## 1. Introduction

## Source details

**Canonical URL:** [1. Introduction](https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026069-source-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2026/english/wpiea2026069-source-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2026/english/wpiea2026069-source-pdf.pdf.json)

---

### Rising insecurity and fiscal responses
- Violent conflict events in sub-Saharan Africa increased significantly over the past two decades, with the average quarterly number of incidents in 2024 being five times higher than in 2000.
- Fatalities from these incidents almost tripled over the period; in 2024, violent conflicts claimed more than 50 thousand lives in the region.
- The Sahel is particularly impacted, hosting over 5.5 million refugees and internally displaced persons.
- Security pressures have reshaped fiscal priorities:
  - In the Sahel, defense spending nearly doubled from 2.7 percent of GDP in 2016 to 4.6 percent in 2023.
- Insecurity has disrupted economic activity, weakened revenue mobilization, and increased demand for development and humanitarian spending, tightening fiscal space and forcing allocation trade-offs.

### Competing hypotheses from the literature
- Guns versus butter (crowding-out): higher military spending reduces fiscal space for investment in physical and human capital.
  - Example: Fan, Liu, and Coyte (2018) find a negative link between military and health spending across nearly 200 countries (2000–2013).
- Complementarity/positive spillovers: military spending can reduce violence, protect infrastructure, enable service delivery, and lower uncertainty, potentially boosting productivity and public services.
  - Examples: Lin et al. (2015) and Njifen and Aneman (2023) find positive links or slight increases in social spending associated with military spending in some samples.
- No systematic link: social and military spending may respond to different drivers and both can expand with overall government expenditure.
  - Examples: Russett (1982); Davis and Chan (1990); Schiff, Gupta and Clements (1998).
- Recent empirical evidence is mixed:
  - Marzian and Trebesch (2025): no significant cuts to social spending from military booms in historical sample of 20 countries (1870–2022).
  - IMF (2026): on average no significant effects on social spending across 165 countries since 1946, but wartime episodes show declines in real-term social spending.
  - Regional heterogeneity in SSA: Njifena and Anemann (2023) find crowding-out of education in high-security-challenge areas (Sahel, Lake Chad Basin); Central and Southern Africa show complementarity in some studies (Elshafei et al. 2025).

### Paper’s objectives and contribution
- Main aim: investigate whether military expenditures exert crowding-out effects on education and health budgets in Sub-Saharan Africa and examine the dynamic fiscal response to military spending shocks.
- Twofold contribution:
  - Identification: employ multiple strategies to isolate exogenous changes in military spending:
    - (i) military spending surge variable based on observed spending patterns;
    - (ii) residuals from a government reaction function (Caldara and Kamps, 2017 style);
    - (iii) instrumental variables (preferred approach).
  - State-dependencies: analyze whether sovereign debt levels, availability of official development assistance (ODA), and the business cycle influence crowding-out.

### Paper structure
- Section 2: data and stylized facts on military, health, and education spending in SSA.
- Section 3: empirical strategy and baseline results.
- Section 4: robustness checks, including alternative definitions and differences between Fragile and Conflict-Affected States (FCS) and non-FCS.
- Section 5: state-dependencies (debt, international aid, business cycle).
- Section 6: conclusions.

### Correlation and sample notes (from preliminary observations)
- Non-causal bivariate evidence: changes in military spending share are on average followed by small declines in health and education spending shares; association is statistically significant but magnitude is modest.
- The estimate "0.07 which implies that about 0.5 percent of the variation in changes in social spending is explained by co-movements with changes in military share."
- Sample construction:
  - Final dataset: an unbalanced panel covering 1990-2023 and 33 countries.
  - Excluded 12 SSA countries due to extremely low military expenditures and insufficient data.
  - Countries listed in the sample subset (coverage-constrained): Cabo Verde, Equatorial Guinea, Eritrea, Eswatini, Lesotho, Seychelles, South Sudan, Zimbabwe, Republic of Congo, The Gambia, and Comoros.
- Data sources:
  - Military expenditure and ammunition imports: SIPRI Military Expenditure Database.
  - Government spending on education and health: Gethin (2024) database (1980–2022), updated to 2023.
  - Macroeconomic variables: IMF WEO (October 2025 vintage), Global Macro Database (Müller et al., 2025).
  - Additional: World Bank WDI (ODA, terms of trade), Bloomberg (VIX), Haver (10-year US Treasury yields), ACLED (fatalities).

### Key empirical approach (preview)
- Method: local projection method (Jordà, 2005) to estimate dynamic effects of military spending shocks on public expenditure shares for education and health.
- Shock identification strategies:
  - (i) military spending surge (top 30th percentile in both country-specific and full-sample distributions);
  - (ii) residuals from a government reaction function controlling for output gap, terms of trade gap, debt/GDP, ODA/GDP, fatalities/POP;
  - (iii) instrumental variables (preferred): value of weapons imports; country’s share of total military spending in SSA; interaction of U.S. producer price index for ammunition with U.S. military equipment imports.
- Baseline local projection regression (schematic):
  - 푦푖,푡+ℎ − 푦푖,푡−1 = 훼푖ℎ + 훽ℎ 푠ℎ표푐푘푡 + 휹 X푖,푡 + Θ푡ℎ + 휀푖,푡+ℎ
  - Main coefficient of interest: 훽ℎ (effects of military spending shocks at horizons h).
- Controls include: two lags of dependent variable and shocks; contemporaneous and two lags of output gap, aid/GDP, terms-of-trade cyclical component, fatalities per capita; two lags of government debt/GDP; global financial controls (VIX, 10-year US Treasury yield).

