0.0626; 0.0235; -0.0540; 0.0422; -0.0172 (standard errors ≈ 1.204–1.218).
- Degree of democracy, t-1: -0.0460**; -0.0469**; -0.0524***; -0.0540***; -0.0466**; -0.0514***; -0.0518***; -0.0483*** (standard errors 0.019).

### Interpretation and robustness
- Main inference: conflict and multiple forms of political instability are positively associated with an increased likelihood of banking crises; coefficients on conflict and political-instability measures are positive and statistically significant (levels of significance vary by measure).
- Model performance: AUROC statistics above 0.73 across specifications indicate reasonable classification performance.
- Robustness checks: results remain consistent across probit, logit, and bias-corrected fixed-effects probit (Fernández-Val and Weidner, 2016), with noted caveat that the latter excludes countries without banking crises.

*This summary preserves the empirical strategy, estimation approach, baseline results, coefficient estimates, predictive-statistic values, and key conditional-probability findings as reported in the source material.*

### 3. A: Probability of conflict

### 3. A: Probability of conflict

### Figure panels and axes (Figure 3. (cont.))
- Panel labels present in the figure continuation:
  - 3. A: Probability of conflict
  - 3. B: Intensity of conflict
  - 3. C: Number of cabinet changes
  - 3. D: Number of changes in effective executive
  - 3. E: Number of anti-government demonstrations

### Axis values and annotated numbers as shown in the figure
- Text strings and numeric ticks appearing in the figure (preserved exactly):
  - .05
  - .1
  - .15
  - Probability of banking crisis
  - 0
  - .2.4
  - .6
  - .81
  - Probability of conflict
  - .05
  - .1
  - .15
  - .2
  - Probability of banking crisis
  - 0
  - 1
  - 2
  - Intensity of conflict
  - .05
  - .1
  - .15
  - .2
  - .25
  - .3
  - Probability of banking crisis
  - 01234
  - Number of changes in effective executive
  - .05
  - .1
  - .15
  - .2
  - .25
  - .3
  - Probability of banking crisis
  - 01234
  - Number of cabinet changes

### Structural note
- The figure is labeled "18" and marked "Figure 3. (cont.)" indicating continuation of Figure 3 across panels.

*Source: wpiea2020041-print-pdf - 3. A: Probability of conflict*

### 3. F: Number of government crises

### 3. F: Number of government crises

### Key empirical findings on conflict, political instability, and banking crises
- Coefficients associated with exchange rate, economic growth, and the degree of democracy are negatively associated with banking crises.
- GDP per capita is not statistically significant in the baseline specification.
- Terms of trade are non-significant determinants of banking crisis.
- Conflict and political instability are positive predictors of banking crises in the full sample.
- Duration matters: conflict and political instability that last only one year have no significant effect, but when they last longer the impact becomes apparent and stronger.
  - Probability of experiencing a banking crisis is 25 percent when the conflict lasts 10 years, against 16.4 percent when it lasts two years.
- The primary transmission channel is fiscal distress: coefficients associated with conflict and political instability become insignificant when interacting with fiscal crisis and including fiscal crisis itself, while fiscal crisis is positive and significant in all columns.

### Subsample results by income group (emerging vs low-income developing countries)
- Conflict and political instability predict banking crises in both emerging economies and low-income economies.
- Emerging markets:
  - Coefficients associated with conflict and political instability variables are all positive and significant in all columns, except for government crises and political assassinations.
  - A conflict raises the probability of a banking crisis to 14.12 percent (when an emerging market is hit by a conflict).
- Low-income developing countries:
  - All coefficients are positive and significant, except those associated with effective changes in the executive, anti-government demonstrations, and general strikes.
  - A conflict raises the probability of a banking crisis to 17.15 percent (when a low-income country is hit by a conflict).
- Sample sizes and model fit reported:
  - Observations for whole-sample specifications: 2,775 and variants.
  - Number of countries in some panels: 91, 92, and 94 (varies by table and panel).
  - AUROC values reported across specifications, e.g., 0.743, 0.756, 0.659, 0.741 (varies by table/panel).

