## CHAPTER 3 THE MACROECONOMICS OF CONFLICTS AND RECOVERY (Online Annex)

## Source details

**Canonical URL:** [CHAPTER 3 THE MACROECONOMICS OF CONFLICTS AND RECOVERY (Online Annex)](https://www.imf.org/-/media/files/publications/weo/2026/april/english/ch3onlineannex.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/weo/2026/april/english/ch3onlineannex.pdf.md)
- [Structured JSON version](/-/media/files/publications/weo/2026/april/english/ch3onlineannex.pdf.json)

---

### Sample construction, conflict coding, and data sources
- Final sample: 183 conflict episodes.
- Exclusions: isolated skirmishes, terrorist attacks, or highly fragmented violence with low casualty counts (systematic reading and coding).
- ChatGPT use for classification:
  - Prompted ChatGPT (version 5.1) with temperature = 0.3 for each conflict; model outputs were manually reviewed for accuracy and consistency.
- Conflict exposure and coding:
  - Country-year indicator = 1 if conflict occurs on a country’s territory in a given year.
  - Separate indicators: major, minor, between states, within states; belligerent countries treated analogously.
  - Spillover exposure: geographic proximity via land border; trade exposure if share of imports from conflict-site economies exceeds the 90th percentile.
  - Battle-death classification: “Minor” = 25–999 battle-related deaths; “Major” = 1,000 and more battle-related deaths.
  - When GED best estimate unavailable, use average of low and high estimates rounded to nearest whole number.
- Primary macro and auxiliary datasets:
  - Global Macro Database (Müller and others 2025); IMF WEO (October 2025 vintage); World Bank WDI; Penn World Table.
  - UCDP GED; PRIO Battledeaths; EM-DAT (major natural disaster = deaths share > 99th percentile); project-level ratings; geocoded aid; night-time lights; subnational corruption; Orbis firm data.
- Data treatment:
  - Combine sources for coverage; review series to eliminate structural breaks; use most recent continuous vintage; left-hand-side variables winsorized at the 1st/99th percentiles.

### Empirical approach (LP‑DiD) and sample restrictions
- Estimation equation (cumulative growth up to horizon h):
  - ln(y_{i,t+h}) − ln(y_{i,t−1}) = β_h · ConflictOnset_{i,t} + δ_h · Controls_{i,t} + θ_t^h + μ_i^h + ε_{i,t}^h, for h = 0, ..., 5
- Key definitions:
  - ConflictOnset_{i,t} = 1 in first year a country becomes conflict-site if preceding five years were without conflict (L = 5); controls include three lags of dependent variable.
  - Year fixed effects θ_t^h; country fixed effects μ_i^h.
  - Sample restricted to conflict onset observations and “clean controls” not affected by conflict within a five-year window (robustness uses L = 10).
  - Horizons h = 0, ..., 4 relabeled as 1, ..., 5 in figures/tables.
- Regression reporting:
  - Standard errors clustered at country level.
  - Multiple specifications and robustness checks: longer horizons, alternative clean-control windows, alternative conflict definitions (battle deaths per capita).

### Main quantitative results — baseline LP‑DiD impacts on output (Column (1), Online Annex Table 3.2.1)
- Dependent variable: cumulative change in output up to horizon h (h = 1 is year of conflict onset).
- Baseline cumulative impacts (horizon → coefficient):
  - h = 1 → -2.795 ***
  - h = 2 → -4.728 ***
  - h = 3 → -5.578 ***
  - h = 4 → -5.766 ***
  - h = 5 → -6.659 ***
  - h = 6 → -7.813 ***
  - h = 7 → -8.756 ***
  - h = 8 → -9.906 ***
  - h = 9 → -10.779 ***
  - h = 10 → -11.135 ***
- Significance notation: *** p<0.01; ** p<0.05; * p<0.1.
- Regression diagnostics for h = 1, Column (1):
  - Number of Observations = 10,725
  - Number of Countries = 191
  - Number of Conflict Onsets = 148
  - R^2 = 0.29
- Robustness summary (Columns (2)–(10)):
  - Negative impacts on output consistent across alternative specifications: 10-year onset window, relaxing/ tightening clean-control rules, restricting to conflicts lasting ≥ five years, excluding economically motivated conflicts, using full conflict sample, and alternative per-capita death thresholds (50 and 100 deaths per million inhabitants).

### Comparison with other shocks (Online Annex Table 3.2.2)
- Same LP‑DiD framework comparing cumulative output changes for conflict, banking crises, currency crises, sovereign debt crises, and major natural disasters.
- Definitions:
  - Banking crisis: significant financial distress + major policy interventions.
  - Currency crisis: nominal depreciation vis-à-vis the U.S. dollar ≥ 30 percent and ≥ 10 percentage points higher than prior year.
  - Sovereign debt crisis: failure to meet principal/interest payments and/or rescheduling on less favourable terms.
  - Major natural disasters: disaster-related deaths as share of population exceed 99th percentile.
- Number of observations and fit for h = 1:
  - Conflict: Number of Observations = 10,648; Number of Crisis Onsets = 160; R^2 = 0.28
  - Banking Crises: Number of Observations = 10,828; Number of Crisis Onsets = 295; R^2 = 0.18
  - Currency Crises: Number of Observations = 10,074; Number of Crisis Onsets = 392; R^2 = 0.18
  - Sovereign Debt Crises: Number of Observations = 10,994; Number of Crisis Onsets = 200; R^2 = 0.18
  - Natural Disasters: Number of Observations = 12,924; Number of Crisis Onsets = 36; R^2 = 0.24
- Overall finding: conflicts generate large cumulative negative effects on output relative to other major shocks.

### Heterogeneity of conflict effects (intensity, duration, type)
- By intensity:
  - Major conflicts (≥1,000 battle‑related deaths) generate largest output losses.
  - Minor conflicts (25–999 deaths) show statistically significant effects at shorter horizons that fade.
- By duration:
  - Short conflicts (up to two years) produce larger immediate impacts than long conflicts (more than two years).
- By type:
  - Between‑state (interstate) conflicts generate initially large but less persistent effects.
  - Within‑state conflicts (intrastate, intrastate with foreign intervention, extrasystemic) are associated with more protracted output losses.

### Spillovers to belligerents, neighbors, and trade partners
- Belligerent countries (participate without hostilities on their territory): effects not statistically significant overall (Online Annex Table 3.2.4).
- Geographic proximity or trade-link exposure:
  - Exposure associated with negative output effects in first two years that subsequently fade.
  - Trade exposure defined by import share from conflict-site economies exceeding the 90th percentile.

### Demand-side components, sectoral and factor outcomes
- Investment:
  - Investment declines sharply after conflict onset and does not recover within five years (Online Annex Table 3.2.5.A).
- Consumption:
  - Private consumption falls; public consumption effect not statistically significant.
- Trade:
  - Imports and exports contract; exports decline more than imports.
- Sectoral value added:
  - Negative and statistically significant across all sectors; industry value added contraction most pronounced (Online Annex Table 3.2.6).
- Production factors and employment:
  - Capital stock declines persistently with no evidence of recovery (Online Annex Table 3.2.5.D).
  - Total factor productivity (TFP) falls in early periods; becomes statistically insignificant at longer horizons.
  - Employment declines significantly across all horizons.
- Human toll:
  - Number of deaths rises sharply following conflict onset.