### Preliminary findings (surge and reaction-function approaches)
- Military spending surges (local projections and event study):
  - A surge crowds-out health spending between years 2 and 4; effects insignificant afterwards.
  - Surge associated with decline in health spending share by up to 0.3 percentage points.
  - Education shows an initial small, not statistically significant crowding-in two years after the shock; event study shows negative post-surge point estimates for education with wider confidence intervals.
  - Pre-surge coefficients fluctuate around zero, indicating absence of systematic pre-trends.
- Reaction-function residual shocks:
  - Crowd-out effects generally neither statistically nor economically significant.
  - A one standard deviation increase in military share leads to decreases in health share of less than 0.5 percentage point after one year.
  - Effect on education share close to zero; some small medium-term crowding-in for education (< 0.5 percentage points) and intermittently statistically significant.

### Instrumental variables (preferred) — instruments and first-stage diagnostics
- Instruments:
  - (i) Value of weapons imports.
  - (ii) Country’s share of total military spending in Sub-Saharan Africa.
  - (iii) U.S. producer price index for ammunition interacted with value of military equipment imports from the U.S.
- First-stage performance:
  - First-stage F-statistics range from 9 to 16 across horizons and specifications.
  - Many specifications exceed the conventional threshold of 10.
  - Kleibergen–Paap and Hansen J tests provide evidence in favor of instrument validity, but weak-identification concerns remain.
  - Pflueger and Wang (2015) weak-instrument test flags potential issues; authors use Finlay, Magnusson, and Schaffer (2013) weak-instrument-robust inference and apply Conley et al. (2012) plausibly exogenous framework allowing each instrument a direct effect on education spending of up to ±5 percent across horizons.

### IV local projections — main quantitative causal findings
- General pattern:
  - Short-term: education shows little short-term effect; health shows short-term declines in some specifications.
  - Medium to long term: significant declines in shares of both education and health.
- Exact IV estimates reported:
  - Education spending:
    - Reduces the share of education spending by over 0.5 percentage points after three years.
    - Reduces the share of education spending by 1.1 percentage point after five years.
  - Health spending:
    - Reduces the share of health spending by about 0.8 percentage points after one year.
    - Reduces the share of health spending by nearly 1.3 percentage points after five years.
- Mechanism explanation:
  - Budget rigidities differ: education has a large share devoted to salaries (rigid short-term), while health allocates a sizable share to operating costs (more flexible and vulnerable to reallocations).

### Key methodological and sensitivity notes
- IV estimates use first-stage fitted values (ΔM̂ i,t) in second-stage local-projection regressions.
- Weak-instrument-robust confidence intervals reported (Finlay, Magnusson and Schaffer (2013)).
- IV inference critically depends on exclusion restriction; baseline IV estimates are sensitive under Conley et al. (2012) plausibly exogenous framework.
- Robustness checks include alternative variable definitions (shares of GDP; real per-capita or per-school-age-child measures), pre-COVID-19 sample checks, alternative instrument selections, and controls for governance, off-budget aid, and IMF programs.

### Robustness and heterogeneity (summary from later sections)
- Alternative definitions:
  - Reaction-function shocks: no economically significant changes in education/GDP after two years; health declines by less than 0.1 percent of GDP after two years.
  - IV: a one-standard-deviation increase in military spending as a share of GDP reduces health expenditure by about 0.5 percentage points after five years (when health is % of GDP).
  - No evidence that increases in total military spending reduce education expenditure per school-age child or health spending per capita in IV results.
- Fragility and conflict (FCS vs non-FCS):
  - In non-FCS countries, military spending shocks clearly crowd out health and education expenditures.
  - In FCS countries, crowding-out less apparent, especially for education — possibly due to donor scaling-up.
- State dependencies (IV smooth-transition analysis):
  - Government debt: high overall debt-to-GDP does not necessarily increase vulnerability of social spending; but high non-concessional external debt/GDP is associated with larger crowding-out of health.
  - Foreign aid (ODA): crowding-out of health more likely when access to ODA is limited; evidence of crowding-in in health when ODA grants are higher.
  - Business cycle: effects do not vary significantly with the output gap.
- Additional controls:
  - Crowding-out effects robust to inclusion of governance (control of corruption), off-budget aid, and IMF program presence, though magnitudes are somewhat reduced.

### Conclusions and policy implications (from paper overview)
- Main quantitative summary:
  - Using IV identification, a one-standard-deviation increase in the share of military spending in total government expenditure reduces the shares of education and health spending by about one percentage points over the medium-term.
  - Crowding-out tends to materialize sooner for health expenditures.
- Heterogeneity:
  - Crowding-out of health is more likely when access to international aid is limited; abundant aid can mitigate or reverse crowding-out.
  - Overall debt levels and the business cycle are not primary determinants of crowding-out; non-concessional external debt matters for health vulnerabilities.
  - Expressing variables as a share of GDP or in real per-capita terms indicates more limited crowding-out, suggesting reallocation of fiscal shares rather than large absolute cuts in some cases.
- Policy recommendations:
  - Implement mechanisms that insulate social sector budgets from security-related fiscal pressures (short-term: social expenditure floors).
  - Pursue medium-term reforms to governance, institutions, and public financial management to enhance predictability of social sector spending.
- Risk warning:
  - Even without large absolute reductions, fiscal reallocation away from social spending is concerning given service delivery gaps and potential negative effects on human capital accumulation, workforce skills, labor productivity, and long-term growth.
- Timeliness:
  - Results are especially relevant amid shifting priorities and reduction of ODA; policymakers should be alert to social repercussions of military expansion, particularly in low-aid environments.