### Spillovers: conflicts and political instability in neighboring countries
- Conflicts and political instability in neighboring countries increase the likelihood of banking crises in a given country, though spillover effects are generally lower than direct effects.
- Specific neighboring-country effects:
  - Number and intensity of conflicts in bordering countries, number of cabinet changes and changes in effective executive, and number of general strikes in bordering countries are associated with higher probability of a banking crisis in the focal country.
  - Example (column 1): a rise in the number of bordering countries affected by conflict from 0 to 3 (the median) increases the likelihood of banking crises from 6.3 percent to 11.2 percent.
  - Number of anti-government demonstrations, government crises, and political assassinations in bordering countries show no significant spillover effects in the reported results.

### Duration of conflict and political instability
- Method: redefine binary indicators equal to 1 if conflict/political instability lasts at least 1, 2, 3, ..., up to 10 years; estimated with Mundlak (1978) estimator.
- Findings:
  - Lasting only one year: no significant effect.
  - Lasting at least two years: conflict coefficient positive and significant at the 1 percent level.
  - Longer durations associated with larger coefficients (stronger effect).
  - Example magnitudes:
    - Conflict lasting 2 years: probability of banking crisis 16.4 percent.
    - Conflict lasting 10 years: probability of banking crisis 25 percent.
  - Anti-government demonstrations and general strikes become insignificant when lasting more than 2 and 3 years respectively, due to reduced sample sizes.

### Transmission channels: role of fiscal crises
- Fiscal crises defined as episodes of extreme fiscal distress from Gerling et al. (2017); fiscal crisis variable is binary (1 under tight budgetary conditions, 0 otherwise).
- Specification includes conflict/political instability, fiscal crisis, and interaction between them.
- Results (Table 7):
  - The effect of conflict and political instability operates mainly through a simultaneous fiscal crisis.
  - When fiscal crisis and the interaction term are included, coefficients on conflict/political instability become insignificant in all columns.
  - Fiscal crisis coefficient is positive and significant in all columns.
  - Interpretation: budgetary constraints and simultaneous fiscal crises are key determinants transmitting conflict/political instability into banking crises.

### Robustness checks
- Alternative data sources:
  - ICRG internal/external conflict and aggregate political risk indicators yield positive and highly significant coefficients (Table A2).
  - Marshall (2017) MEPV civil war binary, civil war score, civil violence score, and total violence index: all positive and strongly significant at 1 percent (Table A3).
  - Correlates of War (COW) data: being in conflict raises probability of a banking crisis from 6.3 percent to 17.7 percent.
  - Assassination of ruling executive (Marshall and Marshall 2018) is positive and strongly significant.
  - Global Terrorism Database (GTD) indicator for terrorism events is positive and strongly significant at the 1 percent level.
  - Political Terror Scale (PTS) coefficient positive and significant at the 10 percent level.
  - Number of coups d’état is not statistically significant in the reported specification.
  - State fragility index and share of deaths caused by conflict (conflict deaths over population) are positive and significant at the 5 percent level.
- Including more covariates (Table A4):
  - Controlling for global conditions (S&P 500 index and US 3-year bond yield) does not change main results; coefficients remain broadly positive and strongly significant.
  - Commodity prices index and natural resource rents (% GDP) do not alter baseline findings.
  - Portfolio investment net inflows (% GDP) and real interest rate included; conflict and political instability remain positive predictors though significance can drop when real interest rate is included (reduced observations).
  - Control of corruption, degree of export diversification, and financial development added; baseline findings persist.
- Alternative econometric methods (Table A5):
  - Probit, logit, and probit fixed effects models produce coefficients on conflict and political instability that are positive and significant across panels.
  - Conclusion: baseline results robust to alternative data sources, additional covariates, and alternative estimators.

### Concluding remarks and policy implications
- Main conclusions:
  - Conflicts and political instability are associated with higher probability of systemic banking crises.
  - Duration and persistence of conflict/political instability amplify the probability of banking crises.
  - Cross-border spillovers exist: conflicts and political instability in neighboring countries raise the probability of banking crises domestically, though with lower magnitude than direct effects.
  - Fiscal crises are the primary transmission channel linking conflict/political instability to banking crises.
- Policy implications:
  - Governments facing conflict and/or political instability should address root causes and mitigate negative effects through appropriate economic policy design and implementation.
  - Creating adequate fiscal space in normal times can reduce the likelihood of fiscal crises and thereby lower the probability of systemic banking crises.
  - Policy makers should monitor conflicts in neighboring countries because their banking systems may suffer negative spillovers even if not directly conflict-afflicted.