### Fiscal, external, financial, and price-level effects
- Government debt (real, using GDP deflator) tends to rise following conflict onset (Online Annex Table 3.2.5.B).
- Exports‑to‑imports ratio declines.
- Personal remittances received (USD) increase (not statistically significant).
- Official development assistance (ODA, USD) rises, especially at longer horizons.
- Capital flows and liabilities:
  - FDI liabilities decline after conflict onset.
  - Portfolio debt liabilities decline after conflict onset.
- Financial openness (Chinn–Ito index) declines after conflict onset (Online Annex Table 3.2.5.C).
- Exchange rate depreciates; reserve assets decline.
- Price level:
  - Consumer prices rise sharply; price level increases by about 30 percent five years after conflict onset (text summary referencing Online Annex Table 3.2.5.C).

### Additional macro patterns and policy-relevant indicators
- Fiscal reallocation: defense spending increases; social spending declines after conflict onset.
- Trade balance: widens in first four years (consistent with import compression thereafter).
- Savings and investment: real gross national savings fall, decline exceeds drop in investment.
- Uncertainty and informality: conflicts associated with higher uncertainty index and greater shadow economy (percent of GDP).
- Monetary policy: short‑term nominal interest rates tend to rise, particularly at medium horizons.

### Scarring effects on individual health outcomes (Online Annex 3.3 / Table 3.3.1)
- Specification: H_{i,t} = β1 YearsInConflict_{i,t} + β2 X_{i,t} + φ_c + ε_{i,t}; YearsInConflict = number of years an individual lived in a conflict‑site country.
- OLS coefficients (standard errors in parentheses); health indicators standardized (mean 0, SD 1):
  - Composite health: Number of Years Living in Conflict = -0.014*** (0.004)
  - Cognitive PC: Number of Years Living in Conflict = -0.010* (0.005)
  - Physical PC: Number of Years Living in Conflict = -0.017*** (0.005)
  - Psychological well‑being: Number of Years Living in Conflict = -0.016*** (0.005)
- Sample sizes and fit:
  - Composite Observations = 755,958; Number of Countries = 41; R² = 0.33
  - Cognitive Observations = 382,436; Number of Countries = 34; R² = 0.38
  - Physical Observations = 628,729; Number of Countries = 30; R² = 0.13
  - Psychological Observations = 345,479; Number of Countries = 30; R² = 0.18

### Postconflict recovery dynamics and empirical design (Online Annex 3.4)
- ConflictTermination_{i,t} = 1 in first year of peace if all five following years remain conflict‑free (L = 5).
- Recovery analysis focuses on first five years of each postconflict episode for policy-driver analysis.
- Postconflict cumulative outcomes (selected point estimates, Online Annex Table 3.4.1; horizons h = 1 to 5):
  - Example point estimates (columns vary by peace type/subsample):
    - h = 1: 1.160*; 1.165*; 1.173**; 1.553*; 1.231
    - h = 2: 2.572**; 1.341; 2.107***; 2.661*; 2.555
    - h = 3: 2.727**; 1.530; 2.281**; 2.952*; 3.426
    - h = 4: 3.361**; 1.789; 2.750**; 3.319*; 4.670
    - h = 5: 3.879**; 1.124; 2.788*; 3.839; 5.309
  - Peace classification:
    - Nonfragile peace = conflict does not restart within first five years.
    - Fragile peace = conflict restarts within first five years.
- Sample and fit examples:
  - Number of Observations (example) = 84
  - R² (example) = 0.10
- Key patterns:
  - Recoveries conditional on maintained peace are positive but slow and uneven.
  - Pooling all terminations (including fragile peace) yields even slower recoveries.
  - Commodity‑exporting nonfragile peace episodes (fuel and primary commodity exports > 50 percent of total exports): initial rebound strong but statistically insignificant later.
  - Additional dynamics: muted uncertainty response; increases in tax revenue collection following conflict termination.

### Policy drivers of postconflict recovery (Online Annex Table 3.5.1)
- Regression: Recovery percentile as dependent variable; policy indicators entered one at a time; controls include episode and horizon fixed effects.
- Key coefficients (robust standard errors in parentheses):
  - Low Inflation: 0.353* (0.181)
  - Stable Inflation: 0.222 (0.283)
  - Stable REER: 0.361* (0.206)
  - Debt Restructuring: 0.461* (0.234)
  - Capacity Development: 0.366*** (0.123)
  - Social Spending: 0.491*** (0.093)
- Number of Observations by column: 693, 684, 545, 737, 336, 431 respectively.
- R2 by column: 0.393, 0.397, 0.398, 0.399, 0.516, 0.490.
- Indicator definitions:
  - Low/stable inflation and stable REER = 1 if measure falls below sample mean (inflation level, inflation volatility, REER volatility).
  - Debt restructuring = 1 if restructuring occurs.
  - Capacity development = log number of participants in IMF training.
- Significance notation: *** p<0.01; ** p<0.05; * p<0.1.

### Case-study selected statistics (Online Annex Table 3.5.2)
- Selected high postconflict recoveries:
  - Countries and conflict durations:
    - Bosnia and Herzegovina: 1992–1995
    - Cambodia: 1989–1998
    - Nepal: 1996–2006
    - Rwanda: 1990–2001
    - Sri Lanka: 1983–2009
    - Côte d’Ivoire: 2010–2011
  - Human casualties (percent): 1.46; 0.08; 0.06; 11.2; 0.43; 0.01
  - Macroeconomic recovery (Real GDP Growth): 24.5; 9.5; 4.5; 8.3; 7.2; 8.4 (percent)
  - Real GDP per capita Growth (average first 5 years): 21.1; 7.7; 3.7; 6.0; 6.4; 5.9 (percent per year)
  - Selected stabilization changes (postconflict relative to conflict period):
    - Inflation rate (difference): -61.8; 2.7; -4.9; -4.7; -1.8 (aligned where available)
    - Inflation volatility (percentage point): -65.3; 0.6; -14.4; -3.3; -1.8
    - GDP growth volatility (percentage point): 9.9; 0.4; -0.5; -15.9; 0.1; -7.1
    - REER appreciation: 1.1; -9.5; 2.3; -0.6; 0.1; 0.9
    - Debt-to-GDP: -3.2; 19.2; 4.2; 8.4; 20.3
    - Aid-to-GDP (postconflict average): 23.4; 8.6; 4.7; 18.5; 0.7; 3.1
    - Remittances-to-GDP: 0.2; 2.6; 20.6; 0.4; 8.0; 0.8
  - Institutions and inclusive policy changes (percentage change, final year vs last conflict year):
    - Institutional quality: 13.7; 2.8; 60.8; 62.0; 41.6 (where provided)
    - Public sector corruption: 0.0; 0.0; -1.2; 0.0; 34.5; -3.7
    - Statistical capacity score: 2.2; 22.0; 11.7; 16.4; 220.8; 111.8
    - Social spending-to-GDP: 2.2; 3.9; 50.6; 6.5; 13.3; 20.3
  - Security and peacebuilding:
    - UN peacekeeping (years in): 5; 0; 0; 0; 0; 5 (aligned where provided)
    - Peacebuilding strategies: justice provisions = 1; political provisions = 1 (as recorded)