*Source: IMF Working Paper — “Military Spending and Crowding-Out Effects: 1. Introduction” (content unit from wpiea2026069-source-pdf).*

### 1. Introduction ........................................................................................................

### 1. Introduction

### Rising insecurity and fiscal responses
- Violent conflict events in sub-Saharan Africa increased significantly over the past two decades, with the average quarterly number of incidents in 2024 being five times higher than in 2000.
- The number of fatalities resulting from these incidents almost tippled over the period. In 2024, violent conflicts claimed more than 50 thousand lives in the region.
- The Sahel region is highlighted as most impacted, hosting over 5.5 million refugees and internally displaced persons.
- Escalating security threats have reshaped fiscal priorities, leading to sharp rises in defense spending in many SSA countries:
  - In the Sahel, defense spending nearly doubled from 2.7 percent of GDP in 2016 to 4.6 percent in 2023.
- Insecurity has also disrupted economic activity, weakened revenue mobilization and increased demand for development and humanitarian spending, forcing difficult allocation decisions within tight fiscal space.

### Competing hypotheses from the literature
- Guns versus butter (crowding-out): higher military spending reduces fiscal space for investment in physical and human capital.
  - Fan, Liu, and Coyte (2018): negative link between military and health spending across nearly 200 countries (2000–2013) using seemingly unrelated regressions and simultaneous equation estimation.
- Complementarity/positive spillovers: military spending can lower violence, protect infrastructure, enable service delivery, and reduce economic uncertainty, potentially boosting productivity and public services.
  - Lin et al. (2015): dynamic panel GMM regressions find a positive link between military spending and government expenditure on education and health for 29 OECD countries (1988–2005).
  - Njifen and Aneman (2023): dynamic panel GMM for SSA countries find military spending slightly increases education spending on average.
- No systematic link: military and social spending may respond to different drivers and overall government expenditure can expand to accommodate both priorities.
  - Russett (1982); Davis and Chan (1990); Schiff, Gupta and Clements (1998) provide examples where social spending did not fall when military spending rose.
- Recent empirical work:
  - Marzian and Trebesch (2025): military spending booms driven by exogenous geopolitical shocks across 20 countries (1870–2022) do not show significant evidence of cuts to social spending.
  - IMF (2026): broader dataset for 165 countries since 1946 finds on average no significant effects on social spending, but wartime episodes show declines in real-term social spending (health, education, social protection).
  - Regional heterogeneity in SSA: Njifena and Anemann (2023) find crowding-out of education in highly security-challenged areas (Sahel, Lake Chad Basin); Central and Southern Africa show instances of complementarity (Elshafei et al. 2025).

### Paper’s objectives and contribution
- Main aim: investigate whether military expenditures exert crowding-out effects on education and health budgets in Sub-Saharan Africa, and examine the dynamic fiscal response to military spending shocks.
- Twofold contribution:
  - Employ multiple identification strategies to isolate exogenous changes in military spending:
    - (i) constructing a military spending surge variable based on observed spending patterns;
    - (ii) using residuals from a government reaction function in the spirit of Caldara and Kamps (2017);
    - (iii) the preferred approach: using instrumental variables.
  - Analyze state-dependencies shaping the relationship between military and social spending, specifically assessing whether sovereign debt levels, the availability of official development assistance (ODA), and the business cycle influence crowding-out.

### Paper structure
- Section 2: data and stylized facts on military, health, and education spending in SSA.
- Section 3: empirical strategy and baseline results.
- Section 4: robustness checks, including alternative definitions and differences between Fragile and Conflict-Affected States (FCS) and non-FCS.
- Section 5: state-dependencies (debt, international aid, business cycle).
- Section 6: conclusions.

---

### 2. Data and Stylized Facts

### Context and regional dynamics
- After relative stability from 2000 to 2005, SSA experienced rapid deterioration in security; the collapse of Libya is cited as a catalytic event linked to proliferation of armed groups in the Sahel.
- Violence can spill across borders via refugee flows, transnational armed networks, illicit trafficking routes, and cross-border ethnic linkages, generating wider regional externalities.
- Since 2010 there have been 32 coup attempts in SSA, with over half being successful.
- Incidence of violent events such as battles, explosions and violence against civilians increased six-fold between 2010 and 2024, with fatalities rising in parallel.
- In 2024 the Sahel accounted for 23 percent of violent events in SSA and 32 percent of fatalities while accounting for only 7 percent of the population.

### Trends in military and social spending
- For SSA as a whole, military spending as a share of total expenditure has trended downward since 1990, falling from 12 percent of total expenditure to close to 7 percent in 2023.
- In contrast, Sahel countries experienced an upward trend in military spending:
  - Military spending in the Sahel rose to about 13 percent of total expenditure, compared to 8 percent in 2014.
- Fiscal pressures: borrowing costs have risen and interest payments have increased in the last decade for SSA countries, tightening fiscal space and raising the likelihood that marginal adjustments will affect education and health.
- Social spending benchmarks and shortfalls:
  - UNESCO education benchmark: 4–6 percent of GDP or 15–20 percent of public expenditure.
  - Average share of education spending has fallen from around 21 percent of total expenditure in 1980 to below 18 percent in 2023.
  - Abuja Declaration health target: at least 15 percent of national budgets to health; since 1980 health spending as a share of total expenditure has remained stable at around 7 percent in SSA.