*Source: 3. F: Number of government crises — wpiea2020041-print-pdf*

### Appendix B1: Definition of Conflict and Political Instability Variables Used in

### Appendix B1: Definition of Conflict and Political Instability Variables Used in Robustness Checks

### Data sources and variable constructs
- International Country Risk Guide (ICRG)
  - Internal conflict index: assessment of political violence and its actual or potential impact on governance; comprises three subcomponents: civil war, terrorism, and civil disorder.
  - External conflict index: assessment of the risk to the incumbent government from foreign action (ranging from non-violent external pressure to violent external pressure); comprises three subcomponents: war, cross-border conflict, and foreign pressures.
  - Aggregate index of conflict: the simple average of the internal and external conflict indices.
  - Paper uses both the indices of internal and external conflicts and the different sub-indices.

- Correlates of Wars (COW)
  - Civil war definition: armed conflict between an internationally recognized state and (mainly) domestic challengers able to mount an organized military opposition to the state.
  - Severity threshold: a war must have caused more than 1,000 battle-related deaths in total and within at least a three-year period.
  - Variable construction: binary variable equal to 1 if the condition is met, and 0 otherwise.

- Political Terror Scale of Amnesty International
  - Measures political terror as violations of physical integrity rights carried out by states or their agents (state-sanctioned killings, torture, disappearances, political imprisonment).
  - Data range: 1 to 5, with higher values representing widespread and systemic violations where murders, disappearances, and torture are a common part of life.

- Global Terrorism Database (GTD) — National Consortium for the Study of Terrorism and Responses to Terrorism (START)
  - Terrorism definition: the threatened or actual use of illegal force and violence by a non-state actor to attain a political, economic, religious or social goal through fear, coercion or intimidation.
  - Three conditions for an incident to be considered terrorism: (i) intentional act; (ii) entails some violence or threat of violence (including violence against property and/or against people); (iii) perpetrator must be sub-national actors.
  - Variable used: the number of terrorism related incidents per year.

- Marshall (2017) — Major Episodes of Political Violence Database (Center for Systemic Peace)
  - Major episode definition: systematic and sustained use of lethal violence by organized groups resulting in at least 500 directly-related deaths over the course of the episode.
  - Variable construction: binary variable equal to 1 if such an event occurs and 0 otherwise.
  - Additional measures: an eleven-point scale score (0-10) assessing intensity of civil war and civil violence.
  - Aggregate total violence score: simple average of the civil war and civil violence scores.

- Marshall and Marshall (2018) — Coup d’état Events Database (Center for Systemic Peace)
  - Coup d’état definition: forceful seizure of executive authority and office by a dissident/opposition faction within the country’s ruling or political elites that results in a substantial change in the executive leadership and the policies of the prior regime.
  - Variables used:
    - Binary variable equal to 1 if a military coup occurred and 0 otherwise.
    - Binary variable equal to 1 if the ruling executive was assassinated and 0 otherwise.

- Marshall and Marshall (2017) — State Fragility Index (Center for Systemic Peace)
  - Composite index based on four dimensions: security, political, economic and social; measures the degree to which a country is vulnerable to political and social violence.
  - Variable range: between 0 and 25, with higher values meaning extreme fragility.

### Key variable types and aggregation rules
- Binary indicators:
  - COW civil war indicator: 1 if war exceeds 1,000 battle-related deaths over at least a three-year period, 0 otherwise.
  - Marshall (2017) major episode indicator: 1 if episode meets the 500-death threshold, 0 otherwise.
  - Coup occurrence: 1 if coup d’état occurs, 0 otherwise.
  - Executive assassination: 1 if ruling executive is assassinated, 0 otherwise.
- Scales and counts:
  - Political Terror Scale: 1 to 5.
  - Major Episodes intensity scale: eleven-point scale (0-10).
  - State Fragility Index: 0 to 25.
  - GTD terrorism measure: number of terrorism related incidents per year.
- Aggregation methods:
  - ICRG aggregate conflict index: simple average of internal and external conflict indices.
  - Marshall total violence score: simple average of civil war and civil violence scores.

*Source: Appendix B1, "Definition of Conflict and Political Instability Variables Used in Robustness Checks," from the cited PDF.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020041-print-pdf.pdf_