### Project outcome analysis (difference‑in‑differences; Online Annex Table 3.6.1)
- Outcome: 1–6 Likert project outcome rating (higher = more satisfactory).
- PostConflict indicator = 1 if project implemented during postconflict period.
- Controls: project size (log), donor local office presence, external/ex‑post evaluations, repeated project indicator; macro covariates averaged over project duration.
- Fixed effects: recipient, donor, approval‑year.
- Baseline and heterogeneity coefficients (standard errors):
  - Baseline postconflict effect (column (1)): 0.101** (0.046)
  - Column (2) (satisfactory supervision): 0.094** (0.046)
  - Column (3) (sectoral heterogeneity): 0.105** (0.047)
  - Column (4) (high vs low investment scale-up): 0.089* (0.050)
  - Supervision expanded: 0.332*** (0.070)
  - Productive sector coefficient: -0.180** (0.074)
  - Education and Health sector: 0.134* (0.081)
  - Infrastructure sector: 0.231*** (0.051)
  - High Scale‑Up: 0.206 (0.131)
  - Low Scale‑Up: 0.101** (0.047)
- Observations up to 8,760; Outcome Mean = 4.277.
- Robust SEs clustered at country-by-approval-year; significance: *** p<0.01; ** p<0.05; * p<0.1.

### Time-varying effects on project outcomes (Online Annex Table 3.6.2)
- Yearly coefficients (standard errors):
  - Year 1: -0.067 (0.189)
  - Year 2: -0.008 (0.148)
  - Year 3: 0.096 (0.095)
  - Year 4: 0.105 (0.072)
  - Year 5: 0.082 (0.067)
  - Year 6: 0.113* (0.060)
  - Year 7: 0.150*** (0.056)
  - Year 8: 0.126** (0.050)
  - Year 9: 0.089* (0.048)
  - Year 10: 0.101** (0.046)
- Observations for each year: 8,760; Outcome Mean = 4.277.

### Subnational analysis — nightlights, aid, and governance (Online Annex 3.7 / Table 3.7.1)
- Data: geocoded ADM2 aid disbursements (22 donors, 1992–2023), geocoded GED events, nightlights intensity, subnational corruption; sample = 35,432 ADM2 across 125 aid-recipient countries.
- Baseline regression: change in log nightlights on lagged nightlights, lagged aid (cumulative 3‑year, lagged one year), Postconflict, Aid × Postconflict, population density, ADM2 fixed effects, and country–year fixed effects.
- Key coefficients (standard errors):
  - Lagged Nighttime Light: -0.163*** (0.002); -0.164*** (0.002); -0.164*** (0.002)
  - Aid: 0.011*** (0.000) uniformly
  - Postconflict: 0.064*** (0.019); 0.213*** (0.079); 0.057*** (0.019) (baseline, high governance-improvement, low governance-improvement subsamples)
  - Aid × Postconflict: 0.005*** (0.002); 0.015 (0.010); 0.005*** (0.002)
- Observations: 976,308; 956,555; 974,934 across columns.
- R2 = 0.18 for reported columns.
- High (low) governance improvement = state-year increases in governance index exceeding (not exceeding) two standard deviations of annual changes.
- Significance: *** p<0.01; ** p<0.05; * p<0.1.

### Firm-level analysis — geolocation, exposure, and results (Online Annex 3.8)
- Firm sample: Orbis firms in countries with at least one conflict episode over 1989–2024 (GED).
- Geolocation: postcode → city → Photon geocoder → cross-checked with World Bank shapefiles; hierarchical matching; supplemented by Nominatim where needed.
- Conflict exposure: aggregate conflict events and battle-related deaths within 20 km of firm location; Postconflict dummy = 1 during five years following a conflict episode within 20 km.
- Baseline regression: ln Y_f,i,s,t = β1 PostConflict_f,i,s,t + β2 X_f,i,s,t + δ_f + φ_s,t + ω_i,t + ε_f,i,s,t
  - Dependent variables: log firm capital, log labor, log quantity‑based TFP (TFPQ).
  - Controls: firm age, size, leverage, export status; fixed effects: firm, sector-by-year, country-by-year.
- Baseline coefficients (standard errors):
  - Postconflict on Capital: -0.046* (0.027)
  - Postconflict on Labor: 0.059** (0.022)
  - Postconflict on TFP: -0.016 (0.025)
- Factor-intensity heterogeneity:
  - Capital Intensive: Capital -0.005 (0.031); Labor 0.075** (0.036); TFP 0.048*** (0.016)
  - Labor Intensive: Capital -0.046* (0.024); Labor 0.080*** (0.008); TFP -0.061* (0.030)
- Observations and fit:
  - Capital: 16,577,311 observations; R2 = 0.96
  - Labor: 21,140,632 observations; R2 = 0.96
  - TFP: 4,923,758 observations; R2 = 0.86
- Heterogeneity by firm characteristics (Online Annex Table 3.8.2):
  - By net worth (standard errors):
    - Low Net Worth:
      - Capital: -0.104** (0.042)
      - Labor: -0.038*** (0.011)
      - TFP: -0.080*** (0.029)
    - High Net Worth:
      - Capital: -0.023 (0.045)
      - Labor: 0.101*** (0.014)
      - TFP: 0.050** (0.018)
  - By export status:
    - Exporter:
      - Capital: -0.019 (0.026)
      - Labor: 0.060*** (0.015)
      - TFP: 0.221*** (0.023)
    - Non-Exporter:
      - Capital: -0.047* (0.028)
      - Labor: 0.059** (0.022)
      - TFP: -0.030 (0.033)
- Interpretation: postconflict firm responses vary by factor intensity, net worth, and exporter status; exporters show positive TFP responses postconflict.