### Data description and sample
- Military expenditure and ammunition imports: SIPRI Military Expenditure Database (official national data and alternative sources).
- Government spending on education and health: Gethin (2024) database of General Government Revenue and Expenditure by Function (1980–2022), updated to 2023.
- Macroeconomic variables: IMF World Economic Outlook (WEO) database (October 2025 vintage) and Global Macroeconomic Database (GMD) of Müller, Xu, Lehbib, and Chen (2025) to fill gaps.
- Additional variables: terms of trade and Official Development Assistance from World Bank WDI; VIX from Bloomberg; 10-year US treasury yields from Haver; fatalities from ACLED.
- Sample construction:
  - Excluded 12 SSA countries due to extremely low military expenditures and insufficient data.
  - Final dataset: an unbalanced panel covering 1990-2023 and 33 countries.

---

### 3. Empirical Strategy and Baseline Specification

### Identification approach and estimation method
- Method: local projection method (Jordà, 2005) to estimate dynamic effects of military spending shocks on public expenditure shares for education and health.
- Three approaches to identify exogenous military spending shocks:
  - (i) military spending surge variable based on observed spending patterns;
  - (ii) residuals from a government reaction function (Caldara and Kamps, 2017 style);
  - (iii) instrumental variables (preferred approach).

### Baseline local projection specification
- Baseline regression:
  - 푦푖,푡+ℎ − 푦푖,푡−1 = 훼푖ℎ + 훽ℎ 푠ℎ표푐푘푡 + 휹 X푖,푡 + Θ푡ℎ + 휀푖,푡+ℎ
- Definitions and key components:
  - 푦푖,푡+ℎ: outcome of interest — share of government spending on education or the share on health relative to total government expenditure (baseline).
  - 훼푖ℎ: country fixed effects.
  - Θ푡ℎ: global financial controls — the VIX volatility index and the 10-year US treasury yield.
  - 푠ℎ표푐푘: military spending shock (variously defined as above).
  - X푖,푡: controls including:
    - Two lags of the dependent variable and the military shocks.
    - Contemporaneous value and two lags of:
      - output gap (HP filter);
      - ratio of aid to GDP;
      - cyclical component of terms of trade (HP filter);
      - number of fatalities linked to conflict events per capita;
    - Two lags of government debt as a share of GDP.
  - Main coefficient of interest: 훽ℎ, capturing effects of military spending shocks at horizons h.
- The specification links to recursive identification in SVAR models (Montiel-Olea et al., 2025).

### Preliminary empirical observation
- A non-causal plot of changes in health and education spending shares against lagged change in military spending shows a negative relationship on average: increases in military spending share are followed by small declines in health and education shares. The association is statistically significant, though the magnitude is modest.

---

*IMF Working Papers — Military Spending and Crowding-Out Effects: 1. Introduction (content unit from the source PDF).*

### 0.07 which implies that about 0.5 percent of the variation in changes in social spending is explained by

### wpiea2026069-source-pdf - 0.07 which implies that about 0.5 percent of the variation in changes in social spending is explained by

### Correlation and sample notes
- The estimate "0.07 which implies that about 0.5 percent of the variation in changes in social spending is explained by co-movements with changes in military share."
- Countries listed in the sample subset: Cabo Verde, Equatorial Guinea, Eritrea, Eswatini, Lesotho, Seychelles, South Sudan, Zimbabwe, Republic of Congo, The Gambia, and Comoros.
- Country coverage was constrained by data availability over a sufficiently large time period.
- Robustness checks include alternative definitions such as government education spending per school age child and government spending on health per capita (both expressed in dollar terms).

### Effects of military spending surges (local projections and event study)
- Definition: Military spending surges are defined as increases in military spending in the top 30th percentile of both the country-specific and the full sample's distribution; a binary variable equals 1 for a surge and 0 otherwise.
- Local projection results:
  - A surge in military spending crowds-out health spending between the years 2 and 4; effects are insignificant afterwards.
  - A surge in the share of military spending is associated with a decline in the share of health spending by up to 0.3 percentage points.
  - For education spending, there is an initial crowding-in effect that is small and not statistically significant two years after the shock.
- Event-study results:
  - Pre-surge coefficients for both health and education spending fluctuate around zero and are generally statistically insignificant, indicating absence of systematic pre-trends.
  - Following the surge (year 0), estimated effects turn negative for both sectors.
  - The decline is more persistent for health spending; education spending shows negative point estimates in several post-surge years with wider confidence intervals.
- Interpretation: Evidence is consistent with reallocation toward defense at the expense of health and education, with varying magnitude and statistical precision across horizons.

### Identifying military spending shocks through a reaction function
- Method: Residuals from country-level regression (Equation 2) of military share on controls are used as "surprise" shocks; residuals are then plugged into dynamic specification (Equation 1).
- Equation 2 controls include:
  - Output gap (Ygap)
  - Terms of trade gap (Totgap)
  - Debt/GDP
  - ODA/GDP
  - Fatalities/POP (number of fatalities per 1000 of population)
- Findings:
  - Crowd-out effects from this approach are typically neither statistically nor economically significant.
  - A one standard deviation increase in the share of military spending leads to decreases in the share of health spending of less than 0.5 percentage point after one year.
  - The effect on the share of education spending is close to zero.
  - Some evidence of crowding-in over the medium-term for education, but effects are small (less than 0.5 percentage points) and only statistically significant for education spending.

### Instrumental variables approach (preferred method) — instruments and first stage
- Instruments used:
  - (i) Value of weapons imports.
  - (ii) Country’s share of total military spending in Sub-Saharan Africa.
  - (iii) U.S. producer price index for ammunition interacted with the value of military equipment imports from the U.S.
- Rationale:
  - Weapons imports are strongly correlated with military expenditure but unlikely to directly affect domestic health or education spending except through military spending.
  - Country’s share in regional military spending captures strategic/regional security dynamics influencing defense allocations but not directly social spending.
  - Interaction of U.S. ammunition PPI with U.S. weapon imports introduces exogenous price variation from global supply conditions.
- First-stage performance:
  - First-stage F-statistics range from 9 to 16 across horizons and specifications.
  - The conventional threshold of 10 is noted; many specifications exceed it.
  - Kleibergen–Paap under-identification test and Hansen J over-identification test provide evidence in favor of instrument validity, but concerns about weak identification remain.
  - Pflueger and Wang (2015) weak-instrument test suggests potential weak identification; authors implement Finlay, Magnusson, and Schaffer (2013) weak-instrument-robust inference and apply a plausibly exogenous Conley et al. (2012) framework allowing each instrument a direct effect on education spending of up to ±5 percent across horizons.

### IV local projections results (dynamic causal effects)
- General pattern:
  - Short-term: one-standard-deviation shock to the share of military spending in total government expenditure does not seem to have an effect in the short-term for education; health shows short-term declines in some specifications.
  - Medium to long term: significant declines in shares of education and health.
- Exact quantitative findings (IV estimates):
  - Education spending:
    - Reduces the share of education spending by over 0.5 percentage points after three years.
    - Reduces the share of education spending by 1.1 percentage point after five years.
  - Health spending:
    - Reduces the share of health spending by about 0.8 percentage points after one year.
    - Reduces the share of health spending by nearly 1.3 percentage points after five years.
- Explanation offered: Differences in budget rigidities — education spending in Sub‑Saharan Africa has a large share devoted to salaries (rigid in the short term) while health budgets allocate a sizable share to operating costs (more flexible and vulnerable to reallocations).

### Key methodological notes and robustness
- Instruments are constrained by data availability but guided by literature and economic rationale.
- First-stage fitted values (ΔM̂ i,t) are used in second-stage local-projection regressions (Equation 4) to estimate dynamic effects on sectoral shares.
- Weak-instrument robust confidence intervals (Finlay, Magnusson and Schaffer (2013)) are used in reporting Figure 6 results.
- Sensitivity: IV inference relies critically on the exclusion restriction; baseline IV estimates are not robust to violations of instrument validity under the Conley et al. (2012) plausibly exogenous framework.

*Source: wpiea2026069-source-pdf - 0.07 which implies that about 0.5 percent of the variation in changes in social spending is explained by*

### 4. Robustness Checks

### 4. Robustness Checks

### Alternative Definitions of Key Variables
- Tested the reaction-function approach with military, education, and health spending expressed:
  - as a share of GDP (instead of share of total spending), and
  - in real terms (variables defined in constant 2023 dollars; education spending as government spending per school age child; health spending as government spending on health per capita).
- Key findings:
  - A one standard deviation shock to military spending is not associated with any (economically and statistically) significant changes in the ratio of education spending to GDP after two years.
  - Health spending declines by less than 0.1 percent of GDP after two years in the reaction-function specification.
  - A one standard deviation real increase in military spending is not linked to significant changes in education and health spending after two years (though there is an economically very small increase in education spending on impact).
- Interpretation: Military shocks identified through the reaction-function approach do not point to large crowding-out effects on education and health spending over the near term.
- Statistical note: Shaded areas in associated figures indicate the 90 percent confidence interval using Driscoll-Kraay standard errors.

### Instrumental Variables (IV) Results and Robustness
- IV exercise (Figure 8) findings:
  - A one-standard-deviation increase in military spending as a share of GDP reduces health expenditure by about 0.5 percentage points after five years (when health is expressed as percentage of GDP).
  - No evidence of significant crowding-out effects of increases in total military spending on:
    - education expenditure per school-age child, or
    - health spending per capita.
- Interpretation:
  - Absolute spending levels may remain stable, but military spending shocks can trigger reallocation of fiscal priorities away from social sectors.
  - Reallocation does not necessarily imply a substantial reduction in absolute resources for delivering social services.
- Possible mechanisms limiting social spending declines:
  - Nominal and per-capita spending “stickiness” (e.g., high share of wage-related expenditures).
  - Overall government expenditure may increase because of positive multiplier effects of military spending on GDP; however, in sub-Saharan Africa, high import content of military spending can cause leakages through higher imports.
  - Military spending may also boost growth by reducing insecurity and economic uncertainty.
- Pandemic check: Re-estimating the IV specification with a sample restricted to the pre-COVID-19 period produced results broadly robust across key variable definitions (results available upon request).
- Statistical note: IV figures use Finlay, Magnusson and Schaffer (2013) weak-instrument-robust confidence intervals.

### Fragility and Conflict (FCS vs non-FCS)
- Estimation results (based on IMF’s fiscal year 2026 FCS classification) show:
  - In non-FCS countries, military spending shocks clearly crowd out health and education expenditures (bottom charts of Figure 9).
  - In FCS countries, crowding-out effects are less apparent, especially for education spending — possibly due to increased donor support accompanying heightened security concerns.
- Additional checks:
  - Results are broadly similar when comparing countries experiencing active conflict with those without conflict.
  - Removing Sahel countries does not significantly alter the results.
- Statistical note: Figures indicate 90 percent confidence intervals using Finlay, Magnusson and Schaffer (2013) weak-instrument-robust confidence intervals.

### Additional Control Variables
- Controlled for:
  - governance (control of corruption),
  - off-budget aid to education and health, and
  - presence of an IMF program.
- Findings (Figure 10):
  - Crowding-out effects remain robust to inclusion of these controls.
  - Magnitude of coefficients is somewhat reduced compared to baseline estimates.
- Interpretation:
  - Off-budget aid can be substantial and may be omitted from baseline statistics; IMF programs may include floors on social sector spending that constrain declines.