### Theoretical model, shocks, and recovery simulations (Online Annex 3.9)
- Model framework: overlapping generations small open economy with heterogeneous households; production with constant returns and labor-augmenting technological progress; net rental rate r_t = marginal product of capital − δ.
- Asset markets: degree of financial openness controlled by elasticity of foreign demand for domestic assets; net foreign asset elasticity positive.
- Conflict shocks modeled:
  - Capital destruction shocks: reduce physical capital stock.
  - Population shocks: calibrated to a three percent decline on impact for simulations (consistent with refugees); casualties in case studies up to 11 percent of population.
    - Losses to working-age cohorts: effects subside as cohorts age out.
    - Losses to young cohorts: deeper, longer scars; output effects can peak years later as cohorts enter workforce.
  - Confidence shocks:
    - Domestic confidence shocks reduce willingness to hold assets → excess supply of assets → upward pressure on interest rates.
    - Foreign confidence shocks raise foreign risk premia → require higher domestic interest rates.
- Transmission mechanisms:
  - Higher interest rates reduce investment; magnitude depends on whether asset demand gap is filled by disinvestment, increased savings, or capital inflows.
  - Small open economies can rely on capital inflows; closed economies adjust via disinvestment → larger output losses.
  - Increase in government debt mimics negative asset demand shock crowding out productive investment.
- Recovery policies and simulation design:
  - Timing: recovery policies initiated five years after conflict onset; policies simulated individually and in combination.
  - Macroeconomic stabilization: shrink confidence‑shock wedges to pre‑conflict levels over eight years.
  - Financing package:
    - Linear tax-to-GDP increase by three percentage points over 15 years of postconflict, fully used for public investment.
    - Public investment efficiency increases by 10 percentage points relative to October 2025 Fiscal Monitor average, boosting output return of tax‑funded recovery by about 1.4 percentage points.
    - Grant support: ~0.5 percent of GDP per year for first five years of recovery.
    - Population recovery: displaced populations return gradually net of natural mortality four years after conflict end.
- Key model insights:
  - Degree of financial openness fundamentally shapes vulnerability: limited international capital access → adjustment via domestic disinvestment → amplified output losses; open economies can mitigate via inflows.
  - Government debt increases and confidence shocks operate through negative asset demand channels crowding out productive investment.
  - Combining temporary tax increases dedicated to public investment, improved public investment efficiency, and grant support can boost marginal products of capital and labor and crowd in private investment.
- Model limitations:
  - Simulations in perfect foresight (limits role of expectations analysis).
  - Focus on real variables; inflationary dynamics not captured.
  - Abstracts from certain real-world complexities while highlighting transmission channels and role of financial openness.

*Source: IMF staff compilation (Online Annexes to CHAPTER 3 THE MACROECONOMICS OF CONFLICTS AND RECOVERY).*

### CHAPTER 3 THE MACROECONOMICS OF CONFLICTS AND RECOVERY

### CHAPTER 3 THE MACROECONOMICS OF CONFLICTS AND RECOVERY

### Sample construction and conflict coding
- Final sample used in the analysis consists of 183 conflict episodes.
- Episodes excluded: isolated skirmishes, terrorist attacks, or highly fragmented violence with low casualty counts (systematic reading and coding).
- Note on ChatGPT use for classification:
  - For each conflict a prompt was posed to ChatGPT (version 5.1) with temperature = 0.3: “You are an expert in history and wars. You are given a description of a conflict, including information on participants, start and end year, and location. Provide a description of the conflict and indicate whether the conflict had a large impact on the economy.” Model outputs were manually reviewed for each conflict to verify accuracy and ensure consistency in classification across episodes.

### Conflict exposure and indicator construction
- Country-year panel mapping:
  - An indicator variable equals one if a conflict occurs on a country’s territory (the “conflict site”) in a given year, and zero otherwise.
  - Separate indicators constructed for major, minor, between states, and within states conflicts.
  - Belligerent countries treated analogously.
- Spillover exposure:
  - Geographic proximity: exposure through land border.
  - Trade linkages: share of imports from conflict-site economies in total imports computed for each country-year; a country is classified as exposed through trade if this share exceeds the 90th percentile of the sample distribution.
- Battle-death classification:
  - “Minor” indicates conflicts with 25–999 battle-related deaths.
  - “Major” indicates conflicts with 1,000 and more battle-related deaths.
- When GED best estimate unavailable, the average of the low and high estimates is used and rounded to the nearest whole number.

### Macroeconomic and auxiliary datasets
- Primary macroeconomic sources:
  - Global Macro Database (Müller and others 2025)
  - IMF’s World Economic Outlook database (October 2025 vintage)
  - World Bank’s World Development Indicators database
  - Penn World Table
- Other datasets and indicators used:
  - UCDP Georeferenced Event Database (GED)
  - PRIO Battledeaths Dataset
  - EM-DAT database for natural disasters (major natural disaster defined when share of people killed, as a percentage of the population, exceeds the 99th percentile of the distribution)
  - Project-level performance ratings (Honig, Lall, and Parks 2023)
  - Geocoded aid data (Bomprezzi and others 2025)
  - Night-time lights intensity (Li and others 2020)
  - Subnational corruption measures (Crombach and Smits 2024)
  - Firm-level data from Orbis
- Data treatment:
  - When same concept available from multiple sources, data are combined to improve country and year coverage.
  - Series reviewed to eliminate structural breaks; for most variables the most recent continuous vintage is used.
  - Left-hand-side variables in regressions are winsorized at the 1st/99th percentiles.

### Empirical approach: Local projection difference-in-differences (LP-DiD)
- Estimation equation (cumulative growth up to horizon h for country i in year t):
  - ln(y_{i,t+h}) − ln(y_{i,t−1}) = β_h · ConflictOnset_{i,t} + δ_h · Controls_{i,t} + θ_t^h + μ_i^h + ε_{i,t}^h, for h = 0, ..., 5
- Key definitions and sample restrictions:
  - ConflictOnset_{i,t} = 1 in the first year when a country becomes a conflict-site as long as the preceding five years were without conflict (ΔConflict_{i,t} = 1 & Conflict_{i,t−j} = 0 for 1 ≤ j ≤ L, where L = 5), and zero otherwise.
  - Controls_{i,t} include three lags of the dependent variable.
  - θ_t^h are year fixed effects; μ_i^h are country fixed effects.
  - Sample restricted to observations that experience conflict onset and “clean controls” not affected by conflict within a five-year window (Conflict_{i,t−j} = 0 for −h ≤ j ≤ L, L = 5). Robustness checks use L = 10.
  - In figures and tables, horizons h = 0, ..., 4 are relabeled as 1, ..., 5.
- Regression reporting:
  - Standard errors clustered at the country level.
  - Results reported for multiple specifications and robustness checks (longer horizons, alternative clean-control restrictions, alternative sample restrictions, alternative definitions of conflict based on battle-related deaths per capita).

### Main quantitative results — impact of conflict onset on output (baseline LP-DiD estimates, Column (1) of Online Annex Table 3.2.1)
- Dependent variable: cumulative change in output up to horizon h, where h = 1 corresponds to the year of conflict onset.
- Baseline cumulative impacts (horizon → coefficient):
  - h = 1 → -2.795 ***
  - h = 2 → -4.728 ***
  - h = 3 → -5.578 ***
  - h = 4 → -5.766 ***
  - h = 5 → -6.659 ***
  - h = 6 → -7.813 ***
  - h = 7 → -8.756 ***
  - h = 8 → -9.906 ***
  - h = 9 → -10.779 ***
  - h = 10 → -11.135 ***
- Significance notation: *** p<0.01; ** p<0.05; * p<0.1.
- Summary of robustness checks (columns (2)–(10)):
  - Across specifications the results consistently show a negative impact of conflict onset on output, with magnitudes and signs broadly consistent with the baseline.
  - Column (2): cumulative effect over a 10-year horizon, maintaining baseline onset definition.
  - Column (3): defines conflict onset using a 10-year window and excludes from clean-control group observations exposed to conflict within that window.
  - Columns (4) and (5): relax clean-control restrictions and report results over 5- and 10-year horizons, respectively, while including additional lags of the conflict-onset indicator.
  - Column (6): restricts sample to conflicts lasting at least five consecutive years.
  - Column (7): excludes conflicts motivated by economic considerations based on a narrative classification of motives.
  - Column (8): reports estimates using the full conflict sample rather than the chapter’s reclassification.
  - Columns (9) and (10): alternative conflict definitions based on battle-related deaths per capita using thresholds of 50 and 100 deaths per million inhabitants, respectively, with onset defined as first year threshold is crossed following five years below it.
- Regression diagnostics reported for h = 1 in Column (1):
  - Number of Observations = 10,725
  - Number of Countries = 191
  - Number of Conflict Onsets = 148
  - R^2 = 0.29