### State Dependencies (IV Approach)
- Framework:
  - Used smooth transition function (Auerbach and Gorodnichenko, 2012) applied to the IV specification to capture nonlinear effects across states.
  - Smooth transition function parameter 휏0 is set to 2.
- Government Debt:
  - High debt-to-GDP ratio does not necessarily make social spending more vulnerable to military spending shocks (Figure 11).
  - When focusing on non-concessional external debt, there is evidence of larger crowding-out effects of military spending on health expenditure when the non-concessional debt to GDP ratio is high.
  - Interpretation: Limited access to concessional financing increases likelihood of crowding-out in the health sector.
- Foreign Aid (ODA):
  - No significant differences in effects on education expenditures across high and low periods of ODA overall.
  - Military spending shocks tend to crowd out health expenditure when access to ODA is limited (Figure 12, top panels).
  - Crowding-out on health is more pronounced when ODA grants to the health sector are lower (Figure 12, bottom panels).
  - Evidence of crowding-in effects in health when access to ODA grants is higher — likely reflecting donor scaling-up to protect social spending in crises.
  - Note: Education and health spending variables include donor financing channeled through the government budget; off-budget donor support is excluded.
- Business Cycle:
  - Effects of military spending on shares of education and health spending do not vary significantly with the business cycle (Figure 13).
  - Whether the economy operates above or below potential does not significantly influence crowding-out or crowding-in effects.
  - Interpretation: Fiscal trade-offs between defense and social spending are shaped primarily by structural factors rather than cyclical conditions.

### Conclusions and Policy Implications
- Main quantitative finding:
  - Using IV identification, a one-standard-deviation increase in the share of military spending in total government expenditure reduces the shares of education and health spending by about one percentage points over the medium-term.
  - Crowding-out effects tend to materialize sooner for health expenditures.
- Heterogeneity:
  - Military spending shocks tend to crowd out health expenditures when access to international aid is limited; no evidence of crowding-out when aid is ample.
  - Overall debt levels and the state of the business cycle are not significant determinants of crowding-out extent.
  - Measuring variables as a share of GDP or in real terms indicates somewhat more limited crowding-out, suggesting military shocks primarily reshape fiscal allocation rather than large absolute cuts.
- Policy recommendations:
  - Implement policy mechanisms that insulate social sector budgets from security-related fiscal pressures (short-term example: social expenditure floors in the budget).
  - Over the medium-term, pursue deeper reforms improving governance, institutions, and public financial management to enhance predictability of social sector spending.
- Risk warning:
  - Even without large absolute reductions, reallocation away from social spending is concerning given service delivery gaps in several countries and potential negative effects on human capital accumulation, workforce skills, labor productivity, and long-term growth potential.
- Timeliness:
  - Findings are especially relevant amid shifting priorities and reduction of ODA; policymakers should be alert to social repercussions of military expansion, particularly in low aid environments.

*Source: IMF working paper — section 4, "Robustness Checks" from the provided PDF content.*

### References

### References

### Bibliographic entries (selected)
- Abdel-Latif, H., David, A. C., Ouedraogo, R. and Specht, M. (2025) “Echoes across Borders: Macroeconomic Spillover Effects of Conflict in Sub-Saharan African Countries” Journal of African Economies, 34(5): 605-638.  
- Auerbach, A. J. and Gorodnichenko, Y. (2012) “Measuring the output responses to fiscal policy”, American Economic Journal Economic Policy, 4 (2): 1–27.  
- Caldara, D., and Kamps, C. (2017) “The analytics of SVARs: A unified framework to measure fiscal multipliers” The Review of Economic Studies, 84(3): 1015–1040.  
- Caballero, J.A. (2016) “Do Surges in International Capital Inflows Influence the Likelihood of Banking Crises” The Economic Journal, 126: 281-316.  
- Conley, G. T., Hansen, B.C. and Rossi. E.P.(2012). “Plausibly Exogenous”. Review of Economics and Statistics 94 (1): 260–272.  
- Davis, D. R., and Chan, S. (1990) “The Security–Welfare Relationship: Longitudinal Evidence from Taiwan” Journal of Peace Research, 27(1): 87–100.  
- Du Bois, C. (2024) “Military Spending and the Burden Sharing Debate” In: Sookermany, A.M. (eds) Handbook of Military Sciences. Springer, Cham.  
- Dunne, J. P., and Smith, R. P. (2019) “Military Expenditure, Investment and Growth” Defence and Peace Economics, 31(6): 601–614.  
- Elshafei, A. S. M. A., Ahmed, D.M. H., El‑Raouf, M. M.A. and El‑Qurashi, M. A. (2025) “The Optimal Orchestration of Military Expenditure for Economic Growth in Africa: A Non‑Linear Threshold and Regional Heterogeneity Analysis” Journal of Cultural Analysis and Social Change 10(4): 1117-1134.  
- Fan, H., W. Liu, W., and Coyte, P. C. (2018) “Do military expenditures crowd-out health expenditures? Evidence from around the world, 2000–2013” Defense and Peace Economics, 29(7): 766–779.  
- Finlay, K., Magnusson, L. and Schaffer, M. E. (2013) "WEAKIV: Stata module to perform weak-instrument-robust tests and confidence intervals for instrumental-variable (IV) estimation of linear, probit and tobit models," Statistical Software Components, Boston College Department of Economics.  
- Gethin, A., (2024) “A new database of general government revenue and expenditure by function, 1980-2022”. unpublished mimeo, available at https://amory-gethin. fr.  
- Ghosh, A., Qureshi, M. Kim, J., Zalduendo, J. (2014) “Surges” Journal of International Economics, 92(2): 266–85.  
- Gleditsch, K. S. (2007) “Transnational dimensions of civil war” Journal of Peace Research, 44(3): 293-309.  
- Gupta, S., Schena, M., & Yousefi, S. R. (2020). “Revisiting IMF Expenditure Conditionality”. Applied Economics, 52(58): 6338–6359.  
- Harris, G., Kelly, M. and Pranowo (1988) “Trade-offs Between Defence and Education/Health Expenditures in Developing Countries” Journal of Peace Research, 25(2): 165–177.  
- International Monetary Fund (2026) “Defense Spending: Macroeconomic Consequences and Trade-Offs” World Economic Outlook, Analytical Chapter, April.  
- Institute for Economics & Peace. Global Terrorism Index 2025: Measuring The Impact of Terrorism, Sydney, March 2025. Available from: http://visionofhumanity.org/resources (accessed 01-26)  
- Jordà, Ò. (2005) “Estimation and inference of impulse responses by local projections” American Economic Review, 95(1): 161–182.  
- Kollias, C., and Paleologou, S.-M. (2011) “Budgetary trade-offs between defence, education and social spending in Greece” Applied Economics Letters, 18(11): 1071-1075.  
- Lemaire, T. Nyankiye, F. and Sode, A. (2023) “Debt Dilemmas in Sub-Saharan Africa: Principles and Trade-Offs” October 2023 Regional Economic Outlook: Sub-Saharan Africa Analytical Note.  
- Lin, E. S., Ali, H. E., and Lu, Y. L. (2015) “Does Military Spending Crowd Out Social Welfare Expenditures? Evidence from a Panel of OECD Countries” Defence and Peace Economics, 26(1): 33–48.  
- Marzian, J. and Trebesch, C. (2025) “Guns and Butter: The Fiscal Consequences of Rearmament and War” Kiel Institute for the World Economy Working Paper No. 2310, December.  
- Montiel-Olea, J. L., Plagborg-Møller, M., Qian, E., Wolf C. K. (2025) “Local projections or VARs? a primer for macroeconomists” NBER Working Paper w33871.  
- Müller, Karsten, Chenzi Xu, Mohamed Lehbib, and Ziliang Chen. (2025) “The Global Macro Database: A New International Macroeconomic Dataset” (Version 2025-12). NBER Working Paper 33714.  
- Njifen, I., and Anemann, A. (2023) “Military Expenditures and Human Capital Development in Sub- Saharan Africa: A System GMM Approach” Development Studies Research, 10(1).  
- Pflueger, C. E., and Wang, S. (2015) “A Robust Test for Weak Instruments in Stata” The Stata Journal, 15: 216-225.  
- Powell, J. M. and Thyne, C. L. (2011) Global Instances of Coups from 1950-Present. Journal of Peace Research, 48(2): 249-259.  
- Russett, B. (1982) “Defense Expenditures and National Well‑Being” American Political Science Review, 76(4): 767–777.  
- Saeed, L. (2025) “The impact of military expenditures on economic growth: A new instrumental variables approach ” Defence and Peace Economics, 36(1): 86–101.  
- Salehyan, I., and Gleditsch, K. S. (2006) “Refugees and the spread of civil war” International Organization, 60(2): 335-366.  
- Schiff, J, Gupta, S. & Clements, B. (1998) “Worldwide Military Spending, 1990–95”, Defence and Peace Economics, 9(3): 237-281.  
- Strazzari, F. (2014) “Libyan arms and regional instability” The International Spectator, 49(3): 54-68.  
- UNESCO (2025) “What You Need to Know about Education Financing” UNESCO. Last updated July 28, 2025. Accessed January 26, 2026.  
- Yesilyurt, M. E., and Elhorst, J. P. (2017) “Impacts of Neighboring Countries on Military Expenditures: A Dynamic Spatial Panel Approach” Journal of Peace Research, 54(6): 777–790.  

### Annex — Variables and source datasets (Annex Table 1)
- Arms Transfers: Annual volume of international transfers of major conventional weapons (imports/exports or bilateral flows). Typically measured using SIPRI TIV (a volume/capability metric, not financial value). Source: SIPRI Arms Transfers Database.  
- Education Spending (% of GDP or % of total expenditures): Government expenditure on education, expressed either as a share of GDP or as a share of total government spending (depending on the series used). Source: Gethin (2024).  
- Debt-to-GDP ratio: Stock of public debt expressed as a percentage of GDP. Source: GMD (Global Macro Database) , IMF WEO.  
- Fatalities per 1,000 people: Total conflict-related deaths recorded in ACLED over a period, divided by population and scaled per 1,000 people. Source: ACLED.  
- Health Spending (% of GDP or % of total expenditures): Government expenditure on health, expressed either as a share of GDP or as a share of total government spending (depending on the series used). Source: Gethin (2024).  
- Military Spending (% of GDP or % of total expenditures): National military expenditure (as defined by SIPRI), expressed as a share of GDP or as a share of total government spending. Source: SIPRI Military Expenditure Database (Milex).  
- Official Development Assistance (% of GDP): Net official development assistance received, scaled by GDP. Source: World Bank WDI.  
- Output Gap (HP-filter): Percent deviation of log real output from its HP-filtered trend (trend computed by authors; gap expressed in percent). Source: IMF WEO, authors’ calculations.  
- Arms/Ammunition Producer Price Index: Producer Price Index for arms/ammunition-related products (BLS PPI series accessed via FRED; exact series depends on whether you use an “industry” or “commodity” PPI for ammunition). Source: FRED (BLS PPI via FRED).  
- Terms of Trade Gap (HP-filter): Percent deviation of the terms-of-trade index from its HP-filtered trend (trend computed by authors; gap expressed in percent). Source: WDI, authors’ calculations.  
- VIX (Volatility Index): Market-implied near-term (≈30-day) expected volatility for the S&P 500 derived from option prices (the “fear gauge”). Source: Bloomberg (VIX).  
- 10-year US Treasury Yield: Yield on the 10-year US Treasury benchmark (typically end-period or average, depending on the series pull). Source: Haver.