### Comparison with other shocks (Online Annex Table 3.2.2)
- The analysis compares conflict impacts with banking crises, currency crises, sovereign debt crises, and severe natural disasters using the same LP-DiD framework and cumulative output changes up to horizon h.
- Definitions:
  - Banking crisis: convergence of significant financial distress and major policy interventions.
  - Currency crisis: nominal depreciation vis-à-vis the U.S. dollar of at least 30 percent and at least 10 percentage points higher than the rate of depreciation in the year before.
  - Sovereign debt crisis: failure to meet principal and/or interest payments on the due date and/or rescheduling debt with less favourable terms.
  - Major natural disasters: disaster-related deaths as a share of the population exceed the 99th percentile of the sample distribution.
- Number of observations and R^2 reported for h = 1 in Table 3.2.2:
  - Conflict: Number of Observations = 10,648; Number of Crisis Onsets = 160; R^2 = 0.28
  - Banking Crises: Number of Observations = 10,828; Number of Crisis Onsets = 295; R^2 = 0.18
  - Currency Crises: Number of Observations = 10,074; Number of Crisis Onsets = 392; R^2 = 0.18
  - Sovereign Debt Crises: Number of Observations = 10,994; Number of Crisis Onsets = 200; R^2 = 0.18
  - Natural Disasters: Number of Observations = 12,924; Number of Crisis Onsets = 36; R^2 = 0.24
- Overall finding: conflicts generate large cumulative negative effects on output; estimates are placed in context by direct comparison to other major shocks using the same empirical framework.

### Micro-level analyses and postconflict recovery datasets
- Postconflict recovery analyses use project-level performance ratings, geocoded aid, night-time lights, subnational corruption measures, and firm-level data to analyze recovery dynamics at more granular levels (Online Annexes 3.6–3.8).
- Panel- and unit-level analyses and samples:
  - Multiple empirical exercises use subsets of countries classified by income group (AEs, EMEs, LICs) and by exercise (Macroeconomic Dynamics Before and After Wars End; Postconflict Recovery Analysis; Long-Term Scarring Effects of Wars on Individuals; Project Outcome Analysis; Subnational Analysis; Firm-Level Analysis).
  - Note: the chapter’s sample lists of economies by exercise and the coding conventions (* conflict onset; ^ first year of nonfragile peace) are provided in Online Annex Table 3.1.2 and accompanying notes.

_Italic line: Source: IMF staff compilation. Note: * indicates conflict onset; ^ indicates first year of nonfragile peace. AEs = advanced economies; EMEs = emerging market economies; LICs = low-income countries._

### CHAPTER 3 THE MACROECONOMICS OF CONFLICTS AND RECOVERY

### CHAPTER 3 THE MACROECONOMICS OF CONFLICTS AND RECOVERY

### Output costs and comparison with other shocks
- Estimated output costs from conflicts exceed those typically associated with economic crises (banking, currency, or debt crises) and those induced by severe natural disasters (LP‑DiD framework; comparability with panel 2 of Figure 3.4).
- Baseline LP‑DiD estimates report cumulative negative output effects following conflict onset across multi‑year horizons (see Online Annex Table 3.2.2 and panel 2 of Figure 3.4).

### Heterogeneity of conflict effects (intensity, duration, type)
- Conflict onset has negative effects across all dimensions considered (Online Annex Table 3.2.3 / panel 3 of Figure 3.4).
- By intensity:
  - Major conflicts (≥1,000 battle‑related deaths) generate the largest output losses.
  - Minor conflicts (25–999 battle‑related deaths) have statistically significant effects at shorter horizons that fade over time.
- By duration:
  - Short conflicts (up to two years) tend to produce larger immediate impacts than long conflicts (more than two years).
- By type:
  - Between‑state (interstate) conflicts generate initially large but less persistent effects.
  - Within‑state conflicts (intrastate, intrastate with foreign intervention, extrasystemic) are associated with more protracted output losses.

### Spillovers to belligerents, neighbors, and trade partners
- Effects on belligerent countries are not statistically significant overall (Online Annex Table 3.2.4).
- Exposure through geographic proximity or trade linkages is associated with negative output effects in the first two years, which subsequently fade.
- Definitions used:
  - Belligerent countries: participate in a conflict without experiencing hostilities on their own territory.
  - Neighboring countries: share a land border with a conflict‑site economy.
  - Trade partners: countries whose import share from conflict‑site economies exceeds the 90th percentile of the distribution.

### Demand‑side components of GDP and sectoral outcomes
- Investment:
  - Investment declines sharply following conflict onset and does not recover within five years (Online Annex Table 3.2.5.A).
- Consumption:
  - Private consumption falls.
  - Public consumption effect is not statistically significant.
- Trade:
  - Both imports and exports contract, with exports experiencing larger declines than imports.
- Sectoral value added:
  - Conflict onset has a negative and statistically significant impact across all sectors, with the contraction in industry value added being the most pronounced (Online Annex Table 3.2.6).

### Fiscal and external sector vulnerabilities
- Government debt (real, using GDP deflator) tends to rise following conflict onset, particularly in early years (Online Annex Table 3.2.5.B).
- Exports‑to‑imports ratio declines, reflecting larger contraction in exports than imports.
- Personal remittances received (USD) increase, although the effect is not statistically significant.
- Official development assistance (ODA, USD) rises, especially at longer horizons.
- Capital flows and liabilities:
  - FDI liabilities (foreign‑owned domestic investment) decline after conflict onset.
  - Portfolio debt liabilities decline after conflict onset.

### Financial, price‑level, and reserve effects
- Financial openness (Chinn–Ito index) declines following conflict onset (Online Annex Table 3.2.5.C).
- Exchange rate depreciates; reserve assets decline.
- Consumer prices rise sharply:
  - The price level increases by about 30 percent five years after conflict onset (text summary referencing Online Annex Table 3.2.5.C).

### Production factors, employment, and human toll
- Capital stock declines persistently with no evidence of recovery (Online Annex Table 3.2.5.D).
- Total factor productivity (TFP) falls in early periods; effect becomes statistically insignificant at longer horizons.
- Employment declines significantly across all horizons.
- Number of deaths rises sharply following conflict onset.

### Additional macroeconomic patterns
- Fiscal reallocation:
  - Defense spending increases after conflict onset.
  - Social spending declines after conflict onset.
- Trade balance:
  - Trade balance widens, but widening is confined to the first four years of conflict, consistent with import compression in subsequent years.
- Savings and investment:
  - Real gross national savings fall, with the decline exceeding the drop in investment.
- Uncertainty and informality:
  - Conflicts are associated with higher uncertainty (uncertainty index) and greater informality (shadow economy as percent of GDP).
- Monetary policy:
  - Monetary authorities tend to raise short‑term nominal interest rates in response to wartime economic pressures, particularly at medium horizons.