### Summary statistics for military spending shocks (Annex Table 2)
- Military Spending Surges: Observations 853; Countries 30; Mean 0.18; Standard deviation 0.39; Minimum 0; Maximum 1.  
- Reaction Function: Observations 560; Countries 34; Mean 0; Standard deviation 0.74; Minimum -4.876.02; Maximum (entry lists as) [value appears combined as "-4.876.02" in source].  
- IV: Observations 582; Countries 30; Mean -0.121.35; Standard deviation -8.6313.9; Minimum (entry presents combined values "-8.6313.9" in source).  

(Note: numeric entries for Reaction Function and IV in the source appear concatenated; preserved exactly as presented.)

### Instrument test results for baseline local projections model (Annex Table 3)
- Education — Weak instrument test and instrument validity test by horizon:
  - Horizon 1: Cragg-Donald Wald F statistic 16.08; Stock-Yogo weak ID F test critical value assuming 10% maximal IV relative bias 9.08; Weak identification: No; Hansen J Chi-sq (2) 5.091; P-val 0.0785; Validity concerns: No.  
  - Horizon 2: Cragg-Donald Wald F statistic 16.08; Stock-Yogo critical value 9.08; Weak identification: No; Hansen J Chi-sq (2) 1.484; P-val 0.4760; Validity concerns: No.  
  - Horizon 3: Cragg-Donald Wald F statistic 13.85; Stock-Yogo critical value 9.08; Weak identification: No; Hansen J Chi-sq (2) 0.258; P-val 0.8792; Validity concerns: No.  
  - Horizon 4: Cragg-Donald Wald F statistic 12.66; Stock-Yogo critical value 9.08; Weak identification: No; Hansen J Chi-sq (2) 0.439; P-val 0.8031; Validity concerns: No.  
  - Horizon 5: Cragg-Donald Wald F statistic 11.81; Stock-Yogo critical value 9.08; Weak identification: No; Hansen J Chi-sq (2) 2.375; P-val 0.3050; Validity concerns: No.  

- Health — Weak instrument test and instrument validity test by horizon:
  - Horizon 1: Cragg-Donald Wald F statistic 11.53; Stock-Yogo critical value 9.08; Weak identification: No; Hansen J Chi-sq (2) 2.224; P-val 0.3289; Validity concerns: No.  
  - Horizon 2: Cragg-Donald Wald F statistic 11.53; Stock-Yogo critical value 9.08; Weak identification: No; Hansen J Chi-sq (2) 2.683; P-val 0.2614; Validity concerns: No.  
  - Horizon 3: Cragg-Donald Wald F statistic 10.43; Stock-Yogo critical value 9.08; Weak identification: No; Hansen J Chi-sq (2) 0.475; P-val 0.7888; Validity concerns: No.  
  - Horizon 4: Cragg-Donald Wald F statistic 9.83; Stock-Yogo critical value 9.08; Weak identification: No; Hansen J Chi-sq (2) 1.000; P-val 0.6066; Validity concerns: No.  
  - Horizon 5: Cragg-Donald Wald F statistic 9.32; Stock-Yogo critical value 9.08; Weak identification: No; Hansen J Chi-sq (2) 2.112; P-val 0.3478; Validity concerns: No.  

### Notes on figures and robustness/sensitivity analysis (Annex Figures)
- Annex Figure 1 (Sensitivity analysis for the smooth transition function): depicts effects of a one-standard-deviation shock to military spending as a share of total government expenditure on the share of education and health spending in two states: (i) high debt to GDP and (ii) low debt to GDP ratio. The parameter 휏0 of the smooth transition function was set at 0.5 in the top panel and 3.5 in the bottom panel. Shaded areas indicate the 90 percent confidence interval using Finlay, Magnusson and Schaffer (2013) weak-instrument-robust confidence intervals.  
- Annex Figure 1.B (Foreign Aid): depicts effects of a one-standard deviation shock to military spending on education and health shares in countries with high ODA (high state) and low ODA (low state). Parameter 휏0 set at 0.5 (top) and 3.5 (bottom). Shaded areas indicate the 90 percent confidence interval using Finlay, Magnusson and Schaffer (2013) weak-instrument-robust confidence intervals.  
- Annex Figure 1.C (Business Cycle): depicts effects of a one-standard deviation shock to military spending on education and health shares in two states: positive output gap (high state) and negative output gap (low state). Parameter 휏0 set at 0.5 (top) and 3.5 (bottom). Shaded areas indicate the 90 percent confidence interval using Finlay, Magnusson and Schaffer (2013) weak-instrument-robust confidence intervals.  
- Annex Figure 2 (Robustness check with alternative selection of instruments): Top panel removes the value of weapons imports among the instruments; middle panel removes the interaction of U.S. ammunition PPI with weapon imports from the USA among the instruments; bottom panel excludes the country’s share in regional military spending. Shaded areas indicate the 90 percent confidence interval using Finlay, Magnusson and Schaffer (2013) weak-instrument-robust confidence intervals.

*Military Spending and Crowding-Out Effects: Evidence from Sub-Saharan Africa — Working Paper No. WP/2026/069*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2026/english/wpiea2026069-source-pdf.pdf_