### Scarring effects on individual health outcomes
- Empirical specification: H_{i,t} = β1 YearsInConflict_{i,t} + β2 X_{i,t} + φ_c + ε_{i,t}; YearsInConflict measures number of years an individual lived in a conflict‑site country over the lifecycle; X includes year of birth, age, gender, education, household wealth; country fixed effects included (Online Annex 3.3 / Table 3.3.1).
- Coefficients (OLS on standardized health indicators; mean 0, SD 1):
  - Composite health: Number of Years Living in Conflict = -0.014*** (standard error 0.004).
  - Cognitive PC: Number of Years Living in Conflict = -0.010* (standard error 0.005).
  - Physical PC: Number of Years Living in Conflict = -0.017*** (standard error 0.005).
  - Psychological well‑being: Number of Years Living in Conflict = -0.016*** (standard error 0.005).
- Samples and fit:
  - Number of Observations: 755,958 (Composite), 382,436 (Cognitive PC), 628,729 (Physical PC), 345,479 (Psych. well‑being).
  - Number of Countries: 41 (Composite), 34 (Cognitive), 30 (Physical), 30 (Psych.).
  - R²: 0.33 (Composite), 0.38 (Cognitive), 0.13 (Physical), 0.18 (Psych.).

### Macroeconomic dynamics after conflict termination (postconflict peace)
- Regression approach mirrors LP‑DiD for conflict onset; ConflictTermination_{i,t} = 1 in first year of peace if all five following years remain conflict‑free (L = 5) (Online Annex 3.4).
- Postconflict cumulative outcomes (selected results, Online Annex Table 3.4.1):
  - First year of peace is associated with positive cumulative changes in output across specifications; example point estimates reported for horizons h = 1 to h = 5 (selected values shown in table):
    - h = 1: 1.160*; 1.165*; 1.173**; 1.553*; 1.231 (columns vary by peace type and subsample).
    - h = 2: 2.572**; 1.341; 2.107***; 2.661*; 2.555.
    - h = 3: 2.727**; 1.530; 2.281**; 2.952*; 3.426.
    - h = 4: 3.361**; 1.789; 2.750**; 3.319*; 4.670.
    - h = 5: 3.879**; 1.124; 2.788*; 3.839; 5.309.
  - Table focuses on nonfragile vs fragile peace:
    - Peace classified as nonfragile if conflict does not restart within first five years, fragile if conflict restarts within that period.
  - Additional postconflict indicators:
    - CPI, capital stock, TFP, employment, uncertainty index, government tax revenue are among the variables analyzed up to horizon h.
- Sample and fit for postconflict estimates:
  - Number of Observations (for peace analysis examples): 84.
  - R² (example): 0.10 for one specification noted in table.

*Source: IMF staff calculations (Online Annexes to CHAPTER 3 THE MACROECONOMICS OF CONFLICTS AND RECOVERY).*

### CHAPTER 3 THE MACROECONOMICS OF CONFLICTS AND RECOVERY

### CHAPTER 3 THE MACROECONOMICS OF CONFLICTS AND RECOVERY

### Empirical design and sample restrictions
- Conflict-termination sample restricted to observations that experience conflict termination and clean controls that are not affected by conflict within a five-year window (퐶표푛푓푙푖푐푡
푖,푡−푗
=0 푓표푟 −ℎ≤푗≤퐿, where 퐿=5).
- Controls include three lags of the dependent variable, year fixed effects (휃
푡
ℎ
), and country fixed effects (휇
푖
ℎ
). 휀
푖,푡
ℎ
 denotes the error term.
- Recovery-window focus: first five years of each post-conflict recovery episode for policy-driver analysis.

### Post-conflict output and macro dynamics (summary of regression results)
- Recoveries are conditional on the maintenance of peace; even when peace endures, recoveries remain slow and uneven.
- Pooling all conflict-termination episodes (unconditional recovery dynamics) yields even slower recoveries.
- Commodity-exporting nonfragile peace episodes (fuel and primary commodity exports > 50 percent of total exports): output initially rebounds strongly but becomes statistically insignificant in subsequent years.
- Additional documented dynamics:
  - Inflation and production factors documented in columns (5)–(8) (panels 2 and 3 of Figure 3.9).
  - Muted response of uncertainty following conflict termination (column (9)).
  - Increases in tax revenue collection following conflict termination (column (10)).

### Drivers of post-conflict recovery — regression specification
- Estimated specification: 푅푒푐표푣푒푟푦
푟,푖,푡
= 훽
1
푃표푙푖푐푖푒푠
푟,푖,푡
+ 훾
푟,푖
+ 훿
푡
+ 휀
푟,푖,푡
  - 푅푒푐표푣푒푟푦
푟,푖,푡
 is the percentile distribution of real GDP growth within each horizon t = 1,...,5.
  - 푃표푙푖푐푖푒푠
푟,푖,푡
 includes low inflation, stable inflation, stable REER, debt restructuring, capacity development, and social spending.
  - 훾
푟,푖
 and 훿
푡
 denote recovery-specific and horizon fixed effects.
- Policy variables entered one at a time to address multicollinearity; alternative specification (percentage growth differential) yields broadly comparable results.

### Policy correlates of post-conflict recovery — key coefficients (Online Annex Table 3.5.1)
- Dependent variable: percentile of the growth distribution at each horizon (t = 1,...,5). Regressions control for postconflict episode and horizon fixed effects.
- Coefficients and robust standard errors (in parentheses):
  - Low Inflation: 0.353* (0.181)
  - Stable Inflation: 0.222 (0.283)
  - Stable REER: 0.361* (0.206)
  - Debt Restructuring: 0.461* (0.234)
  - Capacity Development: 0.366*** (0.123)
  - Social Spending: 0.491*** (0.093)
- Number of Observations by column: 693, 684, 545, 737, 336, 431 respectively.
- R2 by column: 0.393, 0.397, 0.398, 0.399, 0.516, 0.490.
- Indicator definitions:
  - Low inflation, stable inflation, stable REER = indicator variables equal to 1 if measure falls below the sample mean (based on inflation levels, inflation volatility, and REER volatility).
  - Debt restructuring = 1 if a restructuring occurs.
  - Capacity development = log number of participants in IMF training.
- Significance notation: *** p<0.01; ** p<0.05; * p<0.1.

### Case-study summary statistics (selected high post-conflict recoveries; Online Annex Table 3.5.2)
- Countries and conflict durations:
  - Bosnia and Herzegovina: 1992–1995
  - Cambodia: 1989–1998
  - Nepal: 1996–2006
  - Rwanda: 1990–2001
  - Sri Lanka: 1983–2009
  - Côte d’Ivoire: 2010–2011
- Human casualties (percent): 1.46, 0.08, 0.06, 11.2, 0.43, 0.01 (by country order above).
- Macroeconomic recovery (Real GDP Growth): 24.5, 9.5, 4.5, 8.3, 7.2, 8.4 (percent).
- Real GDP per capita Growth (average over first 5 years): 21.1, 7.7, 3.7, 6.0, 6.4, 5.9 (percent per year).
- Selected macro stabilization changes (postconflict relative to conflict period):
  - Inflation rate (difference): -61.8, 2.7, -4.9, -4.7, -1.8 (values aligned with available columns; missing entries in source omitted).
  - Inflation volatility (percentage point): -65.3, 0.6, -14.4, -3.3, -1.8.
  - GDP growth volatility (percentage point): 9.9, 0.4, -0.5, -15.9, 0.1, -7.1.
  - REER appreciation: 1.1, -9.5, 2.3, -0.6, 0.1, 0.9.
  - Debt-to-GDP: -3.2, 19.2, 4.2, 8.4, 20.3 (aligned to available columns).
  - Aid-to-GDP (average during postconflict): 23.4, 8.6, 4.7, 18.5, 0.7, 3.1.
  - Remittances-to-GDP: 0.2, 2.6, 20.6, 0.4, 8.0, 0.8.
- Institutions and inclusive policy changes (percentage change, final year relative to last conflict year):
  - Institutional quality: 13.7, 2.8, 60.8, 62.0, 41.6 (by country order where data provided).
  - Public sector corruption: 0.0, 0.0, -1.2, 0.0, 34.5, -3.7.
  - Statistical capacity score: 2.2, 22.0, 11.7, 16.4, 220.8, 111.8.
  - Social spending-to-GDP: 2.2, 3.9, 50.6, 6.5, 13.3, 20.3.
- Security and peacebuilding:
  - UN peacekeeping (years in): entries include 5, 0, 0, 0, 0, 5 (aligned to table where provided).
  - Peacebuilding strategies: Justice provisions = 1; Political provisions = 1 (as recorded in the table).

### Project outcome analysis — difference-in-differences specification
- Outcome model: 푂푢푡푐표푚푒
푝,푟,푑,푡
= 훽
1
푃표푠푡퐶표푛푓푙푖푐푡
푝,푟,푑,푡
+ 훽
2
푃푟표푗푒푐푡
푝,푟,푑,푡
+ 훽
3
푀푎푐푟표
푝,푟,푑,푡
+ 훾
푟
+ ∅
푑
+ 휑
푡
+ 휀
푝,푟,푑,푡
  - Outcome is 1–6 Likert-scale project outcome rating (higher = more satisfactory).
  - 푃표푠푡퐶표푛푓푙푖푐푡 = 1 if project implemented during post-conflict period.
  - Project-level controls: project size (log), donor local office presence, external evaluation, ex-post evaluation, repeated project indicator.
  - Macro-level covariates averaged over project duration: log GDP per capita, GDP growth, log population, inflation, aid-to-GNI, FDI-to-GDP, index of aid fragmentation.
  - Fixed effects: recipient, donor, approval-year.

### Project outcome regression results (Online Annex Table 3.6.1)
- Baseline and heterogeneity coefficients (standard errors in parentheses):
  - Baseline postconflict effect (column (1)): 0.101** (0.046)
  - Column (2) (satisfactory supervision): 0.094** (0.046)
  - Column (3) (sectoral heterogeneity): 0.105** (0.047)
  - Column (4) (high vs low investment scale-up): 0.089* (0.050)
  - Supervision (column (2) expanded): 0.332*** (0.070)
  - Productive sector (column (3) sector coefficient): -0.180** (0.074)
  - Education and Health sector: 0.134* (0.081)
  - Infrastructure sector: 0.231*** (0.051)
  - High Scale-Up: 0.206 (0.131)
  - Low Scale-Up: 0.101** (0.047)
- Number of observations across specifications: up to 8,760; Outcome Mean: 4.277 (baseline).
- Robust standard errors clustered at the country-by-approval-year level.
- Significance notation: *** p<0.01; ** p<0.05; * p<0.1.

### Time-varying effects on project outcome (Online Annex Table 3.6.2)
- Time-varying effect coefficients by year (standard errors in parentheses):
  - Year 1: -0.067 (0.189)
  - Year 2: -0.008 (0.148)
  - Year 3: 0.096 (0.095)
  - Year 4: 0.105 (0.072)
  - Year 5: 0.082 (0.067)
  - Year 6: 0.113* (0.060)
  - Year 7: 0.150*** (0.056)
  - Year 8: 0.126** (0.050)
  - Year 9: 0.089* (0.048)
  - Year 10: 0.101** (0.046)
- Observations for each year: 8,760; Outcome Mean: 4.277.
- Significance notation: *** p<0.01, ** p<0.05, * p<0.1.

### Subnational analysis — nightlights, aid, and governance (Online Annex 3.7)
- Data sources: geocoded ADM2 aid disbursements from 22 donors (1992–2023), geocoded GED, nightlights intensity, subnational corruption dataset; sample covers 35,432 ADM2 across 125 aid-recipient countries.
- Baseline specification: change in log nightlights (푍
푟,푖,푡
− 푍
푟,푖,푡−1
) regressed on lagged nightlights, lagged aid (cumulative 3-year aid, lagged one year), postconflict dummy, interaction Aid × Postconflict, population density, ADM2 fixed effects, and country–year fixed effects.
- Notes: 0.01 added to nightlights before logs; 1 added to aid to retain zero observations.

### Subnational regression coefficients (Online Annex Table 3.7.1)
- Coefficients and standard errors (in parentheses):
  - Lagged Nighttime Light: -0.163*** (0.002) ; -0.164*** (0.002) ; -0.164*** (0.002) across columns.
  - Aid: 0.011*** (0.000) uniformly.
  - Postconflict: 0.064*** (0.019) ; 0.213*** (0.079) ; 0.057*** (0.019) for baseline, high governance-improvement, low governance-improvement subsamples respectively.
  - Aid × Postconflict: 0.005*** (0.002) ; 0.015 (0.010) ; 0.005*** (0.002).
- Number of observations: 976,308; 956,555; 974,934 across columns.
- R2: 0.18 for all reported columns.
- High (low) governance improvement defined as state-year increases in the governance index exceeding (not exceeding) two standard deviations of annual changes.
- Significance notation: *** p<0.01; ** p<0.05; * p<0.1.

### Firm-level analysis — geolocation, exposure, and specification
- Firm sample drawn from Orbis for countries experiencing at least one conflict episode over 1989–2024 (GED).
- Geolocation process: postcode and city → Photon geocoder → cross-checked with World Bank shapefiles; hierarchical matching (postcode, then city, then city+postcode); supplemented by Nominatim for low-match countries.
- Conflict exposure: for each firm-year, aggregate conflict events and battle-related deaths within a 20-kilometer radius of firm location.
- Exclude observations affected by conflict episodes; regressions exploit within-firm variation.
- Baseline regression: ln Y_f,i,s,t = β1 PostConflict_f,i,s,t + β2 X_f,i,s,t + δ_f + φ_s,t + ω_i,t + ε_f,i,s,t
  - Dependent variables: log firm-level capital, labor, and quantity-based TFP (TFPQ).
  - Postconflict dummy = 1 during five years following a conflict episode within 20 km of firm.
  - Firm-level controls: firm age, size, leverage, export status (depending on specification).
  - Fixed effects: firm, sector-by-year, country-by-year.

### Firm dynamics during postconflict — baseline and factor-intensity heterogeneity (Online Annex Table 3.8.1)
- Baseline coefficients (standard errors in parentheses):
  - Postconflict on Capital: -0.046* (0.027)
  - Postconflict on Labor: 0.059** (0.022)
  - Postconflict on TFP: -0.016 (0.025)
- Factor-intensity interactions (columns 4–6):
  - Capital Intensive: -0.005 (0.031), Labor: 0.075** (0.036), TFP: 0.048*** (0.016)
  - Labor Intensive: -0.046* (0.024), Labor: 0.080*** (0.008), TFP: -0.061* (0.030)
- Number of observations:
  - Capital: 16,577,311
  - Labor: 21,140,632
  - TFP: 4,923,758
- R2: 0.96 for capital and labor; 0.86 for TFP.
- All regressions include firm fixed effects, country-year fixed effects, sector-year fixed effects.
- Significance notation: *** p<0.01; ** p<0.05; * p<0.1.

### Firm dynamics — heterogeneity by firm characteristics (Online Annex Table 3.8.2)
- By net worth (standard errors in parentheses):
  - Low Net Worth:
    - Capital: -0.104** (0.042)
    - Labor: -0.038*** (0.011)
    - TFP: -0.080*** (0.029)
  - High Net Worth:
    - Capital: -0.023 (0.045)
    - Labor: 0.101*** (0.014)
    - TFP: 0.050** (0.018)
- By export status:
  - Exporter:
    - Capital: -0.019 (0.026)
    - Labor: 0.060*** (0.015)
    - TFP: 0.221*** (0.023)
  - Non-Exporter:
    - Capital: -0.047* (0.028)
    - Labor: 0.059** (0.022)
    - TFP: -0.030 (0.033)
- Observations and R2 as in baseline.
- Interpretation: postconflict effects vary by firm net worth and exporter status, with exporters showing positive TFP responses in postconflict settings.

*Source: IMF staff calculations (Online Annexes to Chapter 3, CHAPTER 3 THE MACROECONOMICS OF CONFLICTS AND RECOVERY).*

### Annex Table 3.8.2 reports heterogeneity by firm characteristics, shown in panel 3 of Figure 3.12.

### Annex Table 3.8.2 reports heterogeneity by firm characteristics, shown in panel 3 of Figure 3.12

### Model overview
- Framework: overlapping generations model for a small open economy with heterogeneous households; household problems and production technology follow Auclert and others (2024).
- Population: 푁_t equals the sum across all ages j of individuals 푁_{jt}, growing at rate 1 + 푛_t. Each individual faces an exogenous survival probability; maximum lifespan J = 96 periods.
- Agents choose sequences of consumption and asset holdings to maximize expected lifetime utility.
- Output: single good used for private consumption, government consumption, and investment; final output produced competitively with constant returns to scale and labor-augmenting technological progress.
- Effective labor aggregates labor across all age groups and household types.
- Net rental rate of capital r_t equals the marginal product of capital minus the depreciation rate δ.

### Environment, demographics, and government budget constraint
- Public investment: governed by public investment efficiency which determines how public investment translates into higher productivity for the capital stock.
- Government role: purchases goods, provides transfers, finances via labor taxation and debt issuance; government spending comprises consumption and public investment.
- Financing: government faces a flow budget constraint.

### Asset market clearing and financial frictions
- Financial accounts are frictional, spanning from closed economy to fully open small economy.
- Degree of financial openness controlled by the elasticity of foreign demand for domestic assets.
- Asset market clearing condition: households’ asset demand relative to output equals the sum of government bonds relative to output 퐵/푌, the capital-output ratio, and the net foreign asset position relative to output.
- Net foreign asset elasticity: equals the negative of the partial derivative of (푁퐹퐴/푌) with respect to r and is positive; captures sensitivity of capital flows to domestic interest rates.
- Implication: in a small open economy foreign capital can fill gaps; in a closed economy adjustment occurs through disinvestment.

### Conflict shocks and transmission mechanisms
- Three main categories of conflict shocks:
  - Capital destruction shocks: directly reduce physical capital stock and production capacity.
  - Population shocks: affect age groups differently; calibrated in simulations to a three percent decline on impact (consistent with refugees accounting for the bulk of population losses). Note: casualties alone in some cases amounted to as much as 11 percent of the population (see Online Annex Table 3.5.2).
    - Losses to working-age population: effects gradually subside as cohorts age out.
    - Losses to young populations: deeper and longer-lasting scars; output effects peak several years after the initial shock as depleted cohorts enter working age.
  - Confidence shocks: two channels—domestic confidence shocks reduce households' willingness to hold assets (creating excess supply and pushing interest rates upward); foreign confidence shocks raise foreign risk premia, requiring higher domestic interest rates.
- Mechanisms:
  - Higher interest rates are contractionary via reduced investment (since output depends on capital stock).
  - Magnitude of output decline depends on whether the asset demand gap is filled by reduced investment, increased savings, or capital inflows.
  - In a small open economy, household asset changes have negligible output effects because foreign capital fills the gap; in a closed economy, adjustment through disinvestment substantially harms output.
  - Increase in government debt acts like a negative asset demand shock, crowding out productive capital investment.

### Recovery policies and simulation design
- Timing: recovery policies initiated five years after conflict onset; first simulated in isolation, then together.
- Macroeconomic stabilization: from the scenario combining all shocks, shrink the wedges generated by external and domestic confidence shocks gradually to their pre-conflict levels over eight years.
- Financing package:
  - Linear increase of the tax-to-GDP ratio by three percentage points over 15 years of post-conflict, fully used for public investment.
  - Public investment efficiency increases by 10 percentage points relative to the average level reported in the October 2025 Fiscal Monitor, boosting the output return of the tax-funded recovery by about 1.4 percentage points (consistent with estimates in the October 2025 Fiscal Monitor).
  - Grant support: government has access to grant support amounting to about 0.5 percent of GDP per year during the first five years of post-conflict recovery.
  - Population recovery: population assumed to recover gradually with the return of displaced populations net of natural mortality four years after the end of the conflict (in line with evidence in Box 3.3).

### Key model insights and policy implications
- Degree of financial openness fundamentally shapes vulnerability to conflict shocks:
  - Limited access to international capital markets → confidence shocks force adjustment through domestic disinvestment and capital stock reduction → amplified output losses.
  - Open economies can rely on capital inflows to mitigate asset demand gaps.
- Government debt increases and confidence shocks operate through equivalent mechanisms as negative asset demand shocks that crowd out productive investment.
- Combining fiscal financing (temporary tax increase dedicated to public investment), improvements in public investment efficiency, and grant support can boost marginal products of capital and labor and crowd in private investment, improving recovery prospects.

### Limitations of the model
- Simulations run in perfect foresight, which precludes a detailed analysis of the role of expectations.
- Model focuses on real variables and does not capture potentially inflationary dynamics.
- Simplification: the model abstracts from certain real-world complexities while highlighting transmission channels and the role of financial openness for policy design.

*Source: Online Annex 3.9, ch3onlineannex - Annex Table 3.8.2 reports heterogeneity by firm characteristics, shown in panel 3 of Figure 3.12.*

---


_Source: https://www.imf.org/-/media/files/publications/weo/2026/april/english/ch3onlineannex.pdf_